236 files changed, 6092 insertions, 2036 deletions

diff --git a/.ci/scripts/common/post-upload.sh b/.ci/scripts/common/post-upload.sh
new file mode 100644
index 000000000..bb4e9d328
--- /dev/null
+++ b/.ci/scripts/common/post-upload.sh
@@ -0,0 +1,15 @@
+#!/bin/bash -ex
+
+# Copy documentation
+cp license.txt "$REV_NAME"
+cp README.md "$REV_NAME"
+
+tar $COMPRESSION_FLAGS "$ARCHIVE_NAME" "$REV_NAME"
+
+mv "$REV_NAME" $RELEASE_NAME
+
+7z a "$REV_NAME.7z" $RELEASE_NAME
+
+# move the compiled archive into the artifacts directory to be uploaded by travis releases
+mv "$ARCHIVE_NAME" artifacts/
+mv "$REV_NAME.7z" artifacts/
diff --git a/.ci/scripts/common/pre-upload.sh b/.ci/scripts/common/pre-upload.sh
new file mode 100644
index 000000000..3c2fc79a2
--- /dev/null
+++ b/.ci/scripts/common/pre-upload.sh
@@ -0,0 +1,6 @@
+#!/bin/bash -ex
+
+GITDATE="`git show -s --date=short --format='%ad' | sed 's/-//g'`"
+GITREV="`git show -s --format='%h'`"
+
+mkdir -p artifacts
diff --git a/.ci/scripts/format/docker.sh b/.ci/scripts/format/docker.sh
new file mode 100644
index 000000000..778411e4a
--- /dev/null
+++ b/.ci/scripts/format/docker.sh
@@ -0,0 +1,6 @@
+#!/bin/bash -ex
+
+# Run clang-format
+cd /yuzu
+chmod a+x ./.ci/scripts/format/script.sh
+./.ci/scripts/format/script.sh
diff --git a/.ci/scripts/format/exec.sh b/.ci/scripts/format/exec.sh
new file mode 100644
index 000000000..5d6393b38
--- /dev/null
+++ b/.ci/scripts/format/exec.sh
@@ -0,0 +1,4 @@
+#!/bin/bash -ex
+
+chmod a+x ./.ci/scripts/format/docker.sh
+docker run -v $(pwd):/yuzu yuzuemu/build-environments:linux-clang-format /bin/bash -ex /yuzu/.ci/scripts/format/docker.sh
diff --git a/.ci/scripts/format/script.sh b/.ci/scripts/format/script.sh
new file mode 100644
index 000000000..5ab828d5e
--- /dev/null
+++ b/.ci/scripts/format/script.sh
@@ -0,0 +1,37 @@
+#!/bin/bash -ex
+
+if grep -nrI '\s$' src *.yml *.txt *.md Doxyfile .gitignore .gitmodules .ci* dist/*.desktop \
+                 dist/*.svg dist/*.xml; then
+    echo Trailing whitespace found, aborting
+    exit 1
+fi
+
+# Default clang-format points to default 3.5 version one
+CLANG_FORMAT=clang-format-6.0
+$CLANG_FORMAT --version
+
+if [ "$TRAVIS_EVENT_TYPE" = "pull_request" ]; then
+    # Get list of every file modified in this pull request
+    files_to_lint="$(git diff --name-only --diff-filter=ACMRTUXB $TRAVIS_COMMIT_RANGE | grep '^src/[^.]*[.]\(cpp\|h\)$' || true)"
+else
+    # Check everything for branch pushes
+    files_to_lint="$(find src/ -name '*.cpp' -or -name '*.h')"
+fi
+
+# Turn off tracing for this because it's too verbose
+set +x
+
+for f in $files_to_lint; do
+    d=$(diff -u "$f" <($CLANG_FORMAT "$f") || true)
+    if ! [ -z "$d" ]; then
+        echo "!!! $f not compliant to coding style, here is the fix:"
+        echo "$d"
+        fail=1
+    fi
+done
+
+set -x
+
+if [ "$fail" = 1 ]; then
+    exit 1
+fi
diff --git a/.ci/scripts/linux/docker.sh b/.ci/scripts/linux/docker.sh
new file mode 100644
index 000000000..f538a4081
--- /dev/null
+++ b/.ci/scripts/linux/docker.sh
@@ -0,0 +1,14 @@
+#!/bin/bash -ex
+
+cd /yuzu
+
+ccache -s
+
+mkdir build || true && cd build
+cmake .. -G Ninja -DYUZU_USE_BUNDLED_UNICORN=ON -DYUZU_USE_QT_WEB_ENGINE=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=/usr/lib/ccache/gcc -DCMAKE_CXX_COMPILER=/usr/lib/ccache/g++ -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${ENABLE_COMPATIBILITY_REPORTING:-"OFF"} -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DUSE_DISCORD_PRESENCE=ON
+
+ninja
+
+ccache -s
+
+ctest -VV -C Release
diff --git a/.ci/scripts/linux/exec.sh b/.ci/scripts/linux/exec.sh
new file mode 100644
index 000000000..a5a6c34b9
--- /dev/null
+++ b/.ci/scripts/linux/exec.sh
@@ -0,0 +1,5 @@
+#!/bin/bash -ex
+
+mkdir -p "ccache"  || true
+chmod a+x ./.ci/scripts/linux/docker.sh
+docker run -e ENABLE_COMPATIBILITY_REPORTING -e CCACHE_DIR=/yuzu/ccache -v $(pwd):/yuzu yuzuemu/build-environments:linux-fresh /bin/bash /yuzu/.ci/scripts/linux/docker.sh
diff --git a/.ci/scripts/linux/upload.sh b/.ci/scripts/linux/upload.sh
new file mode 100644
index 000000000..0d131d1dd
--- /dev/null
+++ b/.ci/scripts/linux/upload.sh
@@ -0,0 +1,14 @@
+#!/bin/bash -ex
+
+. .ci/scripts/common/pre-upload.sh
+
+REV_NAME="yuzu-linux-${GITDATE}-${GITREV}"
+ARCHIVE_NAME="${REV_NAME}.tar.xz"
+COMPRESSION_FLAGS="-cJvf"
+
+mkdir "$REV_NAME"
+
+cp build/bin/yuzu-cmd "$REV_NAME"
+cp build/bin/yuzu "$REV_NAME"
+
+. .ci/scripts/common/post-upload.sh
diff --git a/.ci/scripts/merge/apply-patches-by-label.py b/.ci/scripts/merge/apply-patches-by-label.py
new file mode 100644
index 000000000..b346001a5
--- /dev/null
+++ b/.ci/scripts/merge/apply-patches-by-label.py
@@ -0,0 +1,28 @@
+# Download all pull requests as patches that match a specific label
+# Usage: python download-patches-by-label.py <Label to Match> <Root Path Folder to DL to>
+
+import requests, sys, json, urllib3.request, shutil, subprocess
+
+http = urllib3.PoolManager()
+dl_list = {}
+
+def check_individual(labels):
+    for label in labels:
+        if (label["name"] == sys.argv[1]):
+            return True
+    return False
+
+try:
+    url = 'https://api.github.com/repos/yuzu-emu/yuzu/pulls'
+    response = requests.get(url)
+    if (response.ok):
+        j = json.loads(response.content)
+        for pr in j:
+            if (check_individual(pr["labels"])):
+                pn = pr["number"]
+                print("Matched PR# %s" % pn)
+                print(subprocess.check_output(["git", "fetch", "https://github.com/yuzu-emu/yuzu.git", "pull/%s/head:pr-%s" % (pn, pn), "-f"]))
+                print(subprocess.check_output(["git", "merge", "--squash", "pr-%s" % pn]))
+                print(subprocess.check_output(["git", "commit", "-m\"Merge PR %s\"" % pn]))
+except:
+    sys.exit(-1)
diff --git a/.ci/scripts/merge/check-label-presence.py b/.ci/scripts/merge/check-label-presence.py
new file mode 100644
index 000000000..048466d7e
--- /dev/null
+++ b/.ci/scripts/merge/check-label-presence.py
@@ -0,0 +1,18 @@
+# Checks to see if the specified pull request # has the specified tag
+# Usage: python check-label-presence.py <Pull Request ID> <Name of Label>
+
+import requests, json, sys
+
+try:
+    url = 'https://api.github.com/repos/yuzu-emu/yuzu/issues/%s' % sys.argv[1]
+    response = requests.get(url)
+    if (response.ok):
+        j = json.loads(response.content)
+        for label in j["labels"]:
+            if label["name"] == sys.argv[2]:
+                print('##vso[task.setvariable variable=enabletesting;]true')
+                sys.exit()
+except:
+    sys.exit(-1)
+
+print('##vso[task.setvariable variable=enabletesting;]false')
diff --git a/.ci/scripts/merge/yuzubot-git-config.sh b/.ci/scripts/merge/yuzubot-git-config.sh
new file mode 100644
index 000000000..d9d595bbc
--- /dev/null
+++ b/.ci/scripts/merge/yuzubot-git-config.sh
@@ -0,0 +1,2 @@
+git config --global user.email "yuzu@yuzu-emu.org"
+git config --global user.name "yuzubot"
\ No newline at end of file
diff --git a/.ci/scripts/windows/docker.sh b/.ci/scripts/windows/docker.sh
new file mode 100644
index 000000000..f7093363b
--- /dev/null
+++ b/.ci/scripts/windows/docker.sh
@@ -0,0 +1,50 @@
+#!/bin/bash -ex
+
+cd /yuzu
+
+ccache -s
+
+# Dirty hack to trick unicorn makefile into believing we are in a MINGW system
+mv /bin/uname /bin/uname1 && echo -e '#!/bin/sh\necho MINGW64' >> /bin/uname
+chmod +x /bin/uname
+
+# Dirty hack to trick unicorn makefile into believing we have cmd
+echo '' >> /bin/cmd
+chmod +x /bin/cmd
+
+mkdir build || true && cd build
+cmake .. -G Ninja -DCMAKE_TOOLCHAIN_FILE="$(pwd)/../CMakeModules/MinGWCross.cmake" -DUSE_CCACHE=ON -DYUZU_USE_BUNDLED_UNICORN=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DCMAKE_BUILD_TYPE=Release
+ninja
+
+# Clean up the dirty hacks
+rm /bin/uname && mv /bin/uname1 /bin/uname
+rm /bin/cmd
+
+ccache -s
+
+echo "Tests skipped"
+#ctest -VV -C Release
+
+echo 'Prepare binaries...'
+cd ..
+mkdir package
+
+QT_PLATFORM_DLL_PATH='/usr/x86_64-w64-mingw32/lib/qt5/plugins/platforms/'
+find build/ -name "yuzu*.exe" -exec cp {} 'package' \;
+
+# copy Qt plugins
+mkdir package/platforms
+cp "${QT_PLATFORM_DLL_PATH}/qwindows.dll" package/platforms/
+cp -rv "${QT_PLATFORM_DLL_PATH}/../mediaservice/" package/
+cp -rv "${QT_PLATFORM_DLL_PATH}/../imageformats/" package/
+rm -f package/mediaservice/*d.dll
+
+for i in package/*.exe; do
+  # we need to process pdb here, however, cv2pdb
+  # does not work here, so we just simply strip all the debug symbols
+  x86_64-w64-mingw32-strip "${i}"
+done
+
+pip3 install pefile
+python3 .ci/scripts/windows/scan_dll.py package/*.exe "package/"
+python3 .ci/scripts/windows/scan_dll.py package/imageformats/*.dll "package/"
diff --git a/.ci/scripts/windows/exec.sh b/.ci/scripts/windows/exec.sh
new file mode 100644
index 000000000..d6a994856
--- /dev/null
+++ b/.ci/scripts/windows/exec.sh
@@ -0,0 +1,5 @@
+#!/bin/bash -ex
+
+mkdir -p "ccache" || true
+chmod a+x ./.ci/scripts/windows/docker.sh
+docker run -e CCACHE_DIR=/yuzu/ccache -v $(pwd):/yuzu yuzuemu/build-environments:linux-mingw /bin/bash -ex /yuzu/.ci/scripts/windows/docker.sh
diff --git a/.ci/scripts/windows/scan_dll.py b/.ci/scripts/windows/scan_dll.py
new file mode 100644
index 000000000..163183f2e
--- /dev/null
+++ b/.ci/scripts/windows/scan_dll.py
@@ -0,0 +1,106 @@
+import pefile
+import sys
+import re
+import os
+import queue
+import shutil
+
+# constant definitions
+KNOWN_SYS_DLLS = ['WINMM.DLL', 'MSVCRT.DLL', 'VERSION.DLL', 'MPR.DLL',
+                  'DWMAPI.DLL', 'UXTHEME.DLL', 'DNSAPI.DLL', 'IPHLPAPI.DLL']
+# below is for Ubuntu 18.04 with specified PPA enabled, if you are using
+# other distro or different repositories, change the following accordingly
+DLL_PATH = [
+    '/usr/x86_64-w64-mingw32/bin/',
+    '/usr/x86_64-w64-mingw32/lib/',
+    '/usr/lib/gcc/x86_64-w64-mingw32/7.3-posix/'
+]
+
+missing = []
+
+
+def parse_imports(file_name):
+    results = []
+    pe = pefile.PE(file_name, fast_load=True)
+    pe.parse_data_directories()
+
+    for entry in pe.DIRECTORY_ENTRY_IMPORT:
+        current = entry.dll.decode()
+        current_u = current.upper()  # b/c Windows is often case insensitive
+        # here we filter out system dlls
+        # dll w/ names like *32.dll are likely to be system dlls
+        if current_u.upper() not in KNOWN_SYS_DLLS and not re.match(string=current_u, pattern=r'.*32\.DLL'):
+            results.append(current)
+
+    return results
+
+
+def parse_imports_recursive(file_name, path_list=[]):
+    q = queue.Queue()  # create a FIFO queue
+    # file_name can be a string or a list for the convience
+    if isinstance(file_name, str):
+        q.put(file_name)
+    elif isinstance(file_name, list):
+        for i in file_name:
+            q.put(i)
+    full_list = []
+    while q.qsize():
+        current = q.get_nowait()
+        print('> %s' % current)
+        deps = parse_imports(current)
+        # if this dll does not have any import, ignore it
+        if not deps:
+            continue
+        for dep in deps:
+            # the dependency already included in the list, skip
+            if dep in full_list:
+                continue
+            # find the requested dll in the provided paths
+            full_path = find_dll(dep)
+            if not full_path:
+                missing.append(dep)
+                continue
+            full_list.append(dep)
+            q.put(full_path)
+            path_list.append(full_path)
+    return full_list
+
+
+def find_dll(name):
+    for path in DLL_PATH:
+        for root, _, files in os.walk(path):
+            for f in files:
+                if name.lower() == f.lower():
+                    return os.path.join(root, f)
+
+
+def deploy(name, dst, dry_run=False):
+    dlls_path = []
+    parse_imports_recursive(name, dlls_path)
+    for dll_entry in dlls_path:
+        if not dry_run:
+            shutil.copy(dll_entry, dst)
+        else:
+            print('[Dry-Run] Copy %s to %s' % (dll_entry, dst))
+    print('Deploy completed.')
+    return dlls_path
+
+
+def main():
+    if len(sys.argv) < 3:
+        print('Usage: %s [files to examine ...] [target deploy directory]')
+        return 1
+    to_deploy = sys.argv[1:-1]
+    tgt_dir = sys.argv[-1]
+    if not os.path.isdir(tgt_dir):
+        print('%s is not a directory.' % tgt_dir)
+        return 1
+    print('Scanning dependencies...')
+    deploy(to_deploy, tgt_dir)
+    if missing:
+        print('Following DLLs are not found: %s' % ('\n'.join(missing)))
+    return 0
+
+
+if __name__ == '__main__':
+    main()
diff --git a/.ci/scripts/windows/upload.sh b/.ci/scripts/windows/upload.sh
new file mode 100644
index 000000000..de73d3541
--- /dev/null
+++ b/.ci/scripts/windows/upload.sh
@@ -0,0 +1,13 @@
+#!/bin/bash -ex
+
+. .ci/scripts/common/pre-upload.sh
+
+REV_NAME="yuzu-windows-mingw-${GITDATE}-${GITREV}"
+ARCHIVE_NAME="${REV_NAME}.tar.gz"
+COMPRESSION_FLAGS="-czvf"
+
+mkdir "$REV_NAME"
+# get around the permission issues
+cp -r package/* "$REV_NAME"
+
+. .ci/scripts/common/post-upload.sh
diff --git a/.ci/templates/build-single.yml b/.ci/templates/build-single.yml
new file mode 100644
index 000000000..357731eb9
--- /dev/null
+++ b/.ci/templates/build-single.yml
@@ -0,0 +1,23 @@
+parameters:
+  artifactSource: 'true'
+  cache: 'false'
+
+steps:
+- task: DockerInstaller@0
+  displayName: 'Prepare Environment'
+  inputs:
+    dockerVersion: '17.09.0-ce'
+- ${{ if eq(parameters.cache, 'true') }}:
+  - task: CacheBeta@0
+    displayName: 'Cache Build System'
+    inputs:
+      key: yuzu-v1-$(BuildName)-$(BuildSuffix)-$(CacheSuffix)
+      path: $(System.DefaultWorkingDirectory)/ccache
+      cacheHitVar: CACHE_RESTORED
+- script: chmod a+x ./.ci/scripts/$(ScriptFolder)/exec.sh && ./.ci/scripts/$(ScriptFolder)/exec.sh
+  displayName: 'Build'
+- script: chmod a+x ./.ci/scripts/$(ScriptFolder)/upload.sh && RELEASE_NAME=$(BuildName) ./.ci/scripts/$(ScriptFolder)/upload.sh
+  displayName: 'Package Artifacts'
+- publish: artifacts
+  artifact: 'yuzu-$(BuildName)-$(BuildSuffix)'
+  displayName: 'Upload Artifacts'
diff --git a/.ci/templates/build-standard.yml b/.ci/templates/build-standard.yml
new file mode 100644
index 000000000..aa180894e
--- /dev/null
+++ b/.ci/templates/build-standard.yml
@@ -0,0 +1,23 @@
+jobs:
+- job: build
+  displayName: 'standard'
+  pool:
+    vmImage: ubuntu-latest
+  strategy:
+    maxParallel: 10
+    matrix:
+      windows:
+        BuildSuffix: 'windows-mingw'
+        ScriptFolder: 'windows'
+      linux:
+        BuildSuffix: 'linux'
+        ScriptFolder: 'linux'
+  steps:
+  - template: ./sync-source.yml
+    parameters:
+      artifactSource: $(parameters.artifactSource)
+      needSubmodules: 'true'
+  - template: ./build-single.yml
+    parameters:
+      artifactSource: 'false'
+      cache: $(parameters.cache)
\ No newline at end of file
diff --git a/.ci/templates/build-testing.yml b/.ci/templates/build-testing.yml
new file mode 100644
index 000000000..a307addfd
--- /dev/null
+++ b/.ci/templates/build-testing.yml
@@ -0,0 +1,33 @@
+jobs:
+- job: build_test
+  displayName: 'testing'
+  pool:
+    vmImage: ubuntu-latest
+  strategy:
+    maxParallel: 5
+    matrix:
+      windows:
+        BuildSuffix: 'windows-testing'
+        ScriptFolder: 'windows'
+  steps:
+  - script: sudo apt upgrade python3-pip && pip install requests urllib3
+    displayName: 'Prepare Environment'
+  - task: PythonScript@0
+    condition: eq(variables['Build.Reason'], 'PullRequest')
+    displayName: 'Determine Testing Status'
+    inputs:
+      scriptSource: 'filePath'
+      scriptPath: '.ci/scripts/merge/check-label-presence.py'
+      arguments: '$(System.PullRequest.PullRequestNumber) create-testing-build'
+  - ${{ if eq(variables.enabletesting, 'true') }}:
+    - template: ./sync-source.yml
+      parameters:
+        artifactSource: $(parameters.artifactSource)
+        needSubmodules: 'true'
+    - template: ./mergebot.yml
+      parameters:
+        matchLabel: 'testing-merge'
+    - template: ./build-single.yml
+      parameters:
+        artifactSource: 'false'
+        cache: 'false'
diff --git a/.ci/templates/format-check.yml b/.ci/templates/format-check.yml
new file mode 100644
index 000000000..5061f1cb8
--- /dev/null
+++ b/.ci/templates/format-check.yml
@@ -0,0 +1,14 @@
+parameters:
+  artifactSource: 'true'
+
+steps:
+- template: ./sync-source.yml
+  parameters:
+    artifactSource: $(parameters.artifactSource)
+    needSubmodules: 'false'
+- task: DockerInstaller@0
+  displayName: 'Prepare Environment'
+  inputs:
+    dockerVersion: '17.09.0-ce'
+- script: chmod a+x ./.ci/scripts/format/exec.sh && ./.ci/scripts/format/exec.sh
+  displayName: 'Verify Formatting'
diff --git a/.ci/templates/merge.yml b/.ci/templates/merge.yml
new file mode 100644
index 000000000..efc82778a
--- /dev/null
+++ b/.ci/templates/merge.yml
@@ -0,0 +1,46 @@
+jobs:
+- job: merge
+  displayName: 'pull requests'
+  steps:
+  - checkout: self
+    submodules: recursive
+  - template: ./mergebot.yml
+    parameters:
+      matchLabel: '$(BuildName)-merge'
+  - task: ArchiveFiles@2
+    displayName: 'Package Source'
+    inputs:
+      rootFolderOrFile: '$(System.DefaultWorkingDirectory)'
+      includeRootFolder: false
+      archiveType: '7z'
+      archiveFile: '$(Build.ArtifactStagingDirectory)/yuzu-$(BuildName)-source.7z'
+  - task: PublishPipelineArtifact@1
+    displayName: 'Upload Artifacts'
+    inputs:
+      targetPath: '$(Build.ArtifactStagingDirectory)/yuzu-$(BuildName)-source.7z'
+      artifact: 'yuzu-$(BuildName)-source'
+      replaceExistingArchive: true
+- job: upload_source
+  displayName: 'upload'
+  dependsOn: merge
+  steps:
+  - template: ./sync-source.yml
+    parameters:
+      artifactSource: 'true'
+      needSubmodules: 'true'
+  - script: chmod a+x $(System.DefaultWorkingDirectory)/.ci/scripts/merge/yuzubot-git-config.sh && $(System.DefaultWorkingDirectory)/.ci/scripts/merge/yuzubot-git-config.sh
+    displayName: 'Apply Git Configuration'
+  - script: git tag -a $(BuildName)-$(Build.BuildId) -m "yuzu $(BuildName) $(Build.BuildNumber) $(Build.DefinitionName)"
+    displayName: 'Tag Source'
+  - script: git remote add other $(GitRepoPushChangesURL)
+    displayName: 'Register Repository'
+  - script: git push --follow-tags --force other HEAD:$(GitPushBranch)
+    displayName: 'Update Code'
+  - script: git rev-list -n 1 $(BuildName)-$(Build.BuildId) > $(Build.ArtifactStagingDirectory)/tag-commit.sha
+    displayName: 'Calculate Release Point'
+  - task: PublishPipelineArtifact@1
+    displayName: 'Upload Release Point'
+    inputs:
+      targetPath: '$(Build.ArtifactStagingDirectory)/tag-commit.sha'
+      artifact: 'yuzu-$(BuildName)-release-point'
+      replaceExistingArchive: true
\ No newline at end of file
diff --git a/.ci/templates/mergebot.yml b/.ci/templates/mergebot.yml
new file mode 100644
index 000000000..5211efcc6
--- /dev/null
+++ b/.ci/templates/mergebot.yml
@@ -0,0 +1,15 @@
+parameters:
+  matchLabel: 'dummy-merge'
+
+steps:
+  - script: mkdir $(System.DefaultWorkingDirectory)/patches && pip install requests urllib3
+    displayName: 'Prepare Environment'
+  - script: chmod a+x $(System.DefaultWorkingDirectory)/.ci/scripts/merge/yuzubot-git-config.sh && $(System.DefaultWorkingDirectory)/.ci/scripts/merge/yuzubot-git-config.sh
+    displayName: 'Apply Git Configuration'
+  - task: PythonScript@0
+    displayName: 'Discover, Download, and Apply Patches'
+    inputs:
+      scriptSource: 'filePath'
+      scriptPath: '.ci/scripts/merge/apply-patches-by-label.py'
+      arguments: '${{ parameters.matchLabel }} patches'
+      workingDirectory: '$(System.DefaultWorkingDirectory)'
diff --git a/.ci/templates/retrieve-artifact-source.yml b/.ci/templates/retrieve-artifact-source.yml
new file mode 100644
index 000000000..47d217e7b
--- /dev/null
+++ b/.ci/templates/retrieve-artifact-source.yml
@@ -0,0 +1,16 @@
+steps:
+- checkout: none
+- task: DownloadPipelineArtifact@2
+  displayName: 'Download Source'
+  inputs:
+    artifactName: 'yuzu-$(BuildName)-source'
+    buildType: 'current'
+    targetPath: '$(Build.ArtifactStagingDirectory)'
+- script: rm -rf $(System.DefaultWorkingDirectory) && mkdir $(System.DefaultWorkingDirectory)
+  displayName: 'Clean Working Directory'
+- task: ExtractFiles@1
+  displayName: 'Prepare Source'
+  inputs:
+    archiveFilePatterns: '$(Build.ArtifactStagingDirectory)/*.7z'
+    destinationFolder: '$(System.DefaultWorkingDirectory)'
+    cleanDestinationFolder: false
\ No newline at end of file
diff --git a/.ci/templates/retrieve-master-source.yml b/.ci/templates/retrieve-master-source.yml
new file mode 100644
index 000000000..a08a3f926
--- /dev/null
+++ b/.ci/templates/retrieve-master-source.yml
@@ -0,0 +1,11 @@
+parameters:
+  needSubmodules: 'true'
+
+steps:
+- checkout: self
+  displayName: 'Checkout Recursive'
+  submodules: recursive
+#  condition: eq(parameters.needSubmodules, 'true')
+#- checkout: self
+#  displayName: 'Checkout Fast'
+#  condition: ne(parameters.needSubmodules, 'true')
diff --git a/.ci/templates/sync-source.yml b/.ci/templates/sync-source.yml
new file mode 100644
index 000000000..409e1cd83
--- /dev/null
+++ b/.ci/templates/sync-source.yml
@@ -0,0 +1,7 @@
+steps:
+- ${{ if eq(parameters.artifactSource, 'true') }}:
+  - template: ./retrieve-artifact-source.yml
+- ${{ if ne(parameters.artifactSource, 'true') }}:
+  - template: ./retrieve-master-source.yml
+    parameters:
+      needSubmodules: $(parameters.needSubmodules)
\ No newline at end of file
diff --git a/.ci/yuzu-mainline.yml b/.ci/yuzu-mainline.yml
new file mode 100644
index 000000000..2930a8564
--- /dev/null
+++ b/.ci/yuzu-mainline.yml
@@ -0,0 +1,25 @@
+trigger:
+- master
+
+stages:
+- stage: merge
+  displayName: 'merge'
+  jobs:
+  - template: ./templates/merge.yml
+- stage: format
+  dependsOn: merge
+  displayName: 'format'
+  jobs:
+  - job: format
+    displayName: 'clang'
+    pool:
+      vmImage: ubuntu-latest
+    steps:
+    - template: ./templates/format-check.yml
+- stage: build
+  displayName: 'build'
+  dependsOn: format
+  jobs:
+  - template: ./templates/build-standard.yml
+    parameters:
+      cache: 'true'
diff --git a/azure-pipelines.yml b/.ci/yuzu-patreon.yml
index aa912913d..aa912913d 100644
--- a/azure-pipelines.yml
+++ b/.ci/yuzu-patreon.yml
diff --git a/.ci/yuzu-repo-sync.yml b/.ci/yuzu-repo-sync.yml
new file mode 100644
index 000000000..602e298a6
--- /dev/null
+++ b/.ci/yuzu-repo-sync.yml
@@ -0,0 +1,19 @@
+trigger:
+- master
+
+jobs:
+- job: copy
+  displayName: 'Sync Repository'
+  pool:
+    vmImage: 'ubuntu-latest'
+  steps:
+  - script: echo 'https://$(GitUsername):$(GitAccessToken)@dev.azure.com' > $HOME/.git-credentials
+    displayName: 'Load Credentials'
+  - script: git config --global credential.helper store
+    displayName: 'Register Credential Helper'
+  - script: git remote add other $(GitRepoPushChangesURL)
+    displayName: 'Register Repository'
+  - script: git push --force other HEAD:$(GitPushBranch)
+    displayName: 'Update Code'
+  - script: rm -rf $HOME/.git-credentials
+    displayName: 'Clear Cached Credentials'
diff --git a/.ci/yuzu-verify.yml b/.ci/yuzu-verify.yml
new file mode 100644
index 000000000..5492e696a
--- /dev/null
+++ b/.ci/yuzu-verify.yml
@@ -0,0 +1,20 @@
+stages:
+- stage: format
+  displayName: 'format'
+  jobs:
+  - job: format
+    displayName: 'clang'
+    pool:
+      vmImage: ubuntu-latest
+    steps:
+    - template: ./templates/format-check.yml
+      parameters:
+        artifactSource: 'false'
+- stage: build
+  displayName: 'build'
+  dependsOn: format
+  jobs:
+  - template: ./templates/build-standard.yml
+    parameters:
+      cache: 'false'
+  - template: ./templates/build-testing.yml
diff --git a/CMakeModules/GenerateSCMRev.cmake b/CMakeModules/GenerateSCMRev.cmake
index dd65cfe42..a1ace89cb 100644
--- a/CMakeModules/GenerateSCMRev.cmake
+++ b/CMakeModules/GenerateSCMRev.cmake
@@ -81,7 +81,10 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
     "${VIDEO_CORE}/shader/decode/shift.cpp"
     "${VIDEO_CORE}/shader/decode/video.cpp"
+    "${VIDEO_CORE}/shader/decode/warp.cpp"
     "${VIDEO_CORE}/shader/decode/xmad.cpp"
+    "${VIDEO_CORE}/shader/control_flow.cpp"
+    "${VIDEO_CORE}/shader/control_flow.h"
     "${VIDEO_CORE}/shader/decode.cpp"
     "${VIDEO_CORE}/shader/node.h"
     "${VIDEO_CORE}/shader/node_helper.cpp"
diff --git a/README.md b/README.md
index 4b1ea7d7c..430c6dd65 100644
--- a/README.md
+++ b/README.md
@@ -2,6 +2,7 @@ yuzu emulator
 =============
 [![Travis CI Build Status](https://travis-ci.org/yuzu-emu/yuzu.svg?branch=master)](https://travis-ci.org/yuzu-emu/yuzu)
 [![AppVeyor CI Build Status](https://ci.appveyor.com/api/projects/status/77k97svb2usreu68?svg=true)](https://ci.appveyor.com/project/bunnei/yuzu)
+[![Azure Mainline CI Build Status](https://dev.azure.com/yuzu-emu/yuzu/_apis/build/status/yuzu%20mainline?branchName=master)](https://dev.azure.com/yuzu-emu/yuzu/)
 
 yuzu is an experimental open-source emulator for the Nintendo Switch from the creators of [Citra](https://citra-emu.org/).
 
diff --git a/src/audio_core/audio_renderer.cpp b/src/audio_core/audio_renderer.cpp
index 4882a6cd8..da50a0bbc 100644
--- a/src/audio_core/audio_renderer.cpp
+++ b/src/audio_core/audio_renderer.cpp
@@ -73,13 +73,15 @@ private:
     EffectInStatus info{};
 };
 AudioRenderer::AudioRenderer(Core::Timing::CoreTiming& core_timing, AudioRendererParameter params,
-                             Kernel::SharedPtr<Kernel::WritableEvent> buffer_event)
+                             Kernel::SharedPtr<Kernel::WritableEvent> buffer_event,
+                             std::size_t instance_number)
     : worker_params{params}, buffer_event{buffer_event}, voices(params.voice_count),
       effects(params.effect_count) {
 
     audio_out = std::make_unique<AudioCore::AudioOut>();
     stream = audio_out->OpenStream(core_timing, STREAM_SAMPLE_RATE, STREAM_NUM_CHANNELS,
-                                   "AudioRenderer", [=]() { buffer_event->Signal(); });
+                                   fmt::format("AudioRenderer-Instance{}", instance_number),
+                                   [=]() { buffer_event->Signal(); });
     audio_out->StartStream(stream);
 
     QueueMixedBuffer(0);
diff --git a/src/audio_core/audio_renderer.h b/src/audio_core/audio_renderer.h
index b2e5d336c..45afbe759 100644
--- a/src/audio_core/audio_renderer.h
+++ b/src/audio_core/audio_renderer.h
@@ -215,7 +215,8 @@ static_assert(sizeof(UpdateDataHeader) == 0x40, "UpdateDataHeader has wrong size
 class AudioRenderer {
 public:
     AudioRenderer(Core::Timing::CoreTiming& core_timing, AudioRendererParameter params,
-                  Kernel::SharedPtr<Kernel::WritableEvent> buffer_event);
+                  Kernel::SharedPtr<Kernel::WritableEvent> buffer_event,
+                  std::size_t instance_number);
     ~AudioRenderer();
 
     std::vector<u8> UpdateAudioRenderer(const std::vector<u8>& input_params);
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 2554add28..01abdb3bb 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -55,7 +55,10 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
       "${VIDEO_CORE}/shader/decode/shift.cpp"
       "${VIDEO_CORE}/shader/decode/video.cpp"
+      "${VIDEO_CORE}/shader/decode/warp.cpp"
       "${VIDEO_CORE}/shader/decode/xmad.cpp"
+      "${VIDEO_CORE}/shader/control_flow.cpp"
+      "${VIDEO_CORE}/shader/control_flow.h"
       "${VIDEO_CORE}/shader/decode.cpp"
       "${VIDEO_CORE}/shader/node.h"
       "${VIDEO_CORE}/shader/node_helper.cpp"
diff --git a/src/common/alignment.h b/src/common/alignment.h
index 617b14d9b..88d5d3a65 100644
--- a/src/common/alignment.h
+++ b/src/common/alignment.h
@@ -3,6 +3,7 @@
 #pragma once
 
 #include <cstddef>
+#include <memory>
 #include <type_traits>
 
 namespace Common {
@@ -37,4 +38,63 @@ constexpr bool IsWordAligned(T value) {
     return (value & 0b11) == 0;
 }
 
+template <typename T, std::size_t Align = 16>
+class AlignmentAllocator {
+public:
+    using value_type = T;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    using pointer = T*;
+    using const_pointer = const T*;
+
+    using reference = T&;
+    using const_reference = const T&;
+
+public:
+    pointer address(reference r) noexcept {
+        return std::addressof(r);
+    }
+
+    const_pointer address(const_reference r) const noexcept {
+        return std::addressof(r);
+    }
+
+    pointer allocate(size_type n) {
+        return static_cast<pointer>(::operator new (n, std::align_val_t{Align}));
+    }
+
+    void deallocate(pointer p, size_type) {
+        ::operator delete (p, std::align_val_t{Align});
+    }
+
+    void construct(pointer p, const value_type& wert) {
+        new (p) value_type(wert);
+    }
+
+    void destroy(pointer p) {
+        p->~value_type();
+    }
+
+    size_type max_size() const noexcept {
+        return size_type(-1) / sizeof(value_type);
+    }
+
+    template <typename T2>
+    struct rebind {
+        using other = AlignmentAllocator<T2, Align>;
+    };
+
+    bool operator!=(const AlignmentAllocator<T, Align>& other) const noexcept {
+        return !(*this == other);
+    }
+
+    // Returns true if and only if storage allocated from *this
+    // can be deallocated from other, and vice versa.
+    // Always returns true for stateless allocators.
+    bool operator==(const AlignmentAllocator<T, Align>& other) const noexcept {
+        return true;
+    }
+};
+
 } // namespace Common
diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
index f4325f0f8..5462decee 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -111,6 +111,8 @@ add_library(core STATIC
     frontend/scope_acquire_window_context.h
     gdbstub/gdbstub.cpp
     gdbstub/gdbstub.h
+    hardware_interrupt_manager.cpp
+    hardware_interrupt_manager.h
     hle/ipc.h
     hle/ipc_helpers.h
     hle/kernel/address_arbiter.cpp
@@ -372,6 +374,7 @@ add_library(core STATIC
     hle/service/nvdrv/devices/nvmap.h
     hle/service/nvdrv/interface.cpp
     hle/service/nvdrv/interface.h
+    hle/service/nvdrv/nvdata.h
     hle/service/nvdrv/nvdrv.cpp
     hle/service/nvdrv/nvdrv.h
     hle/service/nvdrv/nvmemp.cpp
diff --git a/src/core/arm/arm_interface.h b/src/core/arm/arm_interface.h
index c6691a8e1..45e94e625 100644
--- a/src/core/arm/arm_interface.h
+++ b/src/core/arm/arm_interface.h
@@ -44,13 +44,6 @@ public:
     /// Step CPU by one instruction
     virtual void Step() = 0;
 
-    /// Maps a backing memory region for the CPU
-    virtual void MapBackingMemory(VAddr address, std::size_t size, u8* memory,
-                                  Kernel::VMAPermission perms) = 0;
-
-    /// Unmaps a region of memory that was previously mapped using MapBackingMemory
-    virtual void UnmapMemory(VAddr address, std::size_t size) = 0;
-
     /// Clear all instruction cache
     virtual void ClearInstructionCache() = 0;
 
diff --git a/src/core/arm/dynarmic/arm_dynarmic.cpp b/src/core/arm/dynarmic/arm_dynarmic.cpp
index 44307fa19..f1506b372 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic.cpp
@@ -177,15 +177,6 @@ ARM_Dynarmic::ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor,
 
 ARM_Dynarmic::~ARM_Dynarmic() = default;
 
-void ARM_Dynarmic::MapBackingMemory(u64 address, std::size_t size, u8* memory,
-                                    Kernel::VMAPermission perms) {
-    inner_unicorn.MapBackingMemory(address, size, memory, perms);
-}
-
-void ARM_Dynarmic::UnmapMemory(u64 address, std::size_t size) {
-    inner_unicorn.UnmapMemory(address, size);
-}
-
 void ARM_Dynarmic::SetPC(u64 pc) {
     jit->SetPC(pc);
 }
diff --git a/src/core/arm/dynarmic/arm_dynarmic.h b/src/core/arm/dynarmic/arm_dynarmic.h
index b701e97a3..504d46c68 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.h
+++ b/src/core/arm/dynarmic/arm_dynarmic.h
@@ -23,9 +23,6 @@ public:
     ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index);
     ~ARM_Dynarmic() override;
 
-    void MapBackingMemory(VAddr address, std::size_t size, u8* memory,
-                          Kernel::VMAPermission perms) override;
-    void UnmapMemory(u64 address, std::size_t size) override;
     void SetPC(u64 pc) override;
     u64 GetPC() const override;
     u64 GetReg(int index) const override;
diff --git a/src/core/arm/unicorn/arm_unicorn.cpp b/src/core/arm/unicorn/arm_unicorn.cpp
index 4e07fe8b5..97d5c2a8a 100644
--- a/src/core/arm/unicorn/arm_unicorn.cpp
+++ b/src/core/arm/unicorn/arm_unicorn.cpp
@@ -50,11 +50,14 @@ static void CodeHook(uc_engine* uc, uint64_t address, uint32_t size, void* user_
 
 static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int size, u64 value,
                                void* user_data) {
+    auto* const system = static_cast<System*>(user_data);
+
     ARM_Interface::ThreadContext ctx{};
-    Core::CurrentArmInterface().SaveContext(ctx);
+    system->CurrentArmInterface().SaveContext(ctx);
     ASSERT_MSG(false, "Attempted to read from unmapped memory: 0x{:X}, pc=0x{:X}, lr=0x{:X}", addr,
                ctx.pc, ctx.cpu_registers[30]);
-    return {};
+
+    return false;
 }
 
 ARM_Unicorn::ARM_Unicorn(System& system) : system{system} {
@@ -65,7 +68,7 @@ ARM_Unicorn::ARM_Unicorn(System& system) : system{system} {
 
     uc_hook hook{};
     CHECKED(uc_hook_add(uc, &hook, UC_HOOK_INTR, (void*)InterruptHook, this, 0, -1));
-    CHECKED(uc_hook_add(uc, &hook, UC_HOOK_MEM_INVALID, (void*)UnmappedMemoryHook, this, 0, -1));
+    CHECKED(uc_hook_add(uc, &hook, UC_HOOK_MEM_INVALID, (void*)UnmappedMemoryHook, &system, 0, -1));
     if (GDBStub::IsServerEnabled()) {
         CHECKED(uc_hook_add(uc, &hook, UC_HOOK_CODE, (void*)CodeHook, this, 0, -1));
         last_bkpt_hit = false;
@@ -76,15 +79,6 @@ ARM_Unicorn::~ARM_Unicorn() {
     CHECKED(uc_close(uc));
 }
 
-void ARM_Unicorn::MapBackingMemory(VAddr address, std::size_t size, u8* memory,
-                                   Kernel::VMAPermission perms) {
-    CHECKED(uc_mem_map_ptr(uc, address, size, static_cast<u32>(perms), memory));
-}
-
-void ARM_Unicorn::UnmapMemory(VAddr address, std::size_t size) {
-    CHECKED(uc_mem_unmap(uc, address, size));
-}
-
 void ARM_Unicorn::SetPC(u64 pc) {
     CHECKED(uc_reg_write(uc, UC_ARM64_REG_PC, &pc));
 }
diff --git a/src/core/arm/unicorn/arm_unicorn.h b/src/core/arm/unicorn/arm_unicorn.h
index 34e974b4d..fe2ffd70c 100644
--- a/src/core/arm/unicorn/arm_unicorn.h
+++ b/src/core/arm/unicorn/arm_unicorn.h
@@ -18,9 +18,6 @@ public:
     explicit ARM_Unicorn(System& system);
     ~ARM_Unicorn() override;
 
-    void MapBackingMemory(VAddr address, std::size_t size, u8* memory,
-                          Kernel::VMAPermission perms) override;
-    void UnmapMemory(VAddr address, std::size_t size) override;
     void SetPC(u64 pc) override;
     u64 GetPC() const override;
     u64 GetReg(int index) const override;
diff --git a/src/core/core.cpp b/src/core/core.cpp
index 4aceee785..20d64f3b0 100644
--- a/src/core/core.cpp
+++ b/src/core/core.cpp
@@ -19,6 +19,7 @@
 #include "core/file_sys/vfs_concat.h"
 #include "core/file_sys/vfs_real.h"
 #include "core/gdbstub/gdbstub.h"
+#include "core/hardware_interrupt_manager.h"
 #include "core/hle/kernel/client_port.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
@@ -151,7 +152,7 @@ struct System::Impl {
         if (!renderer->Init()) {
             return ResultStatus::ErrorVideoCore;
         }
-
+        interrupt_manager = std::make_unique<Core::Hardware::InterruptManager>(system);
         gpu_core = VideoCore::CreateGPU(system);
 
         is_powered_on = true;
@@ -298,6 +299,7 @@ struct System::Impl {
     std::unique_ptr<VideoCore::RendererBase> renderer;
     std::unique_ptr<Tegra::GPU> gpu_core;
     std::shared_ptr<Tegra::DebugContext> debug_context;
+    std::unique_ptr<Core::Hardware::InterruptManager> interrupt_manager;
     CpuCoreManager cpu_core_manager;
     bool is_powered_on = false;
 
@@ -444,6 +446,14 @@ const Tegra::GPU& System::GPU() const {
     return *impl->gpu_core;
 }
 
+Core::Hardware::InterruptManager& System::InterruptManager() {
+    return *impl->interrupt_manager;
+}
+
+const Core::Hardware::InterruptManager& System::InterruptManager() const {
+    return *impl->interrupt_manager;
+}
+
 VideoCore::RendererBase& System::Renderer() {
     return *impl->renderer;
 }
diff --git a/src/core/core.h b/src/core/core.h
index 11e73278e..0138d93b0 100644
--- a/src/core/core.h
+++ b/src/core/core.h
@@ -70,6 +70,10 @@ namespace Core::Timing {
 class CoreTiming;
 }
 
+namespace Core::Hardware {
+class InterruptManager;
+}
+
 namespace Core {
 
 class ARM_Interface;
@@ -234,6 +238,12 @@ public:
     /// Provides a constant reference to the core timing instance.
     const Timing::CoreTiming& CoreTiming() const;
 
+    /// Provides a reference to the interrupt manager instance.
+    Core::Hardware::InterruptManager& InterruptManager();
+
+    /// Provides a constant reference to the interrupt manager instance.
+    const Core::Hardware::InterruptManager& InterruptManager() const;
+
     /// Provides a reference to the kernel instance.
     Kernel::KernelCore& Kernel();
 
@@ -327,10 +337,6 @@ private:
     static System s_instance;
 };
 
-inline ARM_Interface& CurrentArmInterface() {
-    return System::GetInstance().CurrentArmInterface();
-}
-
 inline Kernel::Process* CurrentProcess() {
     return System::GetInstance().CurrentProcess();
 }
diff --git a/src/core/core_cpu.cpp b/src/core/core_cpu.cpp
index 99b7d387d..21c410e34 100644
--- a/src/core/core_cpu.cpp
+++ b/src/core/core_cpu.cpp
@@ -53,16 +53,12 @@ bool CpuBarrier::Rendezvous() {
 Cpu::Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_barrier,
          std::size_t core_index)
     : cpu_barrier{cpu_barrier}, core_timing{system.CoreTiming()}, core_index{core_index} {
-    if (Settings::values.cpu_jit_enabled) {
 #ifdef ARCHITECTURE_x86_64
-        arm_interface = std::make_unique<ARM_Dynarmic>(system, exclusive_monitor, core_index);
+    arm_interface = std::make_unique<ARM_Dynarmic>(system, exclusive_monitor, core_index);
 #else
-        arm_interface = std::make_unique<ARM_Unicorn>(system);
-        LOG_WARNING(Core, "CPU JIT requested, but Dynarmic not available");
+    arm_interface = std::make_unique<ARM_Unicorn>(system);
+    LOG_WARNING(Core, "CPU JIT requested, but Dynarmic not available");
 #endif
-    } else {
-        arm_interface = std::make_unique<ARM_Unicorn>(system);
-    }
 
     scheduler = std::make_unique<Kernel::Scheduler>(system, *arm_interface);
 }
@@ -70,15 +66,12 @@ Cpu::Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_ba
 Cpu::~Cpu() = default;
 
 std::unique_ptr<ExclusiveMonitor> Cpu::MakeExclusiveMonitor(std::size_t num_cores) {
-    if (Settings::values.cpu_jit_enabled) {
 #ifdef ARCHITECTURE_x86_64
-        return std::make_unique<DynarmicExclusiveMonitor>(num_cores);
+    return std::make_unique<DynarmicExclusiveMonitor>(num_cores);
 #else
-        return nullptr; // TODO(merry): Passthrough exclusive monitor
+    // TODO(merry): Passthrough exclusive monitor
+    return nullptr;
 #endif
-    } else {
-        return nullptr; // TODO(merry): Passthrough exclusive monitor
-    }
 }
 
 void Cpu::RunLoop(bool tight_loop) {
diff --git a/src/core/file_sys/program_metadata.cpp b/src/core/file_sys/program_metadata.cpp
index eb76174c5..7310b3602 100644
--- a/src/core/file_sys/program_metadata.cpp
+++ b/src/core/file_sys/program_metadata.cpp
@@ -94,6 +94,10 @@ u64 ProgramMetadata::GetFilesystemPermissions() const {
     return aci_file_access.permissions;
 }
 
+u32 ProgramMetadata::GetSystemResourceSize() const {
+    return npdm_header.system_resource_size;
+}
+
 const ProgramMetadata::KernelCapabilityDescriptors& ProgramMetadata::GetKernelCapabilities() const {
     return aci_kernel_capabilities;
 }
diff --git a/src/core/file_sys/program_metadata.h b/src/core/file_sys/program_metadata.h
index 43bf2820a..88ec97d85 100644
--- a/src/core/file_sys/program_metadata.h
+++ b/src/core/file_sys/program_metadata.h
@@ -58,6 +58,7 @@ public:
     u32 GetMainThreadStackSize() const;
     u64 GetTitleID() const;
     u64 GetFilesystemPermissions() const;
+    u32 GetSystemResourceSize() const;
     const KernelCapabilityDescriptors& GetKernelCapabilities() const;
 
     void Print() const;
@@ -76,7 +77,8 @@ private:
         u8 reserved_3;
         u8 main_thread_priority;
         u8 main_thread_cpu;
-        std::array<u8, 8> reserved_4;
+        std::array<u8, 4> reserved_4;
+        u32_le system_resource_size;
         u32_le process_category;
         u32_le main_stack_size;
         std::array<u8, 0x10> application_name;
diff --git a/src/core/hardware_interrupt_manager.cpp b/src/core/hardware_interrupt_manager.cpp
new file mode 100644
index 000000000..c2115db2d
--- /dev/null
+++ b/src/core/hardware_interrupt_manager.cpp
@@ -0,0 +1,30 @@
+// Copyright 2019 Yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/hardware_interrupt_manager.h"
+#include "core/hle/service/nvdrv/interface.h"
+#include "core/hle/service/sm/sm.h"
+
+namespace Core::Hardware {
+
+InterruptManager::InterruptManager(Core::System& system_in) : system(system_in) {
+    gpu_interrupt_event =
+        system.CoreTiming().RegisterEvent("GPUInterrupt", [this](u64 message, s64) {
+            auto nvdrv = system.ServiceManager().GetService<Service::Nvidia::NVDRV>("nvdrv");
+            const u32 syncpt = static_cast<u32>(message >> 32);
+            const u32 value = static_cast<u32>(message);
+            nvdrv->SignalGPUInterruptSyncpt(syncpt, value);
+        });
+}
+
+InterruptManager::~InterruptManager() = default;
+
+void InterruptManager::GPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) {
+    const u64 msg = (static_cast<u64>(syncpoint_id) << 32ULL) | value;
+    system.CoreTiming().ScheduleEvent(10, gpu_interrupt_event, msg);
+}
+
+} // namespace Core::Hardware
diff --git a/src/core/hardware_interrupt_manager.h b/src/core/hardware_interrupt_manager.h
new file mode 100644
index 000000000..494db883a
--- /dev/null
+++ b/src/core/hardware_interrupt_manager.h
@@ -0,0 +1,31 @@
+// Copyright 2019 Yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+
+namespace Core {
+class System;
+}
+
+namespace Core::Timing {
+struct EventType;
+}
+
+namespace Core::Hardware {
+
+class InterruptManager {
+public:
+    explicit InterruptManager(Core::System& system);
+    ~InterruptManager();
+
+    void GPUInterruptSyncpt(u32 syncpoint_id, u32 value);
+
+private:
+    Core::System& system;
+    Core::Timing::EventType* gpu_interrupt_event{};
+};
+
+} // namespace Core::Hardware
diff --git a/src/core/hle/kernel/code_set.h b/src/core/hle/kernel/code_set.h
index 879957dcb..d8ad54030 100644
--- a/src/core/hle/kernel/code_set.h
+++ b/src/core/hle/kernel/code_set.h
@@ -8,6 +8,7 @@
 #include <vector>
 
 #include "common/common_types.h"
+#include "core/hle/kernel/physical_memory.h"
 
 namespace Kernel {
 
@@ -77,7 +78,7 @@ struct CodeSet final {
     }
 
     /// The overall data that backs this code set.
-    std::vector<u8> memory;
+    Kernel::PhysicalMemory memory;
 
     /// The segments that comprise this code set.
     std::array<Segment, 3> segments;
diff --git a/src/core/hle/kernel/physical_memory.h b/src/core/hle/kernel/physical_memory.h
new file mode 100644
index 000000000..090565310
--- /dev/null
+++ b/src/core/hle/kernel/physical_memory.h
@@ -0,0 +1,19 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/alignment.h"
+
+namespace Kernel {
+
+// This encapsulation serves 2 purposes:
+// - First, to encapsulate host physical memory under a single type and set an
+// standard for managing it.
+// - Second to ensure all host backing memory used is aligned to 256 bytes due
+// to strict alignment restrictions on GPU memory.
+
+using PhysicalMemory = std::vector<u8, Common::AlignmentAllocator<u8, 256>>;
+
+} // namespace Kernel
diff --git a/src/core/hle/kernel/process.cpp b/src/core/hle/kernel/process.cpp
index f45ef05f6..e80a12ac3 100644
--- a/src/core/hle/kernel/process.cpp
+++ b/src/core/hle/kernel/process.cpp
@@ -129,20 +129,17 @@ u64 Process::GetTotalPhysicalMemoryAvailable() const {
     return vm_manager.GetTotalPhysicalMemoryAvailable();
 }
 
-u64 Process::GetTotalPhysicalMemoryAvailableWithoutMmHeap() const {
-    // TODO: Subtract the personal heap size from this when the
-    //       personal heap is implemented.
-    return GetTotalPhysicalMemoryAvailable();
+u64 Process::GetTotalPhysicalMemoryAvailableWithoutSystemResource() const {
+    return GetTotalPhysicalMemoryAvailable() - GetSystemResourceSize();
 }
 
 u64 Process::GetTotalPhysicalMemoryUsed() const {
-    return vm_manager.GetCurrentHeapSize() + main_thread_stack_size + code_memory_size;
+    return vm_manager.GetCurrentHeapSize() + main_thread_stack_size + code_memory_size +
+           GetSystemResourceUsage();
 }
 
-u64 Process::GetTotalPhysicalMemoryUsedWithoutMmHeap() const {
-    // TODO: Subtract the personal heap size from this when the
-    //       personal heap is implemented.
-    return GetTotalPhysicalMemoryUsed();
+u64 Process::GetTotalPhysicalMemoryUsedWithoutSystemResource() const {
+    return GetTotalPhysicalMemoryUsed() - GetSystemResourceUsage();
 }
 
 void Process::RegisterThread(const Thread* thread) {
@@ -172,6 +169,7 @@ ResultCode Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
     program_id = metadata.GetTitleID();
     ideal_core = metadata.GetMainThreadCore();
     is_64bit_process = metadata.Is64BitProgram();
+    system_resource_size = metadata.GetSystemResourceSize();
 
     vm_manager.Reset(metadata.GetAddressSpaceType());
 
@@ -186,19 +184,11 @@ ResultCode Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
 }
 
 void Process::Run(s32 main_thread_priority, u64 stack_size) {
-    // The kernel always ensures that the given stack size is page aligned.
-    main_thread_stack_size = Common::AlignUp(stack_size, Memory::PAGE_SIZE);
-
-    // Allocate and map the main thread stack
-    // TODO(bunnei): This is heap area that should be allocated by the kernel and not mapped as part
-    // of the user address space.
-    const VAddr mapping_address = vm_manager.GetTLSIORegionEndAddress() - main_thread_stack_size;
-    vm_manager
-        .MapMemoryBlock(mapping_address, std::make_shared<std::vector<u8>>(main_thread_stack_size),
-                        0, main_thread_stack_size, MemoryState::Stack)
-        .Unwrap();
+    AllocateMainThreadStack(stack_size);
+    tls_region_address = CreateTLSRegion();
 
     vm_manager.LogLayout();
+
     ChangeStatus(ProcessStatus::Running);
 
     SetupMainThread(*this, kernel, main_thread_priority);
@@ -228,6 +218,9 @@ void Process::PrepareForTermination() {
     stop_threads(system.Scheduler(2).GetThreadList());
     stop_threads(system.Scheduler(3).GetThreadList());
 
+    FreeTLSRegion(tls_region_address);
+    tls_region_address = 0;
+
     ChangeStatus(ProcessStatus::Exited);
 }
 
@@ -254,7 +247,7 @@ VAddr Process::CreateTLSRegion() {
         ASSERT(region_address.Succeeded());
 
         const auto map_result = vm_manager.MapMemoryBlock(
-            *region_address, std::make_shared<std::vector<u8>>(Memory::PAGE_SIZE), 0,
+            *region_address, std::make_shared<PhysicalMemory>(Memory::PAGE_SIZE), 0,
             Memory::PAGE_SIZE, MemoryState::ThreadLocal);
         ASSERT(map_result.Succeeded());
 
@@ -284,7 +277,7 @@ void Process::FreeTLSRegion(VAddr tls_address) {
 }
 
 void Process::LoadModule(CodeSet module_, VAddr base_addr) {
-    const auto memory = std::make_shared<std::vector<u8>>(std::move(module_.memory));
+    const auto memory = std::make_shared<PhysicalMemory>(std::move(module_.memory));
 
     const auto MapSegment = [&](const CodeSet::Segment& segment, VMAPermission permissions,
                                 MemoryState memory_state) {
@@ -327,4 +320,16 @@ void Process::ChangeStatus(ProcessStatus new_status) {
     WakeupAllWaitingThreads();
 }
 
+void Process::AllocateMainThreadStack(u64 stack_size) {
+    // The kernel always ensures that the given stack size is page aligned.
+    main_thread_stack_size = Common::AlignUp(stack_size, Memory::PAGE_SIZE);
+
+    // Allocate and map the main thread stack
+    const VAddr mapping_address = vm_manager.GetTLSIORegionEndAddress() - main_thread_stack_size;
+    vm_manager
+        .MapMemoryBlock(mapping_address, std::make_shared<PhysicalMemory>(main_thread_stack_size),
+                        0, main_thread_stack_size, MemoryState::Stack)
+        .Unwrap();
+}
+
 } // namespace Kernel
diff --git a/src/core/hle/kernel/process.h b/src/core/hle/kernel/process.h
index 83ea02bee..c2df451f3 100644
--- a/src/core/hle/kernel/process.h
+++ b/src/core/hle/kernel/process.h
@@ -135,6 +135,11 @@ public:
         return mutex;
     }
 
+    /// Gets the address to the process' dedicated TLS region.
+    VAddr GetTLSRegionAddress() const {
+        return tls_region_address;
+    }
+
     /// Gets the current status of the process
     ProcessStatus GetStatus() const {
         return status;
@@ -168,8 +173,24 @@ public:
         return capabilities.GetPriorityMask();
     }
 
-    u32 IsVirtualMemoryEnabled() const {
-        return is_virtual_address_memory_enabled;
+    /// Gets the amount of secure memory to allocate for memory management.
+    u32 GetSystemResourceSize() const {
+        return system_resource_size;
+    }
+
+    /// Gets the amount of secure memory currently in use for memory management.
+    u32 GetSystemResourceUsage() const {
+        // On hardware, this returns the amount of system resource memory that has
+        // been used by the kernel. This is problematic for Yuzu to emulate, because
+        // system resource memory is used for page tables -- and yuzu doesn't really
+        // have a way to calculate how much memory is required for page tables for
+        // the current process at any given time.
+        // TODO: Is this even worth implementing? Games may retrieve this value via
+        // an SDK function that gets used + available system resource size for debug
+        // or diagnostic purposes. However, it seems unlikely that a game would make
+        // decisions based on how much system memory is dedicated to its page tables.
+        // Is returning a value other than zero wise?
+        return 0;
     }
 
     /// Whether this process is an AArch64 or AArch32 process.
@@ -196,15 +217,15 @@ public:
     u64 GetTotalPhysicalMemoryAvailable() const;
 
     /// Retrieves the total physical memory available to this process in bytes,
-    /// without the size of the personal heap added to it.
-    u64 GetTotalPhysicalMemoryAvailableWithoutMmHeap() const;
+    /// without the size of the personal system resource heap added to it.
+    u64 GetTotalPhysicalMemoryAvailableWithoutSystemResource() const;
 
     /// Retrieves the total physical memory used by this process in bytes.
     u64 GetTotalPhysicalMemoryUsed() const;
 
     /// Retrieves the total physical memory used by this process in bytes,
-    /// without the size of the personal heap added to it.
-    u64 GetTotalPhysicalMemoryUsedWithoutMmHeap() const;
+    /// without the size of the personal system resource heap added to it.
+    u64 GetTotalPhysicalMemoryUsedWithoutSystemResource() const;
 
     /// Gets the list of all threads created with this process as their owner.
     const std::list<const Thread*>& GetThreadList() const {
@@ -280,6 +301,9 @@ private:
     /// a process signal.
     void ChangeStatus(ProcessStatus new_status);
 
+    /// Allocates the main thread stack for the process, given the stack size in bytes.
+    void AllocateMainThreadStack(u64 stack_size);
+
     /// Memory manager for this process.
     Kernel::VMManager vm_manager;
 
@@ -298,12 +322,16 @@ private:
     /// Title ID corresponding to the process
     u64 program_id = 0;
 
+    /// Specifies additional memory to be reserved for the process's memory management by the
+    /// system. When this is non-zero, secure memory is allocated and used for page table allocation
+    /// instead of using the normal global page tables/memory block management.
+    u32 system_resource_size = 0;
+
     /// Resource limit descriptor for this process
     SharedPtr<ResourceLimit> resource_limit;
 
     /// The ideal CPU core for this process, threads are scheduled on this core by default.
     u8 ideal_core = 0;
-    u32 is_virtual_address_memory_enabled = 0;
 
     /// The Thread Local Storage area is allocated as processes create threads,
     /// each TLS area is 0x200 bytes, so one page (0x1000) is split up in 8 parts, and each part
@@ -338,6 +366,9 @@ private:
     /// variable related facilities.
     Mutex mutex;
 
+    /// Address indicating the location of the process' dedicated TLS region.
+    VAddr tls_region_address = 0;
+
     /// Random values for svcGetInfo RandomEntropy
     std::array<u64, RANDOM_ENTROPY_SIZE> random_entropy{};
 
diff --git a/src/core/hle/kernel/shared_memory.cpp b/src/core/hle/kernel/shared_memory.cpp
index f15c5ee36..a815c4eea 100644
--- a/src/core/hle/kernel/shared_memory.cpp
+++ b/src/core/hle/kernel/shared_memory.cpp
@@ -28,7 +28,7 @@ SharedPtr<SharedMemory> SharedMemory::Create(KernelCore& kernel, Process* owner_
     shared_memory->other_permissions = other_permissions;
 
     if (address == 0) {
-        shared_memory->backing_block = std::make_shared<std::vector<u8>>(size);
+        shared_memory->backing_block = std::make_shared<Kernel::PhysicalMemory>(size);
         shared_memory->backing_block_offset = 0;
 
         // Refresh the address mappings for the current process.
@@ -59,8 +59,8 @@ SharedPtr<SharedMemory> SharedMemory::Create(KernelCore& kernel, Process* owner_
 }
 
 SharedPtr<SharedMemory> SharedMemory::CreateForApplet(
-    KernelCore& kernel, std::shared_ptr<std::vector<u8>> heap_block, std::size_t offset, u64 size,
-    MemoryPermission permissions, MemoryPermission other_permissions, std::string name) {
+    KernelCore& kernel, std::shared_ptr<Kernel::PhysicalMemory> heap_block, std::size_t offset,
+    u64 size, MemoryPermission permissions, MemoryPermission other_permissions, std::string name) {
     SharedPtr<SharedMemory> shared_memory(new SharedMemory(kernel));
 
     shared_memory->owner_process = nullptr;
diff --git a/src/core/hle/kernel/shared_memory.h b/src/core/hle/kernel/shared_memory.h
index c2b6155e1..01ca6dcd2 100644
--- a/src/core/hle/kernel/shared_memory.h
+++ b/src/core/hle/kernel/shared_memory.h
@@ -10,6 +10,7 @@
 
 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"
+#include "core/hle/kernel/physical_memory.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/result.h"
 
@@ -62,12 +63,10 @@ public:
      * block.
      * @param name Optional object name, used for debugging purposes.
      */
-    static SharedPtr<SharedMemory> CreateForApplet(KernelCore& kernel,
-                                                   std::shared_ptr<std::vector<u8>> heap_block,
-                                                   std::size_t offset, u64 size,
-                                                   MemoryPermission permissions,
-                                                   MemoryPermission other_permissions,
-                                                   std::string name = "Unknown Applet");
+    static SharedPtr<SharedMemory> CreateForApplet(
+        KernelCore& kernel, std::shared_ptr<Kernel::PhysicalMemory> heap_block, std::size_t offset,
+        u64 size, MemoryPermission permissions, MemoryPermission other_permissions,
+        std::string name = "Unknown Applet");
 
     std::string GetTypeName() const override {
         return "SharedMemory";
@@ -135,7 +134,7 @@ private:
     ~SharedMemory() override;
 
     /// Backing memory for this shared memory block.
-    std::shared_ptr<std::vector<u8>> backing_block;
+    std::shared_ptr<PhysicalMemory> backing_block;
     /// Offset into the backing block for this shared memory.
     std::size_t backing_block_offset = 0;
     /// Size of the memory block. Page-aligned.
diff --git a/src/core/hle/kernel/svc.cpp b/src/core/hle/kernel/svc.cpp
index 332573a95..1fd1a732a 100644
--- a/src/core/hle/kernel/svc.cpp
+++ b/src/core/hle/kernel/svc.cpp
@@ -318,7 +318,14 @@ static ResultCode UnmapMemory(Core::System& system, VAddr dst_addr, VAddr src_ad
         return result;
     }
 
-    return vm_manager.UnmapRange(dst_addr, size);
+    const auto unmap_res = vm_manager.UnmapRange(dst_addr, size);
+
+    // Reprotect the source mapping on success
+    if (unmap_res.IsSuccess()) {
+        ASSERT(vm_manager.ReprotectRange(src_addr, size, VMAPermission::ReadWrite).IsSuccess());
+    }
+
+    return unmap_res;
 }
 
 /// Connect to an OS service given the port name, returns the handle to the port to out
@@ -729,16 +736,16 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
         StackRegionBaseAddr = 14,
         StackRegionSize = 15,
         // 3.0.0+
-        IsVirtualAddressMemoryEnabled = 16,
-        PersonalMmHeapUsage = 17,
+        SystemResourceSize = 16,
+        SystemResourceUsage = 17,
         TitleId = 18,
         // 4.0.0+
         PrivilegedProcessId = 19,
         // 5.0.0+
         UserExceptionContextAddr = 20,
         // 6.0.0+
-        TotalPhysicalMemoryAvailableWithoutMmHeap = 21,
-        TotalPhysicalMemoryUsedWithoutMmHeap = 22,
+        TotalPhysicalMemoryAvailableWithoutSystemResource = 21,
+        TotalPhysicalMemoryUsedWithoutSystemResource = 22,
     };
 
     const auto info_id_type = static_cast<GetInfoType>(info_id);
@@ -756,12 +763,12 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
     case GetInfoType::StackRegionSize:
     case GetInfoType::TotalPhysicalMemoryAvailable:
     case GetInfoType::TotalPhysicalMemoryUsed:
-    case GetInfoType::IsVirtualAddressMemoryEnabled:
-    case GetInfoType::PersonalMmHeapUsage:
+    case GetInfoType::SystemResourceSize:
+    case GetInfoType::SystemResourceUsage:
     case GetInfoType::TitleId:
     case GetInfoType::UserExceptionContextAddr:
-    case GetInfoType::TotalPhysicalMemoryAvailableWithoutMmHeap:
-    case GetInfoType::TotalPhysicalMemoryUsedWithoutMmHeap: {
+    case GetInfoType::TotalPhysicalMemoryAvailableWithoutSystemResource:
+    case GetInfoType::TotalPhysicalMemoryUsedWithoutSystemResource: {
         if (info_sub_id != 0) {
             return ERR_INVALID_ENUM_VALUE;
         }
@@ -822,8 +829,13 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
             *result = process->GetTotalPhysicalMemoryUsed();
             return RESULT_SUCCESS;
 
-        case GetInfoType::IsVirtualAddressMemoryEnabled:
-            *result = process->IsVirtualMemoryEnabled();
+        case GetInfoType::SystemResourceSize:
+            *result = process->GetSystemResourceSize();
+            return RESULT_SUCCESS;
+
+        case GetInfoType::SystemResourceUsage:
+            LOG_WARNING(Kernel_SVC, "(STUBBED) Attempted to query system resource usage");
+            *result = process->GetSystemResourceUsage();
             return RESULT_SUCCESS;
 
         case GetInfoType::TitleId:
@@ -831,17 +843,15 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
             return RESULT_SUCCESS;
 
         case GetInfoType::UserExceptionContextAddr:
-            LOG_WARNING(Kernel_SVC,
-                        "(STUBBED) Attempted to query user exception context address, returned 0");
-            *result = 0;
+            *result = process->GetTLSRegionAddress();
             return RESULT_SUCCESS;
 
-        case GetInfoType::TotalPhysicalMemoryAvailableWithoutMmHeap:
-            *result = process->GetTotalPhysicalMemoryAvailable();
+        case GetInfoType::TotalPhysicalMemoryAvailableWithoutSystemResource:
+            *result = process->GetTotalPhysicalMemoryAvailableWithoutSystemResource();
             return RESULT_SUCCESS;
 
-        case GetInfoType::TotalPhysicalMemoryUsedWithoutMmHeap:
-            *result = process->GetTotalPhysicalMemoryUsedWithoutMmHeap();
+        case GetInfoType::TotalPhysicalMemoryUsedWithoutSystemResource:
+            *result = process->GetTotalPhysicalMemoryUsedWithoutSystemResource();
             return RESULT_SUCCESS;
 
         default:
@@ -946,6 +956,86 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
     }
 }
 
+/// Maps memory at a desired address
+static ResultCode MapPhysicalMemory(Core::System& system, VAddr addr, u64 size) {
+    LOG_DEBUG(Kernel_SVC, "called, addr=0x{:016X}, size=0x{:X}", addr, size);
+
+    if (!Common::Is4KBAligned(addr)) {
+        LOG_ERROR(Kernel_SVC, "Address is not aligned to 4KB, 0x{:016X}", addr);
+        return ERR_INVALID_ADDRESS;
+    }
+
+    if (!Common::Is4KBAligned(size)) {
+        LOG_ERROR(Kernel_SVC, "Size is not aligned to 4KB, 0x{:X}", size);
+        return ERR_INVALID_SIZE;
+    }
+
+    if (size == 0) {
+        LOG_ERROR(Kernel_SVC, "Size is zero");
+        return ERR_INVALID_SIZE;
+    }
+
+    if (!(addr < addr + size)) {
+        LOG_ERROR(Kernel_SVC, "Size causes 64-bit overflow of address");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    Process* const current_process = system.Kernel().CurrentProcess();
+    auto& vm_manager = current_process->VMManager();
+
+    if (current_process->GetSystemResourceSize() == 0) {
+        LOG_ERROR(Kernel_SVC, "System Resource Size is zero");
+        return ERR_INVALID_STATE;
+    }
+
+    if (!vm_manager.IsWithinMapRegion(addr, size)) {
+        LOG_ERROR(Kernel_SVC, "Range not within map region");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    return vm_manager.MapPhysicalMemory(addr, size);
+}
+
+/// Unmaps memory previously mapped via MapPhysicalMemory
+static ResultCode UnmapPhysicalMemory(Core::System& system, VAddr addr, u64 size) {
+    LOG_DEBUG(Kernel_SVC, "called, addr=0x{:016X}, size=0x{:X}", addr, size);
+
+    if (!Common::Is4KBAligned(addr)) {
+        LOG_ERROR(Kernel_SVC, "Address is not aligned to 4KB, 0x{:016X}", addr);
+        return ERR_INVALID_ADDRESS;
+    }
+
+    if (!Common::Is4KBAligned(size)) {
+        LOG_ERROR(Kernel_SVC, "Size is not aligned to 4KB, 0x{:X}", size);
+        return ERR_INVALID_SIZE;
+    }
+
+    if (size == 0) {
+        LOG_ERROR(Kernel_SVC, "Size is zero");
+        return ERR_INVALID_SIZE;
+    }
+
+    if (!(addr < addr + size)) {
+        LOG_ERROR(Kernel_SVC, "Size causes 64-bit overflow of address");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    Process* const current_process = system.Kernel().CurrentProcess();
+    auto& vm_manager = current_process->VMManager();
+
+    if (current_process->GetSystemResourceSize() == 0) {
+        LOG_ERROR(Kernel_SVC, "System Resource Size is zero");
+        return ERR_INVALID_STATE;
+    }
+
+    if (!vm_manager.IsWithinMapRegion(addr, size)) {
+        LOG_ERROR(Kernel_SVC, "Range not within map region");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    return vm_manager.UnmapPhysicalMemory(addr, size);
+}
+
 /// Sets the thread activity
 static ResultCode SetThreadActivity(Core::System& system, Handle handle, u32 activity) {
     LOG_DEBUG(Kernel_SVC, "called, handle=0x{:08X}, activity=0x{:08X}", handle, activity);
@@ -1647,8 +1737,8 @@ static ResultCode SignalProcessWideKey(Core::System& system, VAddr condition_var
 // Wait for an address (via Address Arbiter)
 static ResultCode WaitForAddress(Core::System& system, VAddr address, u32 type, s32 value,
                                  s64 timeout) {
-    LOG_WARNING(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, timeout={}",
-                address, type, value, timeout);
+    LOG_TRACE(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, timeout={}", address,
+              type, value, timeout);
 
     // If the passed address is a kernel virtual address, return invalid memory state.
     if (Memory::IsKernelVirtualAddress(address)) {
@@ -1670,8 +1760,8 @@ static ResultCode WaitForAddress(Core::System& system, VAddr address, u32 type,
 // Signals to an address (via Address Arbiter)
 static ResultCode SignalToAddress(Core::System& system, VAddr address, u32 type, s32 value,
                                   s32 num_to_wake) {
-    LOG_WARNING(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, num_to_wake=0x{:X}",
-                address, type, value, num_to_wake);
+    LOG_TRACE(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, num_to_wake=0x{:X}",
+              address, type, value, num_to_wake);
 
     // If the passed address is a kernel virtual address, return invalid memory state.
     if (Memory::IsKernelVirtualAddress(address)) {
@@ -2303,8 +2393,8 @@ static const FunctionDef SVC_Table[] = {
     {0x29, SvcWrap<GetInfo>, "GetInfo"},
     {0x2A, nullptr, "FlushEntireDataCache"},
     {0x2B, nullptr, "FlushDataCache"},
-    {0x2C, nullptr, "MapPhysicalMemory"},
-    {0x2D, nullptr, "UnmapPhysicalMemory"},
+    {0x2C, SvcWrap<MapPhysicalMemory>, "MapPhysicalMemory"},
+    {0x2D, SvcWrap<UnmapPhysicalMemory>, "UnmapPhysicalMemory"},
     {0x2E, nullptr, "GetFutureThreadInfo"},
     {0x2F, nullptr, "GetLastThreadInfo"},
     {0x30, SvcWrap<GetResourceLimitLimitValue>, "GetResourceLimitLimitValue"},
diff --git a/src/core/hle/kernel/svc_wrap.h b/src/core/hle/kernel/svc_wrap.h
index 865473c6f..c2d8d0dc3 100644
--- a/src/core/hle/kernel/svc_wrap.h
+++ b/src/core/hle/kernel/svc_wrap.h
@@ -32,6 +32,11 @@ void SvcWrap(Core::System& system) {
     FuncReturn(system, func(system, Param(system, 0)).raw);
 }
 
+template <ResultCode func(Core::System&, u64, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1)).raw);
+}
+
 template <ResultCode func(Core::System&, u32)>
 void SvcWrap(Core::System& system) {
     FuncReturn(system, func(system, static_cast<u32>(Param(system, 0))).raw);
diff --git a/src/core/hle/kernel/transfer_memory.cpp b/src/core/hle/kernel/transfer_memory.cpp
index 26c4e5e67..1113c815e 100644
--- a/src/core/hle/kernel/transfer_memory.cpp
+++ b/src/core/hle/kernel/transfer_memory.cpp
@@ -47,7 +47,7 @@ ResultCode TransferMemory::MapMemory(VAddr address, u64 size, MemoryPermission p
         return ERR_INVALID_STATE;
     }
 
-    backing_block = std::make_shared<std::vector<u8>>(size);
+    backing_block = std::make_shared<PhysicalMemory>(size);
 
     const auto map_state = owner_permissions == MemoryPermission::None
                                ? MemoryState::TransferMemoryIsolated
diff --git a/src/core/hle/kernel/transfer_memory.h b/src/core/hle/kernel/transfer_memory.h
index a140b1e2b..6be9dc094 100644
--- a/src/core/hle/kernel/transfer_memory.h
+++ b/src/core/hle/kernel/transfer_memory.h
@@ -8,6 +8,7 @@
 #include <vector>
 
 #include "core/hle/kernel/object.h"
+#include "core/hle/kernel/physical_memory.h"
 
 union ResultCode;
 
@@ -82,7 +83,7 @@ private:
     ~TransferMemory() override;
 
     /// Memory block backing this instance.
-    std::shared_ptr<std::vector<u8>> backing_block;
+    std::shared_ptr<PhysicalMemory> backing_block;
 
     /// The base address for the memory managed by this instance.
     VAddr base_address = 0;
diff --git a/src/core/hle/kernel/vm_manager.cpp b/src/core/hle/kernel/vm_manager.cpp
index 501544090..40cea1e7c 100644
--- a/src/core/hle/kernel/vm_manager.cpp
+++ b/src/core/hle/kernel/vm_manager.cpp
@@ -5,13 +5,15 @@
 #include <algorithm>
 #include <iterator>
 #include <utility>
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/memory_hook.h"
-#include "core/arm/arm_interface.h"
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
 #include "core/hle/kernel/errors.h"
+#include "core/hle/kernel/process.h"
+#include "core/hle/kernel/resource_limit.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
 #include "core/memory_setup.h"
@@ -49,10 +51,14 @@ bool VirtualMemoryArea::CanBeMergedWith(const VirtualMemoryArea& next) const {
         type != next.type) {
         return false;
     }
-    if (type == VMAType::AllocatedMemoryBlock &&
-        (backing_block != next.backing_block || offset + size != next.offset)) {
+    if ((attribute & MemoryAttribute::DeviceMapped) == MemoryAttribute::DeviceMapped) {
+        // TODO: Can device mapped memory be merged sanely?
+        // Not merging it may cause inaccuracies versus hardware when memory layout is queried.
         return false;
     }
+    if (type == VMAType::AllocatedMemoryBlock) {
+        return true;
+    }
     if (type == VMAType::BackingMemory && backing_memory + size != next.backing_memory) {
         return false;
     }
@@ -98,9 +104,9 @@ bool VMManager::IsValidHandle(VMAHandle handle) const {
 }
 
 ResultVal<VMManager::VMAHandle> VMManager::MapMemoryBlock(VAddr target,
-                                                          std::shared_ptr<std::vector<u8>> block,
+                                                          std::shared_ptr<PhysicalMemory> block,
                                                           std::size_t offset, u64 size,
-                                                          MemoryState state) {
+                                                          MemoryState state, VMAPermission perm) {
     ASSERT(block != nullptr);
     ASSERT(offset + size <= block->size());
 
@@ -109,17 +115,8 @@ ResultVal<VMManager::VMAHandle> VMManager::MapMemoryBlock(VAddr target,
     VirtualMemoryArea& final_vma = vma_handle->second;
     ASSERT(final_vma.size == size);
 
-    system.ArmInterface(0).MapBackingMemory(target, size, block->data() + offset,
-                                            VMAPermission::ReadWriteExecute);
-    system.ArmInterface(1).MapBackingMemory(target, size, block->data() + offset,
-                                            VMAPermission::ReadWriteExecute);
-    system.ArmInterface(2).MapBackingMemory(target, size, block->data() + offset,
-                                            VMAPermission::ReadWriteExecute);
-    system.ArmInterface(3).MapBackingMemory(target, size, block->data() + offset,
-                                            VMAPermission::ReadWriteExecute);
-
     final_vma.type = VMAType::AllocatedMemoryBlock;
-    final_vma.permissions = VMAPermission::ReadWrite;
+    final_vma.permissions = perm;
     final_vma.state = state;
     final_vma.backing_block = std::move(block);
     final_vma.offset = offset;
@@ -137,11 +134,6 @@ ResultVal<VMManager::VMAHandle> VMManager::MapBackingMemory(VAddr target, u8* me
     VirtualMemoryArea& final_vma = vma_handle->second;
     ASSERT(final_vma.size == size);
 
-    system.ArmInterface(0).MapBackingMemory(target, size, memory, VMAPermission::ReadWriteExecute);
-    system.ArmInterface(1).MapBackingMemory(target, size, memory, VMAPermission::ReadWriteExecute);
-    system.ArmInterface(2).MapBackingMemory(target, size, memory, VMAPermission::ReadWriteExecute);
-    system.ArmInterface(3).MapBackingMemory(target, size, memory, VMAPermission::ReadWriteExecute);
-
     final_vma.type = VMAType::BackingMemory;
     final_vma.permissions = VMAPermission::ReadWrite;
     final_vma.state = state;
@@ -230,11 +222,6 @@ ResultCode VMManager::UnmapRange(VAddr target, u64 size) {
 
     ASSERT(FindVMA(target)->second.size >= size);
 
-    system.ArmInterface(0).UnmapMemory(target, size);
-    system.ArmInterface(1).UnmapMemory(target, size);
-    system.ArmInterface(2).UnmapMemory(target, size);
-    system.ArmInterface(3).UnmapMemory(target, size);
-
     return RESULT_SUCCESS;
 }
 
@@ -274,7 +261,7 @@ ResultVal<VAddr> VMManager::SetHeapSize(u64 size) {
 
     if (heap_memory == nullptr) {
         // Initialize heap
-        heap_memory = std::make_shared<std::vector<u8>>(size);
+        heap_memory = std::make_shared<PhysicalMemory>(size);
         heap_end = heap_region_base + size;
     } else {
         UnmapRange(heap_region_base, GetCurrentHeapSize());
@@ -308,6 +295,166 @@ ResultVal<VAddr> VMManager::SetHeapSize(u64 size) {
     return MakeResult<VAddr>(heap_region_base);
 }
 
+ResultCode VMManager::MapPhysicalMemory(VAddr target, u64 size) {
+    const auto end_addr = target + size;
+    const auto last_addr = end_addr - 1;
+    VAddr cur_addr = target;
+
+    ResultCode result = RESULT_SUCCESS;
+
+    // Check how much memory we've already mapped.
+    const auto mapped_size_result = SizeOfAllocatedVMAsInRange(target, size);
+    if (mapped_size_result.Failed()) {
+        return mapped_size_result.Code();
+    }
+
+    // If we've already mapped the desired amount, return early.
+    const std::size_t mapped_size = *mapped_size_result;
+    if (mapped_size == size) {
+        return RESULT_SUCCESS;
+    }
+
+    // Check that we can map the memory we want.
+    const auto res_limit = system.CurrentProcess()->GetResourceLimit();
+    const u64 physmem_remaining = res_limit->GetMaxResourceValue(ResourceType::PhysicalMemory) -
+                                  res_limit->GetCurrentResourceValue(ResourceType::PhysicalMemory);
+    if (physmem_remaining < (size - mapped_size)) {
+        return ERR_RESOURCE_LIMIT_EXCEEDED;
+    }
+
+    // Keep track of the memory regions we unmap.
+    std::vector<std::pair<u64, u64>> mapped_regions;
+
+    // Iterate, trying to map memory.
+    {
+        cur_addr = target;
+
+        auto iter = FindVMA(target);
+        ASSERT_MSG(iter != vma_map.end(), "MapPhysicalMemory iter != end");
+
+        while (true) {
+            const auto& vma = iter->second;
+            const auto vma_start = vma.base;
+            const auto vma_end = vma_start + vma.size;
+            const auto vma_last = vma_end - 1;
+
+            // Map the memory block
+            const auto map_size = std::min(end_addr - cur_addr, vma_end - cur_addr);
+            if (vma.state == MemoryState::Unmapped) {
+                const auto map_res =
+                    MapMemoryBlock(cur_addr, std::make_shared<PhysicalMemory>(map_size, 0), 0,
+                                   map_size, MemoryState::Heap, VMAPermission::ReadWrite);
+                result = map_res.Code();
+                if (result.IsError()) {
+                    break;
+                }
+
+                mapped_regions.emplace_back(cur_addr, map_size);
+            }
+
+            // Break once we hit the end of the range.
+            if (last_addr <= vma_last) {
+                break;
+            }
+
+            // Advance to the next block.
+            cur_addr = vma_end;
+            iter = FindVMA(cur_addr);
+            ASSERT_MSG(iter != vma_map.end(), "MapPhysicalMemory iter != end");
+        }
+    }
+
+    // If we failed, unmap memory.
+    if (result.IsError()) {
+        for (const auto [unmap_address, unmap_size] : mapped_regions) {
+            ASSERT_MSG(UnmapRange(unmap_address, unmap_size).IsSuccess(),
+                       "MapPhysicalMemory un-map on error");
+        }
+
+        return result;
+    }
+
+    // Update amount of mapped physical memory.
+    physical_memory_mapped += size - mapped_size;
+
+    return RESULT_SUCCESS;
+}
+
+ResultCode VMManager::UnmapPhysicalMemory(VAddr target, u64 size) {
+    const auto end_addr = target + size;
+    const auto last_addr = end_addr - 1;
+    VAddr cur_addr = target;
+
+    ResultCode result = RESULT_SUCCESS;
+
+    // Check how much memory is currently mapped.
+    const auto mapped_size_result = SizeOfUnmappablePhysicalMemoryInRange(target, size);
+    if (mapped_size_result.Failed()) {
+        return mapped_size_result.Code();
+    }
+
+    // If we've already unmapped all the memory, return early.
+    const std::size_t mapped_size = *mapped_size_result;
+    if (mapped_size == 0) {
+        return RESULT_SUCCESS;
+    }
+
+    // Keep track of the memory regions we unmap.
+    std::vector<std::pair<u64, u64>> unmapped_regions;
+
+    // Try to unmap regions.
+    {
+        cur_addr = target;
+
+        auto iter = FindVMA(target);
+        ASSERT_MSG(iter != vma_map.end(), "UnmapPhysicalMemory iter != end");
+
+        while (true) {
+            const auto& vma = iter->second;
+            const auto vma_start = vma.base;
+            const auto vma_end = vma_start + vma.size;
+            const auto vma_last = vma_end - 1;
+
+            // Unmap the memory block
+            const auto unmap_size = std::min(end_addr - cur_addr, vma_end - cur_addr);
+            if (vma.state == MemoryState::Heap) {
+                result = UnmapRange(cur_addr, unmap_size);
+                if (result.IsError()) {
+                    break;
+                }
+
+                unmapped_regions.emplace_back(cur_addr, unmap_size);
+            }
+
+            // Break once we hit the end of the range.
+            if (last_addr <= vma_last) {
+                break;
+            }
+
+            // Advance to the next block.
+            cur_addr = vma_end;
+            iter = FindVMA(cur_addr);
+            ASSERT_MSG(iter != vma_map.end(), "UnmapPhysicalMemory iter != end");
+        }
+    }
+
+    // If we failed, re-map regions.
+    // TODO: Preserve memory contents?
+    if (result.IsError()) {
+        for (const auto [map_address, map_size] : unmapped_regions) {
+            const auto remap_res =
+                MapMemoryBlock(map_address, std::make_shared<PhysicalMemory>(map_size, 0), 0,
+                               map_size, MemoryState::Heap, VMAPermission::None);
+            ASSERT_MSG(remap_res.Succeeded(), "UnmapPhysicalMemory re-map on error");
+        }
+    }
+
+    // Update mapped amount
+    physical_memory_mapped -= mapped_size;
+
+    return RESULT_SUCCESS;
+}
+
 ResultCode VMManager::MapCodeMemory(VAddr dst_address, VAddr src_address, u64 size) {
     constexpr auto ignore_attribute = MemoryAttribute::LockedForIPC | MemoryAttribute::DeviceMapped;
     const auto src_check_result = CheckRangeState(
@@ -447,7 +594,7 @@ ResultCode VMManager::MirrorMemory(VAddr dst_addr, VAddr src_addr, u64 size, Mem
     ASSERT_MSG(vma_offset + size <= vma->second.size,
                "Shared memory exceeds bounds of mapped block");
 
-    const std::shared_ptr<std::vector<u8>>& backing_block = vma->second.backing_block;
+    const std::shared_ptr<PhysicalMemory>& backing_block = vma->second.backing_block;
     const std::size_t backing_block_offset = vma->second.offset + vma_offset;
 
     CASCADE_RESULT(auto new_vma,
@@ -455,12 +602,12 @@ ResultCode VMManager::MirrorMemory(VAddr dst_addr, VAddr src_addr, u64 size, Mem
     // Protect mirror with permissions from old region
     Reprotect(new_vma, vma->second.permissions);
     // Remove permissions from old region
-    Reprotect(vma, VMAPermission::None);
+    ReprotectRange(src_addr, size, VMAPermission::None);
 
     return RESULT_SUCCESS;
 }
 
-void VMManager::RefreshMemoryBlockMappings(const std::vector<u8>* block) {
+void VMManager::RefreshMemoryBlockMappings(const PhysicalMemory* block) {
     // If this ever proves to have a noticeable performance impact, allow users of the function to
     // specify a specific range of addresses to limit the scan to.
     for (const auto& p : vma_map) {
@@ -588,14 +735,14 @@ VMManager::VMAIter VMManager::SplitVMA(VMAIter vma_handle, u64 offset_in_vma) {
 VMManager::VMAIter VMManager::MergeAdjacent(VMAIter iter) {
     const VMAIter next_vma = std::next(iter);
     if (next_vma != vma_map.end() && iter->second.CanBeMergedWith(next_vma->second)) {
-        iter->second.size += next_vma->second.size;
+        MergeAdjacentVMA(iter->second, next_vma->second);
         vma_map.erase(next_vma);
     }
 
     if (iter != vma_map.begin()) {
         VMAIter prev_vma = std::prev(iter);
         if (prev_vma->second.CanBeMergedWith(iter->second)) {
-            prev_vma->second.size += iter->second.size;
+            MergeAdjacentVMA(prev_vma->second, iter->second);
             vma_map.erase(iter);
             iter = prev_vma;
         }
@@ -604,6 +751,38 @@ VMManager::VMAIter VMManager::MergeAdjacent(VMAIter iter) {
     return iter;
 }
 
+void VMManager::MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryArea& right) {
+    ASSERT(left.CanBeMergedWith(right));
+
+    // Always merge allocated memory blocks, even when they don't share the same backing block.
+    if (left.type == VMAType::AllocatedMemoryBlock &&
+        (left.backing_block != right.backing_block || left.offset + left.size != right.offset)) {
+        // Check if we can save work.
+        if (left.offset == 0 && left.size == left.backing_block->size()) {
+            // Fast case: left is an entire backing block.
+            left.backing_block->insert(left.backing_block->end(),
+                                       right.backing_block->begin() + right.offset,
+                                       right.backing_block->begin() + right.offset + right.size);
+        } else {
+            // Slow case: make a new memory block for left and right.
+            auto new_memory = std::make_shared<PhysicalMemory>();
+            new_memory->insert(new_memory->end(), left.backing_block->begin() + left.offset,
+                               left.backing_block->begin() + left.offset + left.size);
+            new_memory->insert(new_memory->end(), right.backing_block->begin() + right.offset,
+                               right.backing_block->begin() + right.offset + right.size);
+            left.backing_block = new_memory;
+            left.offset = 0;
+        }
+
+        // Page table update is needed, because backing memory changed.
+        left.size += right.size;
+        UpdatePageTableForVMA(left);
+    } else {
+        // Just update the size.
+        left.size += right.size;
+    }
+}
+
 void VMManager::UpdatePageTableForVMA(const VirtualMemoryArea& vma) {
     switch (vma.type) {
     case VMAType::Free:
@@ -778,6 +957,84 @@ VMManager::CheckResults VMManager::CheckRangeState(VAddr address, u64 size, Memo
         std::make_tuple(initial_state, initial_permissions, initial_attributes & ~ignore_mask));
 }
 
+ResultVal<std::size_t> VMManager::SizeOfAllocatedVMAsInRange(VAddr address,
+                                                             std::size_t size) const {
+    const VAddr end_addr = address + size;
+    const VAddr last_addr = end_addr - 1;
+    std::size_t mapped_size = 0;
+
+    VAddr cur_addr = address;
+    auto iter = FindVMA(cur_addr);
+    ASSERT_MSG(iter != vma_map.end(), "SizeOfAllocatedVMAsInRange iter != end");
+
+    while (true) {
+        const auto& vma = iter->second;
+        const VAddr vma_start = vma.base;
+        const VAddr vma_end = vma_start + vma.size;
+        const VAddr vma_last = vma_end - 1;
+
+        // Add size if relevant.
+        if (vma.state != MemoryState::Unmapped) {
+            mapped_size += std::min(end_addr - cur_addr, vma_end - cur_addr);
+        }
+
+        // Break once we hit the end of the range.
+        if (last_addr <= vma_last) {
+            break;
+        }
+
+        // Advance to the next block.
+        cur_addr = vma_end;
+        iter = std::next(iter);
+        ASSERT_MSG(iter != vma_map.end(), "SizeOfAllocatedVMAsInRange iter != end");
+    }
+
+    return MakeResult(mapped_size);
+}
+
+ResultVal<std::size_t> VMManager::SizeOfUnmappablePhysicalMemoryInRange(VAddr address,
+                                                                        std::size_t size) const {
+    const VAddr end_addr = address + size;
+    const VAddr last_addr = end_addr - 1;
+    std::size_t mapped_size = 0;
+
+    VAddr cur_addr = address;
+    auto iter = FindVMA(cur_addr);
+    ASSERT_MSG(iter != vma_map.end(), "SizeOfUnmappablePhysicalMemoryInRange iter != end");
+
+    while (true) {
+        const auto& vma = iter->second;
+        const auto vma_start = vma.base;
+        const auto vma_end = vma_start + vma.size;
+        const auto vma_last = vma_end - 1;
+        const auto state = vma.state;
+        const auto attr = vma.attribute;
+
+        // Memory within region must be free or mapped heap.
+        if (!((state == MemoryState::Heap && attr == MemoryAttribute::None) ||
+              (state == MemoryState::Unmapped))) {
+            return ERR_INVALID_ADDRESS_STATE;
+        }
+
+        // Add size if relevant.
+        if (state != MemoryState::Unmapped) {
+            mapped_size += std::min(end_addr - cur_addr, vma_end - cur_addr);
+        }
+
+        // Break once we hit the end of the range.
+        if (last_addr <= vma_last) {
+            break;
+        }
+
+        // Advance to the next block.
+        cur_addr = vma_end;
+        iter = std::next(iter);
+        ASSERT_MSG(iter != vma_map.end(), "SizeOfUnmappablePhysicalMemoryInRange iter != end");
+    }
+
+    return MakeResult(mapped_size);
+}
+
 u64 VMManager::GetTotalPhysicalMemoryAvailable() const {
     LOG_WARNING(Kernel, "(STUBBED) called");
     return 0xF8000000;
diff --git a/src/core/hle/kernel/vm_manager.h b/src/core/hle/kernel/vm_manager.h
index 9fe6ac3f4..b18cde619 100644
--- a/src/core/hle/kernel/vm_manager.h
+++ b/src/core/hle/kernel/vm_manager.h
@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "common/memory_hook.h"
 #include "common/page_table.h"
+#include "core/hle/kernel/physical_memory.h"
 #include "core/hle/result.h"
 #include "core/memory.h"
 
@@ -290,7 +291,7 @@ struct VirtualMemoryArea {
 
     // Settings for type = AllocatedMemoryBlock
     /// Memory block backing this VMA.
-    std::shared_ptr<std::vector<u8>> backing_block = nullptr;
+    std::shared_ptr<PhysicalMemory> backing_block = nullptr;
     /// Offset into the backing_memory the mapping starts from.
     std::size_t offset = 0;
 
@@ -348,8 +349,9 @@ public:
      * @param size Size of the mapping.
      * @param state MemoryState tag to attach to the VMA.
      */
-    ResultVal<VMAHandle> MapMemoryBlock(VAddr target, std::shared_ptr<std::vector<u8>> block,
-                                        std::size_t offset, u64 size, MemoryState state);
+    ResultVal<VMAHandle> MapMemoryBlock(VAddr target, std::shared_ptr<PhysicalMemory> block,
+                                        std::size_t offset, u64 size, MemoryState state,
+                                        VMAPermission perm = VMAPermission::ReadWrite);
 
     /**
      * Maps an unmanaged host memory pointer at a given address.
@@ -450,6 +452,34 @@ public:
     ///
     ResultVal<VAddr> SetHeapSize(u64 size);
 
+    /// Maps memory at a given address.
+    ///
+    /// @param addr The virtual address to map memory at.
+    /// @param size The amount of memory to map.
+    ///
+    /// @note The destination address must lie within the Map region.
+    ///
+    /// @note This function requires that SystemResourceSize be non-zero,
+    ///       however, this is just because if it were not then the
+    ///       resulting page tables could be exploited on hardware by
+    ///       a malicious program. SystemResource usage does not need
+    ///       to be explicitly checked or updated here.
+    ResultCode MapPhysicalMemory(VAddr target, u64 size);
+
+    /// Unmaps memory at a given address.
+    ///
+    /// @param addr The virtual address to unmap memory at.
+    /// @param size The amount of memory to unmap.
+    ///
+    /// @note The destination address must lie within the Map region.
+    ///
+    /// @note This function requires that SystemResourceSize be non-zero,
+    ///       however, this is just because if it were not then the
+    ///       resulting page tables could be exploited on hardware by
+    ///       a malicious program. SystemResource usage does not need
+    ///       to be explicitly checked or updated here.
+    ResultCode UnmapPhysicalMemory(VAddr target, u64 size);
+
     /// Maps a region of memory as code memory.
     ///
     /// @param dst_address The base address of the region to create the aliasing memory region.
@@ -518,7 +548,7 @@ public:
      * Scans all VMAs and updates the page table range of any that use the given vector as backing
      * memory. This should be called after any operation that causes reallocation of the vector.
      */
-    void RefreshMemoryBlockMappings(const std::vector<u8>* block);
+    void RefreshMemoryBlockMappings(const PhysicalMemory* block);
 
     /// Dumps the address space layout to the log, for debugging
     void LogLayout() const;
@@ -657,6 +687,11 @@ private:
      */
     VMAIter MergeAdjacent(VMAIter vma);
 
+    /**
+     * Merges two adjacent VMAs.
+     */
+    void MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryArea& right);
+
     /// Updates the pages corresponding to this VMA so they match the VMA's attributes.
     void UpdatePageTableForVMA(const VirtualMemoryArea& vma);
 
@@ -701,6 +736,13 @@ private:
                                  MemoryAttribute attribute_mask, MemoryAttribute attribute,
                                  MemoryAttribute ignore_mask) const;
 
+    /// Gets the amount of memory currently mapped (state != Unmapped) in a range.
+    ResultVal<std::size_t> SizeOfAllocatedVMAsInRange(VAddr address, std::size_t size) const;
+
+    /// Gets the amount of memory unmappable by UnmapPhysicalMemory in a range.
+    ResultVal<std::size_t> SizeOfUnmappablePhysicalMemoryInRange(VAddr address,
+                                                                 std::size_t size) const;
+
     /**
      * A map covering the entirety of the managed address space, keyed by the `base` field of each
      * VMA. It must always be modified by splitting or merging VMAs, so that the invariant
@@ -736,12 +778,17 @@ private:
     // the entire virtual address space extents that bound the allocations, including any holes.
     // This makes deallocation and reallocation of holes fast and keeps process memory contiguous
     // in the emulator address space, allowing Memory::GetPointer to be reasonably safe.
-    std::shared_ptr<std::vector<u8>> heap_memory;
+    std::shared_ptr<PhysicalMemory> heap_memory;
 
     // The end of the currently allocated heap. This is not an inclusive
     // end of the range. This is essentially 'base_address + current_size'.
     VAddr heap_end = 0;
 
+    // The current amount of memory mapped via MapPhysicalMemory.
+    // This is used here (and in Nintendo's kernel) only for debugging, and does not impact
+    // any behavior.
+    u64 physical_memory_mapped = 0;
+
     Core::System& system;
 };
 } // namespace Kernel
diff --git a/src/core/hle/service/am/am.cpp b/src/core/hle/service/am/am.cpp
index 9fdcf2965..a192a1f5f 100644
--- a/src/core/hle/service/am/am.cpp
+++ b/src/core/hle/service/am/am.cpp
@@ -266,8 +266,8 @@ ISelfController::ISelfController(std::shared_ptr<NVFlinger::NVFlinger> nvflinger
         {65, nullptr, "ReportUserIsActive"},
         {66, nullptr, "GetCurrentIlluminance"},
         {67, nullptr, "IsIlluminanceAvailable"},
-        {68, nullptr, "SetAutoSleepDisabled"},
-        {69, nullptr, "IsAutoSleepDisabled"},
+        {68, &ISelfController::SetAutoSleepDisabled, "SetAutoSleepDisabled"},
+        {69, &ISelfController::IsAutoSleepDisabled, "IsAutoSleepDisabled"},
         {70, nullptr, "ReportMultimediaError"},
         {71, nullptr, "GetCurrentIlluminanceEx"},
         {80, nullptr, "SetWirelessPriorityMode"},
@@ -454,6 +454,34 @@ void ISelfController::GetIdleTimeDetectionExtension(Kernel::HLERequestContext& c
     rb.Push<u32>(idle_time_detection_extension);
 }
 
+void ISelfController::SetAutoSleepDisabled(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    is_auto_sleep_disabled = rp.Pop<bool>();
+
+    // On the system itself, if the previous state of is_auto_sleep_disabled
+    // differed from the current value passed in, it'd signify the internal
+    // window manager to update (and also increment some statistics like update counts)
+    //
+    // It'd also indicate this change to an idle handling context.
+    //
+    // However, given we're emulating this behavior, most of this can be ignored
+    // and it's sufficient to simply set the member variable for querying via
+    // IsAutoSleepDisabled().
+
+    LOG_DEBUG(Service_AM, "called. is_auto_sleep_disabled={}", is_auto_sleep_disabled);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void ISelfController::IsAutoSleepDisabled(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_AM, "called.");
+
+    IPC::ResponseBuilder rb{ctx, 3};
+    rb.Push(RESULT_SUCCESS);
+    rb.Push(is_auto_sleep_disabled);
+}
+
 void ISelfController::GetAccumulatedSuspendedTickValue(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_AM, "called.");
 
diff --git a/src/core/hle/service/am/am.h b/src/core/hle/service/am/am.h
index 14b010164..6cb582483 100644
--- a/src/core/hle/service/am/am.h
+++ b/src/core/hle/service/am/am.h
@@ -133,6 +133,8 @@ private:
     void SetHandlesRequestToDisplay(Kernel::HLERequestContext& ctx);
     void SetIdleTimeDetectionExtension(Kernel::HLERequestContext& ctx);
     void GetIdleTimeDetectionExtension(Kernel::HLERequestContext& ctx);
+    void SetAutoSleepDisabled(Kernel::HLERequestContext& ctx);
+    void IsAutoSleepDisabled(Kernel::HLERequestContext& ctx);
     void GetAccumulatedSuspendedTickValue(Kernel::HLERequestContext& ctx);
     void GetAccumulatedSuspendedTickChangedEvent(Kernel::HLERequestContext& ctx);
 
@@ -142,6 +144,7 @@ private:
 
     u32 idle_time_detection_extension = 0;
     u64 num_fatal_sections_entered = 0;
+    bool is_auto_sleep_disabled = false;
 };
 
 class ICommonStateGetter final : public ServiceFramework<ICommonStateGetter> {
diff --git a/src/core/hle/service/audio/audio.cpp b/src/core/hle/service/audio/audio.cpp
index 128df7db5..1781bec83 100644
--- a/src/core/hle/service/audio/audio.cpp
+++ b/src/core/hle/service/audio/audio.cpp
@@ -19,16 +19,16 @@
 
 namespace Service::Audio {
 
-void InstallInterfaces(SM::ServiceManager& service_manager) {
+void InstallInterfaces(SM::ServiceManager& service_manager, Core::System& system) {
     std::make_shared<AudCtl>()->InstallAsService(service_manager);
     std::make_shared<AudOutA>()->InstallAsService(service_manager);
-    std::make_shared<AudOutU>()->InstallAsService(service_manager);
+    std::make_shared<AudOutU>(system)->InstallAsService(service_manager);
     std::make_shared<AudInA>()->InstallAsService(service_manager);
     std::make_shared<AudInU>()->InstallAsService(service_manager);
     std::make_shared<AudRecA>()->InstallAsService(service_manager);
     std::make_shared<AudRecU>()->InstallAsService(service_manager);
     std::make_shared<AudRenA>()->InstallAsService(service_manager);
-    std::make_shared<AudRenU>()->InstallAsService(service_manager);
+    std::make_shared<AudRenU>(system)->InstallAsService(service_manager);
     std::make_shared<CodecCtl>()->InstallAsService(service_manager);
     std::make_shared<HwOpus>()->InstallAsService(service_manager);
 
diff --git a/src/core/hle/service/audio/audio.h b/src/core/hle/service/audio/audio.h
index f5bd3bf5f..b6d13912e 100644
--- a/src/core/hle/service/audio/audio.h
+++ b/src/core/hle/service/audio/audio.h
@@ -4,6 +4,10 @@
 
 #pragma once
 
+namespace Core {
+class System;
+}
+
 namespace Service::SM {
 class ServiceManager;
 }
@@ -11,6 +15,6 @@ class ServiceManager;
 namespace Service::Audio {
 
 /// Registers all Audio services with the specified service manager.
-void InstallInterfaces(SM::ServiceManager& service_manager);
+void InstallInterfaces(SM::ServiceManager& service_manager, Core::System& system);
 
 } // namespace Service::Audio
diff --git a/src/core/hle/service/audio/audout_u.cpp b/src/core/hle/service/audio/audout_u.cpp
index 7db6eb08d..fb84a8f13 100644
--- a/src/core/hle/service/audio/audout_u.cpp
+++ b/src/core/hle/service/audio/audout_u.cpp
@@ -40,8 +40,8 @@ enum class AudioState : u32 {
 
 class IAudioOut final : public ServiceFramework<IAudioOut> {
 public:
-    IAudioOut(AudoutParams audio_params, AudioCore::AudioOut& audio_core, std::string&& device_name,
-              std::string&& unique_name)
+    IAudioOut(Core::System& system, AudoutParams audio_params, AudioCore::AudioOut& audio_core,
+              std::string&& device_name, std::string&& unique_name)
         : ServiceFramework("IAudioOut"), audio_core(audio_core),
           device_name(std::move(device_name)), audio_params(audio_params) {
         // clang-format off
@@ -65,7 +65,6 @@ public:
         RegisterHandlers(functions);
 
         // This is the event handle used to check if the audio buffer was released
-        auto& system = Core::System::GetInstance();
         buffer_event = Kernel::WritableEvent::CreateEventPair(
             system.Kernel(), Kernel::ResetType::Manual, "IAudioOutBufferReleased");
 
@@ -212,6 +211,22 @@ private:
     Kernel::EventPair buffer_event;
 };
 
+AudOutU::AudOutU(Core::System& system_) : ServiceFramework("audout:u"), system{system_} {
+    // clang-format off
+    static const FunctionInfo functions[] = {
+        {0, &AudOutU::ListAudioOutsImpl, "ListAudioOuts"},
+        {1, &AudOutU::OpenAudioOutImpl, "OpenAudioOut"},
+        {2, &AudOutU::ListAudioOutsImpl, "ListAudioOutsAuto"},
+        {3, &AudOutU::OpenAudioOutImpl, "OpenAudioOutAuto"},
+    };
+    // clang-format on
+
+    RegisterHandlers(functions);
+    audio_core = std::make_unique<AudioCore::AudioOut>();
+}
+
+AudOutU::~AudOutU() = default;
+
 void AudOutU::ListAudioOutsImpl(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_Audio, "called");
 
@@ -248,7 +263,7 @@ void AudOutU::OpenAudioOutImpl(Kernel::HLERequestContext& ctx) {
 
     std::string unique_name{fmt::format("{}-{}", device_name, audio_out_interfaces.size())};
     auto audio_out_interface = std::make_shared<IAudioOut>(
-        params, *audio_core, std::move(device_name), std::move(unique_name));
+        system, params, *audio_core, std::move(device_name), std::move(unique_name));
 
     IPC::ResponseBuilder rb{ctx, 6, 0, 1};
     rb.Push(RESULT_SUCCESS);
@@ -256,20 +271,9 @@ void AudOutU::OpenAudioOutImpl(Kernel::HLERequestContext& ctx) {
     rb.Push<u32>(params.channel_count);
     rb.Push<u32>(static_cast<u32>(AudioCore::Codec::PcmFormat::Int16));
     rb.Push<u32>(static_cast<u32>(AudioState::Stopped));
-    rb.PushIpcInterface<Audio::IAudioOut>(audio_out_interface);
+    rb.PushIpcInterface<IAudioOut>(audio_out_interface);
 
     audio_out_interfaces.push_back(std::move(audio_out_interface));
 }
 
-AudOutU::AudOutU() : ServiceFramework("audout:u") {
-    static const FunctionInfo functions[] = {{0, &AudOutU::ListAudioOutsImpl, "ListAudioOuts"},
-                                             {1, &AudOutU::OpenAudioOutImpl, "OpenAudioOut"},
-                                             {2, &AudOutU::ListAudioOutsImpl, "ListAudioOutsAuto"},
-                                             {3, &AudOutU::OpenAudioOutImpl, "OpenAudioOutAuto"}};
-    RegisterHandlers(functions);
-    audio_core = std::make_unique<AudioCore::AudioOut>();
-}
-
-AudOutU::~AudOutU() = default;
-
 } // namespace Service::Audio
diff --git a/src/core/hle/service/audio/audout_u.h b/src/core/hle/service/audio/audout_u.h
index aed4c43b2..c9f532ccd 100644
--- a/src/core/hle/service/audio/audout_u.h
+++ b/src/core/hle/service/audio/audout_u.h
@@ -11,6 +11,10 @@ namespace AudioCore {
 class AudioOut;
 }
 
+namespace Core {
+class System;
+}
+
 namespace Kernel {
 class HLERequestContext;
 }
@@ -21,15 +25,17 @@ class IAudioOut;
 
 class AudOutU final : public ServiceFramework<AudOutU> {
 public:
-    AudOutU();
+    explicit AudOutU(Core::System& system_);
     ~AudOutU() override;
 
 private:
+    void ListAudioOutsImpl(Kernel::HLERequestContext& ctx);
+    void OpenAudioOutImpl(Kernel::HLERequestContext& ctx);
+
     std::vector<std::shared_ptr<IAudioOut>> audio_out_interfaces;
     std::unique_ptr<AudioCore::AudioOut> audio_core;
 
-    void ListAudioOutsImpl(Kernel::HLERequestContext& ctx);
-    void OpenAudioOutImpl(Kernel::HLERequestContext& ctx);
+    Core::System& system;
 };
 
 } // namespace Service::Audio
diff --git a/src/core/hle/service/audio/audren_u.cpp b/src/core/hle/service/audio/audren_u.cpp
index 3711e1ea1..5b0b7f17e 100644
--- a/src/core/hle/service/audio/audren_u.cpp
+++ b/src/core/hle/service/audio/audren_u.cpp
@@ -5,6 +5,7 @@
 #include <algorithm>
 #include <array>
 #include <memory>
+#include <string_view>
 
 #include "audio_core/audio_renderer.h"
 #include "common/alignment.h"
@@ -25,7 +26,8 @@ namespace Service::Audio {
 
 class IAudioRenderer final : public ServiceFramework<IAudioRenderer> {
 public:
-    explicit IAudioRenderer(AudioCore::AudioRendererParameter audren_params)
+    explicit IAudioRenderer(Core::System& system, AudioCore::AudioRendererParameter audren_params,
+                            const std::size_t instance_number)
         : ServiceFramework("IAudioRenderer") {
         // clang-format off
         static const FunctionInfo functions[] = {
@@ -45,11 +47,10 @@ public:
         // clang-format on
         RegisterHandlers(functions);
 
-        auto& system = Core::System::GetInstance();
         system_event = Kernel::WritableEvent::CreateEventPair(
             system.Kernel(), Kernel::ResetType::Manual, "IAudioRenderer:SystemEvent");
-        renderer = std::make_unique<AudioCore::AudioRenderer>(system.CoreTiming(), audren_params,
-                                                              system_event.writable);
+        renderer = std::make_unique<AudioCore::AudioRenderer>(
+            system.CoreTiming(), audren_params, system_event.writable, instance_number);
     }
 
 private:
@@ -159,7 +160,8 @@ private:
 
 class IAudioDevice final : public ServiceFramework<IAudioDevice> {
 public:
-    IAudioDevice() : ServiceFramework("IAudioDevice") {
+    explicit IAudioDevice(Core::System& system, u32_le revision_num)
+        : ServiceFramework("IAudioDevice"), revision{revision_num} {
         static const FunctionInfo functions[] = {
             {0, &IAudioDevice::ListAudioDeviceName, "ListAudioDeviceName"},
             {1, &IAudioDevice::SetAudioDeviceOutputVolume, "SetAudioDeviceOutputVolume"},
@@ -177,7 +179,7 @@ public:
         };
         RegisterHandlers(functions);
 
-        auto& kernel = Core::System::GetInstance().Kernel();
+        auto& kernel = system.Kernel();
         buffer_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Automatic,
                                                               "IAudioOutBufferReleasedEvent");
 
@@ -188,15 +190,47 @@ public:
     }
 
 private:
+    using AudioDeviceName = std::array<char, 256>;
+    static constexpr std::array<std::string_view, 4> audio_device_names{{
+        "AudioStereoJackOutput",
+        "AudioBuiltInSpeakerOutput",
+        "AudioTvOutput",
+        "AudioUsbDeviceOutput",
+    }};
+    enum class DeviceType {
+        AHUBHeadphones,
+        AHUBSpeakers,
+        HDA,
+        USBOutput,
+    };
+
     void ListAudioDeviceName(Kernel::HLERequestContext& ctx) {
-        LOG_WARNING(Service_Audio, "(STUBBED) called");
+        LOG_DEBUG(Service_Audio, "called");
 
-        constexpr std::array<char, 15> audio_interface{{"AudioInterface"}};
-        ctx.WriteBuffer(audio_interface);
+        const bool usb_output_supported =
+            IsFeatureSupported(AudioFeatures::AudioUSBDeviceOutput, revision);
+        const std::size_t count = ctx.GetWriteBufferSize() / sizeof(AudioDeviceName);
+
+        std::vector<AudioDeviceName> name_buffer;
+        name_buffer.reserve(audio_device_names.size());
+
+        for (std::size_t i = 0; i < count && i < audio_device_names.size(); i++) {
+            const auto type = static_cast<DeviceType>(i);
+
+            if (!usb_output_supported && type == DeviceType::USBOutput) {
+                continue;
+            }
+
+            const auto& device_name = audio_device_names[i];
+            auto& entry = name_buffer.emplace_back();
+            device_name.copy(entry.data(), device_name.size());
+        }
+
+        ctx.WriteBuffer(name_buffer);
 
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(RESULT_SUCCESS);
-        rb.Push<u32>(1);
+        rb.Push(static_cast<u32>(name_buffer.size()));
     }
 
     void SetAudioDeviceOutputVolume(Kernel::HLERequestContext& ctx) {
@@ -215,12 +249,16 @@ private:
     void GetActiveAudioDeviceName(Kernel::HLERequestContext& ctx) {
         LOG_WARNING(Service_Audio, "(STUBBED) called");
 
-        constexpr std::array<char, 12> audio_interface{{"AudioDevice"}};
-        ctx.WriteBuffer(audio_interface);
+        // Currently set to always be TV audio output.
+        const auto& device_name = audio_device_names[2];
 
-        IPC::ResponseBuilder rb{ctx, 3};
+        AudioDeviceName out_device_name{};
+        device_name.copy(out_device_name.data(), device_name.size());
+
+        ctx.WriteBuffer(out_device_name);
+
+        IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);
-        rb.Push<u32>(1);
     }
 
     void QueryAudioDeviceSystemEvent(Kernel::HLERequestContext& ctx) {
@@ -249,12 +287,13 @@ private:
         rb.PushCopyObjects(audio_output_device_switch_event.readable);
     }
 
+    u32_le revision = 0;
     Kernel::EventPair buffer_event;
     Kernel::EventPair audio_output_device_switch_event;
 
 }; // namespace Audio
 
-AudRenU::AudRenU() : ServiceFramework("audren:u") {
+AudRenU::AudRenU(Core::System& system_) : ServiceFramework("audren:u"), system{system_} {
     // clang-format off
     static const FunctionInfo functions[] = {
         {0, &AudRenU::OpenAudioRenderer, "OpenAudioRenderer"},
@@ -327,7 +366,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     };
 
     // Calculates the portion of the size related to the mix data (and the sorting thereof).
-    const auto calculate_mix_info_size = [this](const AudioCore::AudioRendererParameter& params) {
+    const auto calculate_mix_info_size = [](const AudioCore::AudioRendererParameter& params) {
         // The size of the mixing info data structure.
         constexpr u64 mix_info_size = 0x940;
 
@@ -399,7 +438,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
 
     // Calculates the part of the size related to the splitter context.
     const auto calculate_splitter_context_size =
-        [this](const AudioCore::AudioRendererParameter& params) -> u64 {
+        [](const AudioCore::AudioRendererParameter& params) -> u64 {
         if (!IsFeatureSupported(AudioFeatures::Splitter, params.revision)) {
             return 0;
         }
@@ -446,7 +485,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     };
 
     // Calculates the part of the size related to performance statistics.
-    const auto calculate_perf_size = [this](const AudioCore::AudioRendererParameter& params) {
+    const auto calculate_perf_size = [](const AudioCore::AudioRendererParameter& params) {
         // Extra size value appended to the end of the calculation.
         constexpr u64 appended = 128;
 
@@ -473,78 +512,76 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     };
 
     // Calculates the part of the size that relates to the audio command buffer.
-    const auto calculate_command_buffer_size =
-        [this](const AudioCore::AudioRendererParameter& params) {
-            constexpr u64 alignment = (buffer_alignment_size - 1) * 2;
+    const auto calculate_command_buffer_size = [](const AudioCore::AudioRendererParameter& params) {
+        constexpr u64 alignment = (buffer_alignment_size - 1) * 2;
 
-            if (!IsFeatureSupported(AudioFeatures::VariadicCommandBuffer, params.revision)) {
-                constexpr u64 command_buffer_size = 0x18000;
+        if (!IsFeatureSupported(AudioFeatures::VariadicCommandBuffer, params.revision)) {
+            constexpr u64 command_buffer_size = 0x18000;
 
-                return command_buffer_size + alignment;
-            }
+            return command_buffer_size + alignment;
+        }
 
-            // When the variadic command buffer is supported, this means
-            // the command generator for the audio renderer can issue commands
-            // that are (as one would expect), variable in size. So what we need to do
-            // is determine the maximum possible size for a few command data structures
-            // then multiply them by the amount of present commands indicated by the given
-            // respective audio parameters.
+        // When the variadic command buffer is supported, this means
+        // the command generator for the audio renderer can issue commands
+        // that are (as one would expect), variable in size. So what we need to do
+        // is determine the maximum possible size for a few command data structures
+        // then multiply them by the amount of present commands indicated by the given
+        // respective audio parameters.
 
-            constexpr u64 max_biquad_filters = 2;
-            constexpr u64 max_mix_buffers = 24;
+        constexpr u64 max_biquad_filters = 2;
+        constexpr u64 max_mix_buffers = 24;
 
-            constexpr u64 biquad_filter_command_size = 0x2C;
+        constexpr u64 biquad_filter_command_size = 0x2C;
 
-            constexpr u64 depop_mix_command_size = 0x24;
-            constexpr u64 depop_setup_command_size = 0x50;
+        constexpr u64 depop_mix_command_size = 0x24;
+        constexpr u64 depop_setup_command_size = 0x50;
 
-            constexpr u64 effect_command_max_size = 0x540;
+        constexpr u64 effect_command_max_size = 0x540;
 
-            constexpr u64 mix_command_size = 0x1C;
-            constexpr u64 mix_ramp_command_size = 0x24;
-            constexpr u64 mix_ramp_grouped_command_size = 0x13C;
+        constexpr u64 mix_command_size = 0x1C;
+        constexpr u64 mix_ramp_command_size = 0x24;
+        constexpr u64 mix_ramp_grouped_command_size = 0x13C;
 
-            constexpr u64 perf_command_size = 0x28;
+        constexpr u64 perf_command_size = 0x28;
 
-            constexpr u64 sink_command_size = 0x130;
+        constexpr u64 sink_command_size = 0x130;
 
-            constexpr u64 submix_command_max_size =
-                depop_mix_command_size + (mix_command_size * max_mix_buffers) * max_mix_buffers;
+        constexpr u64 submix_command_max_size =
+            depop_mix_command_size + (mix_command_size * max_mix_buffers) * max_mix_buffers;
 
-            constexpr u64 volume_command_size = 0x1C;
-            constexpr u64 volume_ramp_command_size = 0x20;
+        constexpr u64 volume_command_size = 0x1C;
+        constexpr u64 volume_ramp_command_size = 0x20;
 
-            constexpr u64 voice_biquad_filter_command_size =
-                biquad_filter_command_size * max_biquad_filters;
-            constexpr u64 voice_data_command_size = 0x9C;
-            const u64 voice_command_max_size =
-                (params.splitter_count * depop_setup_command_size) +
-                (voice_data_command_size + voice_biquad_filter_command_size +
-                 volume_ramp_command_size + mix_ramp_grouped_command_size);
+        constexpr u64 voice_biquad_filter_command_size =
+            biquad_filter_command_size * max_biquad_filters;
+        constexpr u64 voice_data_command_size = 0x9C;
+        const u64 voice_command_max_size =
+            (params.splitter_count * depop_setup_command_size) +
+            (voice_data_command_size + voice_biquad_filter_command_size + volume_ramp_command_size +
+             mix_ramp_grouped_command_size);
 
-            // Now calculate the individual elements that comprise the size and add them together.
-            const u64 effect_commands_size = params.effect_count * effect_command_max_size;
+        // Now calculate the individual elements that comprise the size and add them together.
+        const u64 effect_commands_size = params.effect_count * effect_command_max_size;
 
-            const u64 final_mix_commands_size =
-                depop_mix_command_size + volume_command_size * max_mix_buffers;
+        const u64 final_mix_commands_size =
+            depop_mix_command_size + volume_command_size * max_mix_buffers;
 
-            const u64 perf_commands_size =
-                perf_command_size *
-                (CalculateNumPerformanceEntries(params) + max_perf_detail_entries);
+        const u64 perf_commands_size =
+            perf_command_size * (CalculateNumPerformanceEntries(params) + max_perf_detail_entries);
 
-            const u64 sink_commands_size = params.sink_count * sink_command_size;
+        const u64 sink_commands_size = params.sink_count * sink_command_size;
 
-            const u64 splitter_commands_size =
-                params.num_splitter_send_channels * max_mix_buffers * mix_ramp_command_size;
+        const u64 splitter_commands_size =
+            params.num_splitter_send_channels * max_mix_buffers * mix_ramp_command_size;
 
-            const u64 submix_commands_size = params.submix_count * submix_command_max_size;
+        const u64 submix_commands_size = params.submix_count * submix_command_max_size;
 
-            const u64 voice_commands_size = params.voice_count * voice_command_max_size;
+        const u64 voice_commands_size = params.voice_count * voice_command_max_size;
 
-            return effect_commands_size + final_mix_commands_size + perf_commands_size +
-                   sink_commands_size + splitter_commands_size + submix_commands_size +
-                   voice_commands_size + alignment;
-        };
+        return effect_commands_size + final_mix_commands_size + perf_commands_size +
+               sink_commands_size + splitter_commands_size + submix_commands_size +
+               voice_commands_size + alignment;
+    };
 
     IPC::RequestParser rp{ctx};
     const auto params = rp.PopRaw<AudioCore::AudioRendererParameter>();
@@ -577,12 +614,16 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
 }
 
 void AudRenU::GetAudioDeviceService(Kernel::HLERequestContext& ctx) {
-    LOG_DEBUG(Service_Audio, "called");
+    IPC::RequestParser rp{ctx};
+    const u64 aruid = rp.Pop<u64>();
 
-    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
+    LOG_DEBUG(Service_Audio, "called. aruid={:016X}", aruid);
 
+    // Revisionless variant of GetAudioDeviceServiceWithRevisionInfo that
+    // always assumes the initial release revision (REV1).
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<Audio::IAudioDevice>();
+    rb.PushIpcInterface<IAudioDevice>(system, Common::MakeMagic('R', 'E', 'V', '1'));
 }
 
 void AudRenU::OpenAudioRendererAuto(Kernel::HLERequestContext& ctx) {
@@ -592,13 +633,19 @@ void AudRenU::OpenAudioRendererAuto(Kernel::HLERequestContext& ctx) {
 }
 
 void AudRenU::GetAudioDeviceServiceWithRevisionInfo(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_Audio, "(STUBBED) called");
+    struct Parameters {
+        u32 revision;
+        u64 aruid;
+    };
 
-    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
+    IPC::RequestParser rp{ctx};
+    const auto [revision, aruid] = rp.PopRaw<Parameters>();
+
+    LOG_DEBUG(Service_Audio, "called. revision={:08X}, aruid={:016X}", revision, aruid);
 
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<Audio::IAudioDevice>(); // TODO(ogniK): Figure out what is different
-                                                // based on the current revision
+    rb.PushIpcInterface<IAudioDevice>(system, revision);
 }
 
 void AudRenU::OpenAudioRendererImpl(Kernel::HLERequestContext& ctx) {
@@ -607,14 +654,16 @@ void AudRenU::OpenAudioRendererImpl(Kernel::HLERequestContext& ctx) {
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
 
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<IAudioRenderer>(params);
+    rb.PushIpcInterface<IAudioRenderer>(system, params, audren_instance_count++);
 }
 
-bool AudRenU::IsFeatureSupported(AudioFeatures feature, u32_le revision) const {
+bool IsFeatureSupported(AudioFeatures feature, u32_le revision) {
     // Byte swap
     const u32_be version_num = revision - Common::MakeMagic('R', 'E', 'V', '0');
 
     switch (feature) {
+    case AudioFeatures::AudioUSBDeviceOutput:
+        return version_num >= 4U;
     case AudioFeatures::Splitter:
         return version_num >= 2U;
     case AudioFeatures::PerformanceMetricsVersion2:
diff --git a/src/core/hle/service/audio/audren_u.h b/src/core/hle/service/audio/audren_u.h
index 1d3c8df61..4e0ccc792 100644
--- a/src/core/hle/service/audio/audren_u.h
+++ b/src/core/hle/service/audio/audren_u.h
@@ -6,6 +6,10 @@
 
 #include "core/hle/service/service.h"
 
+namespace Core {
+class System;
+}
+
 namespace Kernel {
 class HLERequestContext;
 }
@@ -14,7 +18,7 @@ namespace Service::Audio {
 
 class AudRenU final : public ServiceFramework<AudRenU> {
 public:
-    explicit AudRenU();
+    explicit AudRenU(Core::System& system_);
     ~AudRenU() override;
 
 private:
@@ -26,13 +30,19 @@ private:
 
     void OpenAudioRendererImpl(Kernel::HLERequestContext& ctx);
 
-    enum class AudioFeatures : u32 {
-        Splitter,
-        PerformanceMetricsVersion2,
-        VariadicCommandBuffer,
-    };
+    std::size_t audren_instance_count = 0;
+    Core::System& system;
+};
 
-    bool IsFeatureSupported(AudioFeatures feature, u32_le revision) const;
+// Describes a particular audio feature that may be supported in a particular revision.
+enum class AudioFeatures : u32 {
+    AudioUSBDeviceOutput,
+    Splitter,
+    PerformanceMetricsVersion2,
+    VariadicCommandBuffer,
 };
 
+// Tests if a particular audio feature is supported with a given audio revision.
+bool IsFeatureSupported(AudioFeatures feature, u32_le revision);
+
 } // namespace Service::Audio
diff --git a/src/core/hle/service/friend/friend.cpp b/src/core/hle/service/friend/friend.cpp
index dec541f2e..d1ec12ef9 100644
--- a/src/core/hle/service/friend/friend.cpp
+++ b/src/core/hle/service/friend/friend.cpp
@@ -22,7 +22,7 @@ public:
             {0, nullptr, "GetCompletionEvent"},
             {1, nullptr, "Cancel"},
             {10100, nullptr, "GetFriendListIds"},
-            {10101, nullptr, "GetFriendList"},
+            {10101, &IFriendService::GetFriendList, "GetFriendList"},
             {10102, nullptr, "UpdateFriendInfo"},
             {10110, nullptr, "GetFriendProfileImage"},
             {10200, nullptr, "SendFriendRequestForApplication"},
@@ -99,6 +99,23 @@ public:
     }
 
 private:
+    enum class PresenceFilter : u32 {
+        None = 0,
+        Online = 1,
+        OnlinePlay = 2,
+        OnlineOrOnlinePlay = 3,
+    };
+
+    struct SizedFriendFilter {
+        PresenceFilter presence;
+        u8 is_favorite;
+        u8 same_app;
+        u8 same_app_played;
+        u8 arbitary_app_played;
+        u64 group_id;
+    };
+    static_assert(sizeof(SizedFriendFilter) == 0x10, "SizedFriendFilter is an invalid size");
+
     void DeclareCloseOnlinePlaySession(Kernel::HLERequestContext& ctx) {
         // Stub used by Splatoon 2
         LOG_WARNING(Service_ACC, "(STUBBED) called");
@@ -112,6 +129,22 @@ private:
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);
     }
+
+    void GetFriendList(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto friend_offset = rp.Pop<u32>();
+        const auto uuid = rp.PopRaw<Common::UUID>();
+        [[maybe_unused]] const auto filter = rp.PopRaw<SizedFriendFilter>();
+        const auto pid = rp.Pop<u64>();
+        LOG_WARNING(Service_ACC, "(STUBBED) called, offset={}, uuid={}, pid={}", friend_offset,
+                    uuid.Format(), pid);
+
+        IPC::ResponseBuilder rb{ctx, 3};
+        rb.Push(RESULT_SUCCESS);
+
+        rb.Push<u32>(0); // Friend count
+        // TODO(ogniK): Return a buffer of u64s which are the "NetworkServiceAccountId"
+    }
 };
 
 class INotificationService final : public ServiceFramework<INotificationService> {
diff --git a/src/core/hle/service/ldr/ldr.cpp b/src/core/hle/service/ldr/ldr.cpp
index b839303ac..8ddad8682 100644
--- a/src/core/hle/service/ldr/ldr.cpp
+++ b/src/core/hle/service/ldr/ldr.cpp
@@ -345,14 +345,16 @@ public:
             vm_manager
                 .MirrorMemory(*map_address, nro_address, nro_size, Kernel::MemoryState::ModuleCode)
                 .IsSuccess());
-        ASSERT(vm_manager.UnmapRange(nro_address, nro_size).IsSuccess());
+        ASSERT(vm_manager.ReprotectRange(nro_address, nro_size, Kernel::VMAPermission::None)
+                   .IsSuccess());
 
         if (bss_size > 0) {
             ASSERT(vm_manager
                        .MirrorMemory(*map_address + nro_size, bss_address, bss_size,
                                      Kernel::MemoryState::ModuleCode)
                        .IsSuccess());
-            ASSERT(vm_manager.UnmapRange(bss_address, bss_size).IsSuccess());
+            ASSERT(vm_manager.ReprotectRange(bss_address, bss_size, Kernel::VMAPermission::None)
+                       .IsSuccess());
         }
 
         vm_manager.ReprotectRange(*map_address, header.text_size,
@@ -364,7 +366,8 @@ public:
 
         Core::System::GetInstance().InvalidateCpuInstructionCaches();
 
-        nro.insert_or_assign(*map_address, NROInfo{hash, nro_size + bss_size});
+        nro.insert_or_assign(*map_address,
+                             NROInfo{hash, nro_address, nro_size, bss_address, bss_size});
 
         IPC::ResponseBuilder rb{ctx, 4};
         rb.Push(RESULT_SUCCESS);
@@ -409,9 +412,23 @@ public:
         }
 
         auto& vm_manager = Core::CurrentProcess()->VMManager();
-        const auto& nro_size = iter->second.size;
+        const auto& nro_info = iter->second;
 
-        ASSERT(vm_manager.UnmapRange(nro_address, nro_size).IsSuccess());
+        // Unmap the mirrored memory
+        ASSERT(
+            vm_manager.UnmapRange(nro_address, nro_info.nro_size + nro_info.bss_size).IsSuccess());
+
+        // Reprotect the source memory
+        ASSERT(vm_manager
+                   .ReprotectRange(nro_info.nro_address, nro_info.nro_size,
+                                   Kernel::VMAPermission::ReadWrite)
+                   .IsSuccess());
+        if (nro_info.bss_size > 0) {
+            ASSERT(vm_manager
+                       .ReprotectRange(nro_info.bss_address, nro_info.bss_size,
+                                       Kernel::VMAPermission::ReadWrite)
+                       .IsSuccess());
+        }
 
         Core::System::GetInstance().InvalidateCpuInstructionCaches();
 
@@ -473,7 +490,10 @@ private:
 
     struct NROInfo {
         SHA256Hash hash;
-        u64 size;
+        VAddr nro_address;
+        u64 nro_size;
+        VAddr bss_address;
+        u64 bss_size;
     };
 
     bool initialized = false;
diff --git a/src/core/hle/service/mii/mii.cpp b/src/core/hle/service/mii/mii.cpp
index ce84e25ed..0b3923ad9 100644
--- a/src/core/hle/service/mii/mii.cpp
+++ b/src/core/hle/service/mii/mii.cpp
@@ -48,7 +48,7 @@ public:
             {19, nullptr, "Export"},
             {20, nullptr, "IsBrokenDatabaseWithClearFlag"},
             {21, &IDatabaseService::GetIndex, "GetIndex"},
-            {22, nullptr, "SetInterfaceVersion"},
+            {22, &IDatabaseService::SetInterfaceVersion, "SetInterfaceVersion"},
             {23, nullptr, "Convert"},
         };
         // clang-format on
@@ -350,8 +350,22 @@ private:
         rb.Push(index);
     }
 
+    void SetInterfaceVersion(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        current_interface_version = rp.PopRaw<u32>();
+
+        LOG_DEBUG(Service_Mii, "called, interface_version={:08X}", current_interface_version);
+
+        UNIMPLEMENTED_IF(current_interface_version != 1);
+
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(RESULT_SUCCESS);
+    }
+
     MiiManager db;
 
+    u32 current_interface_version = 0;
+
     // Last read offsets of Get functions
     std::array<u32, 4> offsets{};
 };
diff --git a/src/core/hle/service/ns/pl_u.cpp b/src/core/hle/service/ns/pl_u.cpp
index ad176f89d..2a522136d 100644
--- a/src/core/hle/service/ns/pl_u.cpp
+++ b/src/core/hle/service/ns/pl_u.cpp
@@ -77,7 +77,7 @@ enum class LoadState : u32 {
     Done = 1,
 };
 
-static void DecryptSharedFont(const std::vector<u32>& input, std::vector<u8>& output,
+static void DecryptSharedFont(const std::vector<u32>& input, Kernel::PhysicalMemory& output,
                               std::size_t& offset) {
     ASSERT_MSG(offset + (input.size() * sizeof(u32)) < SHARED_FONT_MEM_SIZE,
                "Shared fonts exceeds 17mb!");
@@ -94,7 +94,7 @@ static void DecryptSharedFont(const std::vector<u32>& input, std::vector<u8>& ou
     offset += transformed_font.size() * sizeof(u32);
 }
 
-static void EncryptSharedFont(const std::vector<u8>& input, std::vector<u8>& output,
+static void EncryptSharedFont(const std::vector<u8>& input, Kernel::PhysicalMemory& output,
                               std::size_t& offset) {
     ASSERT_MSG(offset + input.size() + 8 < SHARED_FONT_MEM_SIZE, "Shared fonts exceeds 17mb!");
     const u32 KEY = EXPECTED_MAGIC ^ EXPECTED_RESULT;
@@ -121,7 +121,7 @@ struct PL_U::Impl {
         return shared_font_regions.at(index);
     }
 
-    void BuildSharedFontsRawRegions(const std::vector<u8>& input) {
+    void BuildSharedFontsRawRegions(const Kernel::PhysicalMemory& input) {
         // As we can derive the xor key we can just populate the offsets
         // based on the shared memory dump
         unsigned cur_offset = 0;
@@ -144,7 +144,7 @@ struct PL_U::Impl {
     Kernel::SharedPtr<Kernel::SharedMemory> shared_font_mem;
 
     /// Backing memory for the shared font data
-    std::shared_ptr<std::vector<u8>> shared_font;
+    std::shared_ptr<Kernel::PhysicalMemory> shared_font;
 
     // Automatically populated based on shared_fonts dump or system archives.
     std::vector<FontRegion> shared_font_regions;
@@ -166,7 +166,7 @@ PL_U::PL_U() : ServiceFramework("pl:u"), impl{std::make_unique<Impl>()} {
     // Rebuild shared fonts from data ncas
     if (nand->HasEntry(static_cast<u64>(FontArchives::Standard),
                        FileSys::ContentRecordType::Data)) {
-        impl->shared_font = std::make_shared<std::vector<u8>>(SHARED_FONT_MEM_SIZE);
+        impl->shared_font = std::make_shared<Kernel::PhysicalMemory>(SHARED_FONT_MEM_SIZE);
         for (auto font : SHARED_FONTS) {
             const auto nca =
                 nand->GetEntry(static_cast<u64>(font.first), FileSys::ContentRecordType::Data);
@@ -207,7 +207,7 @@ PL_U::PL_U() : ServiceFramework("pl:u"), impl{std::make_unique<Impl>()} {
         }
 
     } else {
-        impl->shared_font = std::make_shared<std::vector<u8>>(
+        impl->shared_font = std::make_shared<Kernel::PhysicalMemory>(
             SHARED_FONT_MEM_SIZE); // Shared memory needs to always be allocated and a fixed size
 
         const std::string user_path = FileUtil::GetUserPath(FileUtil::UserPath::SysDataDir);
diff --git a/src/core/hle/service/nvdrv/devices/nvdevice.h b/src/core/hle/service/nvdrv/devices/nvdevice.h
index 4f6042b00..5b8248433 100644
--- a/src/core/hle/service/nvdrv/devices/nvdevice.h
+++ b/src/core/hle/service/nvdrv/devices/nvdevice.h
@@ -8,6 +8,11 @@
 #include "common/bit_field.h"
 #include "common/common_types.h"
 #include "common/swap.h"
+#include "core/hle/service/nvdrv/nvdata.h"
+
+namespace Core {
+class System;
+}
 
 namespace Service::Nvidia::Devices {
 
@@ -15,7 +20,7 @@ namespace Service::Nvidia::Devices {
 /// implement the ioctl interface.
 class nvdevice {
 public:
-    nvdevice() = default;
+    explicit nvdevice(Core::System& system) : system{system} {};
     virtual ~nvdevice() = default;
     union Ioctl {
         u32_le raw;
@@ -33,7 +38,11 @@ public:
      * @param output A buffer where the output data will be written to.
      * @returns The result code of the ioctl.
      */
-    virtual u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) = 0;
+    virtual u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) = 0;
+
+protected:
+    Core::System& system;
 };
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index 20c7c39aa..926a1285d 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -13,10 +13,12 @@
 
 namespace Service::Nvidia::Devices {
 
-nvdisp_disp0::nvdisp_disp0(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvdisp_disp0::nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvdisp_disp0 ::~nvdisp_disp0() = default;
 
-u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl");
     return 0;
 }
@@ -34,9 +36,8 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u3
         addr,      offset,   width, height, stride, static_cast<PixelFormat>(format),
         transform, crop_rect};
 
-    auto& instance = Core::System::GetInstance();
-    instance.GetPerfStats().EndGameFrame();
-    instance.GPU().SwapBuffers(framebuffer);
+    system.GetPerfStats().EndGameFrame();
+    system.GPU().SwapBuffers(&framebuffer);
 }
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
index 12f3ef825..e79e490ff 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
@@ -17,10 +17,11 @@ class nvmap;
 
 class nvdisp_disp0 final : public nvdevice {
 public:
-    explicit nvdisp_disp0(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvdisp_disp0() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
     /// Performs a screen flip, drawing the buffer pointed to by the handle.
     void flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height, u32 stride,
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
index af62d33d2..24ab3f2e9 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
@@ -22,10 +22,12 @@ enum {
 };
 }
 
-nvhost_as_gpu::nvhost_as_gpu(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_as_gpu::nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvhost_as_gpu::~nvhost_as_gpu() = default;
 
-u32 nvhost_as_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_as_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                         IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -65,7 +67,7 @@ u32 nvhost_as_gpu::AllocateSpace(const std::vector<u8>& input, std::vector<u8>&
     LOG_DEBUG(Service_NVDRV, "called, pages={:X}, page_size={:X}, flags={:X}", params.pages,
               params.page_size, params.flags);
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
     const u64 size{static_cast<u64>(params.pages) * static_cast<u64>(params.page_size)};
     if (params.flags & 1) {
         params.offset = gpu.MemoryManager().AllocateSpace(params.offset, size, 1);
@@ -85,7 +87,7 @@ u32 nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& output)
     std::vector<IoctlRemapEntry> entries(num_entries);
     std::memcpy(entries.data(), input.data(), input.size());
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
     for (const auto& entry : entries) {
         LOG_WARNING(Service_NVDRV, "remap entry, offset=0x{:X} handle=0x{:X} pages=0x{:X}",
                     entry.offset, entry.nvmap_handle, entry.pages);
@@ -136,7 +138,7 @@ u32 nvhost_as_gpu::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& ou
     // case to prevent unexpected behavior.
     ASSERT(object->id == params.nvmap_handle);
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
 
     if (params.flags & 1) {
         params.offset = gpu.MemoryManager().MapBufferEx(object->addr, params.offset, object->size);
@@ -173,8 +175,7 @@ u32 nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& ou
         return 0;
     }
 
-    params.offset = Core::System::GetInstance().GPU().MemoryManager().UnmapBuffer(params.offset,
-                                                                                  itr->second.size);
+    params.offset = system.GPU().MemoryManager().UnmapBuffer(params.offset, itr->second.size);
     buffer_mappings.erase(itr->second.offset);
 
     std::memcpy(output.data(), &params, output.size());
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
index eb14b1da8..30ca5f4c3 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
@@ -17,10 +17,11 @@ class nvmap;
 
 class nvhost_as_gpu final : public nvdevice {
 public:
-    explicit nvhost_as_gpu(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_as_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
index b39fb9ef9..9a66a5f88 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
@@ -7,14 +7,20 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
+#include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/devices/nvhost_ctrl.h"
+#include "video_core/gpu.h"
 
 namespace Service::Nvidia::Devices {
 
-nvhost_ctrl::nvhost_ctrl() = default;
+nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface)
+    : nvdevice(system), events_interface{events_interface} {}
 nvhost_ctrl::~nvhost_ctrl() = default;
 
-u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                       IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -22,11 +28,15 @@ u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<
     case IoctlCommand::IocGetConfigCommand:
         return NvOsGetConfigU32(input, output);
     case IoctlCommand::IocCtrlEventWaitCommand:
-        return IocCtrlEventWait(input, output, false);
+        return IocCtrlEventWait(input, output, false, ctrl);
     case IoctlCommand::IocCtrlEventWaitAsyncCommand:
-        return IocCtrlEventWait(input, output, true);
+        return IocCtrlEventWait(input, output, true, ctrl);
     case IoctlCommand::IocCtrlEventRegisterCommand:
         return IocCtrlEventRegister(input, output);
+    case IoctlCommand::IocCtrlEventUnregisterCommand:
+        return IocCtrlEventUnregister(input, output);
+    case IoctlCommand::IocCtrlEventSignalCommand:
+        return IocCtrlEventSignal(input, output);
     }
     UNIMPLEMENTED_MSG("Unimplemented ioctl");
     return 0;
@@ -41,23 +51,137 @@ u32 nvhost_ctrl::NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>&
 }
 
 u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output,
-                                  bool is_async) {
+                                  bool is_async, IoctlCtrl& ctrl) {
     IocCtrlEventWaitParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
-    LOG_WARNING(Service_NVDRV,
-                "(STUBBED) called, syncpt_id={}, threshold={}, timeout={}, is_async={}",
-                params.syncpt_id, params.threshold, params.timeout, is_async);
+    LOG_DEBUG(Service_NVDRV, "syncpt_id={}, threshold={}, timeout={}, is_async={}",
+              params.syncpt_id, params.threshold, params.timeout, is_async);
 
-    // TODO(Subv): Implement actual syncpt waiting.
-    params.value = 0;
+    if (params.syncpt_id >= MaxSyncPoints) {
+        return NvResult::BadParameter;
+    }
+
+    auto& gpu = system.GPU();
+    // This is mostly to take into account unimplemented features. As synced
+    // gpu is always synced.
+    if (!gpu.IsAsync()) {
+        return NvResult::Success;
+    }
+    auto lock = gpu.LockSync();
+    const u32 current_syncpoint_value = gpu.GetSyncpointValue(params.syncpt_id);
+    const s32 diff = current_syncpoint_value - params.threshold;
+    if (diff >= 0) {
+        params.value = current_syncpoint_value;
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Success;
+    }
+    const u32 target_value = current_syncpoint_value - diff;
+
+    if (!is_async) {
+        params.value = 0;
+    }
+
+    if (params.timeout == 0) {
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Timeout;
+    }
+
+    u32 event_id;
+    if (is_async) {
+        event_id = params.value & 0x00FF;
+        if (event_id >= MaxNvEvents) {
+            std::memcpy(output.data(), &params, sizeof(params));
+            return NvResult::BadParameter;
+        }
+    } else {
+        if (ctrl.fresh_call) {
+            const auto result = events_interface.GetFreeEvent();
+            if (result) {
+                event_id = *result;
+            } else {
+                LOG_CRITICAL(Service_NVDRV, "No Free Events available!");
+                event_id = params.value & 0x00FF;
+            }
+        } else {
+            event_id = ctrl.event_id;
+        }
+    }
+
+    EventState status = events_interface.status[event_id];
+    if (event_id < MaxNvEvents || status == EventState::Free || status == EventState::Registered) {
+        events_interface.SetEventStatus(event_id, EventState::Waiting);
+        events_interface.assigned_syncpt[event_id] = params.syncpt_id;
+        events_interface.assigned_value[event_id] = target_value;
+        if (is_async) {
+            params.value = params.syncpt_id << 4;
+        } else {
+            params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000;
+        }
+        params.value |= event_id;
+        events_interface.events[event_id].writable->Clear();
+        gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value);
+        if (!is_async && ctrl.fresh_call) {
+            ctrl.must_delay = true;
+            ctrl.timeout = params.timeout;
+            ctrl.event_id = event_id;
+            return NvResult::Timeout;
+        }
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Timeout;
+    }
     std::memcpy(output.data(), &params, sizeof(params));
-    return 0;
+    return NvResult::BadParameter;
 }
 
 u32 nvhost_ctrl::IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output) {
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-    // TODO(bunnei): Implement this.
-    return 0;
+    IocCtrlEventRegisterParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    const u32 event_id = params.user_event_id & 0x00FF;
+    LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (events_interface.registered[event_id]) {
+        return NvResult::BadParameter;
+    }
+    events_interface.RegisterEvent(event_id);
+    return NvResult::Success;
+}
+
+u32 nvhost_ctrl::IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output) {
+    IocCtrlEventUnregisterParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    const u32 event_id = params.user_event_id & 0x00FF;
+    LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (!events_interface.registered[event_id]) {
+        return NvResult::BadParameter;
+    }
+    events_interface.UnregisterEvent(event_id);
+    return NvResult::Success;
+}
+
+u32 nvhost_ctrl::IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output) {
+    IocCtrlEventSignalParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    // TODO(Blinkhawk): This is normally called when an NvEvents timeout on WaitSynchronization
+    // It is believed from RE to cancel the GPU Event. However, better research is required
+    u32 event_id = params.user_event_id & 0x00FF;
+    LOG_WARNING(Service_NVDRV, "(STUBBED) called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (events_interface.status[event_id] == EventState::Waiting) {
+        auto& gpu = system.GPU();
+        if (gpu.CancelSyncptInterrupt(events_interface.assigned_syncpt[event_id],
+                                      events_interface.assigned_value[event_id])) {
+            events_interface.LiberateEvent(event_id);
+            events_interface.events[event_id].writable->Signal();
+        }
+    }
+    return NvResult::Success;
 }
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
index 6d0de2212..14e6e7e57 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
@@ -8,15 +8,17 @@
 #include <vector>
 #include "common/common_types.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "core/hle/service/nvdrv/nvdrv.h"
 
 namespace Service::Nvidia::Devices {
 
 class nvhost_ctrl final : public nvdevice {
 public:
-    nvhost_ctrl();
+    explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface);
     ~nvhost_ctrl() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
@@ -132,9 +134,16 @@ private:
 
     u32 NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>& output);
 
-    u32 IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async);
+    u32 IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async,
+                         IoctlCtrl& ctrl);
 
     u32 IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output);
+
+    u32 IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output);
+
+    u32 IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output);
+
+    EventInterface& events_interface;
 };
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
index 0e28755bd..988effd90 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
@@ -12,10 +12,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_ctrl_gpu::nvhost_ctrl_gpu() = default;
+nvhost_ctrl_gpu::nvhost_ctrl_gpu(Core::System& system) : nvdevice(system) {}
 nvhost_ctrl_gpu::~nvhost_ctrl_gpu() = default;
 
-u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                           IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -185,7 +186,7 @@ u32 nvhost_ctrl_gpu::GetGpuTime(const std::vector<u8>& input, std::vector<u8>& o
 
     IoctlGetGpuTime params{};
     std::memcpy(&params, input.data(), input.size());
-    const auto ns = Core::Timing::CyclesToNs(Core::System::GetInstance().CoreTiming().GetTicks());
+    const auto ns = Core::Timing::CyclesToNs(system.CoreTiming().GetTicks());
     params.gpu_time = static_cast<u64_le>(ns.count());
     std::memcpy(output.data(), &params, output.size());
     return 0;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
index 240435eea..2b035ae3f 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_ctrl_gpu final : public nvdevice {
 public:
-    nvhost_ctrl_gpu();
+    explicit nvhost_ctrl_gpu(Core::System& system);
     ~nvhost_ctrl_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 8ce7bc7a5..241dac881 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -13,10 +13,12 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_gpu::nvhost_gpu(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvhost_gpu::~nvhost_gpu() = default;
 
-u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -119,8 +121,10 @@ u32 nvhost_gpu::AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& ou
                 params.num_entries, params.flags, params.unk0, params.unk1, params.unk2,
                 params.unk3);
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
+    auto& gpu = system.GPU();
+    params.fence_out.id = assigned_syncpoints;
+    params.fence_out.value = gpu.GetSyncpointValue(assigned_syncpoints);
+    assigned_syncpoints++;
     std::memcpy(output.data(), &params, output.size());
     return 0;
 }
@@ -143,7 +147,7 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags);
+                params.address, params.num_entries, params.flags.raw);
 
     ASSERT_MSG(input.size() == sizeof(IoctlSubmitGpfifo) +
                                    params.num_entries * sizeof(Tegra::CommandListHeader),
@@ -153,10 +157,18 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)],
                 params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().PushGPUEntries(std::move(entries));
+    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
+    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
+
+    auto& gpu = system.GPU();
+    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
+    if (params.flags.increment.Value()) {
+        params.fence_out.value += current_syncpoint_value;
+    } else {
+        params.fence_out.value = current_syncpoint_value;
+    }
+    gpu.PushGPUEntries(std::move(entries));
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
     std::memcpy(output.data(), &params, sizeof(IoctlSubmitGpfifo));
     return 0;
 }
@@ -168,16 +180,24 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output)
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags);
+                params.address, params.num_entries, params.flags.raw);
 
     Tegra::CommandList entries(params.num_entries);
     Memory::ReadBlock(params.address, entries.data(),
                       params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().PushGPUEntries(std::move(entries));
+    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
+    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
+
+    auto& gpu = system.GPU();
+    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
+    if (params.flags.increment.Value()) {
+        params.fence_out.value += current_syncpoint_value;
+    } else {
+        params.fence_out.value = current_syncpoint_value;
+    }
+    gpu.PushGPUEntries(std::move(entries));
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
     std::memcpy(output.data(), &params, output.size());
     return 0;
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
index 62beb5c0c..d2e8fbae9 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
@@ -10,6 +10,7 @@
 #include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 
 namespace Service::Nvidia::Devices {
 
@@ -20,10 +21,11 @@ constexpr u32 NVGPU_IOCTL_CHANNEL_KICKOFF_PB(0x1b);
 
 class nvhost_gpu final : public nvdevice {
 public:
-    explicit nvhost_gpu(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
@@ -113,11 +115,7 @@ private:
     static_assert(sizeof(IoctlGetErrorNotification) == 16,
                   "IoctlGetErrorNotification is incorrect size");
 
-    struct IoctlFence {
-        u32_le id;
-        u32_le value;
-    };
-    static_assert(sizeof(IoctlFence) == 8, "IoctlFence is incorrect size");
+    static_assert(sizeof(Fence) == 8, "Fence is incorrect size");
 
     struct IoctlAllocGpfifoEx {
         u32_le num_entries;
@@ -132,13 +130,13 @@ private:
     static_assert(sizeof(IoctlAllocGpfifoEx) == 32, "IoctlAllocGpfifoEx is incorrect size");
 
     struct IoctlAllocGpfifoEx2 {
-        u32_le num_entries;   // in
-        u32_le flags;         // in
-        u32_le unk0;          // in (1 works)
-        IoctlFence fence_out; // out
-        u32_le unk1;          // in
-        u32_le unk2;          // in
-        u32_le unk3;          // in
+        u32_le num_entries; // in
+        u32_le flags;       // in
+        u32_le unk0;        // in (1 works)
+        Fence fence_out;    // out
+        u32_le unk1;        // in
+        u32_le unk2;        // in
+        u32_le unk3;        // in
     };
     static_assert(sizeof(IoctlAllocGpfifoEx2) == 32, "IoctlAllocGpfifoEx2 is incorrect size");
 
@@ -153,10 +151,16 @@ private:
     struct IoctlSubmitGpfifo {
         u64_le address;     // pointer to gpfifo entry structs
         u32_le num_entries; // number of fence objects being submitted
-        u32_le flags;
-        IoctlFence fence_out; // returned new fence object for others to wait on
-    };
-    static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(IoctlFence),
+        union {
+            u32_le raw;
+            BitField<0, 1, u32_le> add_wait;      // append a wait sync_point to the list
+            BitField<1, 1, u32_le> add_increment; // append an increment to the list
+            BitField<2, 1, u32_le> new_hw_format; // Mostly ignored
+            BitField<8, 1, u32_le> increment;     // increment the returned fence
+        } flags;
+        Fence fence_out; // returned new fence object for others to wait on
+    };
+    static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(Fence),
                   "IoctlSubmitGpfifo is incorrect size");
 
     struct IoctlGetWaitbase {
@@ -184,6 +188,7 @@ private:
     u32 ChannelSetTimeout(const std::vector<u8>& input, std::vector<u8>& output);
 
     std::shared_ptr<nvmap> nvmap_dev;
+    u32 assigned_syncpoints{};
 };
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
index f5e8ea7c3..f572ad30f 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvdec::nvhost_nvdec() = default;
+nvhost_nvdec::nvhost_nvdec(Core::System& system) : nvdevice(system) {}
 nvhost_nvdec::~nvhost_nvdec() = default;
 
-u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
index 0e7b284f8..2710f0511 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_nvdec final : public nvdevice {
 public:
-    nvhost_nvdec();
+    explicit nvhost_nvdec(Core::System& system);
     ~nvhost_nvdec() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
index 3e0951ab0..38282956f 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp
@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvjpg::nvhost_nvjpg() = default;
+nvhost_nvjpg::nvhost_nvjpg(Core::System& system) : nvdevice(system) {}
 nvhost_nvjpg::~nvhost_nvjpg() = default;
 
-u32 nvhost_nvjpg::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_nvjpg::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
index 89fd5e95e..379766693 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h
@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_nvjpg final : public nvdevice {
 public:
-    nvhost_nvjpg();
+    explicit nvhost_nvjpg(Core::System& system);
     ~nvhost_nvjpg() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
index d544f0f31..70e8091db 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_vic::nvhost_vic() = default;
+nvhost_vic::nvhost_vic(Core::System& system) : nvdevice(system) {}
 nvhost_vic::~nvhost_vic() = default;
 
-u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.h b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
index fc24c3f9c..7d111977e 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_vic final : public nvdevice {
 public:
-    nvhost_vic();
+    explicit nvhost_vic(Core::System& system);
     ~nvhost_vic() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
diff --git a/src/core/hle/service/nvdrv/devices/nvmap.cpp b/src/core/hle/service/nvdrv/devices/nvmap.cpp
index 1ec796fc6..223b496b7 100644
--- a/src/core/hle/service/nvdrv/devices/nvmap.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvmap.cpp
@@ -18,7 +18,7 @@ enum {
 };
 }
 
-nvmap::nvmap() = default;
+nvmap::nvmap(Core::System& system) : nvdevice(system) {}
 nvmap::~nvmap() = default;
 
 VAddr nvmap::GetObjectAddress(u32 handle) const {
@@ -28,7 +28,8 @@ VAddr nvmap::GetObjectAddress(u32 handle) const {
     return object->addr;
 }
 
-u32 nvmap::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvmap::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                 IoctlCtrl& ctrl) {
     switch (static_cast<IoctlCommand>(command.raw)) {
     case IoctlCommand::Create:
         return IocCreate(input, output);
diff --git a/src/core/hle/service/nvdrv/devices/nvmap.h b/src/core/hle/service/nvdrv/devices/nvmap.h
index 396230c19..bf4a101c2 100644
--- a/src/core/hle/service/nvdrv/devices/nvmap.h
+++ b/src/core/hle/service/nvdrv/devices/nvmap.h
@@ -16,13 +16,14 @@ namespace Service::Nvidia::Devices {
 
 class nvmap final : public nvdevice {
 public:
-    nvmap();
+    explicit nvmap(Core::System& system);
     ~nvmap() override;
 
     /// Returns the allocated address of an nvmap object given its handle.
     VAddr GetObjectAddress(u32 handle) const;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
     /// Represents an nvmap object.
     struct Object {
diff --git a/src/core/hle/service/nvdrv/interface.cpp b/src/core/hle/service/nvdrv/interface.cpp
index b60fc748b..d5be64ed2 100644
--- a/src/core/hle/service/nvdrv/interface.cpp
+++ b/src/core/hle/service/nvdrv/interface.cpp
@@ -8,12 +8,18 @@
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/thread.h"
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/interface.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 
 namespace Service::Nvidia {
 
+void NVDRV::SignalGPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) {
+    nvdrv->SignalSyncpt(syncpoint_id, value);
+}
+
 void NVDRV::Open(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_NVDRV, "called");
 
@@ -36,11 +42,31 @@ void NVDRV::Ioctl(Kernel::HLERequestContext& ctx) {
 
     std::vector<u8> output(ctx.GetWriteBufferSize());
 
+    IoctlCtrl ctrl{};
+
+    u32 result = nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output, ctrl);
+
+    if (ctrl.must_delay) {
+        ctrl.fresh_call = false;
+        ctx.SleepClientThread(
+            "NVServices::DelayedResponse", ctrl.timeout,
+            [=](Kernel::SharedPtr<Kernel::Thread> thread, Kernel::HLERequestContext& ctx,
+                Kernel::ThreadWakeupReason reason) {
+                IoctlCtrl ctrl2{ctrl};
+                std::vector<u8> output2 = output;
+                u32 result = nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output2, ctrl2);
+                ctx.WriteBuffer(output2);
+                IPC::ResponseBuilder rb{ctx, 3};
+                rb.Push(RESULT_SUCCESS);
+                rb.Push(result);
+            },
+            nvdrv->GetEventWriteable(ctrl.event_id));
+    } else {
+        ctx.WriteBuffer(output);
+    }
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output));
-
-    ctx.WriteBuffer(output);
+    rb.Push(result);
 }
 
 void NVDRV::Close(Kernel::HLERequestContext& ctx) {
@@ -66,13 +92,19 @@ void NVDRV::Initialize(Kernel::HLERequestContext& ctx) {
 void NVDRV::QueryEvent(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     u32 fd = rp.Pop<u32>();
-    u32 event_id = rp.Pop<u32>();
+    // TODO(Blinkhawk): Figure the meaning of the flag at bit 16
+    u32 event_id = rp.Pop<u32>() & 0x000000FF;
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, fd={:X}, event_id={:X}", fd, event_id);
 
     IPC::ResponseBuilder rb{ctx, 3, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushCopyObjects(query_event.readable);
-    rb.Push<u32>(0);
+    if (event_id < MaxNvEvents) {
+        rb.PushCopyObjects(nvdrv->GetEvent(event_id));
+        rb.Push<u32>(NvResult::Success);
+    } else {
+        rb.Push<u32>(0);
+        rb.Push<u32>(NvResult::BadParameter);
+    }
 }
 
 void NVDRV::SetClientPID(Kernel::HLERequestContext& ctx) {
@@ -127,10 +159,6 @@ NVDRV::NVDRV(std::shared_ptr<Module> nvdrv, const char* name)
         {13, &NVDRV::FinishInitialize, "FinishInitialize"},
     };
     RegisterHandlers(functions);
-
-    auto& kernel = Core::System::GetInstance().Kernel();
-    query_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Automatic,
-                                                         "NVDRV::query_event");
 }
 
 NVDRV::~NVDRV() = default;
diff --git a/src/core/hle/service/nvdrv/interface.h b/src/core/hle/service/nvdrv/interface.h
index 5b4889910..10a0ecd52 100644
--- a/src/core/hle/service/nvdrv/interface.h
+++ b/src/core/hle/service/nvdrv/interface.h
@@ -19,6 +19,8 @@ public:
     NVDRV(std::shared_ptr<Module> nvdrv, const char* name);
     ~NVDRV() override;
 
+    void SignalGPUInterruptSyncpt(const u32 syncpoint_id, const u32 value);
+
 private:
     void Open(Kernel::HLERequestContext& ctx);
     void Ioctl(Kernel::HLERequestContext& ctx);
@@ -33,8 +35,6 @@ private:
     std::shared_ptr<Module> nvdrv;
 
     u64 pid{};
-
-    Kernel::EventPair query_event;
 };
 
 } // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvdrv/nvdata.h b/src/core/hle/service/nvdrv/nvdata.h
new file mode 100644
index 000000000..ac03cbc23
--- /dev/null
+++ b/src/core/hle/service/nvdrv/nvdata.h
@@ -0,0 +1,48 @@
+#pragma once
+
+#include <array>
+#include "common/common_types.h"
+
+namespace Service::Nvidia {
+
+constexpr u32 MaxSyncPoints = 192;
+constexpr u32 MaxNvEvents = 64;
+
+struct Fence {
+    s32 id;
+    u32 value;
+};
+
+static_assert(sizeof(Fence) == 8, "Fence has wrong size");
+
+struct MultiFence {
+    u32 num_fences;
+    std::array<Fence, 4> fences;
+};
+
+enum NvResult : u32 {
+    Success = 0,
+    BadParameter = 4,
+    Timeout = 5,
+    ResourceError = 15,
+};
+
+enum class EventState {
+    Free = 0,
+    Registered = 1,
+    Waiting = 2,
+    Busy = 3,
+};
+
+struct IoctlCtrl {
+    // First call done to the servioce for services that call itself again after a call.
+    bool fresh_call{true};
+    // Tells the Ioctl Wrapper that it must delay the IPC response and send the thread to sleep
+    bool must_delay{};
+    // Timeout for the delay
+    s64 timeout{};
+    // NV Event Id
+    s32 event_id{-1};
+};
+
+} // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvdrv/nvdrv.cpp b/src/core/hle/service/nvdrv/nvdrv.cpp
index 6e4b8f2c6..2011a226a 100644
--- a/src/core/hle/service/nvdrv/nvdrv.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv.cpp
@@ -4,7 +4,10 @@
 
 #include <utility>
 
+#include <fmt/format.h>
 #include "core/hle/ipc_helpers.h"
+#include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
 #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h"
 #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h"
@@ -22,8 +25,9 @@
 
 namespace Service::Nvidia {
 
-void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger) {
-    auto module_ = std::make_shared<Module>();
+void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger,
+                       Core::System& system) {
+    auto module_ = std::make_shared<Module>(system);
     std::make_shared<NVDRV>(module_, "nvdrv")->InstallAsService(service_manager);
     std::make_shared<NVDRV>(module_, "nvdrv:a")->InstallAsService(service_manager);
     std::make_shared<NVDRV>(module_, "nvdrv:s")->InstallAsService(service_manager);
@@ -32,17 +36,25 @@ void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger
     nvflinger.SetNVDrvInstance(module_);
 }
 
-Module::Module() {
-    auto nvmap_dev = std::make_shared<Devices::nvmap>();
-    devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(nvmap_dev);
-    devices["/dev/nvhost-gpu"] = std::make_shared<Devices::nvhost_gpu>(nvmap_dev);
-    devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>();
+Module::Module(Core::System& system) {
+    auto& kernel = system.Kernel();
+    for (u32 i = 0; i < MaxNvEvents; i++) {
+        std::string event_label = fmt::format("NVDRV::NvEvent_{}", i);
+        events_interface.events[i] = Kernel::WritableEvent::CreateEventPair(
+            kernel, Kernel::ResetType::Automatic, event_label);
+        events_interface.status[i] = EventState::Free;
+        events_interface.registered[i] = false;
+    }
+    auto nvmap_dev = std::make_shared<Devices::nvmap>(system);
+    devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(system, nvmap_dev);
+    devices["/dev/nvhost-gpu"] = std::make_shared<Devices::nvhost_gpu>(system, nvmap_dev);
+    devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>(system);
     devices["/dev/nvmap"] = nvmap_dev;
-    devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(nvmap_dev);
-    devices["/dev/nvhost-ctrl"] = std::make_shared<Devices::nvhost_ctrl>();
-    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>();
-    devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>();
-    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>();
+    devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(system, nvmap_dev);
+    devices["/dev/nvhost-ctrl"] = std::make_shared<Devices::nvhost_ctrl>(system, events_interface);
+    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system);
+    devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system);
+    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system);
 }
 
 Module::~Module() = default;
@@ -59,12 +71,13 @@ u32 Module::Open(const std::string& device_name) {
     return fd;
 }
 
-u32 Module::Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 Module::Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output,
+                  IoctlCtrl& ctrl) {
     auto itr = open_files.find(fd);
     ASSERT_MSG(itr != open_files.end(), "Tried to talk to an invalid device");
 
     auto& device = itr->second;
-    return device->ioctl({command}, input, output);
+    return device->ioctl({command}, input, output, ctrl);
 }
 
 ResultCode Module::Close(u32 fd) {
@@ -77,4 +90,22 @@ ResultCode Module::Close(u32 fd) {
     return RESULT_SUCCESS;
 }
 
+void Module::SignalSyncpt(const u32 syncpoint_id, const u32 value) {
+    for (u32 i = 0; i < MaxNvEvents; i++) {
+        if (events_interface.assigned_syncpt[i] == syncpoint_id &&
+            events_interface.assigned_value[i] == value) {
+            events_interface.LiberateEvent(i);
+            events_interface.events[i].writable->Signal();
+        }
+    }
+}
+
+Kernel::SharedPtr<Kernel::ReadableEvent> Module::GetEvent(const u32 event_id) const {
+    return events_interface.events[event_id].readable;
+}
+
+Kernel::SharedPtr<Kernel::WritableEvent> Module::GetEventWriteable(const u32 event_id) const {
+    return events_interface.events[event_id].writable;
+}
+
 } // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvdrv/nvdrv.h b/src/core/hle/service/nvdrv/nvdrv.h
index 53564f696..a339ab672 100644
--- a/src/core/hle/service/nvdrv/nvdrv.h
+++ b/src/core/hle/service/nvdrv/nvdrv.h
@@ -8,8 +8,14 @@
 #include <unordered_map>
 #include <vector>
 #include "common/common_types.h"
+#include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/service.h"
 
+namespace Core {
+class System;
+}
+
 namespace Service::NVFlinger {
 class NVFlinger;
 }
@@ -20,16 +26,72 @@ namespace Devices {
 class nvdevice;
 }
 
-struct IoctlFence {
-    u32 id;
-    u32 value;
+struct EventInterface {
+    // Mask representing currently busy events
+    u64 events_mask{};
+    // Each kernel event associated to an NV event
+    std::array<Kernel::EventPair, MaxNvEvents> events;
+    // The status of the current NVEvent
+    std::array<EventState, MaxNvEvents> status{};
+    // Tells if an NVEvent is registered or not
+    std::array<bool, MaxNvEvents> registered{};
+    // When an NVEvent is waiting on GPU interrupt, this is the sync_point
+    // associated with it.
+    std::array<u32, MaxNvEvents> assigned_syncpt{};
+    // This is the value of the GPU interrupt for which the NVEvent is waiting
+    // for.
+    std::array<u32, MaxNvEvents> assigned_value{};
+    // Constant to denote an unasigned syncpoint.
+    static constexpr u32 unassigned_syncpt = 0xFFFFFFFF;
+    std::optional<u32> GetFreeEvent() const {
+        u64 mask = events_mask;
+        for (u32 i = 0; i < MaxNvEvents; i++) {
+            const bool is_free = (mask & 0x1) == 0;
+            if (is_free) {
+                if (status[i] == EventState::Registered || status[i] == EventState::Free) {
+                    return {i};
+                }
+            }
+            mask = mask >> 1;
+        }
+        return {};
+    }
+    void SetEventStatus(const u32 event_id, EventState new_status) {
+        EventState old_status = status[event_id];
+        if (old_status == new_status) {
+            return;
+        }
+        status[event_id] = new_status;
+        if (new_status == EventState::Registered) {
+            registered[event_id] = true;
+        }
+        if (new_status == EventState::Waiting || new_status == EventState::Busy) {
+            events_mask |= (1ULL << event_id);
+        }
+    }
+    void RegisterEvent(const u32 event_id) {
+        registered[event_id] = true;
+        if (status[event_id] == EventState::Free) {
+            status[event_id] = EventState::Registered;
+        }
+    }
+    void UnregisterEvent(const u32 event_id) {
+        registered[event_id] = false;
+        if (status[event_id] == EventState::Registered) {
+            status[event_id] = EventState::Free;
+        }
+    }
+    void LiberateEvent(const u32 event_id) {
+        status[event_id] = registered[event_id] ? EventState::Registered : EventState::Free;
+        events_mask &= ~(1ULL << event_id);
+        assigned_syncpt[event_id] = unassigned_syncpt;
+        assigned_value[event_id] = 0;
+    }
 };
 
-static_assert(sizeof(IoctlFence) == 8, "IoctlFence has wrong size");
-
 class Module final {
 public:
-    Module();
+    Module(Core::System& system);
     ~Module();
 
     /// Returns a pointer to one of the available devices, identified by its name.
@@ -44,10 +106,17 @@ public:
     /// Opens a device node and returns a file descriptor to it.
     u32 Open(const std::string& device_name);
     /// Sends an ioctl command to the specified file descriptor.
-    u32 Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output);
+    u32 Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl);
     /// Closes a device file descriptor and returns operation success.
     ResultCode Close(u32 fd);
 
+    void SignalSyncpt(const u32 syncpoint_id, const u32 value);
+
+    Kernel::SharedPtr<Kernel::ReadableEvent> GetEvent(u32 event_id) const;
+
+    Kernel::SharedPtr<Kernel::WritableEvent> GetEventWriteable(u32 event_id) const;
+
 private:
     /// Id to use for the next open file descriptor.
     u32 next_fd = 1;
@@ -57,9 +126,12 @@ private:
 
     /// Mapping of device node names to their implementation.
     std::unordered_map<std::string, std::shared_ptr<Devices::nvdevice>> devices;
+
+    EventInterface events_interface;
 };
 
 /// Registers all NVDRV services with the specified service manager.
-void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger);
+void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger,
+                       Core::System& system);
 
 } // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvflinger/buffer_queue.cpp b/src/core/hle/service/nvflinger/buffer_queue.cpp
index 5731e815f..e1a07d3ee 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.cpp
+++ b/src/core/hle/service/nvflinger/buffer_queue.cpp
@@ -34,7 +34,8 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
     buffer_wait_event.writable->Signal();
 }
 
-std::optional<u32> BufferQueue::DequeueBuffer(u32 width, u32 height) {
+std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::DequeueBuffer(u32 width,
+                                                                                       u32 height) {
     auto itr = std::find_if(queue.begin(), queue.end(), [&](const Buffer& buffer) {
         // Only consider free buffers. Buffers become free once again after they've been Acquired
         // and Released by the compositor, see the NVFlinger::Compose method.
@@ -51,7 +52,7 @@ std::optional<u32> BufferQueue::DequeueBuffer(u32 width, u32 height) {
     }
 
     itr->status = Buffer::Status::Dequeued;
-    return itr->slot;
+    return {{itr->slot, &itr->multi_fence}};
 }
 
 const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
@@ -63,7 +64,8 @@ const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
 }
 
 void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
-                              const Common::Rectangle<int>& crop_rect) {
+                              const Common::Rectangle<int>& crop_rect, u32 swap_interval,
+                              Service::Nvidia::MultiFence& multi_fence) {
     auto itr = std::find_if(queue.begin(), queue.end(),
                             [&](const Buffer& buffer) { return buffer.slot == slot; });
     ASSERT(itr != queue.end());
@@ -71,12 +73,21 @@ void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
     itr->status = Buffer::Status::Queued;
     itr->transform = transform;
     itr->crop_rect = crop_rect;
+    itr->swap_interval = swap_interval;
+    itr->multi_fence = multi_fence;
+    queue_sequence.push_back(slot);
 }
 
 std::optional<std::reference_wrapper<const BufferQueue::Buffer>> BufferQueue::AcquireBuffer() {
-    auto itr = std::find_if(queue.begin(), queue.end(), [](const Buffer& buffer) {
-        return buffer.status == Buffer::Status::Queued;
-    });
+    auto itr = queue.end();
+    // Iterate to find a queued buffer matching the requested slot.
+    while (itr == queue.end() && !queue_sequence.empty()) {
+        u32 slot = queue_sequence.front();
+        itr = std::find_if(queue.begin(), queue.end(), [&slot](const Buffer& buffer) {
+            return buffer.status == Buffer::Status::Queued && buffer.slot == slot;
+        });
+        queue_sequence.pop_front();
+    }
     if (itr == queue.end())
         return {};
     itr->status = Buffer::Status::Acquired;
diff --git a/src/core/hle/service/nvflinger/buffer_queue.h b/src/core/hle/service/nvflinger/buffer_queue.h
index e1ccb6171..356bedb81 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.h
+++ b/src/core/hle/service/nvflinger/buffer_queue.h
@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <list>
 #include <optional>
 #include <vector>
 
@@ -12,6 +13,7 @@
 #include "common/swap.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 
 namespace Service::NVFlinger {
 
@@ -68,13 +70,17 @@ public:
         IGBPBuffer igbp_buffer;
         BufferTransformFlags transform;
         Common::Rectangle<int> crop_rect;
+        u32 swap_interval;
+        Service::Nvidia::MultiFence multi_fence;
     };
 
     void SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer);
-    std::optional<u32> DequeueBuffer(u32 width, u32 height);
+    std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> DequeueBuffer(u32 width,
+                                                                              u32 height);
     const IGBPBuffer& RequestBuffer(u32 slot) const;
     void QueueBuffer(u32 slot, BufferTransformFlags transform,
-                     const Common::Rectangle<int>& crop_rect);
+                     const Common::Rectangle<int>& crop_rect, u32 swap_interval,
+                     Service::Nvidia::MultiFence& multi_fence);
     std::optional<std::reference_wrapper<const Buffer>> AcquireBuffer();
     void ReleaseBuffer(u32 slot);
     u32 Query(QueryType type);
@@ -92,6 +98,7 @@ private:
     u64 layer_id;
 
     std::vector<Buffer> queue;
+    std::list<u32> queue_sequence;
     Kernel::EventPair buffer_wait_event;
 };
 
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index 3c5c53e24..f9db79370 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -37,15 +37,14 @@ NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing) : core_timing{core_t
     displays.emplace_back(4, "Null");
 
     // Schedule the screen composition events
-    const auto ticks = Settings::values.force_30fps_mode ? frame_ticks_30fps : frame_ticks;
-
-    composition_event = core_timing.RegisterEvent(
-        "ScreenComposition", [this, ticks](u64 userdata, s64 cycles_late) {
-            Compose();
-            this->core_timing.ScheduleEvent(ticks - cycles_late, composition_event);
-        });
-
-    core_timing.ScheduleEvent(ticks, composition_event);
+    composition_event = core_timing.RegisterEvent("ScreenComposition", [this](u64 userdata,
+                                                                              s64 cycles_late) {
+        Compose();
+        const auto ticks = Settings::values.force_30fps_mode ? frame_ticks_30fps : GetNextTicks();
+        this->core_timing.ScheduleEvent(std::max<s64>(0LL, ticks - cycles_late), composition_event);
+    });
+
+    core_timing.ScheduleEvent(frame_ticks, composition_event);
 }
 
 NVFlinger::~NVFlinger() {
@@ -206,8 +205,14 @@ void NVFlinger::Compose() {
                      igbp_buffer.width, igbp_buffer.height, igbp_buffer.stride,
                      buffer->get().transform, buffer->get().crop_rect);
 
+        swap_interval = buffer->get().swap_interval;
         buffer_queue.ReleaseBuffer(buffer->get().slot);
     }
 }
 
+s64 NVFlinger::GetNextTicks() const {
+    constexpr s64 max_hertz = 120LL;
+    return (Core::Timing::BASE_CLOCK_RATE * (1LL << swap_interval)) / max_hertz;
+}
+
 } // namespace Service::NVFlinger
diff --git a/src/core/hle/service/nvflinger/nvflinger.h b/src/core/hle/service/nvflinger/nvflinger.h
index c0a83fffb..988be8726 100644
--- a/src/core/hle/service/nvflinger/nvflinger.h
+++ b/src/core/hle/service/nvflinger/nvflinger.h
@@ -74,6 +74,8 @@ public:
     /// finished.
     void Compose();
 
+    s64 GetNextTicks() const;
+
 private:
     /// Finds the display identified by the specified ID.
     VI::Display* FindDisplay(u64 display_id);
@@ -98,6 +100,8 @@ private:
     /// layers.
     u32 next_buffer_queue_id = 1;
 
+    u32 swap_interval = 1;
+
     /// Event that handles screen composition.
     Core::Timing::EventType* composition_event;
 
diff --git a/src/core/hle/service/pm/pm.cpp b/src/core/hle/service/pm/pm.cpp
index ebcc41a43..fe6b5f798 100644
--- a/src/core/hle/service/pm/pm.cpp
+++ b/src/core/hle/service/pm/pm.cpp
@@ -3,11 +3,44 @@
 // Refer to the license.txt file included.
 
 #include "core/hle/ipc_helpers.h"
+#include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/process.h"
 #include "core/hle/service/pm/pm.h"
 #include "core/hle/service/service.h"
 
 namespace Service::PM {
 
+namespace {
+
+constexpr ResultCode ERROR_PROCESS_NOT_FOUND{ErrorModule::PM, 1};
+
+constexpr u64 NO_PROCESS_FOUND_PID{0};
+
+std::optional<Kernel::SharedPtr<Kernel::Process>> SearchProcessList(
+    const std::vector<Kernel::SharedPtr<Kernel::Process>>& process_list,
+    std::function<bool(const Kernel::SharedPtr<Kernel::Process>&)> predicate) {
+    const auto iter = std::find_if(process_list.begin(), process_list.end(), predicate);
+
+    if (iter == process_list.end()) {
+        return std::nullopt;
+    }
+
+    return *iter;
+}
+
+void GetApplicationPidGeneric(Kernel::HLERequestContext& ctx,
+                              const std::vector<Kernel::SharedPtr<Kernel::Process>>& process_list) {
+    const auto process = SearchProcessList(process_list, [](const auto& process) {
+        return process->GetProcessID() == Kernel::Process::ProcessIDMin;
+    });
+
+    IPC::ResponseBuilder rb{ctx, 4};
+    rb.Push(RESULT_SUCCESS);
+    rb.Push(process.has_value() ? (*process)->GetProcessID() : NO_PROCESS_FOUND_PID);
+}
+
+} // Anonymous namespace
+
 class BootMode final : public ServiceFramework<BootMode> {
 public:
     explicit BootMode() : ServiceFramework{"pm:bm"} {
@@ -41,14 +74,15 @@ private:
 
 class DebugMonitor final : public ServiceFramework<DebugMonitor> {
 public:
-    explicit DebugMonitor() : ServiceFramework{"pm:dmnt"} {
+    explicit DebugMonitor(const Kernel::KernelCore& kernel)
+        : ServiceFramework{"pm:dmnt"}, kernel(kernel) {
         // clang-format off
         static const FunctionInfo functions[] = {
             {0, nullptr, "GetDebugProcesses"},
             {1, nullptr, "StartDebugProcess"},
-            {2, nullptr, "GetTitlePid"},
+            {2, &DebugMonitor::GetTitlePid, "GetTitlePid"},
             {3, nullptr, "EnableDebugForTitleId"},
-            {4, nullptr, "GetApplicationPid"},
+            {4, &DebugMonitor::GetApplicationPid, "GetApplicationPid"},
             {5, nullptr, "EnableDebugForApplication"},
             {6, nullptr, "DisableDebug"},
         };
@@ -56,21 +90,77 @@ public:
 
         RegisterHandlers(functions);
     }
+
+private:
+    void GetTitlePid(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto title_id = rp.PopRaw<u64>();
+
+        LOG_DEBUG(Service_PM, "called, title_id={:016X}", title_id);
+
+        const auto process =
+            SearchProcessList(kernel.GetProcessList(), [title_id](const auto& process) {
+                return process->GetTitleID() == title_id;
+            });
+
+        if (!process.has_value()) {
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERROR_PROCESS_NOT_FOUND);
+            return;
+        }
+
+        IPC::ResponseBuilder rb{ctx, 4};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push((*process)->GetProcessID());
+    }
+
+    void GetApplicationPid(Kernel::HLERequestContext& ctx) {
+        LOG_DEBUG(Service_PM, "called");
+        GetApplicationPidGeneric(ctx, kernel.GetProcessList());
+    }
+
+    const Kernel::KernelCore& kernel;
 };
 
 class Info final : public ServiceFramework<Info> {
 public:
-    explicit Info() : ServiceFramework{"pm:info"} {
+    explicit Info(const std::vector<Kernel::SharedPtr<Kernel::Process>>& process_list)
+        : ServiceFramework{"pm:info"}, process_list(process_list) {
         static const FunctionInfo functions[] = {
-            {0, nullptr, "GetTitleId"},
+            {0, &Info::GetTitleId, "GetTitleId"},
         };
         RegisterHandlers(functions);
     }
+
+private:
+    void GetTitleId(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto process_id = rp.PopRaw<u64>();
+
+        LOG_DEBUG(Service_PM, "called, process_id={:016X}", process_id);
+
+        const auto process = SearchProcessList(process_list, [process_id](const auto& process) {
+            return process->GetProcessID() == process_id;
+        });
+
+        if (!process.has_value()) {
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERROR_PROCESS_NOT_FOUND);
+            return;
+        }
+
+        IPC::ResponseBuilder rb{ctx, 4};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push((*process)->GetTitleID());
+    }
+
+    const std::vector<Kernel::SharedPtr<Kernel::Process>>& process_list;
 };
 
 class Shell final : public ServiceFramework<Shell> {
 public:
-    explicit Shell() : ServiceFramework{"pm:shell"} {
+    explicit Shell(const Kernel::KernelCore& kernel)
+        : ServiceFramework{"pm:shell"}, kernel(kernel) {
         // clang-format off
         static const FunctionInfo functions[] = {
             {0, nullptr, "LaunchProcess"},
@@ -79,21 +169,31 @@ public:
             {3, nullptr, "GetProcessEventWaiter"},
             {4, nullptr, "GetProcessEventType"},
             {5, nullptr, "NotifyBootFinished"},
-            {6, nullptr, "GetApplicationPid"},
+            {6, &Shell::GetApplicationPid, "GetApplicationPid"},
             {7, nullptr, "BoostSystemMemoryResourceLimit"},
             {8, nullptr, "EnableAdditionalSystemThreads"},
+            {9, nullptr, "GetUnimplementedEventHandle"},
         };
         // clang-format on
 
         RegisterHandlers(functions);
     }
+
+private:
+    void GetApplicationPid(Kernel::HLERequestContext& ctx) {
+        LOG_DEBUG(Service_PM, "called");
+        GetApplicationPidGeneric(ctx, kernel.GetProcessList());
+    }
+
+    const Kernel::KernelCore& kernel;
 };
 
-void InstallInterfaces(SM::ServiceManager& sm) {
-    std::make_shared<BootMode>()->InstallAsService(sm);
-    std::make_shared<DebugMonitor>()->InstallAsService(sm);
-    std::make_shared<Info>()->InstallAsService(sm);
-    std::make_shared<Shell>()->InstallAsService(sm);
+void InstallInterfaces(Core::System& system) {
+    std::make_shared<BootMode>()->InstallAsService(system.ServiceManager());
+    std::make_shared<DebugMonitor>(system.Kernel())->InstallAsService(system.ServiceManager());
+    std::make_shared<Info>(system.Kernel().GetProcessList())
+        ->InstallAsService(system.ServiceManager());
+    std::make_shared<Shell>(system.Kernel())->InstallAsService(system.ServiceManager());
 }
 
 } // namespace Service::PM
diff --git a/src/core/hle/service/pm/pm.h b/src/core/hle/service/pm/pm.h
index cc8d3f215..852e7050c 100644
--- a/src/core/hle/service/pm/pm.h
+++ b/src/core/hle/service/pm/pm.h
@@ -4,8 +4,8 @@
 
 #pragma once
 
-namespace Service::SM {
-class ServiceManager;
+namespace Core {
+class System;
 }
 
 namespace Service::PM {
@@ -16,6 +16,6 @@ enum class SystemBootMode {
 };
 
 /// Registers all PM services with the specified service manager.
-void InstallInterfaces(SM::ServiceManager& service_manager);
+void InstallInterfaces(Core::System& system);
 
 } // namespace Service::PM
diff --git a/src/core/hle/service/service.cpp b/src/core/hle/service/service.cpp
index 952c03e27..3a0f8c3f6 100644
--- a/src/core/hle/service/service.cpp
+++ b/src/core/hle/service/service.cpp
@@ -206,7 +206,7 @@ void Init(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system) {
     AM::InstallInterfaces(*sm, nv_flinger, system);
     AOC::InstallInterfaces(*sm);
     APM::InstallInterfaces(system);
-    Audio::InstallInterfaces(*sm);
+    Audio::InstallInterfaces(*sm, system);
     BCAT::InstallInterfaces(*sm);
     BPC::InstallInterfaces(*sm);
     BtDrv::InstallInterfaces(*sm);
@@ -236,12 +236,12 @@ void Init(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system) {
     NIM::InstallInterfaces(*sm);
     NPNS::InstallInterfaces(*sm);
     NS::InstallInterfaces(*sm);
-    Nvidia::InstallInterfaces(*sm, *nv_flinger);
+    Nvidia::InstallInterfaces(*sm, *nv_flinger, system);
     PCIe::InstallInterfaces(*sm);
     PCTL::InstallInterfaces(*sm);
     PCV::InstallInterfaces(*sm);
     PlayReport::InstallInterfaces(*sm);
-    PM::InstallInterfaces(*sm);
+    PM::InstallInterfaces(system);
     PSC::InstallInterfaces(*sm);
     PSM::InstallInterfaces(*sm);
     Set::InstallInterfaces(*sm);
diff --git a/src/core/hle/service/vi/vi.cpp b/src/core/hle/service/vi/vi.cpp
index f1fa6ccd1..199b30635 100644
--- a/src/core/hle/service/vi/vi.cpp
+++ b/src/core/hle/service/vi/vi.cpp
@@ -21,6 +21,7 @@
 #include "core/hle/kernel/readable_event.h"
 #include "core/hle/kernel/thread.h"
 #include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "core/hle/service/nvflinger/nvflinger.h"
@@ -328,32 +329,22 @@ public:
     Data data;
 };
 
-struct BufferProducerFence {
-    u32 is_valid;
-    std::array<Nvidia::IoctlFence, 4> fences;
-};
-static_assert(sizeof(BufferProducerFence) == 36, "BufferProducerFence has wrong size");
-
 class IGBPDequeueBufferResponseParcel : public Parcel {
 public:
-    explicit IGBPDequeueBufferResponseParcel(u32 slot) : slot(slot) {}
+    explicit IGBPDequeueBufferResponseParcel(u32 slot, Service::Nvidia::MultiFence& multi_fence)
+        : slot(slot), multi_fence(multi_fence) {}
     ~IGBPDequeueBufferResponseParcel() override = default;
 
 protected:
     void SerializeData() override {
-        // TODO(Subv): Find out how this Fence is used.
-        BufferProducerFence fence = {};
-        fence.is_valid = 1;
-        for (auto& fence_ : fence.fences)
-            fence_.id = -1;
-
         Write(slot);
         Write<u32_le>(1);
-        WriteObject(fence);
+        WriteObject(multi_fence);
         Write<u32_le>(0);
     }
 
     u32_le slot;
+    Service::Nvidia::MultiFence multi_fence;
 };
 
 class IGBPRequestBufferRequestParcel : public Parcel {
@@ -400,12 +391,6 @@ public:
         data = Read<Data>();
     }
 
-    struct Fence {
-        u32_le id;
-        u32_le value;
-    };
-    static_assert(sizeof(Fence) == 8, "Fence has wrong size");
-
     struct Data {
         u32_le slot;
         INSERT_PADDING_WORDS(3);
@@ -418,15 +403,15 @@ public:
         s32_le scaling_mode;
         NVFlinger::BufferQueue::BufferTransformFlags transform;
         u32_le sticky_transform;
-        INSERT_PADDING_WORDS(2);
-        u32_le fence_is_valid;
-        std::array<Fence, 2> fences;
+        INSERT_PADDING_WORDS(1);
+        u32_le swap_interval;
+        Service::Nvidia::MultiFence multi_fence;
 
         Common::Rectangle<int> GetCropRect() const {
             return {crop_left, crop_top, crop_right, crop_bottom};
         }
     };
-    static_assert(sizeof(Data) == 80, "ParcelData has wrong size");
+    static_assert(sizeof(Data) == 96, "ParcelData has wrong size");
 
     Data data;
 };
@@ -547,11 +532,11 @@ private:
             IGBPDequeueBufferRequestParcel request{ctx.ReadBuffer()};
             const u32 width{request.data.width};
             const u32 height{request.data.height};
-            std::optional<u32> slot = buffer_queue.DequeueBuffer(width, height);
+            auto result = buffer_queue.DequeueBuffer(width, height);
 
-            if (slot) {
+            if (result) {
                 // Buffer is available
-                IGBPDequeueBufferResponseParcel response{*slot};
+                IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                 ctx.WriteBuffer(response.Serialize());
             } else {
                 // Wait the current thread until a buffer becomes available
@@ -561,10 +546,10 @@ private:
                         Kernel::ThreadWakeupReason reason) {
                         // Repeat TransactParcel DequeueBuffer when a buffer is available
                         auto& buffer_queue = nv_flinger->FindBufferQueue(id);
-                        std::optional<u32> slot = buffer_queue.DequeueBuffer(width, height);
-                        ASSERT_MSG(slot != std::nullopt, "Could not dequeue buffer.");
+                        auto result = buffer_queue.DequeueBuffer(width, height);
+                        ASSERT_MSG(result != std::nullopt, "Could not dequeue buffer.");
 
-                        IGBPDequeueBufferResponseParcel response{*slot};
+                        IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                         ctx.WriteBuffer(response.Serialize());
                         IPC::ResponseBuilder rb{ctx, 2};
                         rb.Push(RESULT_SUCCESS);
@@ -582,7 +567,8 @@ private:
             IGBPQueueBufferRequestParcel request{ctx.ReadBuffer()};
 
             buffer_queue.QueueBuffer(request.data.slot, request.data.transform,
-                                     request.data.GetCropRect());
+                                     request.data.GetCropRect(), request.data.swap_interval,
+                                     request.data.multi_fence);
 
             IGBPQueueBufferResponseParcel response{1280, 720};
             ctx.WriteBuffer(response.Serialize());
diff --git a/src/core/loader/elf.cpp b/src/core/loader/elf.cpp
index 6d4b02375..f1795fdd6 100644
--- a/src/core/loader/elf.cpp
+++ b/src/core/loader/elf.cpp
@@ -295,7 +295,7 @@ Kernel::CodeSet ElfReader::LoadInto(VAddr vaddr) {
         }
     }
 
-    std::vector<u8> program_image(total_image_size);
+    Kernel::PhysicalMemory program_image(total_image_size);
     std::size_t current_image_position = 0;
 
     Kernel::CodeSet codeset;
diff --git a/src/core/loader/kip.cpp b/src/core/loader/kip.cpp
index 70051c13a..474b55cb1 100644
--- a/src/core/loader/kip.cpp
+++ b/src/core/loader/kip.cpp
@@ -69,7 +69,7 @@ AppLoader::LoadResult AppLoader_KIP::Load(Kernel::Process& process) {
 
     const VAddr base_address = process.VMManager().GetCodeRegionBaseAddress();
     Kernel::CodeSet codeset;
-    std::vector<u8> program_image;
+    Kernel::PhysicalMemory program_image;
 
     const auto load_segment = [&program_image](Kernel::CodeSet::Segment& segment,
                                                const std::vector<u8>& data, u32 offset) {
diff --git a/src/core/loader/nro.cpp b/src/core/loader/nro.cpp
index 6a0ca389b..e92e2e06e 100644
--- a/src/core/loader/nro.cpp
+++ b/src/core/loader/nro.cpp
@@ -143,7 +143,7 @@ static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
     }
 
     // Build program image
-    std::vector<u8> program_image(PageAlignSize(nro_header.file_size));
+    Kernel::PhysicalMemory program_image(PageAlignSize(nro_header.file_size));
     std::memcpy(program_image.data(), data.data(), program_image.size());
     if (program_image.size() != PageAlignSize(nro_header.file_size)) {
         return {};
diff --git a/src/core/loader/nso.cpp b/src/core/loader/nso.cpp
index 29311404a..70c90109f 100644
--- a/src/core/loader/nso.cpp
+++ b/src/core/loader/nso.cpp
@@ -89,7 +89,7 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
 
     // Build program image
     Kernel::CodeSet codeset;
-    std::vector<u8> program_image;
+    Kernel::PhysicalMemory program_image;
     for (std::size_t i = 0; i < nso_header.segments.size(); ++i) {
         std::vector<u8> data =
             file.ReadBytes(nso_header.segments_compressed_size[i], nso_header.segments[i].offset);
diff --git a/src/core/settings.cpp b/src/core/settings.cpp
index 63aa59690..0dd1632ac 100644
--- a/src/core/settings.cpp
+++ b/src/core/settings.cpp
@@ -85,7 +85,6 @@ void LogSettings() {
     LogSetting("System_RngSeed", Settings::values.rng_seed.value_or(0));
     LogSetting("System_CurrentUser", Settings::values.current_user);
     LogSetting("System_LanguageIndex", Settings::values.language_index);
-    LogSetting("Core_CpuJitEnabled", Settings::values.cpu_jit_enabled);
     LogSetting("Core_UseMultiCore", Settings::values.use_multi_core);
     LogSetting("Renderer_UseResolutionFactor", Settings::values.resolution_factor);
     LogSetting("Renderer_UseFrameLimit", Settings::values.use_frame_limit);
diff --git a/src/core/settings.h b/src/core/settings.h
index acf18d653..6638ce8f9 100644
--- a/src/core/settings.h
+++ b/src/core/settings.h
@@ -378,7 +378,6 @@ struct Values {
     std::atomic_bool is_device_reload_pending{true};
 
     // Core
-    bool cpu_jit_enabled;
     bool use_multi_core;
 
     // Data Storage
diff --git a/src/core/telemetry_session.cpp b/src/core/telemetry_session.cpp
index 98f49042a..793d102d3 100644
--- a/src/core/telemetry_session.cpp
+++ b/src/core/telemetry_session.cpp
@@ -168,7 +168,6 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader) {
     AddField(Telemetry::FieldType::UserConfig, "Audio_SinkId", Settings::values.sink_id);
     AddField(Telemetry::FieldType::UserConfig, "Audio_EnableAudioStretching",
              Settings::values.enable_audio_stretching);
-    AddField(Telemetry::FieldType::UserConfig, "Core_UseCpuJit", Settings::values.cpu_jit_enabled);
     AddField(Telemetry::FieldType::UserConfig, "Core_UseMultiCore",
              Settings::values.use_multi_core);
     AddField(Telemetry::FieldType::UserConfig, "Renderer_ResolutionFactor",
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 6839abe71..e2f85c5f1 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -1,4 +1,7 @@
 add_library(video_core STATIC
+    buffer_cache/buffer_block.h
+    buffer_cache/buffer_cache.h
+    buffer_cache/map_interval.h
     dma_pusher.cpp
     dma_pusher.h
     debug_utils/debug_utils.cpp
@@ -43,8 +46,6 @@ add_library(video_core STATIC
     renderer_opengl/gl_device.h
     renderer_opengl/gl_framebuffer_cache.cpp
     renderer_opengl/gl_framebuffer_cache.h
-    renderer_opengl/gl_global_cache.cpp
-    renderer_opengl/gl_global_cache.h
     renderer_opengl/gl_rasterizer.cpp
     renderer_opengl/gl_rasterizer.h
     renderer_opengl/gl_resource_manager.cpp
@@ -101,8 +102,11 @@ add_library(video_core STATIC
     shader/decode/integer_set.cpp
     shader/decode/half_set.cpp
     shader/decode/video.cpp
+    shader/decode/warp.cpp
     shader/decode/xmad.cpp
     shader/decode/other.cpp
+    shader/control_flow.cpp
+    shader/control_flow.h
     shader/decode.cpp
     shader/node_helper.cpp
     shader/node_helper.h
diff --git a/src/video_core/buffer_cache/buffer_block.h b/src/video_core/buffer_cache/buffer_block.h
new file mode 100644
index 000000000..4b9193182
--- /dev/null
+++ b/src/video_core/buffer_cache/buffer_block.h
@@ -0,0 +1,76 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <unordered_set>
+#include <utility>
+
+#include "common/alignment.h"
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+
+namespace VideoCommon {
+
+class BufferBlock {
+public:
+    bool Overlaps(const CacheAddr start, const CacheAddr end) const {
+        return (cache_addr < end) && (cache_addr_end > start);
+    }
+
+    bool IsInside(const CacheAddr other_start, const CacheAddr other_end) const {
+        return cache_addr <= other_start && other_end <= cache_addr_end;
+    }
+
+    u8* GetWritableHostPtr() const {
+        return FromCacheAddr(cache_addr);
+    }
+
+    u8* GetWritableHostPtr(std::size_t offset) const {
+        return FromCacheAddr(cache_addr + offset);
+    }
+
+    std::size_t GetOffset(const CacheAddr in_addr) {
+        return static_cast<std::size_t>(in_addr - cache_addr);
+    }
+
+    CacheAddr GetCacheAddr() const {
+        return cache_addr;
+    }
+
+    CacheAddr GetCacheAddrEnd() const {
+        return cache_addr_end;
+    }
+
+    void SetCacheAddr(const CacheAddr new_addr) {
+        cache_addr = new_addr;
+        cache_addr_end = new_addr + size;
+    }
+
+    std::size_t GetSize() const {
+        return size;
+    }
+
+    void SetEpoch(u64 new_epoch) {
+        epoch = new_epoch;
+    }
+
+    u64 GetEpoch() {
+        return epoch;
+    }
+
+protected:
+    explicit BufferBlock(CacheAddr cache_addr, const std::size_t size) : size{size} {
+        SetCacheAddr(cache_addr);
+    }
+    ~BufferBlock() = default;
+
+private:
+    CacheAddr cache_addr{};
+    CacheAddr cache_addr_end{};
+    std::size_t size{};
+    u64 epoch{};
+};
+
+} // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
new file mode 100644
index 000000000..2442ddfd6
--- /dev/null
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -0,0 +1,447 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <mutex>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+#include <vector>
+
+#include "common/alignment.h"
+#include "common/common_types.h"
+#include "core/core.h"
+#include "video_core/buffer_cache/buffer_block.h"
+#include "video_core/buffer_cache/map_interval.h"
+#include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace VideoCommon {
+
+using MapInterval = std::shared_ptr<MapIntervalBase>;
+
+template <typename TBuffer, typename TBufferType, typename StreamBuffer>
+class BufferCache {
+public:
+    using BufferInfo = std::pair<const TBufferType*, u64>;
+
+    BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
+                            bool is_written = false) {
+        std::lock_guard lock{mutex};
+
+        auto& memory_manager = system.GPU().MemoryManager();
+        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
+        if (!host_ptr) {
+            return {GetEmptyBuffer(size), 0};
+        }
+        const auto cache_addr = ToCacheAddr(host_ptr);
+
+        // Cache management is a big overhead, so only cache entries with a given size.
+        // TODO: Figure out which size is the best for given games.
+        constexpr std::size_t max_stream_size = 0x800;
+        if (size < max_stream_size) {
+            if (!is_written && !IsRegionWritten(cache_addr, cache_addr + size - 1)) {
+                return StreamBufferUpload(host_ptr, size, alignment);
+            }
+        }
+
+        auto block = GetBlock(cache_addr, size);
+        auto map = MapAddress(block, gpu_addr, cache_addr, size);
+        if (is_written) {
+            map->MarkAsModified(true, GetModifiedTicks());
+            if (!map->IsWritten()) {
+                map->MarkAsWritten(true);
+                MarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
+            }
+        } else {
+            if (map->IsWritten()) {
+                WriteBarrier();
+            }
+        }
+
+        const u64 offset = static_cast<u64>(block->GetOffset(cache_addr));
+
+        return {ToHandle(block), offset};
+    }
+
+    /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
+    BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
+                                std::size_t alignment = 4) {
+        std::lock_guard lock{mutex};
+        return StreamBufferUpload(raw_pointer, size, alignment);
+    }
+
+    void Map(std::size_t max_size) {
+        std::lock_guard lock{mutex};
+
+        std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
+        buffer_offset = buffer_offset_base;
+    }
+
+    /// Finishes the upload stream, returns true on bindings invalidation.
+    bool Unmap() {
+        std::lock_guard lock{mutex};
+
+        stream_buffer->Unmap(buffer_offset - buffer_offset_base);
+        return std::exchange(invalidated, false);
+    }
+
+    void TickFrame() {
+        ++epoch;
+        while (!pending_destruction.empty()) {
+            if (pending_destruction.front()->GetEpoch() + 1 > epoch) {
+                break;
+            }
+            pending_destruction.pop_front();
+        }
+    }
+
+    /// Write any cached resources overlapping the specified region back to memory
+    void FlushRegion(CacheAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        std::vector<MapInterval> objects = GetMapsInRange(addr, size);
+        std::sort(objects.begin(), objects.end(), [](const MapInterval& a, const MapInterval& b) {
+            return a->GetModificationTick() < b->GetModificationTick();
+        });
+        for (auto& object : objects) {
+            if (object->IsModified() && object->IsRegistered()) {
+                FlushMap(object);
+            }
+        }
+    }
+
+    /// Mark the specified region as being invalidated
+    void InvalidateRegion(CacheAddr addr, u64 size) {
+        std::lock_guard lock{mutex};
+
+        std::vector<MapInterval> objects = GetMapsInRange(addr, size);
+        for (auto& object : objects) {
+            if (object->IsRegistered()) {
+                Unregister(object);
+            }
+        }
+    }
+
+    virtual const TBufferType* GetEmptyBuffer(std::size_t size) = 0;
+
+protected:
+    explicit BufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
+                         std::unique_ptr<StreamBuffer> stream_buffer)
+        : rasterizer{rasterizer}, system{system}, stream_buffer{std::move(stream_buffer)},
+          stream_buffer_handle{this->stream_buffer->GetHandle()} {}
+
+    ~BufferCache() = default;
+
+    virtual const TBufferType* ToHandle(const TBuffer& storage) = 0;
+
+    virtual void WriteBarrier() = 0;
+
+    virtual TBuffer CreateBlock(CacheAddr cache_addr, std::size_t size) = 0;
+
+    virtual void UploadBlockData(const TBuffer& buffer, std::size_t offset, std::size_t size,
+                                 const u8* data) = 0;
+
+    virtual void DownloadBlockData(const TBuffer& buffer, std::size_t offset, std::size_t size,
+                                   u8* data) = 0;
+
+    virtual void CopyBlock(const TBuffer& src, const TBuffer& dst, std::size_t src_offset,
+                           std::size_t dst_offset, std::size_t size) = 0;
+
+    /// Register an object into the cache
+    void Register(const MapInterval& new_map, bool inherit_written = false) {
+        const CacheAddr cache_ptr = new_map->GetStart();
+        const std::optional<VAddr> cpu_addr =
+            system.GPU().MemoryManager().GpuToCpuAddress(new_map->GetGpuAddress());
+        if (!cache_ptr || !cpu_addr) {
+            LOG_CRITICAL(HW_GPU, "Failed to register buffer with unmapped gpu_address 0x{:016x}",
+                         new_map->GetGpuAddress());
+            return;
+        }
+        const std::size_t size = new_map->GetEnd() - new_map->GetStart();
+        new_map->SetCpuAddress(*cpu_addr);
+        new_map->MarkAsRegistered(true);
+        const IntervalType interval{new_map->GetStart(), new_map->GetEnd()};
+        mapped_addresses.insert({interval, new_map});
+        rasterizer.UpdatePagesCachedCount(*cpu_addr, size, 1);
+        if (inherit_written) {
+            MarkRegionAsWritten(new_map->GetStart(), new_map->GetEnd() - 1);
+            new_map->MarkAsWritten(true);
+        }
+    }
+
+    /// Unregisters an object from the cache
+    void Unregister(MapInterval& map) {
+        const std::size_t size = map->GetEnd() - map->GetStart();
+        rasterizer.UpdatePagesCachedCount(map->GetCpuAddress(), size, -1);
+        map->MarkAsRegistered(false);
+        if (map->IsWritten()) {
+            UnmarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
+        }
+        const IntervalType delete_interval{map->GetStart(), map->GetEnd()};
+        mapped_addresses.erase(delete_interval);
+    }
+
+private:
+    MapInterval CreateMap(const CacheAddr start, const CacheAddr end, const GPUVAddr gpu_addr) {
+        return std::make_shared<MapIntervalBase>(start, end, gpu_addr);
+    }
+
+    MapInterval MapAddress(const TBuffer& block, const GPUVAddr gpu_addr,
+                           const CacheAddr cache_addr, const std::size_t size) {
+
+        std::vector<MapInterval> overlaps = GetMapsInRange(cache_addr, size);
+        if (overlaps.empty()) {
+            const CacheAddr cache_addr_end = cache_addr + size;
+            MapInterval new_map = CreateMap(cache_addr, cache_addr_end, gpu_addr);
+            u8* host_ptr = FromCacheAddr(cache_addr);
+            UploadBlockData(block, block->GetOffset(cache_addr), size, host_ptr);
+            Register(new_map);
+            return new_map;
+        }
+
+        const CacheAddr cache_addr_end = cache_addr + size;
+        if (overlaps.size() == 1) {
+            MapInterval& current_map = overlaps[0];
+            if (current_map->IsInside(cache_addr, cache_addr_end)) {
+                return current_map;
+            }
+        }
+        CacheAddr new_start = cache_addr;
+        CacheAddr new_end = cache_addr_end;
+        bool write_inheritance = false;
+        bool modified_inheritance = false;
+        // Calculate new buffer parameters
+        for (auto& overlap : overlaps) {
+            new_start = std::min(overlap->GetStart(), new_start);
+            new_end = std::max(overlap->GetEnd(), new_end);
+            write_inheritance |= overlap->IsWritten();
+            modified_inheritance |= overlap->IsModified();
+        }
+        GPUVAddr new_gpu_addr = gpu_addr + new_start - cache_addr;
+        for (auto& overlap : overlaps) {
+            Unregister(overlap);
+        }
+        UpdateBlock(block, new_start, new_end, overlaps);
+        MapInterval new_map = CreateMap(new_start, new_end, new_gpu_addr);
+        if (modified_inheritance) {
+            new_map->MarkAsModified(true, GetModifiedTicks());
+        }
+        Register(new_map, write_inheritance);
+        return new_map;
+    }
+
+    void UpdateBlock(const TBuffer& block, CacheAddr start, CacheAddr end,
+                     std::vector<MapInterval>& overlaps) {
+        const IntervalType base_interval{start, end};
+        IntervalSet interval_set{};
+        interval_set.add(base_interval);
+        for (auto& overlap : overlaps) {
+            const IntervalType subtract{overlap->GetStart(), overlap->GetEnd()};
+            interval_set.subtract(subtract);
+        }
+        for (auto& interval : interval_set) {
+            std::size_t size = interval.upper() - interval.lower();
+            if (size > 0) {
+                u8* host_ptr = FromCacheAddr(interval.lower());
+                UploadBlockData(block, block->GetOffset(interval.lower()), size, host_ptr);
+            }
+        }
+    }
+
+    std::vector<MapInterval> GetMapsInRange(CacheAddr addr, std::size_t size) {
+        if (size == 0) {
+            return {};
+        }
+
+        std::vector<MapInterval> objects{};
+        const IntervalType interval{addr, addr + size};
+        for (auto& pair : boost::make_iterator_range(mapped_addresses.equal_range(interval))) {
+            objects.push_back(pair.second);
+        }
+
+        return objects;
+    }
+
+    /// Returns a ticks counter used for tracking when cached objects were last modified
+    u64 GetModifiedTicks() {
+        return ++modified_ticks;
+    }
+
+    void FlushMap(MapInterval map) {
+        std::size_t size = map->GetEnd() - map->GetStart();
+        TBuffer block = blocks[map->GetStart() >> block_page_bits];
+        u8* host_ptr = FromCacheAddr(map->GetStart());
+        DownloadBlockData(block, block->GetOffset(map->GetStart()), size, host_ptr);
+        map->MarkAsModified(false, 0);
+    }
+
+    BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size,
+                                  std::size_t alignment) {
+        AlignBuffer(alignment);
+        const std::size_t uploaded_offset = buffer_offset;
+        std::memcpy(buffer_ptr, raw_pointer, size);
+
+        buffer_ptr += size;
+        buffer_offset += size;
+        return {&stream_buffer_handle, uploaded_offset};
+    }
+
+    void AlignBuffer(std::size_t alignment) {
+        // Align the offset, not the mapped pointer
+        const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment);
+        buffer_ptr += offset_aligned - buffer_offset;
+        buffer_offset = offset_aligned;
+    }
+
+    TBuffer EnlargeBlock(TBuffer buffer) {
+        const std::size_t old_size = buffer->GetSize();
+        const std::size_t new_size = old_size + block_page_size;
+        const CacheAddr cache_addr = buffer->GetCacheAddr();
+        TBuffer new_buffer = CreateBlock(cache_addr, new_size);
+        CopyBlock(buffer, new_buffer, 0, 0, old_size);
+        buffer->SetEpoch(epoch);
+        pending_destruction.push_back(buffer);
+        const CacheAddr cache_addr_end = cache_addr + new_size - 1;
+        u64 page_start = cache_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        while (page_start <= page_end) {
+            blocks[page_start] = new_buffer;
+            ++page_start;
+        }
+        return new_buffer;
+    }
+
+    TBuffer MergeBlocks(TBuffer first, TBuffer second) {
+        const std::size_t size_1 = first->GetSize();
+        const std::size_t size_2 = second->GetSize();
+        const CacheAddr first_addr = first->GetCacheAddr();
+        const CacheAddr second_addr = second->GetCacheAddr();
+        const CacheAddr new_addr = std::min(first_addr, second_addr);
+        const std::size_t new_size = size_1 + size_2;
+        TBuffer new_buffer = CreateBlock(new_addr, new_size);
+        CopyBlock(first, new_buffer, 0, new_buffer->GetOffset(first_addr), size_1);
+        CopyBlock(second, new_buffer, 0, new_buffer->GetOffset(second_addr), size_2);
+        first->SetEpoch(epoch);
+        second->SetEpoch(epoch);
+        pending_destruction.push_back(first);
+        pending_destruction.push_back(second);
+        const CacheAddr cache_addr_end = new_addr + new_size - 1;
+        u64 page_start = new_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        while (page_start <= page_end) {
+            blocks[page_start] = new_buffer;
+            ++page_start;
+        }
+        return new_buffer;
+    }
+
+    TBuffer GetBlock(const CacheAddr cache_addr, const std::size_t size) {
+        TBuffer found{};
+        const CacheAddr cache_addr_end = cache_addr + size - 1;
+        u64 page_start = cache_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        while (page_start <= page_end) {
+            auto it = blocks.find(page_start);
+            if (it == blocks.end()) {
+                if (found) {
+                    found = EnlargeBlock(found);
+                } else {
+                    const CacheAddr start_addr = (page_start << block_page_bits);
+                    found = CreateBlock(start_addr, block_page_size);
+                    blocks[page_start] = found;
+                }
+            } else {
+                if (found) {
+                    if (found == it->second) {
+                        ++page_start;
+                        continue;
+                    }
+                    found = MergeBlocks(found, it->second);
+                } else {
+                    found = it->second;
+                }
+            }
+            ++page_start;
+        }
+        return found;
+    }
+
+    void MarkRegionAsWritten(const CacheAddr start, const CacheAddr end) {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            auto it = written_pages.find(page_start);
+            if (it != written_pages.end()) {
+                it->second = it->second + 1;
+            } else {
+                written_pages[page_start] = 1;
+            }
+            page_start++;
+        }
+    }
+
+    void UnmarkRegionAsWritten(const CacheAddr start, const CacheAddr end) {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            auto it = written_pages.find(page_start);
+            if (it != written_pages.end()) {
+                if (it->second > 1) {
+                    it->second = it->second - 1;
+                } else {
+                    written_pages.erase(it);
+                }
+            }
+            page_start++;
+        }
+    }
+
+    bool IsRegionWritten(const CacheAddr start, const CacheAddr end) const {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            if (written_pages.count(page_start) > 0) {
+                return true;
+            }
+            page_start++;
+        }
+        return false;
+    }
+
+    VideoCore::RasterizerInterface& rasterizer;
+    Core::System& system;
+    std::unique_ptr<StreamBuffer> stream_buffer;
+
+    TBufferType stream_buffer_handle{};
+
+    bool invalidated = false;
+
+    u8* buffer_ptr = nullptr;
+    u64 buffer_offset = 0;
+    u64 buffer_offset_base = 0;
+
+    using IntervalSet = boost::icl::interval_set<CacheAddr>;
+    using IntervalCache = boost::icl::interval_map<CacheAddr, MapInterval>;
+    using IntervalType = typename IntervalCache::interval_type;
+    IntervalCache mapped_addresses{};
+
+    static constexpr u64 write_page_bit{11};
+    std::unordered_map<u64, u32> written_pages{};
+
+    static constexpr u64 block_page_bits{21};
+    static constexpr u64 block_page_size{1 << block_page_bits};
+    std::unordered_map<u64, TBuffer> blocks{};
+
+    std::list<TBuffer> pending_destruction{};
+    u64 epoch{};
+    u64 modified_ticks{};
+
+    std::recursive_mutex mutex;
+};
+
+} // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/map_interval.h b/src/video_core/buffer_cache/map_interval.h
new file mode 100644
index 000000000..3a104d5cd
--- /dev/null
+++ b/src/video_core/buffer_cache/map_interval.h
@@ -0,0 +1,89 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+
+namespace VideoCommon {
+
+class MapIntervalBase {
+public:
+    MapIntervalBase(const CacheAddr start, const CacheAddr end, const GPUVAddr gpu_addr)
+        : start{start}, end{end}, gpu_addr{gpu_addr} {}
+
+    void SetCpuAddress(VAddr new_cpu_addr) {
+        cpu_addr = new_cpu_addr;
+    }
+
+    VAddr GetCpuAddress() const {
+        return cpu_addr;
+    }
+
+    GPUVAddr GetGpuAddress() const {
+        return gpu_addr;
+    }
+
+    bool IsInside(const CacheAddr other_start, const CacheAddr other_end) const {
+        return (start <= other_start && other_end <= end);
+    }
+
+    bool operator==(const MapIntervalBase& rhs) const {
+        return std::tie(start, end) == std::tie(rhs.start, rhs.end);
+    }
+
+    bool operator!=(const MapIntervalBase& rhs) const {
+        return !operator==(rhs);
+    }
+
+    void MarkAsRegistered(const bool registered) {
+        is_registered = registered;
+    }
+
+    bool IsRegistered() const {
+        return is_registered;
+    }
+
+    CacheAddr GetStart() const {
+        return start;
+    }
+
+    CacheAddr GetEnd() const {
+        return end;
+    }
+
+    void MarkAsModified(const bool is_modified_, const u64 tick) {
+        is_modified = is_modified_;
+        ticks = tick;
+    }
+
+    bool IsModified() const {
+        return is_modified;
+    }
+
+    u64 GetModificationTick() const {
+        return ticks;
+    }
+
+    void MarkAsWritten(const bool is_written_) {
+        is_written = is_written_;
+    }
+
+    bool IsWritten() const {
+        return is_written;
+    }
+
+private:
+    CacheAddr start;
+    CacheAddr end;
+    GPUVAddr gpu_addr;
+    VAddr cpu_addr{};
+    bool is_written{};
+    bool is_modified{};
+    bool is_registered{};
+    u64 ticks{};
+};
+
+} // namespace VideoCommon
diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index 3175579cc..0094fd715 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -22,7 +22,7 @@ void DmaPusher::DispatchCalls() {
     MICROPROFILE_SCOPE(DispatchCalls);
 
     // On entering GPU code, assume all memory may be touched by the ARM core.
-    gpu.Maxwell3D().dirty_flags.OnMemoryWrite();
+    gpu.Maxwell3D().dirty.OnMemoryWrite();
 
     dma_pushbuffer_subindex = 0;
 
@@ -31,6 +31,7 @@ void DmaPusher::DispatchCalls() {
             break;
         }
     }
+    gpu.FlushCommands();
 }
 
 bool DmaPusher::Step() {
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 0ee228e28..98a8b5337 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -10,8 +10,7 @@
 
 namespace Tegra::Engines {
 
-Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
-    : rasterizer{rasterizer}, memory_manager{memory_manager} {}
+Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {}
 
 void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
     ASSERT_MSG(method_call.method < Regs::NUM_REGS,
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 05421d185..0901cf2fa 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -33,7 +33,7 @@ namespace Tegra::Engines {
 
 class Fermi2D final {
 public:
-    explicit Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager);
+    explicit Fermi2D(VideoCore::RasterizerInterface& rasterizer);
     ~Fermi2D() = default;
 
     /// Write the value to the register identified by method.
@@ -145,7 +145,6 @@ public:
 
 private:
     VideoCore::RasterizerInterface& rasterizer;
-    MemoryManager& memory_manager;
 
     /// Performs the copy from the source surface to the destination surface as configured in the
     /// registers.
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index 7404a8163..08586d33c 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -37,7 +37,7 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+            system.GPU().Maxwell3D().dirty.OnMemoryWrite();
         }
         break;
     }
@@ -50,13 +50,14 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
 }
 
 void KeplerCompute::ProcessLaunch() {
-
     const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
     memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
                                    LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32));
 
-    const GPUVAddr code_loc = regs.code_loc.Address() + launch_description.program_start;
-    LOG_WARNING(HW_GPU, "Compute Kernel Execute at Address 0x{:016x}, STUBBED", code_loc);
+    const GPUVAddr code_addr = regs.code_loc.Address() + launch_description.program_start;
+    LOG_TRACE(HW_GPU, "Compute invocation launched at address 0x{:016x}", code_addr);
+
+    rasterizer.DispatchCompute(code_addr);
 }
 
 } // namespace Tegra::Engines
diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp
index 0561f676c..fa4a7c5c1 100644
--- a/src/video_core/engines/kepler_memory.cpp
+++ b/src/video_core/engines/kepler_memory.cpp
@@ -15,7 +15,7 @@
 namespace Tegra::Engines {
 
 KeplerMemory::KeplerMemory(Core::System& system, MemoryManager& memory_manager)
-    : system{system}, memory_manager{memory_manager}, upload_state{memory_manager, regs.upload} {}
+    : system{system}, upload_state{memory_manager, regs.upload} {}
 
 KeplerMemory::~KeplerMemory() = default;
 
@@ -34,7 +34,7 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+            system.GPU().Maxwell3D().dirty.OnMemoryWrite();
         }
         break;
     }
diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h
index f3bc675a9..e0e25c321 100644
--- a/src/video_core/engines/kepler_memory.h
+++ b/src/video_core/engines/kepler_memory.h
@@ -65,7 +65,6 @@ public:
 
 private:
     Core::System& system;
-    MemoryManager& memory_manager;
     Upload::State upload_state;
 };
 
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 8755b8af4..f5158d219 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -22,6 +22,7 @@ Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& raste
                      MemoryManager& memory_manager)
     : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager},
       macro_interpreter{*this}, upload_state{memory_manager, regs.upload} {
+    InitDirtySettings();
     InitializeRegisterDefaults();
 }
 
@@ -69,6 +70,10 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.stencil_back_func_mask = 0xFFFFFFFF;
     regs.stencil_back_mask = 0xFFFFFFFF;
 
+    regs.depth_test_func = Regs::ComparisonOp::Always;
+    regs.cull.front_face = Regs::Cull::FrontFace::CounterClockWise;
+    regs.cull.cull_face = Regs::Cull::CullFace::Back;
+
     // TODO(Rodrigo): Most games do not set a point size. I think this is a case of a
     // register carrying a default value. Assume it's OpenGL's default (1).
     regs.point_size = 1.0f;
@@ -86,21 +91,168 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.rt_separate_frag_data = 1;
 }
 
+#define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name))
+
+void Maxwell3D::InitDirtySettings() {
+    const auto set_block = [this](const u32 start, const u32 range, const u8 position) {
+        const auto start_itr = dirty_pointers.begin() + start;
+        const auto end_itr = start_itr + range;
+        std::fill(start_itr, end_itr, position);
+    };
+    dirty.regs.fill(true);
+
+    // Init Render Targets
+    constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
+    constexpr u32 rt_start_reg = MAXWELL3D_REG_INDEX(rt);
+    constexpr u32 rt_end_reg = rt_start_reg + registers_per_rt * 8;
+    u32 rt_dirty_reg = DIRTY_REGS_POS(render_target);
+    for (u32 rt_reg = rt_start_reg; rt_reg < rt_end_reg; rt_reg += registers_per_rt) {
+        set_block(rt_reg, registers_per_rt, rt_dirty_reg);
+        rt_dirty_reg++;
+    }
+    constexpr u32 depth_buffer_flag = DIRTY_REGS_POS(depth_buffer);
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_enable)] = depth_buffer_flag;
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_width)] = depth_buffer_flag;
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_height)] = depth_buffer_flag;
+    constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
+    constexpr u32 zeta_reg = MAXWELL3D_REG_INDEX(zeta);
+    set_block(zeta_reg, registers_in_zeta, depth_buffer_flag);
+
+    // Init Vertex Arrays
+    constexpr u32 vertex_array_start = MAXWELL3D_REG_INDEX(vertex_array);
+    constexpr u32 vertex_array_size = sizeof(regs.vertex_array[0]) / sizeof(u32);
+    constexpr u32 vertex_array_end = vertex_array_start + vertex_array_size * Regs::NumVertexArrays;
+    u32 va_reg = DIRTY_REGS_POS(vertex_array);
+    u32 vi_reg = DIRTY_REGS_POS(vertex_instance);
+    for (u32 vertex_reg = vertex_array_start; vertex_reg < vertex_array_end;
+         vertex_reg += vertex_array_size) {
+        set_block(vertex_reg, 3, va_reg);
+        // The divisor concerns vertex array instances
+        dirty_pointers[vertex_reg + 3] = vi_reg;
+        va_reg++;
+        vi_reg++;
+    }
+    constexpr u32 vertex_limit_start = MAXWELL3D_REG_INDEX(vertex_array_limit);
+    constexpr u32 vertex_limit_size = sizeof(regs.vertex_array_limit[0]) / sizeof(u32);
+    constexpr u32 vertex_limit_end = vertex_limit_start + vertex_limit_size * Regs::NumVertexArrays;
+    va_reg = DIRTY_REGS_POS(vertex_array);
+    for (u32 vertex_reg = vertex_limit_start; vertex_reg < vertex_limit_end;
+         vertex_reg += vertex_limit_size) {
+        set_block(vertex_reg, vertex_limit_size, va_reg);
+        va_reg++;
+    }
+    constexpr u32 vertex_instance_start = MAXWELL3D_REG_INDEX(instanced_arrays);
+    constexpr u32 vertex_instance_size =
+        sizeof(regs.instanced_arrays.is_instanced[0]) / sizeof(u32);
+    constexpr u32 vertex_instance_end =
+        vertex_instance_start + vertex_instance_size * Regs::NumVertexArrays;
+    vi_reg = DIRTY_REGS_POS(vertex_instance);
+    for (u32 vertex_reg = vertex_instance_start; vertex_reg < vertex_instance_end;
+         vertex_reg += vertex_instance_size) {
+        set_block(vertex_reg, vertex_instance_size, vi_reg);
+        vi_reg++;
+    }
+    set_block(MAXWELL3D_REG_INDEX(vertex_attrib_format), regs.vertex_attrib_format.size(),
+              DIRTY_REGS_POS(vertex_attrib_format));
+
+    // Init Shaders
+    constexpr u32 shader_registers_count =
+        sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
+    set_block(MAXWELL3D_REG_INDEX(shader_config[0]), shader_registers_count,
+              DIRTY_REGS_POS(shaders));
+
+    // State
+
+    // Viewport
+    constexpr u32 viewport_dirty_reg = DIRTY_REGS_POS(viewport);
+    constexpr u32 viewport_start = MAXWELL3D_REG_INDEX(viewports);
+    constexpr u32 viewport_size = sizeof(regs.viewports) / sizeof(u32);
+    set_block(viewport_start, viewport_size, viewport_dirty_reg);
+    constexpr u32 view_volume_start = MAXWELL3D_REG_INDEX(view_volume_clip_control);
+    constexpr u32 view_volume_size = sizeof(regs.view_volume_clip_control) / sizeof(u32);
+    set_block(view_volume_start, view_volume_size, viewport_dirty_reg);
+
+    // Viewport transformation
+    constexpr u32 viewport_trans_start = MAXWELL3D_REG_INDEX(viewport_transform);
+    constexpr u32 viewport_trans_size = sizeof(regs.viewport_transform) / sizeof(u32);
+    set_block(viewport_trans_start, viewport_trans_size, DIRTY_REGS_POS(viewport_transform));
+
+    // Cullmode
+    constexpr u32 cull_mode_start = MAXWELL3D_REG_INDEX(cull);
+    constexpr u32 cull_mode_size = sizeof(regs.cull) / sizeof(u32);
+    set_block(cull_mode_start, cull_mode_size, DIRTY_REGS_POS(cull_mode));
+
+    // Screen y control
+    dirty_pointers[MAXWELL3D_REG_INDEX(screen_y_control)] = DIRTY_REGS_POS(screen_y_control);
+
+    // Primitive Restart
+    constexpr u32 primitive_restart_start = MAXWELL3D_REG_INDEX(primitive_restart);
+    constexpr u32 primitive_restart_size = sizeof(regs.primitive_restart) / sizeof(u32);
+    set_block(primitive_restart_start, primitive_restart_size, DIRTY_REGS_POS(primitive_restart));
+
+    // Depth Test
+    constexpr u32 depth_test_dirty_reg = DIRTY_REGS_POS(depth_test);
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_enable)] = depth_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_write_enabled)] = depth_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_func)] = depth_test_dirty_reg;
+
+    // Stencil Test
+    constexpr u32 stencil_test_dirty_reg = DIRTY_REGS_POS(stencil_test);
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_enable)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_func)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_ref)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_fail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zfail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zpass)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_two_side_enable)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_func)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_ref)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_fail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zfail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zpass)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_mask)] = stencil_test_dirty_reg;
+
+    // Color Mask
+    constexpr u32 color_mask_dirty_reg = DIRTY_REGS_POS(color_mask);
+    dirty_pointers[MAXWELL3D_REG_INDEX(color_mask_common)] = color_mask_dirty_reg;
+    set_block(MAXWELL3D_REG_INDEX(color_mask), sizeof(regs.color_mask) / sizeof(u32),
+              color_mask_dirty_reg);
+    // Blend State
+    constexpr u32 blend_state_dirty_reg = DIRTY_REGS_POS(blend_state);
+    set_block(MAXWELL3D_REG_INDEX(blend_color), sizeof(regs.blend_color) / sizeof(u32),
+              blend_state_dirty_reg);
+    dirty_pointers[MAXWELL3D_REG_INDEX(independent_blend_enable)] = blend_state_dirty_reg;
+    set_block(MAXWELL3D_REG_INDEX(blend), sizeof(regs.blend) / sizeof(u32), blend_state_dirty_reg);
+    set_block(MAXWELL3D_REG_INDEX(independent_blend), sizeof(regs.independent_blend) / sizeof(u32),
+              blend_state_dirty_reg);
+
+    // Scissor State
+    constexpr u32 scissor_test_dirty_reg = DIRTY_REGS_POS(scissor_test);
+    set_block(MAXWELL3D_REG_INDEX(scissor_test), sizeof(regs.scissor_test) / sizeof(u32),
+              scissor_test_dirty_reg);
+
+    // Polygon Offset
+    constexpr u32 polygon_offset_dirty_reg = DIRTY_REGS_POS(polygon_offset);
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_fill_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_line_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_point_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_units)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_factor)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_clamp)] = polygon_offset_dirty_reg;
+}
+
 void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
     // Reset the current macro.
     executing_macro = 0;
 
     // Lookup the macro offset
-    const u32 entry{(method - MacroRegistersStart) >> 1};
-    const auto& search{macro_offsets.find(entry)};
-    if (search == macro_offsets.end()) {
-        LOG_CRITICAL(HW_GPU, "macro not found for method 0x{:X}!", method);
-        UNREACHABLE();
-        return;
-    }
+    const u32 entry = ((method - MacroRegistersStart) >> 1) % macro_positions.size();
 
     // Execute the current macro.
-    macro_interpreter.Execute(search->second, std::move(parameters));
+    macro_interpreter.Execute(macro_positions[entry], std::move(parameters));
 }
 
 void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
@@ -108,6 +260,14 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
     const u32 method = method_call.method;
 
+    if (method == cb_data_state.current) {
+        regs.reg_array[method] = method_call.argument;
+        ProcessCBData(method_call.argument);
+        return;
+    } else if (cb_data_state.current != null_cb_data) {
+        FinishCBData();
+    }
+
     // It is an error to write to a register other than the current macro's ARG register before it
     // has finished execution.
     if (executing_macro != 0) {
@@ -143,49 +303,19 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
     if (regs.reg_array[method] != method_call.argument) {
         regs.reg_array[method] = method_call.argument;
-        // Color buffers
-        constexpr u32 first_rt_reg = MAXWELL3D_REG_INDEX(rt);
-        constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
-        if (method >= first_rt_reg &&
-            method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
-            const std::size_t rt_index = (method - first_rt_reg) / registers_per_rt;
-            dirty_flags.color_buffer.set(rt_index);
-        }
-
-        // Zeta buffer
-        constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
-        if (method == MAXWELL3D_REG_INDEX(zeta_enable) ||
-            method == MAXWELL3D_REG_INDEX(zeta_width) ||
-            method == MAXWELL3D_REG_INDEX(zeta_height) ||
-            (method >= MAXWELL3D_REG_INDEX(zeta) &&
-             method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
-            dirty_flags.zeta_buffer = true;
-        }
-
-        // Shader
-        constexpr u32 shader_registers_count =
-            sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
-        if (method >= MAXWELL3D_REG_INDEX(shader_config[0]) &&
-            method < MAXWELL3D_REG_INDEX(shader_config[0]) + shader_registers_count) {
-            dirty_flags.shaders = true;
-        }
-
-        // Vertex format
-        if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
-            method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
-            dirty_flags.vertex_attrib_format = true;
-        }
-
-        // Vertex buffer
-        if (method >= MAXWELL3D_REG_INDEX(vertex_array) &&
-            method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
-        } else if (method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
-                   method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
-        } else if (method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
-                   method < MAXWELL3D_REG_INDEX(instanced_arrays) + Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set(method - MAXWELL3D_REG_INDEX(instanced_arrays));
+        const std::size_t dirty_reg = dirty_pointers[method];
+        if (dirty_reg) {
+            dirty.regs[dirty_reg] = true;
+            if (dirty_reg >= DIRTY_REGS_POS(vertex_array) &&
+                dirty_reg < DIRTY_REGS_POS(vertex_array_buffers)) {
+                dirty.vertex_array_buffers = true;
+            } else if (dirty_reg >= DIRTY_REGS_POS(vertex_instance) &&
+                       dirty_reg < DIRTY_REGS_POS(vertex_instances)) {
+                dirty.vertex_instances = true;
+            } else if (dirty_reg >= DIRTY_REGS_POS(render_target) &&
+                       dirty_reg < DIRTY_REGS_POS(render_settings)) {
+                dirty.render_settings = true;
+            }
         }
     }
 
@@ -214,7 +344,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
-        ProcessCBData(method_call.argument);
+        StartCBData(method);
         break;
     }
     case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): {
@@ -249,6 +379,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         ProcessQueryGet();
         break;
     }
+    case MAXWELL3D_REG_INDEX(condition.mode): {
+        ProcessQueryCondition();
+        break;
+    }
     case MAXWELL3D_REG_INDEX(sync_info): {
         ProcessSyncPoint();
         break;
@@ -261,7 +395,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            dirty_flags.OnMemoryWrite();
+            dirty.OnMemoryWrite();
         }
         break;
     }
@@ -281,7 +415,7 @@ void Maxwell3D::ProcessMacroUpload(u32 data) {
 }
 
 void Maxwell3D::ProcessMacroBind(u32 data) {
-    macro_offsets[regs.macros.entry] = data;
+    macro_positions[regs.macros.entry++] = data;
 }
 
 void Maxwell3D::ProcessQueryGet() {
@@ -302,6 +436,7 @@ void Maxwell3D::ProcessQueryGet() {
         result = regs.query.query_sequence;
         break;
     default:
+        result = 1;
         UNIMPLEMENTED_MSG("Unimplemented query select type {}",
                           static_cast<u32>(regs.query.query_get.select.Value()));
     }
@@ -333,7 +468,6 @@ void Maxwell3D::ProcessQueryGet() {
             query_result.timestamp = system.CoreTiming().GetTicks();
             memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result));
         }
-        dirty_flags.OnMemoryWrite();
         break;
     }
     default:
@@ -342,12 +476,52 @@ void Maxwell3D::ProcessQueryGet() {
     }
 }
 
+void Maxwell3D::ProcessQueryCondition() {
+    const GPUVAddr condition_address{regs.condition.Address()};
+    switch (regs.condition.mode) {
+    case Regs::ConditionMode::Always: {
+        execute_on = true;
+        break;
+    }
+    case Regs::ConditionMode::Never: {
+        execute_on = false;
+        break;
+    }
+    case Regs::ConditionMode::ResNonZero: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on = cmp.initial_sequence != 0U && cmp.initial_mode != 0U;
+        break;
+    }
+    case Regs::ConditionMode::Equal: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on =
+            cmp.initial_sequence == cmp.current_sequence && cmp.initial_mode == cmp.current_mode;
+        break;
+    }
+    case Regs::ConditionMode::NotEqual: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on =
+            cmp.initial_sequence != cmp.current_sequence || cmp.initial_mode != cmp.current_mode;
+        break;
+    }
+    default: {
+        UNIMPLEMENTED_MSG("Uninplemented Condition Mode!");
+        execute_on = true;
+        break;
+    }
+    }
+}
+
 void Maxwell3D::ProcessSyncPoint() {
     const u32 sync_point = regs.sync_info.sync_point.Value();
     const u32 increment = regs.sync_info.increment.Value();
-    const u32 cache_flush = regs.sync_info.unknown.Value();
-    LOG_DEBUG(HW_GPU, "Syncpoint set {}, increment: {}, unk: {}", sync_point, increment,
-              cache_flush);
+    [[maybe_unused]] const u32 cache_flush = regs.sync_info.unknown.Value();
+    if (increment) {
+        system.GPU().IncrementSyncPoint(sync_point);
+    }
 }
 
 void Maxwell3D::DrawArrays() {
@@ -405,23 +579,39 @@ void Maxwell3D::ProcessCBBind(Regs::ShaderStage stage) {
 }
 
 void Maxwell3D::ProcessCBData(u32 value) {
+    const u32 id = cb_data_state.id;
+    cb_data_state.buffer[id][cb_data_state.counter] = value;
+    // Increment the current buffer position.
+    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
+    cb_data_state.counter++;
+}
+
+void Maxwell3D::StartCBData(u32 method) {
+    constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+    cb_data_state.start_pos = regs.const_buffer.cb_pos;
+    cb_data_state.id = method - first_cb_data;
+    cb_data_state.current = method;
+    cb_data_state.counter = 0;
+    ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
+}
+
+void Maxwell3D::FinishCBData() {
     // Write the input value to the current const buffer at the current position.
     const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
     ASSERT(buffer_address != 0);
 
     // Don't allow writing past the end of the buffer.
-    ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
+    ASSERT(regs.const_buffer.cb_pos <= regs.const_buffer.cb_size);
 
-    const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos};
+    const GPUVAddr address{buffer_address + cb_data_state.start_pos};
+    const std::size_t size = regs.const_buffer.cb_pos - cb_data_state.start_pos;
 
-    u8* ptr{memory_manager.GetPointer(address)};
-    rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32));
-    memory_manager.Write<u32>(address, value);
+    const u32 id = cb_data_state.id;
+    memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
+    dirty.OnMemoryWrite();
 
-    dirty_flags.OnMemoryWrite();
-
-    // Increment the current buffer position.
-    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
+    cb_data_state.id = null_cb_data;
+    cb_data_state.current = null_cb_data;
 }
 
 Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
@@ -430,10 +620,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
     Texture::TICEntry tic_entry;
     memory_manager.ReadBlockUnsafe(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry));
 
-    const auto r_type{tic_entry.r_type.Value()};
-    const auto g_type{tic_entry.g_type.Value()};
-    const auto b_type{tic_entry.b_type.Value()};
-    const auto a_type{tic_entry.a_type.Value()};
+    [[maybe_unused]] const auto r_type{tic_entry.r_type.Value()};
+    [[maybe_unused]] const auto g_type{tic_entry.g_type.Value()};
+    [[maybe_unused]] const auto b_type{tic_entry.b_type.Value()};
+    [[maybe_unused]] const auto a_type{tic_entry.a_type.Value()};
 
     // TODO(Subv): Different data types for separate components are not supported
     DEBUG_ASSERT(r_type == g_type && r_type == b_type && r_type == a_type);
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 13e314944..0184342a0 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -67,6 +67,7 @@ public:
         static constexpr std::size_t MaxShaderStage = 5;
         // Maximum number of const buffers per shader stage.
         static constexpr std::size_t MaxConstBuffers = 18;
+        static constexpr std::size_t MaxConstBufferSize = 0x10000;
 
         enum class QueryMode : u32 {
             Write = 0,
@@ -89,6 +90,20 @@ public:
 
         enum class QuerySelect : u32 {
             Zero = 0,
+            TimeElapsed = 2,
+            TransformFeedbackPrimitivesGenerated = 11,
+            PrimitivesGenerated = 18,
+            SamplesPassed = 21,
+            TransformFeedbackUnknown = 26,
+        };
+
+        struct QueryCompare {
+            u32 initial_sequence;
+            u32 initial_mode;
+            u32 unknown1;
+            u32 unknown2;
+            u32 current_sequence;
+            u32 current_mode;
         };
 
         enum class QuerySyncCondition : u32 {
@@ -96,6 +111,14 @@ public:
             GreaterThan = 1,
         };
 
+        enum class ConditionMode : u32 {
+            Never = 0,
+            Always = 1,
+            ResNonZero = 2,
+            Equal = 3,
+            NotEqual = 4,
+        };
+
         enum class ShaderProgram : u32 {
             VertexA = 0,
             VertexB = 1,
@@ -814,7 +837,18 @@ public:
                     BitField<4, 1, u32> alpha_to_one;
                 } multisample_control;
 
-                INSERT_PADDING_WORDS(0x7);
+                INSERT_PADDING_WORDS(0x4);
+
+                struct {
+                    u32 address_high;
+                    u32 address_low;
+                    ConditionMode mode;
+
+                    GPUVAddr Address() const {
+                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                                     address_low);
+                    }
+                } condition;
 
                 struct {
                     u32 tsc_address_high;
@@ -1123,23 +1157,77 @@ public:
 
     State state{};
 
-    struct DirtyFlags {
-        std::bitset<8> color_buffer{0xFF};
-        std::bitset<32> vertex_array{0xFFFFFFFF};
+    struct DirtyRegs {
+        static constexpr std::size_t NUM_REGS = 256;
+        union {
+            struct {
+                bool null_dirty;
+
+                // Vertex Attributes
+                bool vertex_attrib_format;
+
+                // Vertex Arrays
+                std::array<bool, 32> vertex_array;
+
+                bool vertex_array_buffers;
+
+                // Vertex Instances
+                std::array<bool, 32> vertex_instance;
 
-        bool vertex_attrib_format = true;
-        bool zeta_buffer = true;
-        bool shaders = true;
+                bool vertex_instances;
+
+                // Render Targets
+                std::array<bool, 8> render_target;
+                bool depth_buffer;
+
+                bool render_settings;
+
+                // Shaders
+                bool shaders;
+
+                // Rasterizer State
+                bool viewport;
+                bool clip_coefficient;
+                bool cull_mode;
+                bool primitive_restart;
+                bool depth_test;
+                bool stencil_test;
+                bool blend_state;
+                bool scissor_test;
+                bool transform_feedback;
+                bool color_mask;
+                bool polygon_offset;
+
+                // Complementary
+                bool viewport_transform;
+                bool screen_y_control;
+
+                bool memory_general;
+            };
+            std::array<bool, NUM_REGS> regs;
+        };
+
+        void ResetVertexArrays() {
+            vertex_array.fill(true);
+            vertex_array_buffers = true;
+        }
+
+        void ResetRenderTargets() {
+            depth_buffer = true;
+            render_target.fill(true);
+            render_settings = true;
+        }
 
         void OnMemoryWrite() {
-            zeta_buffer = true;
             shaders = true;
-            color_buffer.set();
-            vertex_array.set();
+            memory_general = true;
+            ResetRenderTargets();
+            ResetVertexArrays();
         }
-    };
 
-    DirtyFlags dirty_flags;
+    } dirty{};
+
+    std::array<u8, Regs::NUM_REGS> dirty_pointers{};
 
     /// Reads a register value located at the input method address
     u32 GetRegisterValue(u32 method) const;
@@ -1168,6 +1256,10 @@ public:
         return macro_memory;
     }
 
+    bool ShouldExecute() const {
+        return execute_on;
+    }
+
 private:
     void InitializeRegisterDefaults();
 
@@ -1178,7 +1270,7 @@ private:
     MemoryManager& memory_manager;
 
     /// Start offsets of each macro in macro_memory
-    std::unordered_map<u32, u32> macro_offsets;
+    std::array<u32, 0x80> macro_positions = {};
 
     /// Memory for macro code
     MacroMemory macro_memory;
@@ -1191,14 +1283,27 @@ private:
     /// Interpreter for the macro codes uploaded to the GPU.
     MacroInterpreter macro_interpreter;
 
+    static constexpr u32 null_cb_data = 0xFFFFFFFF;
+    struct {
+        std::array<std::array<u32, 0x4000>, 16> buffer;
+        u32 current{null_cb_data};
+        u32 id{null_cb_data};
+        u32 start_pos{};
+        u32 counter{};
+    } cb_data_state;
+
     Upload::State upload_state;
 
+    bool execute_on{true};
+
     /// Retrieves information about a specific TIC entry from the TIC buffer.
     Texture::TICEntry GetTICEntry(u32 tic_index) const;
 
     /// Retrieves information about a specific TSC entry from the TSC buffer.
     Texture::TSCEntry GetTSCEntry(u32 tsc_index) const;
 
+    void InitDirtySettings();
+
     /**
      * Call a macro on this engine.
      * @param method Method to call
@@ -1218,11 +1323,16 @@ private:
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
 
+    // Handles Conditional Rendering
+    void ProcessQueryCondition();
+
     /// Handles writes to syncing register.
     void ProcessSyncPoint();
 
     /// Handles a write to the CB_DATA[i] register.
+    void StartCBData(u32 method);
     void ProcessCBData(u32 value);
+    void FinishCBData();
 
     /// Handles a write to the CB_BIND register.
     void ProcessCBBind(Regs::ShaderStage stage);
@@ -1289,6 +1399,7 @@ ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
 ASSERT_REG_POSITION(point_size, 0x546);
 ASSERT_REG_POSITION(zeta_enable, 0x54E);
 ASSERT_REG_POSITION(multisample_control, 0x54F);
+ASSERT_REG_POSITION(condition, 0x554);
 ASSERT_REG_POSITION(tsc, 0x557);
 ASSERT_REG_POSITION(polygon_offset_factor, 0x55b);
 ASSERT_REG_POSITION(tic, 0x55D);
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index afb9578d0..ad8453c5f 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -5,18 +5,17 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
+#include "core/settings.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/memory_manager.h"
-#include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_base.h"
 #include "video_core/textures/decoders.h"
 
 namespace Tegra::Engines {
 
-MaxwellDMA::MaxwellDMA(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
-                       MemoryManager& memory_manager)
-    : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager} {}
+MaxwellDMA::MaxwellDMA(Core::System& system, MemoryManager& memory_manager)
+    : system{system}, memory_manager{memory_manager} {}
 
 void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
     ASSERT_MSG(method_call.method < Regs::NUM_REGS,
@@ -38,7 +37,7 @@ void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
 }
 
 void MaxwellDMA::HandleCopy() {
-    LOG_WARNING(HW_GPU, "Requested a DMA copy");
+    LOG_TRACE(HW_GPU, "Requested a DMA copy");
 
     const GPUVAddr source = regs.src_address.Address();
     const GPUVAddr dest = regs.dst_address.Address();
@@ -58,7 +57,7 @@ void MaxwellDMA::HandleCopy() {
     }
 
     // All copies here update the main memory, so mark all rasterizer states as invalid.
-    system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+    system.GPU().Maxwell3D().dirty.OnMemoryWrite();
 
     if (regs.exec.is_dst_linear && regs.exec.is_src_linear) {
         // When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D
@@ -84,13 +83,17 @@ void MaxwellDMA::HandleCopy() {
     ASSERT(regs.exec.enable_2d == 1);
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
-        ASSERT(regs.src_params.size_z == 1);
+        ASSERT(regs.src_params.BlockDepth() == 0);
         // If the input is tiled and the output is linear, deswizzle the input and copy it over.
-        const u32 src_bytes_per_pixel = regs.src_pitch / regs.src_params.size_x;
+        const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
         const std::size_t src_size = Texture::CalculateSize(
-            true, src_bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y,
+            true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y,
             regs.src_params.size_z, regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
 
+        const std::size_t src_layer_size = Texture::CalculateSize(
+            true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y, 1,
+            regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
+
         const std::size_t dst_size = regs.dst_pitch * regs.y_count;
 
         if (read_buffer.size() < src_size) {
@@ -104,23 +107,23 @@ void MaxwellDMA::HandleCopy() {
         memory_manager.ReadBlock(source, read_buffer.data(), src_size);
         memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
 
-        Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
-                                  regs.src_params.size_x, src_bytes_per_pixel, read_buffer.data(),
-                                  write_buffer.data(), regs.src_params.BlockHeight(),
-                                  regs.src_params.pos_x, regs.src_params.pos_y);
+        Texture::UnswizzleSubrect(
+            regs.x_count, regs.y_count, regs.dst_pitch, regs.src_params.size_x, bytes_per_pixel,
+            read_buffer.data() + src_layer_size * regs.src_params.pos_z, write_buffer.data(),
+            regs.src_params.BlockHeight(), regs.src_params.pos_x, regs.src_params.pos_y);
 
         memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
     } else {
         ASSERT(regs.dst_params.BlockDepth() == 0);
 
-        const u32 src_bytes_per_pixel = regs.src_pitch / regs.x_count;
+        const u32 bytes_per_pixel = regs.src_pitch / regs.x_count;
 
         const std::size_t dst_size = Texture::CalculateSize(
-            true, src_bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y,
+            true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y,
             regs.dst_params.size_z, regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
 
         const std::size_t dst_layer_size = Texture::CalculateSize(
-            true, src_bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, 1,
+            true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, 1,
             regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
 
         const std::size_t src_size = regs.src_pitch * regs.y_count;
@@ -133,14 +136,19 @@ void MaxwellDMA::HandleCopy() {
             write_buffer.resize(dst_size);
         }
 
-        memory_manager.ReadBlock(source, read_buffer.data(), src_size);
-        memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
+        if (Settings::values.use_accurate_gpu_emulation) {
+            memory_manager.ReadBlock(source, read_buffer.data(), src_size);
+            memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
+        } else {
+            memory_manager.ReadBlockUnsafe(source, read_buffer.data(), src_size);
+            memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
+        }
 
         // If the input is linear and the output is tiled, swizzle the input and copy it over.
-        Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x,
-                                src_bytes_per_pixel,
-                                write_buffer.data() + dst_layer_size * regs.dst_params.pos_z,
-                                read_buffer.data(), regs.dst_params.BlockHeight());
+        Texture::SwizzleSubrect(
+            regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x, bytes_per_pixel,
+            write_buffer.data() + dst_layer_size * regs.dst_params.pos_z, read_buffer.data(),
+            regs.dst_params.BlockHeight(), regs.dst_params.pos_x, regs.dst_params.pos_y);
 
         memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
     }
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index 17b015ca7..93808a9bb 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -20,10 +20,6 @@ namespace Tegra {
 class MemoryManager;
 }
 
-namespace VideoCore {
-class RasterizerInterface;
-}
-
 namespace Tegra::Engines {
 
 /**
@@ -33,8 +29,7 @@ namespace Tegra::Engines {
 
 class MaxwellDMA final {
 public:
-    explicit MaxwellDMA(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
-                        MemoryManager& memory_manager);
+    explicit MaxwellDMA(Core::System& system, MemoryManager& memory_manager);
     ~MaxwellDMA() = default;
 
     /// Write the value to the register identified by method.
@@ -180,8 +175,6 @@ public:
 private:
     Core::System& system;
 
-    VideoCore::RasterizerInterface& rasterizer;
-
     MemoryManager& memory_manager;
 
     std::vector<u8> read_buffer;
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 404d4f5aa..c3678b9ea 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -78,7 +78,7 @@ union Attribute {
     constexpr explicit Attribute(u64 value) : value(value) {}
 
     enum class Index : u64 {
-        PointSize = 6,
+        LayerViewportPointSize = 6,
         Position = 7,
         Attribute_0 = 8,
         Attribute_31 = 39,
@@ -538,6 +538,12 @@ enum class PhysicalAttributeDirection : u64 {
     Output = 1,
 };
 
+enum class VoteOperation : u64 {
+    All = 0, // allThreadsNV
+    Any = 1, // anyThreadNV
+    Eq = 2,  // allThreadsEqualNV
+};
+
 union Instruction {
     Instruction& operator=(const Instruction& instr) {
         value = instr.value;
@@ -560,6 +566,18 @@ union Instruction {
     BitField<48, 16, u64> opcode;
 
     union {
+        BitField<8, 5, ConditionCode> cc;
+        BitField<13, 1, u64> trigger;
+    } nop;
+
+    union {
+        BitField<48, 2, VoteOperation> operation;
+        BitField<45, 3, u64> dest_pred;
+        BitField<39, 3, u64> value;
+        BitField<42, 1, u64> negate_value;
+    } vote;
+
+    union {
         BitField<8, 8, Register> gpr;
         BitField<20, 24, s64> offset;
     } gmem;
@@ -868,6 +886,7 @@ union Instruction {
     union {
         BitField<0, 3, u64> pred0;
         BitField<3, 3, u64> pred3;
+        BitField<6, 1, u64> neg_b;
         BitField<7, 1, u64> abs_a;
         BitField<39, 3, u64> pred39;
         BitField<42, 1, u64> neg_pred;
@@ -931,8 +950,6 @@ union Instruction {
     } csetp;
 
     union {
-        BitField<35, 4, PredCondition> cond;
-        BitField<49, 1, u64> h_and;
         BitField<6, 1, u64> ftz;
         BitField<45, 2, PredOperation> op;
         BitField<3, 3, u64> pred3;
@@ -940,9 +957,21 @@ union Instruction {
         BitField<43, 1, u64> negate_a;
         BitField<44, 1, u64> abs_a;
         BitField<47, 2, HalfType> type_a;
-        BitField<31, 1, u64> negate_b;
-        BitField<30, 1, u64> abs_b;
-        BitField<28, 2, HalfType> type_b;
+        union {
+            BitField<35, 4, PredCondition> cond;
+            BitField<49, 1, u64> h_and;
+            BitField<31, 1, u64> negate_b;
+            BitField<30, 1, u64> abs_b;
+            BitField<28, 2, HalfType> type_b;
+        } reg;
+        union {
+            BitField<56, 1, u64> negate_b;
+            BitField<54, 1, u64> abs_b;
+        } cbuf;
+        union {
+            BitField<49, 4, PredCondition> cond;
+            BitField<53, 1, u64> h_and;
+        } cbuf_and_imm;
         BitField<42, 1, u64> neg_pred;
         BitField<39, 3, u64> pred39;
     } hsetp2;
@@ -991,7 +1020,6 @@ union Instruction {
     } iset;
 
     union {
-        BitField<41, 2, u64> selector; // i2i and i2f only
         BitField<45, 1, u64> negate_a;
         BitField<49, 1, u64> abs_a;
         BitField<10, 2, Register::Size> src_size;
@@ -1008,8 +1036,6 @@ union Instruction {
         } f2i;
 
         union {
-            BitField<8, 2, Register::Size> src_size;
-            BitField<10, 2, Register::Size> dst_size;
             BitField<39, 4, u64> rounding;
             // H0, H1 extract for F16 missing
             BitField<41, 1, u64> selector; // Guessed as some games set it, TODO: reverse this value
@@ -1019,6 +1045,13 @@ union Instruction {
             }
         } f2f;
 
+        union {
+            BitField<41, 2, u64> selector;
+        } int_src;
+
+        union {
+            BitField<41, 1, u64> selector;
+        } float_src;
     } conversion;
 
     union {
@@ -1278,6 +1311,7 @@ union Instruction {
     union {
         BitField<49, 1, u64> nodep_flag;
         BitField<53, 4, u64> texture_info;
+        BitField<59, 1, u64> fp32_flag;
 
         TextureType GetTextureType() const {
             // The TLDS instruction has a weird encoding for the texture type.
@@ -1368,6 +1402,20 @@ union Instruction {
     } bra;
 
     union {
+        BitField<20, 24, u64> target;
+        BitField<5, 1, u64> constant_buffer;
+
+        s32 GetBranchExtend() const {
+            // Sign extend the branch target offset
+            u32 mask = 1U << (24 - 1);
+            u32 value = static_cast<u32>(target);
+            // The branch offset is relative to the next instruction and is stored in bytes, so
+            // divide it by the size of an instruction and add 1 to it.
+            return static_cast<s32>((value ^ mask) - mask) / sizeof(Instruction) + 1;
+        }
+    } brx;
+
+    union {
         BitField<39, 1, u64> emit; // EmitVertex
         BitField<40, 1, u64> cut;  // EndPrimitive
     } out;
@@ -1459,11 +1507,13 @@ public:
         SYNC,
         BRK,
         DEPBAR,
+        VOTE,
         BFE_C,
         BFE_R,
         BFE_IMM,
         BFI_IMM_R,
         BRA,
+        BRX,
         PBK,
         LD_A,
         LD_L,
@@ -1490,6 +1540,7 @@ public:
         TMML,   // Texture Mip Map Level
         SUST,   // Surface Store
         EXIT,
+        NOP,
         IPA,
         OUT_R, // Emit vertex/primitive
         ISBERD,
@@ -1532,7 +1583,9 @@ public:
         HFMA2_RC,
         HFMA2_RR,
         HFMA2_IMM_R,
+        HSETP2_C,
         HSETP2_R,
+        HSETP2_IMM,
         HSET2_R,
         POPC_C,
         POPC_R,
@@ -1617,6 +1670,7 @@ public:
         Hfma2,
         Flow,
         Synch,
+        Warp,
         Memory,
         Texture,
         Image,
@@ -1738,10 +1792,12 @@ private:
             INST("111000101001----", Id::SSY, Type::Flow, "SSY"),
             INST("111000101010----", Id::PBK, Type::Flow, "PBK"),
             INST("111000100100----", Id::BRA, Type::Flow, "BRA"),
+            INST("111000100101----", Id::BRX, Type::Flow, "BRX"),
             INST("1111000011111---", Id::SYNC, Type::Flow, "SYNC"),
             INST("111000110100---", Id::BRK, Type::Flow, "BRK"),
             INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
             INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
+            INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
             INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
             INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"),
             INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"),
@@ -1760,12 +1816,13 @@ private:
             INST("1101111101010---", Id::TXQ_B, Type::Texture, "TXQ_B"),
             INST("1101-00---------", Id::TEXS, Type::Texture, "TEXS"),
             INST("11011100--11----", Id::TLD, Type::Texture, "TLD"),
-            INST("1101101---------", Id::TLDS, Type::Texture, "TLDS"),
+            INST("1101-01---------", Id::TLDS, Type::Texture, "TLDS"),
             INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"),
             INST("1101111100------", Id::TLD4S, Type::Texture, "TLD4S"),
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
             INST("11101011001-----", Id::SUST, Type::Image, "SUST"),
+            INST("0101000010110---", Id::NOP, Type::Trivial, "NOP"),
             INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),
             INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"),
@@ -1814,7 +1871,9 @@ private:
             INST("01100---1-------", Id::HFMA2_RC, Type::Hfma2, "HFMA2_RC"),
             INST("0101110100000---", Id::HFMA2_RR, Type::Hfma2, "HFMA2_RR"),
             INST("01110---0-------", Id::HFMA2_IMM_R, Type::Hfma2, "HFMA2_R_IMM"),
-            INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP_R"),
+            INST("0111111-1-------", Id::HSETP2_C, Type::HalfSetPredicate, "HSETP2_C"),
+            INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"),
+            INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"),
             INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 52706505b..2c47541cb 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -17,26 +17,15 @@
 
 namespace Tegra {
 
-u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
-    switch (format) {
-    case PixelFormat::ABGR8:
-    case PixelFormat::BGRA8:
-        return 4;
-    default:
-        return 4;
-    }
-
-    UNREACHABLE();
-}
-
-GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{renderer} {
+GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
+    : system{system}, renderer{renderer}, is_async{is_async} {
     auto& rasterizer{renderer.Rasterizer()};
-    memory_manager = std::make_unique<Tegra::MemoryManager>(rasterizer);
+    memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
     dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
-    fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
+    fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer);
     kepler_compute = std::make_unique<Engines::KeplerCompute>(system, rasterizer, *memory_manager);
-    maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, rasterizer, *memory_manager);
+    maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, *memory_manager);
     kepler_memory = std::make_unique<Engines::KeplerMemory>(system, *memory_manager);
 }
 
@@ -50,6 +39,14 @@ const Engines::Maxwell3D& GPU::Maxwell3D() const {
     return *maxwell_3d;
 }
 
+Engines::KeplerCompute& GPU::KeplerCompute() {
+    return *kepler_compute;
+}
+
+const Engines::KeplerCompute& GPU::KeplerCompute() const {
+    return *kepler_compute;
+}
+
 MemoryManager& GPU::MemoryManager() {
     return *memory_manager;
 }
@@ -66,6 +63,55 @@ const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
+void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
+    syncpoints[syncpoint_id]++;
+    std::lock_guard lock{sync_mutex};
+    if (!syncpt_interrupts[syncpoint_id].empty()) {
+        u32 value = syncpoints[syncpoint_id].load();
+        auto it = syncpt_interrupts[syncpoint_id].begin();
+        while (it != syncpt_interrupts[syncpoint_id].end()) {
+            if (value >= *it) {
+                TriggerCpuInterrupt(syncpoint_id, *it);
+                it = syncpt_interrupts[syncpoint_id].erase(it);
+                continue;
+            }
+            it++;
+        }
+    }
+}
+
+u32 GPU::GetSyncpointValue(const u32 syncpoint_id) const {
+    return syncpoints[syncpoint_id].load();
+}
+
+void GPU::RegisterSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+    auto& interrupt = syncpt_interrupts[syncpoint_id];
+    bool contains = std::any_of(interrupt.begin(), interrupt.end(),
+                                [value](u32 in_value) { return in_value == value; });
+    if (contains) {
+        return;
+    }
+    syncpt_interrupts[syncpoint_id].emplace_back(value);
+}
+
+bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+    std::lock_guard lock{sync_mutex};
+    auto& interrupt = syncpt_interrupts[syncpoint_id];
+    const auto iter =
+        std::find_if(interrupt.begin(), interrupt.end(),
+                     [value](u32 interrupt_value) { return value == interrupt_value; });
+
+    if (iter == interrupt.end()) {
+        return false;
+    }
+    interrupt.erase(iter);
+    return true;
+}
+
+void GPU::FlushCommands() {
+    renderer.Rasterizer().FlushCommands();
+}
+
 u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
     ASSERT(format != RenderTargetFormat::NONE);
 
@@ -143,12 +189,12 @@ enum class BufferMethods {
     NotifyIntr = 0x8,
     WrcacheFlush = 0x9,
     Unk28 = 0xA,
-    Unk2c = 0xB,
+    UnkCacheFlush = 0xB,
     RefCnt = 0x14,
     SemaphoreAcquire = 0x1A,
     SemaphoreRelease = 0x1B,
-    Unk70 = 0x1C,
-    Unk74 = 0x1D,
+    FenceValue = 0x1C,
+    FenceAction = 0x1D,
     Unk78 = 0x1E,
     Unk7c = 0x1F,
     Yield = 0x20,
@@ -194,6 +240,10 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
     case BufferMethods::SemaphoreAddressLow:
     case BufferMethods::SemaphoreSequence:
     case BufferMethods::RefCnt:
+    case BufferMethods::UnkCacheFlush:
+    case BufferMethods::WrcacheFlush:
+    case BufferMethods::FenceValue:
+    case BufferMethods::FenceAction:
         break;
     case BufferMethods::SemaphoreTrigger: {
         ProcessSemaphoreTriggerMethod();
@@ -204,21 +254,11 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
         LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented");
         break;
     }
-    case BufferMethods::WrcacheFlush: {
-        // TODO(Kmather73): Research and implement this method.
-        LOG_ERROR(HW_GPU, "Special puller engine method WrcacheFlush not implemented");
-        break;
-    }
     case BufferMethods::Unk28: {
         // TODO(Kmather73): Research and implement this method.
         LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented");
         break;
     }
-    case BufferMethods::Unk2c: {
-        // TODO(Kmather73): Research and implement this method.
-        LOG_ERROR(HW_GPU, "Special puller engine method Unk2c not implemented");
-        break;
-    }
     case BufferMethods::SemaphoreAcquire: {
         ProcessSemaphoreAcquire();
         break;
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index fe6628923..78bc0601a 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -5,8 +5,12 @@
 #pragma once
 
 #include <array>
+#include <atomic>
+#include <list>
 #include <memory>
+#include <mutex>
 #include "common/common_types.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "video_core/dma_pusher.h"
 
@@ -15,6 +19,10 @@ inline CacheAddr ToCacheAddr(const void* host_ptr) {
     return reinterpret_cast<CacheAddr>(host_ptr);
 }
 
+inline u8* FromCacheAddr(CacheAddr cache_addr) {
+    return reinterpret_cast<u8*>(cache_addr);
+}
+
 namespace Core {
 class System;
 }
@@ -87,14 +95,10 @@ class DebugContext;
 struct FramebufferConfig {
     enum class PixelFormat : u32 {
         ABGR8 = 1,
+        RGB565 = 4,
         BGRA8 = 5,
     };
 
-    /**
-     * Returns the number of bytes per pixel.
-     */
-    static u32 BytesPerPixel(PixelFormat format);
-
     VAddr address;
     u32 offset;
     u32 width;
@@ -127,7 +131,7 @@ class MemoryManager;
 
 class GPU {
 public:
-    explicit GPU(Core::System& system, VideoCore::RendererBase& renderer);
+    explicit GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async);
 
     virtual ~GPU();
 
@@ -149,12 +153,20 @@ public:
     /// Calls a GPU method.
     void CallMethod(const MethodCall& method_call);
 
+    void FlushCommands();
+
     /// Returns a reference to the Maxwell3D GPU engine.
     Engines::Maxwell3D& Maxwell3D();
 
     /// Returns a const reference to the Maxwell3D GPU engine.
     const Engines::Maxwell3D& Maxwell3D() const;
 
+    /// Returns a reference to the KeplerCompute GPU engine.
+    Engines::KeplerCompute& KeplerCompute();
+
+    /// Returns a reference to the KeplerCompute GPU engine.
+    const Engines::KeplerCompute& KeplerCompute() const;
+
     /// Returns a reference to the GPU memory manager.
     Tegra::MemoryManager& MemoryManager();
 
@@ -164,6 +176,22 @@ public:
     /// Returns a reference to the GPU DMA pusher.
     Tegra::DmaPusher& DmaPusher();
 
+    void IncrementSyncPoint(u32 syncpoint_id);
+
+    u32 GetSyncpointValue(u32 syncpoint_id) const;
+
+    void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+    bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+    std::unique_lock<std::mutex> LockSync() {
+        return std::unique_lock{sync_mutex};
+    }
+
+    bool IsAsync() const {
+        return is_async;
+    }
+
     /// Returns a const reference to the GPU DMA pusher.
     const Tegra::DmaPusher& DmaPusher() const;
 
@@ -194,7 +222,12 @@ public:
 
                 u32 semaphore_acquire;
                 u32 semaphore_release;
-                INSERT_PADDING_WORDS(0xE4);
+                u32 fence_value;
+                union {
+                    BitField<4, 4, u32> operation;
+                    BitField<8, 8, u32> id;
+                } fence_action;
+                INSERT_PADDING_WORDS(0xE2);
 
                 // Puller state
                 u32 acquire_mode;
@@ -216,8 +249,7 @@ public:
     virtual void PushGPUEntries(Tegra::CommandList&& entries) = 0;
 
     /// Swap buffers (render frame)
-    virtual void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) = 0;
+    virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     virtual void FlushRegion(CacheAddr addr, u64 size) = 0;
@@ -228,6 +260,9 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
     virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
+protected:
+    virtual void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const = 0;
+
 private:
     void ProcessBindMethod(const MethodCall& method_call);
     void ProcessSemaphoreTriggerMethod();
@@ -245,6 +280,7 @@ private:
 
 protected:
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
+    Core::System& system;
     VideoCore::RendererBase& renderer;
 
 private:
@@ -262,6 +298,14 @@ private:
     std::unique_ptr<Engines::MaxwellDMA> maxwell_dma;
     /// Inline memory engine
     std::unique_ptr<Engines::KeplerMemory> kepler_memory;
+
+    std::array<std::atomic<u32>, Service::Nvidia::MaxSyncPoints> syncpoints{};
+
+    std::array<std::list<u32>, Service::Nvidia::MaxSyncPoints> syncpt_interrupts;
+
+    std::mutex sync_mutex;
+
+    const bool is_async;
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
@@ -274,6 +318,8 @@ ASSERT_REG_POSITION(semaphore_trigger, 0x7);
 ASSERT_REG_POSITION(reference_count, 0x14);
 ASSERT_REG_POSITION(semaphore_acquire, 0x1A);
 ASSERT_REG_POSITION(semaphore_release, 0x1B);
+ASSERT_REG_POSITION(fence_value, 0x1C);
+ASSERT_REG_POSITION(fence_action, 0x1D);
 
 ASSERT_REG_POSITION(acquire_mode, 0x100);
 ASSERT_REG_POSITION(acquire_source, 0x101);
diff --git a/src/video_core/gpu_asynch.cpp b/src/video_core/gpu_asynch.cpp
index d4e2553a9..f2a3a390e 100644
--- a/src/video_core/gpu_asynch.cpp
+++ b/src/video_core/gpu_asynch.cpp
@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "core/core.h"
+#include "core/hardware_interrupt_manager.h"
 #include "video_core/gpu_asynch.h"
 #include "video_core/gpu_thread.h"
 #include "video_core/renderer_base.h"
@@ -9,7 +11,7 @@
 namespace VideoCommon {
 
 GPUAsynch::GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer)
-    : GPU(system, renderer), gpu_thread{system} {}
+    : GPU(system, renderer, true), gpu_thread{system} {}
 
 GPUAsynch::~GPUAsynch() = default;
 
@@ -21,9 +23,8 @@ void GPUAsynch::PushGPUEntries(Tegra::CommandList&& entries) {
     gpu_thread.SubmitList(std::move(entries));
 }
 
-void GPUAsynch::SwapBuffers(
-    std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
-    gpu_thread.SwapBuffers(std::move(framebuffer));
+void GPUAsynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    gpu_thread.SwapBuffers(framebuffer);
 }
 
 void GPUAsynch::FlushRegion(CacheAddr addr, u64 size) {
@@ -38,4 +39,9 @@ void GPUAsynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     gpu_thread.FlushAndInvalidateRegion(addr, size);
 }
 
+void GPUAsynch::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
+    auto& interrupt_manager = system.InterruptManager();
+    interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
+}
+
 } // namespace VideoCommon
diff --git a/src/video_core/gpu_asynch.h b/src/video_core/gpu_asynch.h
index 30be74cba..a12f9bac4 100644
--- a/src/video_core/gpu_asynch.h
+++ b/src/video_core/gpu_asynch.h
@@ -14,19 +14,21 @@ class RendererBase;
 namespace VideoCommon {
 
 /// Implementation of GPU interface that runs the GPU asynchronously
-class GPUAsynch : public Tegra::GPU {
+class GPUAsynch final : public Tegra::GPU {
 public:
     explicit GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer);
     ~GPUAsynch() override;
 
     void Start() override;
     void PushGPUEntries(Tegra::CommandList&& entries) override;
-    void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) override;
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
 
+protected:
+    void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;
+
 private:
     GPUThread::ThreadManager gpu_thread;
 };
diff --git a/src/video_core/gpu_synch.cpp b/src/video_core/gpu_synch.cpp
index 45e43b1dc..d48221077 100644
--- a/src/video_core/gpu_synch.cpp
+++ b/src/video_core/gpu_synch.cpp
@@ -8,7 +8,7 @@
 namespace VideoCommon {
 
 GPUSynch::GPUSynch(Core::System& system, VideoCore::RendererBase& renderer)
-    : GPU(system, renderer) {}
+    : GPU(system, renderer, false) {}
 
 GPUSynch::~GPUSynch() = default;
 
@@ -19,9 +19,8 @@ void GPUSynch::PushGPUEntries(Tegra::CommandList&& entries) {
     dma_pusher->DispatchCalls();
 }
 
-void GPUSynch::SwapBuffers(
-    std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
-    renderer.SwapBuffers(std::move(framebuffer));
+void GPUSynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    renderer.SwapBuffers(framebuffer);
 }
 
 void GPUSynch::FlushRegion(CacheAddr addr, u64 size) {
diff --git a/src/video_core/gpu_synch.h b/src/video_core/gpu_synch.h
index 3031fcf72..5eb1c461c 100644
--- a/src/video_core/gpu_synch.h
+++ b/src/video_core/gpu_synch.h
@@ -13,18 +13,21 @@ class RendererBase;
 namespace VideoCommon {
 
 /// Implementation of GPU interface that runs the GPU synchronously
-class GPUSynch : public Tegra::GPU {
+class GPUSynch final : public Tegra::GPU {
 public:
     explicit GPUSynch(Core::System& system, VideoCore::RendererBase& renderer);
     ~GPUSynch() override;
 
     void Start() override;
     void PushGPUEntries(Tegra::CommandList&& entries) override;
-    void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) override;
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+
+protected:
+    void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
+                             [[maybe_unused]] u32 value) const override {}
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 3f0939ec9..5f039e4fd 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -21,7 +21,8 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
     MicroProfileOnThreadCreate("GpuThread");
 
     // Wait for first GPU command before acquiring the window context
-    state.WaitForCommands();
+    while (state.queue.Empty())
+        ;
 
     // If emulation was stopped during disk shader loading, abort before trying to acquire context
     if (!state.is_running) {
@@ -32,14 +33,13 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
 
     CommandDataContainer next;
     while (state.is_running) {
-        state.WaitForCommands();
         while (!state.queue.Empty()) {
             state.queue.Pop(next);
             if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
                 dma_pusher.Push(std::move(submit_list->entries));
                 dma_pusher.DispatchCalls();
             } else if (const auto data = std::get_if<SwapBuffersCommand>(&next.data)) {
-                renderer.SwapBuffers(std::move(data->framebuffer));
+                renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr);
             } else if (const auto data = std::get_if<FlushRegionCommand>(&next.data)) {
                 renderer.Rasterizer().FlushRegion(data->addr, data->size);
             } else if (const auto data = std::get_if<InvalidateRegionCommand>(&next.data)) {
@@ -49,8 +49,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
             } else {
                 UNREACHABLE();
             }
-            state.signaled_fence = next.fence;
-            state.TrySynchronize();
+            state.signaled_fence.store(next.fence);
         }
     }
 }
@@ -79,9 +78,9 @@ void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
     system.CoreTiming().ScheduleEvent(synchronization_ticks, synchronization_event, fence);
 }
 
-void ThreadManager::SwapBuffers(
-    std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
-    PushCommand(SwapBuffersCommand(std::move(framebuffer)));
+void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    PushCommand(SwapBuffersCommand(framebuffer ? *framebuffer
+                                               : std::optional<const Tegra::FramebufferConfig>{}));
 }
 
 void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
@@ -89,12 +88,7 @@ void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
 }
 
 void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) {
-    if (state.queue.Empty()) {
-        // It's quicker to invalidate a single region on the CPU if the queue is already empty
-        system.Renderer().Rasterizer().InvalidateRegion(addr, size);
-    } else {
-        PushCommand(InvalidateRegionCommand(addr, size));
-    }
+    system.Renderer().Rasterizer().InvalidateRegion(addr, size);
 }
 
 void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
@@ -105,22 +99,13 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
-    state.SignalCommands();
     return fence;
 }
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 void SynchState::WaitForSynchronization(u64 fence) {
-    if (signaled_fence >= fence) {
-        return;
-    }
-
-    // Wait for the GPU to be idle (all commands to be executed)
-    {
-        MICROPROFILE_SCOPE(GPU_wait);
-        std::unique_lock lock{synchronization_mutex};
-        synchronization_condition.wait(lock, [this, fence] { return signaled_fence >= fence; });
-    }
+    while (signaled_fence.load() < fence)
+        ;
 }
 
 } // namespace VideoCommon::GPUThread
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
index 05a168a72..3ae0ec9f3 100644
--- a/src/video_core/gpu_thread.h
+++ b/src/video_core/gpu_thread.h
@@ -88,41 +88,9 @@ struct CommandDataContainer {
 /// Struct used to synchronize the GPU thread
 struct SynchState final {
     std::atomic_bool is_running{true};
-    std::atomic_int queued_frame_count{};
-    std::mutex synchronization_mutex;
-    std::mutex commands_mutex;
-    std::condition_variable commands_condition;
-    std::condition_variable synchronization_condition;
-
-    /// Returns true if the gap in GPU commands is small enough that we can consider the CPU and GPU
-    /// synchronized. This is entirely empirical.
-    bool IsSynchronized() const {
-        constexpr std::size_t max_queue_gap{5};
-        return queue.Size() <= max_queue_gap;
-    }
-
-    void TrySynchronize() {
-        if (IsSynchronized()) {
-            std::lock_guard lock{synchronization_mutex};
-            synchronization_condition.notify_one();
-        }
-    }
 
     void WaitForSynchronization(u64 fence);
 
-    void SignalCommands() {
-        if (queue.Empty()) {
-            return;
-        }
-
-        commands_condition.notify_one();
-    }
-
-    void WaitForCommands() {
-        std::unique_lock lock{commands_mutex};
-        commands_condition.wait(lock, [this] { return !queue.Empty(); });
-    }
-
     using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
     CommandQueue queue;
     u64 last_fence{};
@@ -142,8 +110,7 @@ public:
     void SubmitList(Tegra::CommandList&& entries);
 
     /// Swap buffers (render frame)
-    void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer);
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     void FlushRegion(CacheAddr addr, u64 size);
diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp
index c766ed692..9f59a2dc1 100644
--- a/src/video_core/macro_interpreter.cpp
+++ b/src/video_core/macro_interpreter.cpp
@@ -4,14 +4,18 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro_interpreter.h"
 
+MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
+
 namespace Tegra {
 
 MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
 
 void MacroInterpreter::Execute(u32 offset, std::vector<u32> parameters) {
+    MICROPROFILE_SCOPE(MacroInterp);
     Reset();
     registers[1] = parameters[0];
     this->parameters = std::move(parameters);
diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp
index 322453116..bffae940c 100644
--- a/src/video_core/memory_manager.cpp
+++ b/src/video_core/memory_manager.cpp
@@ -5,13 +5,17 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
+#include "core/hle/kernel/process.h"
+#include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 
 namespace Tegra {
 
-MemoryManager::MemoryManager(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {
+MemoryManager::MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
+    : rasterizer{rasterizer}, system{system} {
     std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
     std::fill(page_table.attributes.begin(), page_table.attributes.end(),
               Common::PageType::Unmapped);
@@ -49,6 +53,11 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
     const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};
 
     MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr, size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::DeviceMapped)
+               .IsSuccess());
 
     return gpu_addr;
 }
@@ -59,7 +68,11 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size)
     const u64 aligned_size{Common::AlignUp(size, page_size)};
 
     MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
-
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr, size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::DeviceMapped)
+               .IsSuccess());
     return gpu_addr;
 }
 
@@ -68,9 +81,16 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
 
     const u64 aligned_size{Common::AlignUp(size, page_size)};
     const CacheAddr cache_addr{ToCacheAddr(GetPointer(gpu_addr))};
+    const auto cpu_addr = GpuToCpuAddress(gpu_addr);
+    ASSERT(cpu_addr);
 
     rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size);
     UnmapRange(gpu_addr, aligned_size);
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr.value(), size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::None)
+               .IsSuccess());
 
     return gpu_addr;
 }
diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h
index 43a84bd52..aea010087 100644
--- a/src/video_core/memory_manager.h
+++ b/src/video_core/memory_manager.h
@@ -14,6 +14,10 @@ namespace VideoCore {
 class RasterizerInterface;
 }
 
+namespace Core {
+class System;
+}
+
 namespace Tegra {
 
 /**
@@ -47,7 +51,7 @@ struct VirtualMemoryArea {
 
 class MemoryManager final {
 public:
-    explicit MemoryManager(VideoCore::RasterizerInterface& rasterizer);
+    explicit MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer);
     ~MemoryManager();
 
     GPUVAddr AllocateSpace(u64 size, u64 align);
@@ -173,6 +177,8 @@ private:
     Common::PageTable page_table{page_bits};
     VMAMap vma_map;
     VideoCore::RasterizerInterface& rasterizer;
+
+    Core::System& system;
 };
 
 } // namespace Tegra
diff --git a/src/video_core/morton.cpp b/src/video_core/morton.cpp
index 3e91cbc83..084f85e67 100644
--- a/src/video_core/morton.cpp
+++ b/src/video_core/morton.cpp
@@ -25,8 +25,8 @@ static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth
 
     // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
     // pixel values.
-    const u32 tile_size_x{GetDefaultBlockWidth(format)};
-    const u32 tile_size_y{GetDefaultBlockHeight(format)};
+    constexpr u32 tile_size_x{GetDefaultBlockWidth(format)};
+    constexpr u32 tile_size_y{GetDefaultBlockHeight(format)};
 
     if constexpr (morton_to_linear) {
         Tegra::Texture::UnswizzleTexture(buffer, addr, tile_size_x, tile_size_y, bytes_per_pixel,
@@ -186,99 +186,6 @@ static MortonCopyFn GetSwizzleFunction(MortonSwizzleMode mode, Surface::PixelFor
     return morton_to_linear_fns[static_cast<std::size_t>(format)];
 }
 
-static u32 MortonInterleave128(u32 x, u32 y) {
-    // 128x128 Z-Order coordinate from 2D coordinates
-    static constexpr u32 xlut[] = {
-        0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042,
-        0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809,
-        0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000,
-        0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043,
-        0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a,
-        0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001,
-        0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048,
-        0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b,
-        0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002,
-        0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049,
-        0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840,
-        0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003,
-        0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a,
-        0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841,
-        0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008,
-        0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b,
-        0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842,
-        0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009,
-        0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800,
-        0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843,
-        0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a,
-        0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801,
-        0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848,
-        0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b,
-        0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802,
-        0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849,
-        0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040,
-        0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803,
-        0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a,
-        0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041,
-        0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808,
-        0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b,
-        0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042,
-        0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809,
-        0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b,
-    };
-    static constexpr u32 ylut[] = {
-        0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090,
-        0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124,
-        0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200,
-        0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294,
-        0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330,
-        0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404,
-        0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0,
-        0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534,
-        0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610,
-        0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4,
-        0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780,
-        0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014,
-        0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0,
-        0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184,
-        0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220,
-        0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4,
-        0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390,
-        0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424,
-        0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500,
-        0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594,
-        0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630,
-        0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704,
-        0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0,
-        0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034,
-        0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110,
-        0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4,
-        0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280,
-        0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314,
-        0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0,
-        0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484,
-        0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520,
-        0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4,
-        0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690,
-        0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724,
-        0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4,
-    };
-    return xlut[x % 128] + ylut[y % 128];
-}
-
-static u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) {
-    // Calculates the offset of the position of the pixel in Morton order
-    // Framebuffer images are split into 128x128 tiles.
-
-    constexpr u32 block_height = 128;
-    const u32 coarse_x = x & ~127;
-
-    const u32 i = MortonInterleave128(x, y);
-
-    const u32 offset = coarse_x * block_height;
-
-    return (i + offset) * bytes_per_pixel;
-}
-
 void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stride,
                    u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
                    u8* buffer, u8* addr) {
@@ -286,23 +193,4 @@ void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stri
                                      tile_width_spacing, buffer, addr);
 }
 
-void MortonCopyPixels128(MortonSwizzleMode mode, u32 width, u32 height, u32 bytes_per_pixel,
-                         u32 linear_bytes_per_pixel, u8* morton_data, u8* linear_data) {
-    const bool morton_to_linear = mode == MortonSwizzleMode::MortonToLinear;
-    u8* data_ptrs[2];
-    for (u32 y = 0; y < height; ++y) {
-        for (u32 x = 0; x < width; ++x) {
-            const u32 coarse_y = y & ~127;
-            const u32 morton_offset =
-                GetMortonOffset128(x, y, bytes_per_pixel) + coarse_y * width * bytes_per_pixel;
-            const u32 linear_pixel_index = (x + y * width) * linear_bytes_per_pixel;
-
-            data_ptrs[morton_to_linear ? 1 : 0] = morton_data + morton_offset;
-            data_ptrs[morton_to_linear ? 0 : 1] = &linear_data[linear_pixel_index];
-
-            std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
-        }
-    }
-}
-
 } // namespace VideoCore
diff --git a/src/video_core/morton.h b/src/video_core/morton.h
index ee5b45555..b714a7e3f 100644
--- a/src/video_core/morton.h
+++ b/src/video_core/morton.h
@@ -15,7 +15,4 @@ void MortonSwizzle(MortonSwizzleMode mode, VideoCore::Surface::PixelFormat forma
                    u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
                    u8* buffer, u8* addr);
 
-void MortonCopyPixels128(MortonSwizzleMode mode, u32 width, u32 height, u32 bytes_per_pixel,
-                         u32 linear_bytes_per_pixel, u8* morton_data, u8* linear_data);
-
 } // namespace VideoCore
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 5ee4f8e8e..6b3f2d50a 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -34,6 +34,9 @@ public:
     /// Clear the current framebuffer
     virtual void Clear() = 0;
 
+    /// Dispatches a compute shader invocation
+    virtual void DispatchCompute(GPUVAddr code_addr) = 0;
+
     /// Notify rasterizer that all caches should be flushed to Switch memory
     virtual void FlushAll() = 0;
 
@@ -47,6 +50,12 @@ public:
     /// and invalidated
     virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
+    /// Notify the rasterizer to send all written commands to the host GPU.
+    virtual void FlushCommands() = 0;
+
+    /// Notify rasterizer that a frame is about to finish
+    virtual void TickFrame() = 0;
+
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                        const Tegra::Engines::Fermi2D::Regs::Surface& dst,
diff --git a/src/video_core/renderer_base.h b/src/video_core/renderer_base.h
index 1d54c3723..af1bebc4f 100644
--- a/src/video_core/renderer_base.h
+++ b/src/video_core/renderer_base.h
@@ -36,8 +36,7 @@ public:
     virtual ~RendererBase();
 
     /// Swap buffers (render frame)
-    virtual void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) = 0;
+    virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0;
 
     /// Initialize the renderer
     virtual bool Init() = 0;
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
index 2b9bd142e..f8a807c84 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
@@ -2,103 +2,71 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <cstring>
 #include <memory>
 
-#include "common/alignment.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
+#include <glad/glad.h>
+
+#include "common/assert.h"
+#include "common/microprofile.h"
+#include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
 
 namespace OpenGL {
 
-CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
-                                     std::size_t alignment, u8* host_ptr)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, size{size}, offset{offset},
-      alignment{alignment} {}
-
-OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
-    : RasterizerCache{rasterizer}, stream_buffer(size, true) {}
-
-GLintptr OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment,
-                                      bool cache) {
-    std::lock_guard lock{mutex};
-    auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-
-    // Cache management is a big overhead, so only cache entries with a given size.
-    // TODO: Figure out which size is the best for given games.
-    cache &= size >= 2048;
-
-    const auto& host_ptr{memory_manager.GetPointer(gpu_addr)};
-    if (cache) {
-        auto entry = TryGet(host_ptr);
-        if (entry) {
-            if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
-                return entry->GetOffset();
-            }
-            Unregister(entry);
-        }
-    }
-
-    AlignBuffer(alignment);
-    const GLintptr uploaded_offset = buffer_offset;
-
-    if (!host_ptr) {
-        return uploaded_offset;
-    }
-
-    std::memcpy(buffer_ptr, host_ptr, size);
-    buffer_ptr += size;
-    buffer_offset += size;
-
-    if (cache) {
-        auto entry = std::make_shared<CachedBufferEntry>(
-            *memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr);
-        Register(entry);
-    }
-
-    return uploaded_offset;
+MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128));
+
+CachedBufferBlock::CachedBufferBlock(CacheAddr cache_addr, const std::size_t size)
+    : VideoCommon::BufferBlock{cache_addr, size} {
+    gl_buffer.Create();
+    glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
 }
 
-GLintptr OGLBufferCache::UploadHostMemory(const void* raw_pointer, std::size_t size,
-                                          std::size_t alignment) {
-    std::lock_guard lock{mutex};
-    AlignBuffer(alignment);
-    std::memcpy(buffer_ptr, raw_pointer, size);
-    const GLintptr uploaded_offset = buffer_offset;
+CachedBufferBlock::~CachedBufferBlock() = default;
+
+OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
+                               std::size_t stream_size)
+    : VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>{
+          rasterizer, system, std::make_unique<OGLStreamBuffer>(stream_size, true)} {}
+
+OGLBufferCache::~OGLBufferCache() = default;
 
-    buffer_ptr += size;
-    buffer_offset += size;
-    return uploaded_offset;
+Buffer OGLBufferCache::CreateBlock(CacheAddr cache_addr, std::size_t size) {
+    return std::make_shared<CachedBufferBlock>(cache_addr, size);
 }
 
-bool OGLBufferCache::Map(std::size_t max_size) {
-    bool invalidate;
-    std::tie(buffer_ptr, buffer_offset_base, invalidate) =
-        stream_buffer.Map(static_cast<GLsizeiptr>(max_size), 4);
-    buffer_offset = buffer_offset_base;
+void OGLBufferCache::WriteBarrier() {
+    glMemoryBarrier(GL_ALL_BARRIER_BITS);
+}
+
+const GLuint* OGLBufferCache::ToHandle(const Buffer& buffer) {
+    return buffer->GetHandle();
+}
 
-    if (invalidate) {
-        InvalidateAll();
-    }
-    return invalidate;
+const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
+    static const GLuint null_buffer = 0;
+    return &null_buffer;
 }
 
-void OGLBufferCache::Unmap() {
-    stream_buffer.Unmap(buffer_offset - buffer_offset_base);
+void OGLBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                     const u8* data) {
+    glNamedBufferSubData(*buffer->GetHandle(), static_cast<GLintptr>(offset),
+                         static_cast<GLsizeiptr>(size), data);
 }
 
-GLuint OGLBufferCache::GetHandle() const {
-    return stream_buffer.GetHandle();
+void OGLBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                       u8* data) {
+    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
+    glGetNamedBufferSubData(*buffer->GetHandle(), static_cast<GLintptr>(offset),
+                            static_cast<GLsizeiptr>(size), data);
 }
 
-void OGLBufferCache::AlignBuffer(std::size_t alignment) {
-    // Align the offset, not the mapped pointer
-    const GLintptr offset_aligned =
-        static_cast<GLintptr>(Common::AlignUp(static_cast<std::size_t>(buffer_offset), alignment));
-    buffer_ptr += offset_aligned - buffer_offset;
-    buffer_offset = offset_aligned;
+void OGLBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                               std::size_t dst_offset, std::size_t size) {
+    glCopyNamedBufferSubData(*src->GetHandle(), *dst->GetHandle(),
+                             static_cast<GLintptr>(src_offset), static_cast<GLintptr>(dst_offset),
+                             static_cast<GLsizeiptr>(size));
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index f2347581b..022e7bfa9 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -4,80 +4,63 @@
 
 #pragma once
 
-#include <cstddef>
 #include <memory>
-#include <tuple>
 
 #include "common/common_types.h"
+#include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
+namespace Core {
+class System;
+}
+
 namespace OpenGL {
 
+class OGLStreamBuffer;
 class RasterizerOpenGL;
 
-class CachedBufferEntry final : public RasterizerCacheObject {
-public:
-    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
-                               std::size_t alignment, u8* host_ptr);
-
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
+class CachedBufferBlock;
 
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    std::size_t GetSize() const {
-        return size;
-    }
+using Buffer = std::shared_ptr<CachedBufferBlock>;
 
-    GLintptr GetOffset() const {
-        return offset;
-    }
+class CachedBufferBlock : public VideoCommon::BufferBlock {
+public:
+    explicit CachedBufferBlock(CacheAddr cache_addr, const std::size_t size);
+    ~CachedBufferBlock();
 
-    std::size_t GetAlignment() const {
-        return alignment;
+    const GLuint* GetHandle() const {
+        return &gl_buffer.handle;
     }
 
 private:
-    VAddr cpu_addr{};
-    std::size_t size{};
-    GLintptr offset{};
-    std::size_t alignment{};
+    OGLBuffer gl_buffer{};
 };
 
-class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
+class OGLBufferCache final : public VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer> {
 public:
-    explicit OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size);
-
-    /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
-    /// allocated.
-    GLintptr UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
-                          bool cache = true);
+    explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
+                            std::size_t stream_size);
+    ~OGLBufferCache();
 
-    /// Uploads from a host memory. Returns host's buffer offset where it's been allocated.
-    GLintptr UploadHostMemory(const void* raw_pointer, std::size_t size, std::size_t alignment = 4);
+    const GLuint* GetEmptyBuffer(std::size_t) override;
 
-    bool Map(std::size_t max_size);
-    void Unmap();
+protected:
+    Buffer CreateBlock(CacheAddr cache_addr, std::size_t size) override;
 
-    GLuint GetHandle() const;
+    void WriteBarrier() override;
 
-protected:
-    void AlignBuffer(std::size_t alignment);
+    const GLuint* ToHandle(const Buffer& buffer) override;
 
-    // We do not have to flush this cache as things in it are never modified by us.
-    void FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& object) override {}
+    void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                         const u8* data) override;
 
-private:
-    OGLStreamBuffer stream_buffer;
+    void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                           u8* data) override;
 
-    u8* buffer_ptr = nullptr;
-    GLintptr buffer_offset = 0;
-    GLintptr buffer_offset_base = 0;
+    void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                   std::size_t dst_offset, std::size_t size) override;
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index a48e14d2e..03d434b28 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -24,8 +24,12 @@ T GetInteger(GLenum pname) {
 
 Device::Device() {
     uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
+    shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
     max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
     max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
+    has_warp_intrinsics = GLAD_GL_NV_gpu_shader5 && GLAD_GL_NV_shader_thread_group &&
+                          GLAD_GL_NV_shader_thread_shuffle;
+    has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
     has_variable_aoffi = TestVariableAoffi();
     has_component_indexing_bug = TestComponentIndexingBug();
 }
@@ -34,6 +38,8 @@ Device::Device(std::nullptr_t) {
     uniform_buffer_alignment = 0;
     max_vertex_attributes = 16;
     max_varyings = 15;
+    has_warp_intrinsics = true;
+    has_vertex_viewport_layer = true;
     has_variable_aoffi = true;
     has_component_indexing_bug = false;
 }
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index 8c8c93760..3ef7c6dd8 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -18,6 +18,10 @@ public:
         return uniform_buffer_alignment;
     }
 
+    std::size_t GetShaderStorageBufferAlignment() const {
+        return shader_storage_alignment;
+    }
+
     u32 GetMaxVertexAttributes() const {
         return max_vertex_attributes;
     }
@@ -26,6 +30,14 @@ public:
         return max_varyings;
     }
 
+    bool HasWarpIntrinsics() const {
+        return has_warp_intrinsics;
+    }
+
+    bool HasVertexViewportLayer() const {
+        return has_vertex_viewport_layer;
+    }
+
     bool HasVariableAoffi() const {
         return has_variable_aoffi;
     }
@@ -39,8 +51,11 @@ private:
     static bool TestComponentIndexingBug();
 
     std::size_t uniform_buffer_alignment{};
+    std::size_t shader_storage_alignment{};
     u32 max_vertex_attributes{};
     u32 max_varyings{};
+    bool has_warp_intrinsics{};
+    bool has_vertex_viewport_layer{};
     bool has_variable_aoffi{};
     bool has_component_indexing_bug{};
 };
diff --git a/src/video_core/renderer_opengl/gl_global_cache.cpp b/src/video_core/renderer_opengl/gl_global_cache.cpp
deleted file mode 100644
index d5e385151..000000000
--- a/src/video_core/renderer_opengl/gl_global_cache.cpp
+++ /dev/null
@@ -1,102 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <glad/glad.h>
-
-#include "common/logging/log.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
-#include "video_core/renderer_opengl/gl_rasterizer.h"
-#include "video_core/renderer_opengl/gl_shader_decompiler.h"
-#include "video_core/renderer_opengl/utils.h"
-
-namespace OpenGL {
-
-CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, host_ptr{host_ptr}, size{size},
-      max_size{max_size} {
-    buffer.Create();
-    LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory");
-}
-
-CachedGlobalRegion::~CachedGlobalRegion() = default;
-
-void CachedGlobalRegion::Reload(u32 size_) {
-    size = size_;
-    if (size > max_size) {
-        size = max_size;
-        LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the supported size {}!", size_,
-                     max_size);
-    }
-    glNamedBufferData(buffer.handle, size, host_ptr, GL_STREAM_DRAW);
-}
-
-void CachedGlobalRegion::Flush() {
-    LOG_DEBUG(Render_OpenGL, "Flushing {} bytes to CPU memory address 0x{:16}", size, cpu_addr);
-    glGetNamedBufferSubData(buffer.handle, 0, static_cast<GLsizeiptr>(size), host_ptr);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
-    const auto search{reserve.find(addr)};
-    if (search == reserve.end()) {
-        return {};
-    }
-    return search->second;
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr,
-                                                              u32 size) {
-    GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
-    if (!region) {
-        // No reserved surface available, create a new one and reserve it
-        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-        const auto cpu_addr{memory_manager.GpuToCpuAddress(addr)};
-        ASSERT(cpu_addr);
-
-        region = std::make_shared<CachedGlobalRegion>(*cpu_addr, host_ptr, size, max_ssbo_size);
-        ReserveGlobalRegion(region);
-    }
-    region->Reload(size);
-    return region;
-}
-
-void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
-    reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
-}
-
-GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
-    : RasterizerCache{rasterizer} {
-    GLint max_ssbo_size_;
-    glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size_);
-    max_ssbo_size = static_cast<u32>(max_ssbo_size_);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
-    const GLShader::GlobalMemoryEntry& global_region,
-    Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
-    std::lock_guard lock{mutex};
-
-    auto& gpu{Core::System::GetInstance().GPU()};
-    auto& memory_manager{gpu.MemoryManager()};
-    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
-    const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
-                    global_region.GetCbufOffset()};
-    const auto actual_addr{memory_manager.Read<u64>(addr)};
-    const auto size{memory_manager.Read<u32>(addr + 8)};
-
-    // Look up global region in the cache based on address
-    const auto& host_ptr{memory_manager.GetPointer(actual_addr)};
-    GlobalRegion region{TryGet(host_ptr)};
-
-    if (!region) {
-        // No global region found - create a new one
-        region = GetUncachedGlobalRegion(actual_addr, host_ptr, size);
-        Register(region);
-    }
-
-    return region;
-}
-
-} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_global_cache.h b/src/video_core/renderer_opengl/gl_global_cache.h
deleted file mode 100644
index 2d467a240..000000000
--- a/src/video_core/renderer_opengl/gl_global_cache.h
+++ /dev/null
@@ -1,82 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <memory>
-#include <unordered_map>
-
-#include <glad/glad.h>
-
-#include "common/assert.h"
-#include "common/common_types.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-
-namespace OpenGL {
-
-namespace GLShader {
-class GlobalMemoryEntry;
-}
-
-class RasterizerOpenGL;
-class CachedGlobalRegion;
-using GlobalRegion = std::shared_ptr<CachedGlobalRegion>;
-
-class CachedGlobalRegion final : public RasterizerCacheObject {
-public:
-    explicit CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size);
-    ~CachedGlobalRegion();
-
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    /// Gets the GL program handle for the buffer
-    GLuint GetBufferHandle() const {
-        return buffer.handle;
-    }
-
-    /// Reloads the global region from guest memory
-    void Reload(u32 size_);
-
-    void Flush();
-
-private:
-    VAddr cpu_addr{};
-    u8* host_ptr{};
-    u32 size{};
-    u32 max_size{};
-
-    OGLBuffer buffer;
-};
-
-class GlobalRegionCacheOpenGL final : public RasterizerCache<GlobalRegion> {
-public:
-    explicit GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer);
-
-    /// Gets the current specified shader stage program
-    GlobalRegion GetGlobalRegion(const GLShader::GlobalMemoryEntry& descriptor,
-                                 Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
-
-protected:
-    void FlushObjectInner(const GlobalRegion& object) override {
-        object->Flush();
-    }
-
-private:
-    GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
-    GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, u32 size);
-    void ReserveGlobalRegion(GlobalRegion region);
-
-    std::unordered_map<CacheAddr, GlobalRegion> reserve;
-    u32 max_ssbo_size{};
-};
-
-} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index f45a3c5ef..bb09ecd52 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -4,6 +4,7 @@
 
 #include <algorithm>
 #include <array>
+#include <bitset>
 #include <memory>
 #include <string>
 #include <string_view>
@@ -19,7 +20,9 @@
 #include "core/core.h"
 #include "core/hle/kernel/process.h"
 #include "core/settings.h"
+#include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
@@ -80,16 +83,31 @@ struct DrawParameters {
     }
 };
 
+static std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
+                                      const GLShader::ConstBufferEntry& entry) {
+    if (!entry.IsIndirect()) {
+        return entry.GetSize();
+    }
+
+    if (buffer.size > Maxwell::MaxConstBufferSize) {
+        LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size,
+                    Maxwell::MaxConstBufferSize);
+        return Maxwell::MaxConstBufferSize;
+    }
+
+    return buffer.size;
+}
+
 RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window,
                                    ScreenInfo& info)
     : texture_cache{system, *this, device}, shader_cache{*this, system, emu_window, device},
-      global_cache{*this}, system{system}, screen_info{info},
-      buffer_cache(*this, STREAM_BUFFER_SIZE) {
+      system{system}, screen_info{info}, buffer_cache{*this, system, STREAM_BUFFER_SIZE} {
     OpenGLState::ApplyDefaultState();
 
     shader_program_manager = std::make_unique<GLShader::ProgramManager>();
     state.draw.shader_program = 0;
     state.Apply();
+    clear_framebuffer.Create();
 
     LOG_DEBUG(Render_OpenGL, "Sync fixed function OpenGL state here");
     CheckExtensions();
@@ -109,10 +127,10 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
     auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
-    if (!gpu.dirty_flags.vertex_attrib_format) {
+    if (!gpu.dirty.vertex_attrib_format) {
         return state.draw.vertex_array;
     }
-    gpu.dirty_flags.vertex_attrib_format = false;
+    gpu.dirty.vertex_attrib_format = false;
 
     MICROPROFILE_SCOPE(OpenGL_VAO);
 
@@ -129,8 +147,6 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
         state.draw.vertex_array = vao;
         state.ApplyVertexArrayState();
 
-        glVertexArrayElementBuffer(vao, buffer_cache.GetHandle());
-
         // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
         // Enables the first 16 vertex attributes always, as we don't know which ones are actually
         // used until shader time. Note, Tegra technically supports 32, but we're capping this to 16
@@ -168,7 +184,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
     }
 
     // Rebinding the VAO invalidates the vertex buffer bindings.
-    gpu.dirty_flags.vertex_array.set();
+    gpu.dirty.ResetVertexArrays();
 
     state.draw.vertex_array = vao_entry.handle;
     return vao_entry.handle;
@@ -176,17 +192,20 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
 
 void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
     auto& gpu = system.GPU().Maxwell3D();
-    const auto& regs = gpu.regs;
-
-    if (gpu.dirty_flags.vertex_array.none())
+    if (!gpu.dirty.vertex_array_buffers)
         return;
+    gpu.dirty.vertex_array_buffers = false;
+
+    const auto& regs = gpu.regs;
 
     MICROPROFILE_SCOPE(OpenGL_VB);
 
     // Upload all guest vertex arrays sequentially to our buffer
     for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        if (!gpu.dirty_flags.vertex_array[index])
+        if (!gpu.dirty.vertex_array[index])
             continue;
+        gpu.dirty.vertex_array[index] = false;
+        gpu.dirty.vertex_instance[index] = false;
 
         const auto& vertex_array = regs.vertex_array[index];
         if (!vertex_array.IsEnabled())
@@ -197,11 +216,11 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
 
         ASSERT(end > start);
         const u64 size = end - start + 1;
-        const GLintptr vertex_buffer_offset = buffer_cache.UploadMemory(start, size);
+        const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size);
 
         // Bind the vertex array to the buffer at the current offset.
-        glVertexArrayVertexBuffer(vao, index, buffer_cache.GetHandle(), vertex_buffer_offset,
-                                  vertex_array.stride);
+        vertex_array_pushbuffer.SetVertexBuffer(index, vertex_buffer, vertex_buffer_offset,
+                                                vertex_array.stride);
 
         if (regs.instanced_arrays.IsInstancingEnabled(index) && vertex_array.divisor != 0) {
             // Enable vertex buffer instancing with the specified divisor.
@@ -211,11 +230,47 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
             glVertexArrayBindingDivisor(vao, index, 0);
         }
     }
+}
 
-    gpu.dirty_flags.vertex_array.reset();
+void RasterizerOpenGL::SetupVertexInstances(GLuint vao) {
+    auto& gpu = system.GPU().Maxwell3D();
+
+    if (!gpu.dirty.vertex_instances)
+        return;
+    gpu.dirty.vertex_instances = false;
+
+    const auto& regs = gpu.regs;
+    // Upload all guest vertex arrays sequentially to our buffer
+    for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
+        if (!gpu.dirty.vertex_instance[index])
+            continue;
+
+        gpu.dirty.vertex_instance[index] = false;
+
+        if (regs.instanced_arrays.IsInstancingEnabled(index) &&
+            regs.vertex_array[index].divisor != 0) {
+            // Enable vertex buffer instancing with the specified divisor.
+            glVertexArrayBindingDivisor(vao, index, regs.vertex_array[index].divisor);
+        } else {
+            // Disable the vertex buffer instancing.
+            glVertexArrayBindingDivisor(vao, index, 0);
+        }
+    }
 }
 
-DrawParameters RasterizerOpenGL::SetupDraw() {
+GLintptr RasterizerOpenGL::SetupIndexBuffer() {
+    if (accelerate_draw != AccelDraw::Indexed) {
+        return 0;
+    }
+    MICROPROFILE_SCOPE(OpenGL_Index);
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    const std::size_t size = CalculateIndexBufferSize();
+    const auto [buffer, offset] = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
+    vertex_array_pushbuffer.SetIndexBuffer(buffer);
+    return offset;
+}
+
+DrawParameters RasterizerOpenGL::SetupDraw(GLintptr index_buffer_offset) {
     const auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
     const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
@@ -227,11 +282,9 @@ DrawParameters RasterizerOpenGL::SetupDraw() {
     params.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology);
 
     if (is_indexed) {
-        MICROPROFILE_SCOPE(OpenGL_Index);
         params.index_format = MaxwellToGL::IndexFormat(regs.index_array.format);
         params.count = regs.index_array.count;
-        params.index_buffer_offset =
-            buffer_cache.UploadMemory(regs.index_array.IndexStart(), CalculateIndexBufferSize());
+        params.index_buffer_offset = index_buffer_offset;
         params.base_vertex = static_cast<GLint>(regs.vb_element_base);
     } else {
         params.count = regs.vertex_buffer.count;
@@ -247,10 +300,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
     BaseBindings base_bindings;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
 
-    // Prepare packed bindings
-    bind_ubo_pushbuffer.Setup(base_bindings.cbuf);
-    bind_ssbo_pushbuffer.Setup(base_bindings.gmem);
-
     for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         const auto& shader_config = gpu.regs.shader_config[index];
         const Maxwell::ShaderProgram program{static_cast<Maxwell::ShaderProgram>(index)};
@@ -271,18 +320,17 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
 
         GLShader::MaxwellUniformData ubo{};
         ubo.SetFromRegs(gpu, stage);
-        const GLintptr offset =
+        const auto [buffer, offset] =
             buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment());
 
         // Bind the emulation info buffer
-        bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset,
-                                 static_cast<GLsizeiptr>(sizeof(ubo)));
+        bind_ubo_pushbuffer.Push(buffer, offset, static_cast<GLsizeiptr>(sizeof(ubo)));
 
         Shader shader{shader_cache.GetStageProgram(program)};
 
-        const auto stage_enum{static_cast<Maxwell::ShaderStage>(stage)};
+        const auto stage_enum = static_cast<Maxwell::ShaderStage>(stage);
         SetupDrawConstBuffers(stage_enum, shader);
-        SetupGlobalRegions(stage_enum, shader);
+        SetupDrawGlobalMemory(stage_enum, shader);
         const auto texture_buffer_usage{SetupTextures(stage_enum, shader, base_bindings)};
 
         const ProgramVariant variant{base_bindings, primitive_mode, texture_buffer_usage};
@@ -321,12 +369,9 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         base_bindings = next_bindings;
     }
 
-    bind_ubo_pushbuffer.Bind();
-    bind_ssbo_pushbuffer.Bind();
-
     SyncClipEnabled(clip_distances);
 
-    gpu.dirty_flags.shaders = false;
+    gpu.dirty.shaders = false;
 }
 
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@@ -409,13 +454,13 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
 
     const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
                                                  single_color_target};
-    if (fb_config_state == current_framebuffer_config_state &&
-        gpu.dirty_flags.color_buffer.none() && !gpu.dirty_flags.zeta_buffer) {
+    if (fb_config_state == current_framebuffer_config_state && !gpu.dirty.render_settings) {
         // Only skip if the previous ConfigureFramebuffers call was from the same kind (multiple or
         // single color targets). This is done because the guest registers may not change but the
         // host framebuffer may contain different attachments
         return current_depth_stencil_usage;
     }
+    gpu.dirty.render_settings = false;
     current_framebuffer_config_state = fb_config_state;
 
     texture_cache.GuardRenderTargets(true);
@@ -504,13 +549,71 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
     return current_depth_stencil_usage = {static_cast<bool>(depth_surface), fbkey.stencil_enable};
 }
 
+void RasterizerOpenGL::ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
+                                                 bool using_depth_fb, bool using_stencil_fb) {
+    auto& gpu = system.GPU().Maxwell3D();
+    const auto& regs = gpu.regs;
+
+    texture_cache.GuardRenderTargets(true);
+    View color_surface{};
+    if (using_color_fb) {
+        color_surface = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT, false);
+    }
+    View depth_surface{};
+    if (using_depth_fb || using_stencil_fb) {
+        depth_surface = texture_cache.GetDepthBufferSurface(false);
+    }
+    texture_cache.GuardRenderTargets(false);
+
+    current_state.draw.draw_framebuffer = clear_framebuffer.handle;
+    current_state.ApplyFramebufferState();
+
+    if (color_surface) {
+        color_surface->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER);
+    } else {
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+    }
+
+    if (depth_surface) {
+        const auto& params = depth_surface->GetSurfaceParams();
+        switch (params.type) {
+        case VideoCore::Surface::SurfaceType::Depth: {
+            depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+            break;
+        }
+        case VideoCore::Surface::SurfaceType::DepthStencil: {
+            depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
+            break;
+        }
+        default: { UNIMPLEMENTED(); }
+        }
+    } else {
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
+                               0);
+    }
+}
+
 void RasterizerOpenGL::Clear() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    const auto& maxwell3d = system.GPU().Maxwell3D();
+
+    if (!maxwell3d.ShouldExecute()) {
+        return;
+    }
+
+    const auto& regs = maxwell3d.regs;
     bool use_color{};
     bool use_depth{};
     bool use_stencil{};
 
-    OpenGLState clear_state;
+    OpenGLState prev_state{OpenGLState::GetCurState()};
+    SCOPE_EXIT({
+        prev_state.AllDirty();
+        prev_state.Apply();
+    });
+
+    OpenGLState clear_state{OpenGLState::GetCurState()};
+    clear_state.SetDefaultViewports();
     if (regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B ||
         regs.clear_buffers.A) {
         use_color = true;
@@ -530,6 +633,7 @@ void RasterizerOpenGL::Clear() {
         // true.
         clear_state.depth.test_enabled = true;
         clear_state.depth.test_func = GL_ALWAYS;
+        clear_state.depth.write_mask = GL_TRUE;
     }
     if (regs.clear_buffers.S) {
         ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear stencil but buffer is not enabled!");
@@ -566,8 +670,9 @@ void RasterizerOpenGL::Clear() {
         return;
     }
 
-    const auto [clear_depth, clear_stencil] = ConfigureFramebuffers(
-        clear_state, use_color, use_depth || use_stencil, false, regs.clear_buffers.RT.Value());
+    ConfigureClearFramebuffer(clear_state, use_color, use_depth, use_stencil);
+
+    SyncViewport(clear_state);
     if (regs.clear_flags.scissor) {
         SyncScissorTest(clear_state);
     }
@@ -576,21 +681,18 @@ void RasterizerOpenGL::Clear() {
         clear_state.EmulateViewportWithScissor();
     }
 
-    clear_state.ApplyColorMask();
-    clear_state.ApplyDepth();
-    clear_state.ApplyStencilTest();
-    clear_state.ApplyViewport();
-    clear_state.ApplyFramebufferState();
+    clear_state.AllDirty();
+    clear_state.Apply();
 
     if (use_color) {
-        glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color);
+        glClearBufferfv(GL_COLOR, 0, regs.clear_color);
     }
 
-    if (clear_depth && clear_stencil) {
+    if (use_depth && use_stencil) {
         glClearBufferfi(GL_DEPTH_STENCIL, 0, regs.clear_depth, regs.clear_stencil);
-    } else if (clear_depth) {
+    } else if (use_depth) {
         glClearBufferfv(GL_DEPTH, 0, &regs.clear_depth);
-    } else if (clear_stencil) {
+    } else if (use_stencil) {
         glClearBufferiv(GL_STENCIL, 0, &regs.clear_stencil);
     }
 }
@@ -601,7 +703,10 @@ void RasterizerOpenGL::DrawArrays() {
 
     MICROPROFILE_SCOPE(OpenGL_Drawing);
     auto& gpu = system.GPU().Maxwell3D();
-    const auto& regs = gpu.regs;
+
+    if (!gpu.ShouldExecute()) {
+        return;
+    }
 
     SyncColorMask();
     SyncFragmentColorClampState();
@@ -634,26 +739,47 @@ void RasterizerOpenGL::DrawArrays() {
                       Maxwell::MaxShaderStage;
 
     // Add space for at least 18 constant buffers
-    buffer_size +=
-        Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment());
+    buffer_size += Maxwell::MaxConstBuffers *
+                   (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
 
-    const bool invalidate = buffer_cache.Map(buffer_size);
-    if (invalidate) {
-        // As all cached buffers are invalidated, we need to recheck their state.
-        gpu.dirty_flags.vertex_array.set();
-    }
+    // Prepare the vertex array.
+    buffer_cache.Map(buffer_size);
 
+    // Prepare vertex array format.
     const GLuint vao = SetupVertexFormat();
+    vertex_array_pushbuffer.Setup(vao);
+
+    // Upload vertex and index data.
     SetupVertexBuffer(vao);
+    SetupVertexInstances(vao);
+    const GLintptr index_buffer_offset = SetupIndexBuffer();
+
+    // Setup draw parameters. It will automatically choose what glDraw* method to use.
+    const DrawParameters params = SetupDraw(index_buffer_offset);
 
-    DrawParameters params = SetupDraw();
+    // Prepare packed bindings.
+    bind_ubo_pushbuffer.Setup(0);
+    bind_ssbo_pushbuffer.Setup(0);
+
+    // Setup shaders and their used resources.
     texture_cache.GuardSamplers(true);
     SetupShaders(params.primitive_mode);
     texture_cache.GuardSamplers(false);
 
     ConfigureFramebuffers(state);
 
-    buffer_cache.Unmap();
+    // Signal the buffer cache that we are not going to upload more things.
+    const bool invalidate = buffer_cache.Unmap();
+
+    // Now that we are no longer uploading data, we can safely bind the buffers to OpenGL.
+    vertex_array_pushbuffer.Bind();
+    bind_ubo_pushbuffer.Bind();
+    bind_ssbo_pushbuffer.Bind();
+
+    if (invalidate) {
+        // As all cached buffers are invalidated, we need to recheck their state.
+        gpu.dirty.ResetVertexArrays();
+    }
 
     shader_program_manager->ApplyTo(state);
     state.Apply();
@@ -665,6 +791,46 @@ void RasterizerOpenGL::DrawArrays() {
     params.DispatchDraw();
 
     accelerate_draw = AccelDraw::Disabled;
+    gpu.dirty.memory_general = false;
+}
+
+void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
+    if (!GLAD_GL_ARB_compute_variable_group_size) {
+        LOG_ERROR(Render_OpenGL, "Compute is currently not supported on this device due to the "
+                                 "lack of GL_ARB_compute_variable_group_size");
+        return;
+    }
+
+    auto kernel = shader_cache.GetComputeKernel(code_addr);
+    const auto [program, next_bindings] = kernel->GetProgramHandle({});
+    state.draw.shader_program = program;
+    state.draw.program_pipeline = 0;
+
+    const std::size_t buffer_size =
+        Tegra::Engines::KeplerCompute::NumConstBuffers *
+        (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
+    buffer_cache.Map(buffer_size);
+
+    bind_ubo_pushbuffer.Setup(0);
+    bind_ssbo_pushbuffer.Setup(0);
+
+    SetupComputeConstBuffers(kernel);
+    SetupComputeGlobalMemory(kernel);
+
+    // TODO(Rodrigo): Bind images and samplers
+
+    buffer_cache.Unmap();
+
+    bind_ubo_pushbuffer.Bind();
+    bind_ssbo_pushbuffer.Bind();
+
+    state.ApplyShaderProgram();
+    state.ApplyProgramPipeline();
+
+    const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
+    glDispatchComputeGroupSizeARB(launch_desc.grid_dim_x, launch_desc.grid_dim_y,
+                                  launch_desc.grid_dim_z, launch_desc.block_dim_x,
+                                  launch_desc.block_dim_y, launch_desc.block_dim_z);
 }
 
 void RasterizerOpenGL::FlushAll() {}
@@ -675,7 +841,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
         return;
     }
     texture_cache.FlushRegion(addr, size);
-    global_cache.FlushRegion(addr, size);
+    buffer_cache.FlushRegion(addr, size);
 }
 
 void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
@@ -685,7 +851,6 @@ void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     }
     texture_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
-    global_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
 }
 
@@ -696,6 +861,14 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     InvalidateRegion(addr, size);
 }
 
+void RasterizerOpenGL::FlushCommands() {
+    glFlush();
+}
+
+void RasterizerOpenGL::TickFrame() {
+    buffer_cache.TickFrame();
+}
+
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                              const Tegra::Engines::Fermi2D::Regs::Surface& dst,
                                              const Tegra::Engines::Fermi2D::Config& copy_config) {
@@ -737,14 +910,25 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 void RasterizerOpenGL::SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
                                              const Shader& shader) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
-    const auto stage_index = static_cast<std::size_t>(stage);
-    const auto& shader_stage = system.GPU().Maxwell3D().state.shader_stages[stage_index];
-    const auto& entries = shader->GetShaderEntries().const_buffers;
+    const auto& stages = system.GPU().Maxwell3D().state.shader_stages;
+    const auto& shader_stage = stages[static_cast<std::size_t>(stage)];
+    for (const auto& entry : shader->GetShaderEntries().const_buffers) {
+        const auto& buffer = shader_stage.const_buffers[entry.GetIndex()];
+        SetupConstBuffer(buffer, entry);
+    }
+}
 
-    // Upload only the enabled buffers from the 16 constbuffers of each shader stage
-    for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
-        const auto& entry = entries[bindpoint];
-        SetupConstBuffer(shader_stage.const_buffers[entry.GetIndex()], entry);
+void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
+    MICROPROFILE_SCOPE(OpenGL_UBO);
+    const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
+    for (const auto& entry : kernel->GetShaderEntries().const_buffers) {
+        const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
+        const std::bitset<8> mask = launch_desc.memory_config.const_buffer_enable_mask.Value();
+        Tegra::Engines::ConstBufferInfo buffer;
+        buffer.address = config.Address();
+        buffer.size = config.size;
+        buffer.enabled = mask[entry.GetIndex()];
+        SetupConstBuffer(buffer, entry);
     }
 }
 
@@ -752,49 +936,52 @@ void RasterizerOpenGL::SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& b
                                         const GLShader::ConstBufferEntry& entry) {
     if (!buffer.enabled) {
         // Set values to zero to unbind buffers
-        bind_ubo_pushbuffer.Push(0, 0, 0);
+        bind_ubo_pushbuffer.Push(buffer_cache.GetEmptyBuffer(sizeof(float)), 0, sizeof(float));
         return;
     }
 
-    std::size_t size;
-    if (entry.IsIndirect()) {
-        // Buffer is accessed indirectly, so upload the entire thing
-        size = buffer.size;
-
-        if (size > MaxConstbufferSize) {
-            LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", size,
-                        MaxConstbufferSize);
-            size = MaxConstbufferSize;
-        }
-    } else {
-        // Buffer is accessed directly, upload just what we use
-        size = entry.GetSize();
-    }
-
     // Align the actual size so it ends up being a multiple of vec4 to meet the OpenGL std140
     // UBO alignment requirements.
-    size = Common::AlignUp(size, sizeof(GLvec4));
-    ASSERT_MSG(size <= MaxConstbufferSize, "Constant buffer is too big");
+    const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4));
 
-    const std::size_t alignment = device.GetUniformBufferAlignment();
-    const GLintptr offset = buffer_cache.UploadMemory(buffer.address, size, alignment);
-    bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset, size);
+    const auto alignment = device.GetUniformBufferAlignment();
+    const auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment);
+    bind_ubo_pushbuffer.Push(cbuf, offset, size);
 }
 
-void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                                          const Shader& shader) {
-    const auto& entries = shader->GetShaderEntries().global_memory_entries;
-    for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
-        const auto& entry{entries[bindpoint]};
-        const auto& region{global_cache.GetGlobalRegion(entry, stage)};
-        if (entry.IsWritten()) {
-            region->MarkAsModified(true, global_cache);
-        }
-        bind_ssbo_pushbuffer.Push(region->GetBufferHandle(), 0,
-                                  static_cast<GLsizeiptr>(region->GetSizeInBytes()));
+void RasterizerOpenGL::SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+                                             const Shader& shader) {
+    auto& gpu{system.GPU()};
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
+    for (const auto& entry : shader->GetShaderEntries().global_memory_entries) {
+        const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
+        const auto gpu_addr{memory_manager.Read<u64>(addr)};
+        const auto size{memory_manager.Read<u32>(addr + 8)};
+        SetupGlobalMemory(entry, gpu_addr, size);
+    }
+}
+
+void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
+    auto& gpu{system.GPU()};
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
+    for (const auto& entry : kernel->GetShaderEntries().global_memory_entries) {
+        const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()};
+        const auto gpu_addr{memory_manager.Read<u64>(addr)};
+        const auto size{memory_manager.Read<u32>(addr + 8)};
+        SetupGlobalMemory(entry, gpu_addr, size);
     }
 }
 
+void RasterizerOpenGL::SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry,
+                                         GPUVAddr gpu_addr, std::size_t size) {
+    const auto alignment{device.GetShaderStorageBufferAlignment()};
+    const auto [ssbo, buffer_offset] =
+        buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.IsWritten());
+    bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
+}
+
 TextureBufferUsage RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
                                                    BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
@@ -883,10 +1070,11 @@ void RasterizerOpenGL::SyncClipCoef() {
 }
 
 void RasterizerOpenGL::SyncCullMode() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
 
-    state.cull.enabled = regs.cull.enabled != 0;
+    const auto& regs = maxwell3d.regs;
 
+    state.cull.enabled = regs.cull.enabled != 0;
     if (state.cull.enabled) {
         state.cull.front_face = MaxwellToGL::FrontFace(regs.cull.front_face);
         state.cull.mode = MaxwellToGL::CullFace(regs.cull.cull_face);
@@ -919,16 +1107,21 @@ void RasterizerOpenGL::SyncDepthTestState() {
     state.depth.test_enabled = regs.depth_test_enable != 0;
     state.depth.write_mask = regs.depth_write_enabled ? GL_TRUE : GL_FALSE;
 
-    if (!state.depth.test_enabled)
+    if (!state.depth.test_enabled) {
         return;
+    }
 
     state.depth.test_func = MaxwellToGL::ComparisonOp(regs.depth_test_func);
 }
 
 void RasterizerOpenGL::SyncStencilTestState() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
-    state.stencil.test_enabled = regs.stencil_enable != 0;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.stencil_test) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
 
+    state.stencil.test_enabled = regs.stencil_enable != 0;
     if (!regs.stencil_enable) {
         return;
     }
@@ -957,10 +1150,17 @@ void RasterizerOpenGL::SyncStencilTestState() {
         state.stencil.back.action_depth_fail = GL_KEEP;
         state.stencil.back.action_depth_pass = GL_KEEP;
     }
+    state.MarkDirtyStencilState();
+    maxwell3d.dirty.stencil_test = false;
 }
 
 void RasterizerOpenGL::SyncColorMask() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.color_mask) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
+
     const std::size_t count =
         regs.independent_blend_enable ? Tegra::Engines::Maxwell3D::Regs::NumRenderTargets : 1;
     for (std::size_t i = 0; i < count; i++) {
@@ -971,6 +1171,9 @@ void RasterizerOpenGL::SyncColorMask() {
         dest.blue_enabled = (source.B == 0) ? GL_FALSE : GL_TRUE;
         dest.alpha_enabled = (source.A == 0) ? GL_FALSE : GL_TRUE;
     }
+
+    state.MarkDirtyColorMask();
+    maxwell3d.dirty.color_mask = false;
 }
 
 void RasterizerOpenGL::SyncMultiSampleState() {
@@ -985,7 +1188,11 @@ void RasterizerOpenGL::SyncFragmentColorClampState() {
 }
 
 void RasterizerOpenGL::SyncBlendState() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.blend_state) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
 
     state.blend_color.red = regs.blend_color.r;
     state.blend_color.green = regs.blend_color.g;
@@ -1008,6 +1215,8 @@ void RasterizerOpenGL::SyncBlendState() {
         for (std::size_t i = 1; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) {
             state.blend[i].enabled = false;
         }
+        maxwell3d.dirty.blend_state = false;
+        state.MarkDirtyBlendState();
         return;
     }
 
@@ -1024,6 +1233,9 @@ void RasterizerOpenGL::SyncBlendState() {
         blend.src_a_func = MaxwellToGL::BlendFunc(src.factor_source_a);
         blend.dst_a_func = MaxwellToGL::BlendFunc(src.factor_dest_a);
     }
+
+    state.MarkDirtyBlendState();
+    maxwell3d.dirty.blend_state = false;
 }
 
 void RasterizerOpenGL::SyncLogicOpState() {
@@ -1075,13 +1287,21 @@ void RasterizerOpenGL::SyncPointState() {
 }
 
 void RasterizerOpenGL::SyncPolygonOffset() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.polygon_offset) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
+
     state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0;
     state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0;
     state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0;
     state.polygon_offset.units = regs.polygon_offset_units;
     state.polygon_offset.factor = regs.polygon_offset_factor;
     state.polygon_offset.clamp = regs.polygon_offset_clamp;
+
+    state.MarkDirtyPolygonOffset();
+    maxwell3d.dirty.polygon_offset = false;
 }
 
 void RasterizerOpenGL::SyncAlphaTest() {
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index bf67e3a70..9d20a4fbf 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -24,7 +24,6 @@
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_sampler_cache.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
@@ -59,10 +58,13 @@ public:
 
     void DrawArrays() override;
     void Clear() override;
+    void DispatchCompute(GPUVAddr code_addr) override;
     void FlushAll() override;
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushCommands() override;
+    void TickFrame() override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
@@ -73,11 +75,6 @@ public:
     void LoadDiskResources(const std::atomic_bool& stop_loading,
                            const VideoCore::DiskResourceLoadCallback& callback) override;
 
-    /// Maximum supported size that a constbuffer can have in bytes.
-    static constexpr std::size_t MaxConstbufferSize = 0x10000;
-    static_assert(MaxConstbufferSize % sizeof(GLvec4) == 0,
-                  "The maximum size of a constbuffer must be a multiple of the size of GLvec4");
-
 private:
     struct FramebufferConfigState {
         bool using_color_fb{};
@@ -98,30 +95,45 @@ private:
 
     /**
      * Configures the color and depth framebuffer states.
-     * @param must_reconfigure If true, tells the framebuffer to skip the cache and reconfigure
-     * again. Used by the texture cache to solve texception conflicts
-     * @param use_color_fb If true, configure color framebuffers.
-     * @param using_depth_fb If true, configure the depth/stencil framebuffer.
-     * @param preserve_contents If true, tries to preserve data from a previously used framebuffer.
+     *
+     * @param current_state       The current OpenGL state.
+     * @param using_color_fb      If true, configure color framebuffers.
+     * @param using_depth_fb      If true, configure the depth/stencil framebuffer.
+     * @param preserve_contents   If true, tries to preserve data from a previously used
+     *                            framebuffer.
      * @param single_color_target Specifies if a single color buffer target should be used.
+     *
      * @returns If depth (first) or stencil (second) are being stored in the bound zeta texture
-     * (requires using_depth_fb to be true)
+     *          (requires using_depth_fb to be true)
      */
     std::pair<bool, bool> ConfigureFramebuffers(
-        OpenGLState& current_state, bool use_color_fb = true, bool using_depth_fb = true,
+        OpenGLState& current_state, bool using_color_fb = true, bool using_depth_fb = true,
         bool preserve_contents = true, std::optional<std::size_t> single_color_target = {});
 
+    void ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
+                                   bool using_depth_fb, bool using_stencil_fb);
+
     /// Configures the current constbuffers to use for the draw command.
     void SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
                                const Shader& shader);
 
+    /// Configures the current constbuffers to use for the kernel invocation.
+    void SetupComputeConstBuffers(const Shader& kernel);
+
     /// Configures a constant buffer.
     void SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& buffer,
                           const GLShader::ConstBufferEntry& entry);
 
     /// Configures the current global memory entries to use for the draw command.
-    void SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                            const Shader& shader);
+    void SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+                               const Shader& shader);
+
+    /// Configures the current global memory entries to use for the kernel invocation.
+    void SetupComputeGlobalMemory(const Shader& kernel);
+
+    /// Configures a constant buffer.
+    void SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
+                           std::size_t size);
 
     /// Configures the current textures to use for the draw command. Returns shaders texture buffer
     /// usage.
@@ -189,7 +201,6 @@ private:
 
     TextureCacheOpenGL texture_cache;
     ShaderCacheOpenGL shader_cache;
-    GlobalRegionCacheOpenGL global_cache;
     SamplerCacheOpenGL sampler_cache;
     FramebufferCacheOpenGL framebuffer_cache;
 
@@ -208,6 +219,7 @@ private:
     static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
     OGLBufferCache buffer_cache;
 
+    VertexArrayPushBuffer vertex_array_pushbuffer;
     BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER};
     BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER};
 
@@ -219,14 +231,19 @@ private:
     GLuint SetupVertexFormat();
 
     void SetupVertexBuffer(GLuint vao);
+    void SetupVertexInstances(GLuint vao);
 
-    DrawParameters SetupDraw();
+    GLintptr SetupIndexBuffer();
+
+    DrawParameters SetupDraw(GLintptr index_buffer_offset);
 
     void SetupShaders(GLenum primitive_mode);
 
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;
 
+    OGLFramebuffer clear_framebuffer;
+
     using CachedPageMap = boost::icl::interval_map<u64, int>;
     CachedPageMap cached_pages;
 };
diff --git a/src/video_core/renderer_opengl/gl_sampler_cache.h b/src/video_core/renderer_opengl/gl_sampler_cache.h
index defbc2d81..34ee37f00 100644
--- a/src/video_core/renderer_opengl/gl_sampler_cache.h
+++ b/src/video_core/renderer_opengl/gl_sampler_cache.h
@@ -17,9 +17,9 @@ public:
     ~SamplerCacheOpenGL();
 
 protected:
-    OGLSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const;
+    OGLSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
 
-    GLuint ToSamplerType(const OGLSampler& sampler) const;
+    GLuint ToSamplerType(const OGLSampler& sampler) const override;
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index f9b2b03a0..cf6a5cddf 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -23,13 +23,13 @@ namespace OpenGL {
 
 using VideoCommon::Shader::ProgramCode;
 
-// One UBO is always reserved for emulation values
-constexpr u32 RESERVED_UBOS = 1;
+// One UBO is always reserved for emulation values on staged shaders
+constexpr u32 STAGE_RESERVED_UBOS = 1;
 
 struct UnspecializedShader {
     std::string code;
     GLShader::ShaderEntries entries;
-    Maxwell::ShaderProgram program_type;
+    ProgramType program_type;
 };
 
 namespace {
@@ -55,15 +55,17 @@ ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr g
 }
 
 /// Gets the shader type from a Maxwell program type
-constexpr GLenum GetShaderType(Maxwell::ShaderProgram program_type) {
+constexpr GLenum GetShaderType(ProgramType program_type) {
     switch (program_type) {
-    case Maxwell::ShaderProgram::VertexA:
-    case Maxwell::ShaderProgram::VertexB:
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
         return GL_VERTEX_SHADER;
-    case Maxwell::ShaderProgram::Geometry:
+    case ProgramType::Geometry:
         return GL_GEOMETRY_SHADER;
-    case Maxwell::ShaderProgram::Fragment:
+    case ProgramType::Fragment:
         return GL_FRAGMENT_SHADER;
+    case ProgramType::Compute:
+        return GL_COMPUTE_SHADER;
     default:
         return GL_NONE;
     }
@@ -100,6 +102,25 @@ constexpr std::tuple<const char*, const char*, u32> GetPrimitiveDescription(GLen
     }
 }
 
+ProgramType GetProgramType(Maxwell::ShaderProgram program) {
+    switch (program) {
+    case Maxwell::ShaderProgram::VertexA:
+        return ProgramType::VertexA;
+    case Maxwell::ShaderProgram::VertexB:
+        return ProgramType::VertexB;
+    case Maxwell::ShaderProgram::TesselationControl:
+        return ProgramType::TessellationControl;
+    case Maxwell::ShaderProgram::TesselationEval:
+        return ProgramType::TessellationEval;
+    case Maxwell::ShaderProgram::Geometry:
+        return ProgramType::Geometry;
+    case Maxwell::ShaderProgram::Fragment:
+        return ProgramType::Fragment;
+    }
+    UNREACHABLE();
+    return {};
+}
+
 /// Calculates the size of a program stream
 std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
     constexpr std::size_t start_offset = 10;
@@ -128,11 +149,13 @@ std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
 }
 
 /// Hashes one (or two) program streams
-u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode& code,
-                        const ProgramCode& code_b) {
-    u64 unique_identifier =
-        Common::CityHash64(reinterpret_cast<const char*>(code.data()), CalculateProgramSize(code));
-    if (program_type != Maxwell::ShaderProgram::VertexA) {
+u64 GetUniqueIdentifier(ProgramType program_type, const ProgramCode& code,
+                        const ProgramCode& code_b, std::size_t size_a = 0, std::size_t size_b = 0) {
+    if (size_a == 0) {
+        size_a = CalculateProgramSize(code);
+    }
+    u64 unique_identifier = Common::CityHash64(reinterpret_cast<const char*>(code.data()), size_a);
+    if (program_type != ProgramType::VertexA) {
         return unique_identifier;
     }
     // VertexA programs include two programs
@@ -140,50 +163,69 @@ u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode&
     std::size_t seed = 0;
     boost::hash_combine(seed, unique_identifier);
 
-    const u64 identifier_b = Common::CityHash64(reinterpret_cast<const char*>(code_b.data()),
-                                                CalculateProgramSize(code_b));
+    if (size_b == 0) {
+        size_b = CalculateProgramSize(code_b);
+    }
+    const u64 identifier_b =
+        Common::CityHash64(reinterpret_cast<const char*>(code_b.data()), size_b);
     boost::hash_combine(seed, identifier_b);
     return static_cast<u64>(seed);
 }
 
 /// Creates an unspecialized program from code streams
-GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgram program_type,
+GLShader::ProgramResult CreateProgram(const Device& device, ProgramType program_type,
                                       ProgramCode program_code, ProgramCode program_code_b) {
     GLShader::ShaderSetup setup(program_code);
-    if (program_type == Maxwell::ShaderProgram::VertexA) {
+    setup.program.size_a = CalculateProgramSize(program_code);
+    setup.program.size_b = 0;
+    if (program_type == ProgramType::VertexA) {
         // VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
         // Conventional HW does not support this, so we combine VertexA and VertexB into one
         // stage here.
         setup.SetProgramB(program_code_b);
+        setup.program.size_b = CalculateProgramSize(program_code_b);
     }
-    setup.program.unique_identifier =
-        GetUniqueIdentifier(program_type, program_code, program_code_b);
+    setup.program.unique_identifier = GetUniqueIdentifier(
+        program_type, program_code, program_code_b, setup.program.size_a, setup.program.size_b);
 
     switch (program_type) {
-    case Maxwell::ShaderProgram::VertexA:
-    case Maxwell::ShaderProgram::VertexB:
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
         return GLShader::GenerateVertexShader(device, setup);
-    case Maxwell::ShaderProgram::Geometry:
+    case ProgramType::Geometry:
         return GLShader::GenerateGeometryShader(device, setup);
-    case Maxwell::ShaderProgram::Fragment:
+    case ProgramType::Fragment:
         return GLShader::GenerateFragmentShader(device, setup);
+    case ProgramType::Compute:
+        return GLShader::GenerateComputeShader(device, setup);
     default:
-        LOG_CRITICAL(HW_GPU, "Unimplemented program_type={}", static_cast<u32>(program_type));
-        UNREACHABLE();
+        UNIMPLEMENTED_MSG("Unimplemented program_type={}", static_cast<u32>(program_type));
         return {};
     }
 }
 
 CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
-                               Maxwell::ShaderProgram program_type, const ProgramVariant& variant,
+                               ProgramType program_type, const ProgramVariant& variant,
                                bool hint_retrievable = false) {
     auto base_bindings{variant.base_bindings};
     const auto primitive_mode{variant.primitive_mode};
     const auto texture_buffer_usage{variant.texture_buffer_usage};
 
     std::string source = "#version 430 core\n"
-                         "#extension GL_ARB_separate_shader_objects : enable\n\n";
-    source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
+                         "#extension GL_ARB_separate_shader_objects : enable\n"
+                         "#extension GL_NV_gpu_shader5 : enable\n"
+                         "#extension GL_NV_shader_thread_group : enable\n";
+    if (entries.shader_viewport_layer_array) {
+        source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
+    }
+    if (program_type == ProgramType::Compute) {
+        source += "#extension GL_ARB_compute_variable_group_size : require\n";
+    }
+    source += '\n';
+
+    if (program_type != ProgramType::Compute) {
+        source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
+    }
 
     for (const auto& cbuf : entries.const_buffers) {
         source +=
@@ -207,17 +249,24 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
         if (!texture_buffer_usage.test(i)) {
             continue;
         }
-        source += fmt::format("#define SAMPLER_{}_IS_BUFFER", i);
+        source += fmt::format("#define SAMPLER_{}_IS_BUFFER\n", i);
+    }
+    if (texture_buffer_usage.any()) {
+        source += '\n';
     }
 
-    if (program_type == Maxwell::ShaderProgram::Geometry) {
+    if (program_type == ProgramType::Geometry) {
         const auto [glsl_topology, debug_name, max_vertices] =
             GetPrimitiveDescription(primitive_mode);
 
-        source += "layout (" + std::string(glsl_topology) + ") in;\n";
+        source += "layout (" + std::string(glsl_topology) + ") in;\n\n";
         source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n';
     }
+    if (program_type == ProgramType::Compute) {
+        source += "layout (local_size_variable) in;\n";
+    }
 
+    source += '\n';
     source += code;
 
     OGLShader shader;
@@ -244,9 +293,9 @@ std::set<GLenum> GetSupportedFormats() {
 
 } // Anonymous namespace
 
-CachedShader::CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
+CachedShader::CachedShader(const ShaderParameters& params, ProgramType program_type,
                            GLShader::ProgramResult result)
-    : RasterizerCacheObject{params.host_ptr}, host_ptr{params.host_ptr}, cpu_addr{params.cpu_addr},
+    : RasterizerCacheObject{params.host_ptr}, cpu_addr{params.cpu_addr},
       unique_identifier{params.unique_identifier}, program_type{program_type},
       disk_cache{params.disk_cache}, precompiled_programs{params.precompiled_programs},
       entries{result.second}, code{std::move(result.first)}, shader_length{entries.shader_length} {}
@@ -257,29 +306,50 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
                                            ProgramCode&& program_code_b) {
     const auto code_size{CalculateProgramSize(program_code)};
     const auto code_size_b{CalculateProgramSize(program_code_b)};
-    auto result{CreateProgram(params.device, program_type, program_code, program_code_b)};
+    auto result{
+        CreateProgram(params.device, GetProgramType(program_type), program_code, program_code_b)};
     if (result.first.empty()) {
         // TODO(Rodrigo): Unimplemented shader stages hit here, avoid using these for now
         return {};
     }
 
     params.disk_cache.SaveRaw(ShaderDiskCacheRaw(
-        params.unique_identifier, program_type, static_cast<u32>(code_size / sizeof(u64)),
-        static_cast<u32>(code_size_b / sizeof(u64)), std::move(program_code),
-        std::move(program_code_b)));
+        params.unique_identifier, GetProgramType(program_type),
+        static_cast<u32>(code_size / sizeof(u64)), static_cast<u32>(code_size_b / sizeof(u64)),
+        std::move(program_code), std::move(program_code_b)));
 
-    return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, GetProgramType(program_type), std::move(result)));
 }
 
 Shader CachedShader::CreateStageFromCache(const ShaderParameters& params,
                                           Maxwell::ShaderProgram program_type,
                                           GLShader::ProgramResult result) {
-    return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, GetProgramType(program_type), std::move(result)));
+}
+
+Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code) {
+    auto result{CreateProgram(params.device, ProgramType::Compute, code, {})};
+
+    const auto code_size{CalculateProgramSize(code)};
+    params.disk_cache.SaveRaw(ShaderDiskCacheRaw(params.unique_identifier, ProgramType::Compute,
+                                                 static_cast<u32>(code_size / sizeof(u64)), 0,
+                                                 std::move(code), {}));
+
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, ProgramType::Compute, std::move(result)));
+}
+
+Shader CachedShader::CreateKernelFromCache(const ShaderParameters& params,
+                                           GLShader::ProgramResult result) {
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, ProgramType::Compute, std::move(result)));
 }
 
 std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVariant& variant) {
     GLuint handle{};
-    if (program_type == Maxwell::ShaderProgram::Geometry) {
+    if (program_type == ProgramType::Geometry) {
         handle = GetGeometryShader(variant);
     } else {
         const auto [entry, is_cache_miss] = programs.try_emplace(variant);
@@ -297,8 +367,11 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVar
         handle = program->handle;
     }
 
-    auto base_bindings{variant.base_bindings};
-    base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + RESERVED_UBOS;
+    auto base_bindings = variant.base_bindings;
+    base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size());
+    if (program_type != ProgramType::Compute) {
+        base_bindings.cbuf += STAGE_RESERVED_UBOS;
+    }
     base_bindings.gmem += static_cast<u32>(entries.global_memory_entries.size());
     base_bindings.sampler += static_cast<u32>(entries.samplers.size());
 
@@ -561,7 +634,7 @@ std::unordered_map<u64, UnspecializedShader> ShaderCacheOpenGL::GenerateUnspecia
 }
 
 Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
-    if (!system.GPU().Maxwell3D().dirty_flags.shaders) {
+    if (!system.GPU().Maxwell3D().dirty.shaders) {
         return last_shaders[static_cast<std::size_t>(program)];
     }
 
@@ -578,13 +651,15 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     // No shader found - create a new one
     ProgramCode program_code{GetShaderCode(memory_manager, program_addr, host_ptr)};
     ProgramCode program_code_b;
-    if (program == Maxwell::ShaderProgram::VertexA) {
+    const bool is_program_a{program == Maxwell::ShaderProgram::VertexA};
+    if (is_program_a) {
         const GPUVAddr program_addr_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
         program_code_b = GetShaderCode(memory_manager, program_addr_b,
                                        memory_manager.GetPointer(program_addr_b));
     }
 
-    const auto unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
+    const auto unique_identifier =
+        GetUniqueIdentifier(GetProgramType(program), program_code, program_code_b);
     const auto cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
     const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
                                   host_ptr,   unique_identifier};
@@ -601,4 +676,30 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     return last_shaders[static_cast<std::size_t>(program)] = shader;
 }
 
+Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
+    auto& memory_manager{system.GPU().MemoryManager()};
+    const auto host_ptr{memory_manager.GetPointer(code_addr)};
+    auto kernel = TryGet(host_ptr);
+    if (kernel) {
+        return kernel;
+    }
+
+    // No kernel found - create a new one
+    auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
+    const auto unique_identifier{GetUniqueIdentifier(ProgramType::Compute, code, {})};
+    const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
+    const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
+                                  host_ptr,   unique_identifier};
+
+    const auto found = precompiled_shaders.find(unique_identifier);
+    if (found == precompiled_shaders.end()) {
+        kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
+    } else {
+        kernel = CachedShader::CreateKernelFromCache(params, found->second);
+    }
+
+    Register(kernel);
+    return kernel;
+}
+
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index bbb53cdf4..2c8faf855 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -61,6 +61,11 @@ public:
                                        Maxwell::ShaderProgram program_type,
                                        GLShader::ProgramResult result);
 
+    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code);
+
+    static Shader CreateKernelFromCache(const ShaderParameters& params,
+                                        GLShader::ProgramResult result);
+
     VAddr GetCpuAddr() const override {
         return cpu_addr;
     }
@@ -78,7 +83,7 @@ public:
     std::tuple<GLuint, BaseBindings> GetProgramHandle(const ProgramVariant& variant);
 
 private:
-    explicit CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
+    explicit CachedShader(const ShaderParameters& params, ProgramType program_type,
                           GLShader::ProgramResult result);
 
     // Geometry programs. These are needed because GLSL needs an input topology but it's not
@@ -101,10 +106,9 @@ private:
 
     ShaderDiskCacheUsage GetUsage(const ProgramVariant& variant) const;
 
-    u8* host_ptr{};
     VAddr cpu_addr{};
     u64 unique_identifier{};
-    Maxwell::ShaderProgram program_type{};
+    ProgramType program_type{};
     ShaderDiskCacheOpenGL& disk_cache;
     const PrecompiledPrograms& precompiled_programs;
 
@@ -132,6 +136,9 @@ public:
     /// Gets the current specified shader stage program
     Shader GetStageProgram(Maxwell::ShaderProgram program);
 
+    /// Gets a compute kernel in the passed address
+    Shader GetComputeKernel(GPUVAddr code_addr);
+
 protected:
     // We do not have to flush this cache as things in it are never modified by us.
     void FlushObjectInner(const Shader& object) override {}
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 5f2f1510c..359d58cbe 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -14,6 +14,7 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common_types.h"
+#include "common/logging/log.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
@@ -36,7 +37,6 @@ using namespace std::string_literals;
 using namespace VideoCommon::Shader;
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using ShaderStage = Tegra::Engines::Maxwell3D::Regs::ShaderStage;
 using Operation = const OperationNode&;
 
 enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat };
@@ -46,7 +46,7 @@ using TextureArgument = std::pair<Type, Node>;
 using TextureIR = std::variant<TextureAoffi, TextureArgument>;
 
 constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
-    static_cast<u32>(RasterizerOpenGL::MaxConstbufferSize) / (4 * sizeof(float));
+    static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
 
 class ShaderWriter {
 public:
@@ -161,9 +161,13 @@ std::string FlowStackTopName(MetaStackClass stack) {
     return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack));
 }
 
+constexpr bool IsVertexShader(ProgramType stage) {
+    return stage == ProgramType::VertexA || stage == ProgramType::VertexB;
+}
+
 class GLSLDecompiler final {
 public:
-    explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ShaderStage stage,
+    explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ProgramType stage,
                             std::string suffix)
         : device{device}, ir{ir}, stage{stage}, suffix{suffix}, header{ir.GetHeader()} {}
 
@@ -191,10 +195,12 @@ public:
 
         // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems
         // unlikely that shaders will use 20 nested SSYs and PBKs.
-        constexpr u32 FLOW_STACK_SIZE = 20;
-        for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) {
-            code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE);
-            code.AddLine("uint {} = 0u;", FlowStackTopName(stack));
+        if (!ir.IsFlowStackDisabled()) {
+            constexpr u32 FLOW_STACK_SIZE = 20;
+            for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) {
+                code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE);
+                code.AddLine("uint {} = 0u;", FlowStackTopName(stack));
+            }
         }
 
         code.AddLine("while (true) {{");
@@ -244,24 +250,22 @@ public:
                                                        usage.is_read, usage.is_written);
         }
         entries.clip_distances = ir.GetClipDistances();
+        entries.shader_viewport_layer_array =
+            IsVertexShader(stage) && (ir.UsesLayer() || ir.UsesViewportIndex());
         entries.shader_length = ir.GetLength();
         return entries;
     }
 
 private:
-    using OperationDecompilerFn = std::string (GLSLDecompiler::*)(Operation);
-    using OperationDecompilersArray =
-        std::array<OperationDecompilerFn, static_cast<std::size_t>(OperationCode::Amount)>;
-
     void DeclareVertex() {
-        if (stage != ShaderStage::Vertex)
+        if (!IsVertexShader(stage))
             return;
 
         DeclareVertexRedeclarations();
     }
 
     void DeclareGeometry() {
-        if (stage != ShaderStage::Geometry) {
+        if (stage != ProgramType::Geometry) {
             return;
         }
 
@@ -280,22 +284,35 @@ private:
     }
 
     void DeclareVertexRedeclarations() {
-        bool clip_distances_declared = false;
-
         code.AddLine("out gl_PerVertex {{");
         ++code.scope;
 
         code.AddLine("vec4 gl_Position;");
 
-        for (const auto o : ir.GetOutputAttributes()) {
-            if (o == Attribute::Index::PointSize)
-                code.AddLine("float gl_PointSize;");
-            if (!clip_distances_declared && (o == Attribute::Index::ClipDistances0123 ||
-                                             o == Attribute::Index::ClipDistances4567)) {
+        for (const auto attribute : ir.GetOutputAttributes()) {
+            if (attribute == Attribute::Index::ClipDistances0123 ||
+                attribute == Attribute::Index::ClipDistances4567) {
                 code.AddLine("float gl_ClipDistance[];");
-                clip_distances_declared = true;
+                break;
             }
         }
+        if (!IsVertexShader(stage) || device.HasVertexViewportLayer()) {
+            if (ir.UsesLayer()) {
+                code.AddLine("int gl_Layer;");
+            }
+            if (ir.UsesViewportIndex()) {
+                code.AddLine("int gl_ViewportIndex;");
+            }
+        } else if ((ir.UsesLayer() || ir.UsesViewportIndex()) && IsVertexShader(stage) &&
+                   !device.HasVertexViewportLayer()) {
+            LOG_ERROR(
+                Render_OpenGL,
+                "GL_ARB_shader_viewport_layer_array is not available and its required by a shader");
+        }
+
+        if (ir.UsesPointSize()) {
+            code.AddLine("float gl_PointSize;");
+        }
 
         --code.scope;
         code.AddLine("}};");
@@ -323,11 +340,16 @@ private:
     }
 
     void DeclareLocalMemory() {
-        if (const u64 local_memory_size = header.GetLocalMemorySize(); local_memory_size > 0) {
-            const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
-            code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
-            code.AddNewLine();
+        // TODO(Rodrigo): Unstub kernel local memory size and pass it from a register at
+        // specialization time.
+        const u64 local_memory_size =
+            stage == ProgramType::Compute ? 0x400 : header.GetLocalMemorySize();
+        if (local_memory_size == 0) {
+            return;
         }
+        const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
+        code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
+        code.AddNewLine();
     }
 
     void DeclareInternalFlags() {
@@ -381,12 +403,12 @@ private:
         const u32 location{GetGenericAttributeIndex(index)};
 
         std::string name{GetInputAttribute(index)};
-        if (stage == ShaderStage::Geometry) {
+        if (stage == ProgramType::Geometry) {
             name = "gs_" + name + "[]";
         }
 
         std::string suffix;
-        if (stage == ShaderStage::Fragment) {
+        if (stage == ProgramType::Fragment) {
             const auto input_mode{header.ps.GetAttributeUse(location)};
             if (skip_unused && input_mode == AttributeUse::Unused) {
                 return;
@@ -398,7 +420,7 @@ private:
     }
 
     void DeclareOutputAttributes() {
-        if (ir.HasPhysicalAttributes() && stage != ShaderStage::Fragment) {
+        if (ir.HasPhysicalAttributes() && stage != ProgramType::Fragment) {
             for (u32 i = 0; i < GetNumPhysicalVaryings(); ++i) {
                 DeclareOutputAttribute(ToGenericAttribute(i));
             }
@@ -520,7 +542,7 @@ private:
                 constexpr u32 element_stride{4};
                 const u32 address{generic_base + index * generic_stride + element * element_stride};
 
-                const bool declared{stage != ShaderStage::Fragment ||
+                const bool declared{stage != ProgramType::Fragment ||
                                     header.ps.GetAttributeUse(index) != AttributeUse::Unused};
                 const std::string value{declared ? ReadAttribute(attribute, element) : "0"};
                 code.AddLine("case 0x{:x}: return {};", address, value);
@@ -543,7 +565,7 @@ private:
                 case Tegra::Shader::ImageType::Texture1D:
                     return "image1D";
                 case Tegra::Shader::ImageType::TextureBuffer:
-                    return "bufferImage";
+                    return "imageBuffer";
                 case Tegra::Shader::ImageType::Texture1DArray:
                     return "image1DArray";
                 case Tegra::Shader::ImageType::Texture2D:
@@ -624,7 +646,7 @@ private:
         }
 
         if (const auto abuf = std::get_if<AbufNode>(&*node)) {
-            UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ShaderStage::Geometry,
+            UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ProgramType::Geometry,
                                  "Physical attributes in geometry shaders are not implemented");
             if (abuf->IsPhysicalBuffer()) {
                 return fmt::format("readPhysicalAttribute(ftou({}))",
@@ -679,6 +701,9 @@ private:
         }
 
         if (const auto lmem = std::get_if<LmemNode>(&*node)) {
+            if (stage == ProgramType::Compute) {
+                LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
+            }
             return fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
         }
 
@@ -708,7 +733,7 @@ private:
 
     std::string ReadAttribute(Attribute::Index attribute, u32 element, const Node& buffer = {}) {
         const auto GeometryPass = [&](std::string_view name) {
-            if (stage == ShaderStage::Geometry && buffer) {
+            if (stage == ProgramType::Geometry && buffer) {
                 // TODO(Rodrigo): Guard geometry inputs against out of bound reads. Some games
                 // set an 0x80000000 index for those and the shader fails to build. Find out why
                 // this happens and what's its intent.
@@ -720,10 +745,10 @@ private:
         switch (attribute) {
         case Attribute::Index::Position:
             switch (stage) {
-            case ShaderStage::Geometry:
+            case ProgramType::Geometry:
                 return fmt::format("gl_in[ftou({})].gl_Position{}", Visit(buffer),
                                    GetSwizzle(element));
-            case ShaderStage::Fragment:
+            case ProgramType::Fragment:
                 return element == 3 ? "1.0f" : ("gl_FragCoord"s + GetSwizzle(element));
             default:
                 UNREACHABLE();
@@ -744,7 +769,7 @@ private:
             // TODO(Subv): Find out what the values are for the first two elements when inside a
             // vertex shader, and what's the value of the fourth element when inside a Tess Eval
             // shader.
-            ASSERT(stage == ShaderStage::Vertex);
+            ASSERT(IsVertexShader(stage));
             switch (element) {
             case 2:
                 // Config pack's first value is instance_id.
@@ -756,7 +781,7 @@ private:
             return "0";
         case Attribute::Index::FrontFacing:
             // TODO(Subv): Find out what the values are for the other elements.
-            ASSERT(stage == ShaderStage::Fragment);
+            ASSERT(stage == ProgramType::Fragment);
             switch (element) {
             case 3:
                 return "itof(gl_FrontFacing ? -1 : 0)";
@@ -778,7 +803,7 @@ private:
             return value;
         }
         // There's a bug in NVidia's proprietary drivers that makes precise fail on fragment shaders
-        const std::string precise = stage != ShaderStage::Fragment ? "precise " : "";
+        const std::string precise = stage != ProgramType::Fragment ? "precise " : "";
 
         const std::string temporary = code.GenerateTemporary();
         code.AddLine("{}float {} = {};", precise, temporary, value);
@@ -803,6 +828,45 @@ private:
         return CastOperand(VisitOperand(operation, operand_index), type);
     }
 
+    std::optional<std::pair<std::string, bool>> GetOutputAttribute(const AbufNode* abuf) {
+        switch (const auto attribute = abuf->GetIndex()) {
+        case Attribute::Index::Position:
+            return std::make_pair("gl_Position"s + GetSwizzle(abuf->GetElement()), false);
+        case Attribute::Index::LayerViewportPointSize:
+            switch (abuf->GetElement()) {
+            case 0:
+                UNIMPLEMENTED();
+                return {};
+            case 1:
+                if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
+                    return {};
+                }
+                return std::make_pair("gl_Layer", true);
+            case 2:
+                if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
+                    return {};
+                }
+                return std::make_pair("gl_ViewportIndex", true);
+            case 3:
+                UNIMPLEMENTED_MSG("Requires some state changes for gl_PointSize to work in shader");
+                return std::make_pair("gl_PointSize", false);
+            }
+            return {};
+        case Attribute::Index::ClipDistances0123:
+            return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), false);
+        case Attribute::Index::ClipDistances4567:
+            return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4),
+                                  false);
+        default:
+            if (IsGenericAttribute(attribute)) {
+                return std::make_pair(
+                    GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()), false);
+            }
+            UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
+            return {};
+        }
+    }
+
     std::string CastOperand(const std::string& value, Type type) const {
         switch (type) {
         case Type::Bool:
@@ -999,6 +1063,8 @@ private:
         const Node& src = operation[1];
 
         std::string target;
+        bool is_integer = false;
+
         if (const auto gpr = std::get_if<GprNode>(&*dest)) {
             if (gpr->GetIndex() == Register::ZeroIndex) {
                 // Writing to Register::ZeroIndex is a no op
@@ -1007,27 +1073,16 @@ private:
             target = GetRegister(gpr->GetIndex());
         } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) {
             UNIMPLEMENTED_IF(abuf->IsPhysicalBuffer());
-
-            target = [&]() -> std::string {
-                switch (const auto attribute = abuf->GetIndex(); abuf->GetIndex()) {
-                case Attribute::Index::Position:
-                    return "gl_Position"s + GetSwizzle(abuf->GetElement());
-                case Attribute::Index::PointSize:
-                    return "gl_PointSize";
-                case Attribute::Index::ClipDistances0123:
-                    return fmt::format("gl_ClipDistance[{}]", abuf->GetElement());
-                case Attribute::Index::ClipDistances4567:
-                    return fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4);
-                default:
-                    if (IsGenericAttribute(attribute)) {
-                        return GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement());
-                    }
-                    UNIMPLEMENTED_MSG("Unhandled output attribute: {}",
-                                      static_cast<u32>(attribute));
-                    return "0";
-                }
-            }();
+            const auto result = GetOutputAttribute(abuf);
+            if (!result) {
+                return {};
+            }
+            target = result->first;
+            is_integer = result->second;
         } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
+            if (stage == ProgramType::Compute) {
+                LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
+            }
             target = fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
         } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
             const std::string real = Visit(gmem->GetRealAddress());
@@ -1038,7 +1093,11 @@ private:
             UNREACHABLE_MSG("Assign called without a proper target");
         }
 
-        code.AddLine("{} = {};", target, Visit(src));
+        if (is_integer) {
+            code.AddLine("{} = ftoi({});", target, Visit(src));
+        } else {
+            code.AddLine("{} = {};", target, Visit(src));
+        }
         return {};
     }
 
@@ -1077,6 +1136,16 @@ private:
                                Type::Float);
     }
 
+    std::string FCastHalf0(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::HalfFloat);
+        return fmt::format("({})[0]", op_a);
+    }
+
+    std::string FCastHalf1(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::HalfFloat);
+        return fmt::format("({})[1]", op_a);
+    }
+
     template <Type type>
     std::string Min(Operation operation) {
         return GenerateBinaryCall(operation, "min", type, type, type);
@@ -1233,6 +1302,11 @@ private:
         return ApplyPrecise(operation, BitwiseCastResult(clamped, Type::HalfFloat));
     }
 
+    std::string HCastFloat(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::Float);
+        return fmt::format("fromHalf2(vec2({}, 0.0f))", op_a);
+    }
+
     std::string HUnpack(Operation operation) {
         const std::string operand{VisitOperand(operation, 0, Type::HalfFloat)};
         const auto value = [&]() -> std::string {
@@ -1351,14 +1425,10 @@ private:
         return fmt::format("{}[{}]", pair, VisitOperand(operation, 1, Type::Uint));
     }
 
-    std::string LogicalAll2(Operation operation) {
+    std::string LogicalAnd2(Operation operation) {
         return GenerateUnary(operation, "all", Type::Bool, Type::Bool2);
     }
 
-    std::string LogicalAny2(Operation operation) {
-        return GenerateUnary(operation, "any", Type::Bool, Type::Bool2);
-    }
-
     template <bool with_nan>
     std::string GenerateHalfComparison(Operation operation, const std::string& compare_op) {
         const std::string comparison{GenerateBinaryCall(operation, compare_op, Type::Bool2,
@@ -1555,6 +1625,14 @@ private:
         return {};
     }
 
+    std::string BranchIndirect(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::Uint);
+
+        code.AddLine("jmp_to = {};", op_a);
+        code.AddLine("break;");
+        return {};
+    }
+
     std::string PushFlowStack(Operation operation) {
         const auto stack = std::get<MetaStackClass>(operation.GetMeta());
         const auto target = std::get_if<ImmediateNode>(&*operation[0]);
@@ -1573,7 +1651,7 @@ private:
     }
 
     std::string Exit(Operation operation) {
-        if (stage != ShaderStage::Fragment) {
+        if (stage != ProgramType::Fragment) {
             code.AddLine("return;");
             return {};
         }
@@ -1624,7 +1702,7 @@ private:
     }
 
     std::string EmitVertex(Operation operation) {
-        ASSERT_MSG(stage == ShaderStage::Geometry,
+        ASSERT_MSG(stage == ProgramType::Geometry,
                    "EmitVertex is expected to be used in a geometry shader.");
 
         // If a geometry shader is attached, it will always flip (it's the last stage before
@@ -1635,7 +1713,7 @@ private:
     }
 
     std::string EndPrimitive(Operation operation) {
-        ASSERT_MSG(stage == ShaderStage::Geometry,
+        ASSERT_MSG(stage == ProgramType::Geometry,
                    "EndPrimitive is expected to be used in a geometry shader.");
 
         code.AddLine("EndPrimitive();");
@@ -1657,7 +1735,49 @@ private:
         return "utof(gl_WorkGroupID"s + GetSwizzle(element) + ')';
     }
 
-    static constexpr OperationDecompilersArray operation_decompilers = {
+    std::string BallotThread(Operation operation) {
+        const std::string value = VisitOperand(operation, 0, Type::Bool);
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia warp intrinsics are not available and its required by a shader");
+            // Stub on non-Nvidia devices by simulating all threads voting the same as the active
+            // one.
+            return fmt::format("utof({} ? 0xFFFFFFFFU : 0U)", value);
+        }
+        return fmt::format("utof(ballotThreadNV({}))", value);
+    }
+
+    std::string Vote(Operation operation, const char* func) {
+        const std::string value = VisitOperand(operation, 0, Type::Bool);
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia vote intrinsics are not available and its required by a shader");
+            // Stub with a warp size of one.
+            return value;
+        }
+        return fmt::format("{}({})", func, value);
+    }
+
+    std::string VoteAll(Operation operation) {
+        return Vote(operation, "allThreadsNV");
+    }
+
+    std::string VoteAny(Operation operation) {
+        return Vote(operation, "anyThreadNV");
+    }
+
+    std::string VoteEqual(Operation operation) {
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia vote intrinsics are not available and its required by a shader");
+            // We must return true here since a stub for a theoretical warp size of 1 will always
+            // return an equal result for all its votes.
+            return "true";
+        }
+        return Vote(operation, "allThreadsEqualNV");
+    }
+
+    static constexpr std::array operation_decompilers = {
         &GLSLDecompiler::Assign,
 
         &GLSLDecompiler::Select,
@@ -1669,6 +1789,8 @@ private:
         &GLSLDecompiler::Negate<Type::Float>,
         &GLSLDecompiler::Absolute<Type::Float>,
         &GLSLDecompiler::FClamp,
+        &GLSLDecompiler::FCastHalf0,
+        &GLSLDecompiler::FCastHalf1,
         &GLSLDecompiler::Min<Type::Float>,
         &GLSLDecompiler::Max<Type::Float>,
         &GLSLDecompiler::FCos,
@@ -1729,6 +1851,7 @@ private:
         &GLSLDecompiler::Absolute<Type::HalfFloat>,
         &GLSLDecompiler::HNegate,
         &GLSLDecompiler::HClamp,
+        &GLSLDecompiler::HCastFloat,
         &GLSLDecompiler::HUnpack,
         &GLSLDecompiler::HMergeF32,
         &GLSLDecompiler::HMergeH0,
@@ -1741,8 +1864,7 @@ private:
         &GLSLDecompiler::LogicalXor,
         &GLSLDecompiler::LogicalNegate,
         &GLSLDecompiler::LogicalPick2,
-        &GLSLDecompiler::LogicalAll2,
-        &GLSLDecompiler::LogicalAny2,
+        &GLSLDecompiler::LogicalAnd2,
 
         &GLSLDecompiler::LogicalLessThan<Type::Float>,
         &GLSLDecompiler::LogicalEqual<Type::Float>,
@@ -1789,6 +1911,7 @@ private:
         &GLSLDecompiler::ImageStore,
 
         &GLSLDecompiler::Branch,
+        &GLSLDecompiler::BranchIndirect,
         &GLSLDecompiler::PushFlowStack,
         &GLSLDecompiler::PopFlowStack,
         &GLSLDecompiler::Exit,
@@ -1804,7 +1927,13 @@ private:
         &GLSLDecompiler::WorkGroupId<0>,
         &GLSLDecompiler::WorkGroupId<1>,
         &GLSLDecompiler::WorkGroupId<2>,
+
+        &GLSLDecompiler::BallotThread,
+        &GLSLDecompiler::VoteAll,
+        &GLSLDecompiler::VoteAny,
+        &GLSLDecompiler::VoteEqual,
     };
+    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
     std::string GetRegister(u32 index) const {
         return GetDeclarationWithSuffix(index, "gpr");
@@ -1869,7 +1998,7 @@ private:
     }
 
     u32 GetNumPhysicalInputAttributes() const {
-        return stage == ShaderStage::Vertex ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
+        return IsVertexShader(stage) ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
     }
 
     u32 GetNumPhysicalAttributes() const {
@@ -1882,7 +2011,7 @@ private:
 
     const Device& device;
     const ShaderIR& ir;
-    const ShaderStage stage;
+    const ProgramType stage;
     const std::string suffix;
     const Header header;
 
@@ -1913,7 +2042,7 @@ std::string GetCommonDeclarations() {
         MAX_CONSTBUFFER_ELEMENTS);
 }
 
-ProgramResult Decompile(const Device& device, const ShaderIR& ir, Maxwell::ShaderStage stage,
+ProgramResult Decompile(const Device& device, const ShaderIR& ir, ProgramType stage,
                         const std::string& suffix) {
     GLSLDecompiler decompiler(device, ir, stage, suffix);
     decompiler.Decompile();
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h
index 14d11c7fc..2ea02f5bf 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.h
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h
@@ -12,14 +12,26 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/shader/shader_ir.h"
 
-namespace OpenGL {
-class Device;
-}
-
 namespace VideoCommon::Shader {
 class ShaderIR;
 }
 
+namespace OpenGL {
+
+class Device;
+
+enum class ProgramType : u32 {
+    VertexA = 0,
+    VertexB = 1,
+    TessellationControl = 2,
+    TessellationEval = 3,
+    Geometry = 4,
+    Fragment = 5,
+    Compute = 6
+};
+
+} // namespace OpenGL
+
 namespace OpenGL::GLShader {
 
 struct ShaderEntries;
@@ -78,12 +90,13 @@ struct ShaderEntries {
     std::vector<ImageEntry> images;
     std::vector<GlobalMemoryEntry> global_memory_entries;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
+    bool shader_viewport_layer_array{};
     std::size_t shader_length{};
 };
 
 std::string GetCommonDeclarations();
 
 ProgramResult Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
-                        Maxwell::ShaderStage stage, const std::string& suffix);
+                        ProgramType stage, const std::string& suffix);
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
index 10688397b..969fe9ced 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
@@ -51,7 +51,7 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() {
 
 } // namespace
 
-ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
                                        u32 program_code_size, u32 program_code_size_b,
                                        ProgramCode program_code, ProgramCode program_code_b)
     : unique_identifier{unique_identifier}, program_type{program_type},
@@ -373,6 +373,12 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
         }
     }
 
+    bool shader_viewport_layer_array{};
+    if (!LoadObjectFromPrecompiled(shader_viewport_layer_array)) {
+        return {};
+    }
+    entry.entries.shader_viewport_layer_array = shader_viewport_layer_array;
+
     u64 shader_length{};
     if (!LoadObjectFromPrecompiled(shader_length)) {
         return {};
@@ -445,6 +451,10 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
         }
     }
 
+    if (!SaveObjectToPrecompiled(entries.shader_viewport_layer_array)) {
+        return false;
+    }
+
     if (!SaveObjectToPrecompiled(static_cast<u64>(entries.shader_length))) {
         return false;
     }
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
index 4f296dda6..cc8bbd61e 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
@@ -18,7 +18,6 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "core/file_sys/vfs_vector.h"
-#include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
 
 namespace Core {
@@ -34,14 +33,11 @@ namespace OpenGL {
 struct ShaderDiskCacheUsage;
 struct ShaderDiskCacheDump;
 
-using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
-
 using ProgramCode = std::vector<u64>;
-using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-
+using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
 using TextureBufferUsage = std::bitset<64>;
 
-/// Allocated bindings used by an OpenGL shader program.
+/// Allocated bindings used by an OpenGL shader program
 struct BaseBindings {
     u32 cbuf{};
     u32 gmem{};
@@ -126,7 +122,7 @@ namespace OpenGL {
 /// Describes a shader how it's used by the guest GPU
 class ShaderDiskCacheRaw {
 public:
-    explicit ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
+    explicit ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
                                 u32 program_code_size, u32 program_code_size_b,
                                 ProgramCode program_code, ProgramCode program_code_b);
     ShaderDiskCacheRaw();
@@ -141,30 +137,13 @@ public:
     }
 
     bool HasProgramA() const {
-        return program_type == Maxwell::ShaderProgram::VertexA;
+        return program_type == ProgramType::VertexA;
     }
 
-    Maxwell::ShaderProgram GetProgramType() const {
+    ProgramType GetProgramType() const {
         return program_type;
     }
 
-    Maxwell::ShaderStage GetProgramStage() const {
-        switch (program_type) {
-        case Maxwell::ShaderProgram::VertexA:
-        case Maxwell::ShaderProgram::VertexB:
-            return Maxwell::ShaderStage::Vertex;
-        case Maxwell::ShaderProgram::TesselationControl:
-            return Maxwell::ShaderStage::TesselationControl;
-        case Maxwell::ShaderProgram::TesselationEval:
-            return Maxwell::ShaderStage::TesselationEval;
-        case Maxwell::ShaderProgram::Geometry:
-            return Maxwell::ShaderStage::Geometry;
-        case Maxwell::ShaderProgram::Fragment:
-            return Maxwell::ShaderStage::Fragment;
-        }
-        UNREACHABLE();
-    }
-
     const ProgramCode& GetProgramCode() const {
         return program_code;
     }
@@ -175,7 +154,7 @@ public:
 
 private:
     u64 unique_identifier{};
-    Maxwell::ShaderProgram program_type{};
+    ProgramType program_type{};
     u32 program_code_size{};
     u32 program_code_size_b{};
 
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index 9148629ec..3a8d9e1da 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -14,7 +14,8 @@ using Tegra::Engines::Maxwell3D;
 using VideoCommon::Shader::ProgramCode;
 using VideoCommon::Shader::ShaderIR;
 
-static constexpr u32 PROGRAM_OFFSET{10};
+static constexpr u32 PROGRAM_OFFSET = 10;
+static constexpr u32 COMPUTE_OFFSET = 0;
 
 ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup) {
     const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
@@ -29,17 +30,15 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config {
 };
 
 )";
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Vertex, "vertex");
 
+    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
+    const auto stage = setup.IsDualProgram() ? ProgramType::VertexA : ProgramType::VertexB;
+    ProgramResult program = Decompile(device, program_ir, stage, "vertex");
     out += program.first;
 
     if (setup.IsDualProgram()) {
-        const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET);
-        ProgramResult program_b =
-            Decompile(device, program_ir_b, Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b");
-
+        const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b);
+        ProgramResult program_b = Decompile(device, program_ir_b, ProgramType::VertexB, "vertex_b");
         out += program_b.first;
     }
 
@@ -80,9 +79,9 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config {
 };
 
 )";
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Geometry, "geometry");
+
+    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Geometry, "geometry");
     out += program.first;
 
     out += R"(
@@ -115,10 +114,8 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform fs_config {
 };
 
 )";
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Fragment, "fragment");
-
+    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Fragment, "fragment");
     out += program.first;
 
     out += R"(
@@ -130,4 +127,22 @@ void main() {
     return {std::move(out), std::move(program.second)};
 }
 
+ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup) {
+    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
+
+    std::string out = "// Shader Unique Id: CS" + id + "\n\n";
+    out += GetCommonDeclarations();
+
+    const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a);
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Compute, "compute");
+    out += program.first;
+
+    out += R"(
+void main() {
+    execute_compute();
+}
+)";
+    return {std::move(out), std::move(program.second)};
+}
+
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h
index 0536c8a03..3833e88ab 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.h
+++ b/src/video_core/renderer_opengl/gl_shader_gen.h
@@ -27,6 +27,8 @@ struct ShaderSetup {
         ProgramCode code;
         ProgramCode code_b; // Used for dual vertex shaders
         u64 unique_identifier;
+        std::size_t size_a;
+        std::size_t size_b;
     } program;
 
     /// Used in scenarios where we have a dual vertex shaders
@@ -52,4 +54,7 @@ ProgramResult GenerateGeometryShader(const Device& device, const ShaderSetup& se
 /// Generates the GLSL fragment shader program source code for the given FS program
 ProgramResult GenerateFragmentShader(const Device& device, const ShaderSetup& setup);
 
+/// Generates the GLSL compute shader program source code for the given CS program
+ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup);
+
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_util.cpp b/src/video_core/renderer_opengl/gl_shader_util.cpp
index 5f3fe067e..9e74eda0d 100644
--- a/src/video_core/renderer_opengl/gl_shader_util.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_util.cpp
@@ -10,21 +10,25 @@
 
 namespace OpenGL::GLShader {
 
-GLuint LoadShader(const char* source, GLenum type) {
-    const char* debug_type;
+namespace {
+const char* GetStageDebugName(GLenum type) {
     switch (type) {
     case GL_VERTEX_SHADER:
-        debug_type = "vertex";
-        break;
+        return "vertex";
     case GL_GEOMETRY_SHADER:
-        debug_type = "geometry";
-        break;
+        return "geometry";
     case GL_FRAGMENT_SHADER:
-        debug_type = "fragment";
-        break;
-    default:
-        UNREACHABLE();
+        return "fragment";
+    case GL_COMPUTE_SHADER:
+        return "compute";
     }
+    UNIMPLEMENTED();
+    return "unknown";
+}
+} // Anonymous namespace
+
+GLuint LoadShader(const char* source, GLenum type) {
+    const char* debug_type = GetStageDebugName(type);
     const GLuint shader_id = glCreateShader(type);
     glShaderSource(shader_id, 1, &source, nullptr);
     LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type);
diff --git a/src/video_core/renderer_opengl/gl_state.cpp b/src/video_core/renderer_opengl/gl_state.cpp
index d86e137ac..f4777d0b0 100644
--- a/src/video_core/renderer_opengl/gl_state.cpp
+++ b/src/video_core/renderer_opengl/gl_state.cpp
@@ -6,8 +6,11 @@
 #include <glad/glad.h>
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/renderer_opengl/gl_state.h"
 
+MICROPROFILE_DEFINE(OpenGL_State, "OpenGL", "State Change", MP_RGB(192, 128, 128));
+
 namespace OpenGL {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
@@ -162,6 +165,25 @@ OpenGLState::OpenGLState() {
     alpha_test.ref = 0.0f;
 }
 
+void OpenGLState::SetDefaultViewports() {
+    for (auto& item : viewports) {
+        item.x = 0;
+        item.y = 0;
+        item.width = 0;
+        item.height = 0;
+        item.depth_range_near = 0.0f;
+        item.depth_range_far = 1.0f;
+        item.scissor.enabled = false;
+        item.scissor.x = 0;
+        item.scissor.y = 0;
+        item.scissor.width = 0;
+        item.scissor.height = 0;
+    }
+
+    depth_clamp.far_plane = false;
+    depth_clamp.near_plane = false;
+}
+
 void OpenGLState::ApplyDefaultState() {
     glEnable(GL_BLEND);
     glDisable(GL_FRAMEBUFFER_SRGB);
@@ -523,7 +545,8 @@ void OpenGLState::ApplySamplers() const {
     }
 }
 
-void OpenGLState::Apply() const {
+void OpenGLState::Apply() {
+    MICROPROFILE_SCOPE(OpenGL_State);
     ApplyFramebufferState();
     ApplyVertexArrayState();
     ApplyShaderProgram();
@@ -532,19 +555,31 @@ void OpenGLState::Apply() const {
     ApplyPointSize();
     ApplyFragmentColorClamp();
     ApplyMultisample();
+    if (dirty.color_mask) {
+        ApplyColorMask();
+        dirty.color_mask = false;
+    }
     ApplyDepthClamp();
-    ApplyColorMask();
     ApplyViewport();
-    ApplyStencilTest();
+    if (dirty.stencil_state) {
+        ApplyStencilTest();
+        dirty.stencil_state = false;
+    }
     ApplySRgb();
     ApplyCulling();
     ApplyDepth();
     ApplyPrimitiveRestart();
-    ApplyBlending();
+    if (dirty.blend_state) {
+        ApplyBlending();
+        dirty.blend_state = false;
+    }
     ApplyLogicOp();
     ApplyTextures();
     ApplySamplers();
-    ApplyPolygonOffset();
+    if (dirty.polygon_offset) {
+        ApplyPolygonOffset();
+        dirty.polygon_offset = false;
+    }
     ApplyAlphaTest();
 }
 
diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h
index b0140495d..fdf9a8a12 100644
--- a/src/video_core/renderer_opengl/gl_state.h
+++ b/src/video_core/renderer_opengl/gl_state.h
@@ -195,8 +195,9 @@ public:
         s_rgb_used = false;
     }
 
+    void SetDefaultViewports();
     /// Apply this state as the current OpenGL state
-    void Apply() const;
+    void Apply();
 
     void ApplyFramebufferState() const;
     void ApplyVertexArrayState() const;
@@ -237,11 +238,41 @@ public:
     /// Viewport does not affects glClearBuffer so emulate viewport using scissor test
     void EmulateViewportWithScissor();
 
+    void MarkDirtyBlendState() {
+        dirty.blend_state = true;
+    }
+
+    void MarkDirtyStencilState() {
+        dirty.stencil_state = true;
+    }
+
+    void MarkDirtyPolygonOffset() {
+        dirty.polygon_offset = true;
+    }
+
+    void MarkDirtyColorMask() {
+        dirty.color_mask = true;
+    }
+
+    void AllDirty() {
+        dirty.blend_state = true;
+        dirty.stencil_state = true;
+        dirty.polygon_offset = true;
+        dirty.color_mask = true;
+    }
+
 private:
     static OpenGLState cur_state;
 
     // Workaround for sRGB problems caused by QT not supporting srgb output
     static bool s_rgb_used;
+    struct {
+        bool blend_state;
+        bool stencil_state;
+        bool viewport_state;
+        bool polygon_offset;
+        bool color_mask;
+    } dirty{};
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 08ae1a429..4f135fe03 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -31,6 +31,8 @@ using VideoCore::Surface::SurfaceType;
 
 MICROPROFILE_DEFINE(OpenGL_Texture_Upload, "OpenGL", "Texture Upload", MP_RGB(128, 192, 128));
 MICROPROFILE_DEFINE(OpenGL_Texture_Download, "OpenGL", "Texture Download", MP_RGB(128, 192, 128));
+MICROPROFILE_DEFINE(OpenGL_Texture_Buffer_Copy, "OpenGL", "Texture Buffer Copy",
+                    MP_RGB(128, 192, 128));
 
 namespace {
 
@@ -135,7 +137,6 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
 const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) {
     ASSERT(static_cast<std::size_t>(pixel_format) < tex_format_tuples.size());
     const auto& format{tex_format_tuples[static_cast<std::size_t>(pixel_format)]};
-    ASSERT(component_type == format.component_type);
     return format;
 }
 
@@ -183,6 +184,9 @@ GLint GetSwizzleSource(SwizzleSource source) {
 }
 
 void ApplyTextureDefaults(const SurfaceParams& params, GLuint texture) {
+    if (params.IsBuffer()) {
+        return;
+    }
     glTextureParameteri(texture, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     glTextureParameteri(texture, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     glTextureParameteri(texture, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
@@ -207,6 +211,7 @@ OGLTexture CreateTexture(const SurfaceParams& params, GLenum target, GLenum inte
         glNamedBufferStorage(texture_buffer.handle, params.width * params.GetBytesPerPixel(),
                              nullptr, GL_DYNAMIC_STORAGE_BIT);
         glTextureBuffer(texture.handle, internal_format, texture_buffer.handle);
+        break;
     case SurfaceTarget::Texture2D:
     case SurfaceTarget::TextureCubemap:
         glTextureStorage2D(texture.handle, params.emulated_levels, internal_format, params.width,
@@ -483,11 +488,15 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
     const auto& dst_params{dst_view->GetSurfaceParams()};
 
     OpenGLState prev_state{OpenGLState::GetCurState()};
-    SCOPE_EXIT({ prev_state.Apply(); });
+    SCOPE_EXIT({
+        prev_state.AllDirty();
+        prev_state.Apply();
+    });
 
     OpenGLState state;
     state.draw.read_framebuffer = src_framebuffer.handle;
     state.draw.draw_framebuffer = dst_framebuffer.handle;
+    state.AllDirty();
     state.Apply();
 
     u32 buffers{};
@@ -535,6 +544,7 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
 }
 
 void TextureCacheOpenGL::BufferCopy(Surface& src_surface, Surface& dst_surface) {
+    MICROPROFILE_SCOPE(OpenGL_Texture_Buffer_Copy);
     const auto& src_params = src_surface->GetSurfaceParams();
     const auto& dst_params = dst_surface->GetSurfaceParams();
     UNIMPLEMENTED_IF(src_params.num_levels > 1 || dst_params.num_levels > 1);
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index ff6ab6988..21324488a 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -51,7 +51,7 @@ public:
     }
 
 protected:
-    void DecorateSurfaceName();
+    void DecorateSurfaceName() override;
 
     View CreateView(const ViewParams& view_key) override;
     View CreateViewInner(const ViewParams& view_key, bool is_proxy);
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index b142521ec..af9684839 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -101,21 +101,19 @@ RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::Syst
 
 RendererOpenGL::~RendererOpenGL() = default;
 
-/// Swap buffers (render frame)
-void RendererOpenGL::SwapBuffers(
-    std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
-
+void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     system.GetPerfStats().EndSystemFrame();
 
     // Maintain the rasterizer's state as a priority
     OpenGLState prev_state = OpenGLState::GetCurState();
+    state.AllDirty();
     state.Apply();
 
     if (framebuffer) {
         // If framebuffer is provided, reload it from memory to a texture
-        if (screen_info.texture.width != (GLsizei)framebuffer->get().width ||
-            screen_info.texture.height != (GLsizei)framebuffer->get().height ||
-            screen_info.texture.pixel_format != framebuffer->get().pixel_format) {
+        if (screen_info.texture.width != static_cast<GLsizei>(framebuffer->width) ||
+            screen_info.texture.height != static_cast<GLsizei>(framebuffer->height) ||
+            screen_info.texture.pixel_format != framebuffer->pixel_format) {
             // Reallocate texture if the framebuffer size has changed.
             // This is expected to not happen very often and hence should not be a
             // performance problem.
@@ -130,6 +128,8 @@ void RendererOpenGL::SwapBuffers(
 
         DrawScreen(render_window.GetFramebufferLayout());
 
+        rasterizer->TickFrame();
+
         render_window.SwapBuffers();
     }
 
@@ -139,6 +139,7 @@ void RendererOpenGL::SwapBuffers(
     system.GetPerfStats().BeginSystemFrame();
 
     // Restore the rasterizer state
+    prev_state.AllDirty();
     prev_state.Apply();
 }
 
@@ -146,43 +147,43 @@ void RendererOpenGL::SwapBuffers(
  * Loads framebuffer from emulated memory into the active OpenGL texture.
  */
 void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuffer) {
-    const u32 bytes_per_pixel{Tegra::FramebufferConfig::BytesPerPixel(framebuffer.pixel_format)};
-    const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel};
-    const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset};
-
     // Framebuffer orientation handling
     framebuffer_transform_flags = framebuffer.transform_flags;
     framebuffer_crop_rect = framebuffer.crop_rect;
 
-    // Ensure no bad interactions with GL_UNPACK_ALIGNMENT, which by default
-    // only allows rows to have a memory alignement of 4.
-    ASSERT(framebuffer.stride % 4 == 0);
-
-    if (!rasterizer->AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride)) {
-        // Reset the screen info's display texture to its own permanent texture
-        screen_info.display_texture = screen_info.texture.resource.handle;
-
-        rasterizer->FlushRegion(ToCacheAddr(Memory::GetPointer(framebuffer_addr)), size_in_bytes);
-
-        constexpr u32 linear_bpp = 4;
-        VideoCore::MortonCopyPixels128(VideoCore::MortonSwizzleMode::MortonToLinear,
-                                       framebuffer.width, framebuffer.height, bytes_per_pixel,
-                                       linear_bpp, Memory::GetPointer(framebuffer_addr),
-                                       gl_framebuffer_data.data());
-
-        glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride));
+    const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset};
+    if (rasterizer->AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride)) {
+        return;
+    }
 
-        // Update existing texture
-        // TODO: Test what happens on hardware when you change the framebuffer dimensions so that
-        //       they differ from the LCD resolution.
-        // TODO: Applications could theoretically crash yuzu here by specifying too large
-        //       framebuffer sizes. We should make sure that this cannot happen.
-        glTextureSubImage2D(screen_info.texture.resource.handle, 0, 0, 0, framebuffer.width,
-                            framebuffer.height, screen_info.texture.gl_format,
-                            screen_info.texture.gl_type, gl_framebuffer_data.data());
+    // Reset the screen info's display texture to its own permanent texture
+    screen_info.display_texture = screen_info.texture.resource.handle;
 
-        glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
-    }
+    const auto pixel_format{
+        VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)};
+    const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)};
+    const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel};
+    const auto host_ptr{Memory::GetPointer(framebuffer_addr)};
+    rasterizer->FlushRegion(ToCacheAddr(host_ptr), size_in_bytes);
+
+    // TODO(Rodrigo): Read this from HLE
+    constexpr u32 block_height_log2 = 4;
+    VideoCore::MortonSwizzle(VideoCore::MortonSwizzleMode::MortonToLinear, pixel_format,
+                             framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1,
+                             gl_framebuffer_data.data(), host_ptr);
+
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride));
+
+    // Update existing texture
+    // TODO: Test what happens on hardware when you change the framebuffer dimensions so that
+    //       they differ from the LCD resolution.
+    // TODO: Applications could theoretically crash yuzu here by specifying too large
+    //       framebuffer sizes. We should make sure that this cannot happen.
+    glTextureSubImage2D(screen_info.texture.resource.handle, 0, 0, 0, framebuffer.width,
+                        framebuffer.height, screen_info.texture.gl_format,
+                        screen_info.texture.gl_type, gl_framebuffer_data.data());
+
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
 }
 
 /**
@@ -205,6 +206,7 @@ void RendererOpenGL::InitOpenGLObjects() {
     // Link shaders and get variable locations
     shader.CreateFromSource(vertex_shader, nullptr, fragment_shader);
     state.draw.shader_program = shader.handle;
+    state.AllDirty();
     state.Apply();
     uniform_modelview_matrix = glGetUniformLocation(shader.handle, "modelview_matrix");
     uniform_color_texture = glGetUniformLocation(shader.handle, "color_texture");
@@ -262,7 +264,6 @@ void RendererOpenGL::CreateRasterizer() {
     if (rasterizer) {
         return;
     }
-    // Initialize sRGB Usage
     OpenGLState::ClearsRGBUsed();
     rasterizer = std::make_unique<RasterizerOpenGL>(system, emu_window, screen_info);
 }
@@ -273,22 +274,29 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
     texture.height = framebuffer.height;
     texture.pixel_format = framebuffer.pixel_format;
 
+    const auto pixel_format{
+        VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)};
+    const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)};
+    gl_framebuffer_data.resize(texture.width * texture.height * bytes_per_pixel);
+
     GLint internal_format;
     switch (framebuffer.pixel_format) {
     case Tegra::FramebufferConfig::PixelFormat::ABGR8:
         internal_format = GL_RGBA8;
         texture.gl_format = GL_RGBA;
         texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
-        gl_framebuffer_data.resize(texture.width * texture.height * 4);
+        break;
+    case Tegra::FramebufferConfig::PixelFormat::RGB565:
+        internal_format = GL_RGB565;
+        texture.gl_format = GL_RGB;
+        texture.gl_type = GL_UNSIGNED_SHORT_5_6_5;
         break;
     default:
         internal_format = GL_RGBA8;
         texture.gl_format = GL_RGBA;
         texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
-        gl_framebuffer_data.resize(texture.width * texture.height * 4);
-        LOG_CRITICAL(Render_OpenGL, "Unknown framebuffer pixel format: {}",
-                     static_cast<u32>(framebuffer.pixel_format));
-        UNREACHABLE();
+        UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}",
+                          static_cast<u32>(framebuffer.pixel_format));
     }
 
     texture.resource.Release();
@@ -338,12 +346,14 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
     // Workaround brigthness problems in SMO by enabling sRGB in the final output
     // if it has been used in the frame. Needed because of this bug in QT: QTBUG-50987
     state.framebuffer_srgb.enabled = OpenGLState::GetsRGBUsed();
+    state.AllDirty();
     state.Apply();
     glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), vertices.data());
     glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
     // Restore default state
     state.framebuffer_srgb.enabled = false;
     state.texture_units[0].texture = 0;
+    state.AllDirty();
     state.Apply();
     // Clear sRGB state for the next frame
     OpenGLState::ClearsRGBUsed();
@@ -388,6 +398,7 @@ void RendererOpenGL::CaptureScreenshot() {
     GLuint old_read_fb = state.draw.read_framebuffer;
     GLuint old_draw_fb = state.draw.draw_framebuffer;
     state.draw.read_framebuffer = state.draw.draw_framebuffer = screenshot_framebuffer.handle;
+    state.AllDirty();
     state.Apply();
 
     Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
@@ -407,6 +418,7 @@ void RendererOpenGL::CaptureScreenshot() {
     screenshot_framebuffer.Release();
     state.draw.read_framebuffer = old_read_fb;
     state.draw.draw_framebuffer = old_draw_fb;
+    state.AllDirty();
     state.Apply();
     glDeleteRenderbuffers(1, &renderbuffer);
 
diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h
index 4aebf2321..9bd086368 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.h
+++ b/src/video_core/renderer_opengl/renderer_opengl.h
@@ -43,14 +43,13 @@ struct ScreenInfo {
     TextureInfo texture;
 };
 
-class RendererOpenGL : public VideoCore::RendererBase {
+class RendererOpenGL final : public VideoCore::RendererBase {
 public:
     explicit RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system);
     ~RendererOpenGL() override;
 
     /// Swap buffers (render frame)
-    void SwapBuffers(
-        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) override;
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
 
     /// Initialize the renderer
     bool Init() override;
diff --git a/src/video_core/renderer_opengl/utils.cpp b/src/video_core/renderer_opengl/utils.cpp
index 68c36988d..c504a2c1a 100644
--- a/src/video_core/renderer_opengl/utils.cpp
+++ b/src/video_core/renderer_opengl/utils.cpp
@@ -13,29 +13,67 @@
 
 namespace OpenGL {
 
+VertexArrayPushBuffer::VertexArrayPushBuffer() = default;
+
+VertexArrayPushBuffer::~VertexArrayPushBuffer() = default;
+
+void VertexArrayPushBuffer::Setup(GLuint vao_) {
+    vao = vao_;
+    index_buffer = nullptr;
+    vertex_buffers.clear();
+}
+
+void VertexArrayPushBuffer::SetIndexBuffer(const GLuint* buffer) {
+    index_buffer = buffer;
+}
+
+void VertexArrayPushBuffer::SetVertexBuffer(GLuint binding_index, const GLuint* buffer,
+                                            GLintptr offset, GLsizei stride) {
+    vertex_buffers.push_back(Entry{binding_index, buffer, offset, stride});
+}
+
+void VertexArrayPushBuffer::Bind() {
+    if (index_buffer) {
+        glVertexArrayElementBuffer(vao, *index_buffer);
+    }
+
+    // TODO(Rodrigo): Find a way to ARB_multi_bind this
+    for (const auto& entry : vertex_buffers) {
+        glVertexArrayVertexBuffer(vao, entry.binding_index, *entry.buffer, entry.offset,
+                                  entry.stride);
+    }
+}
+
 BindBuffersRangePushBuffer::BindBuffersRangePushBuffer(GLenum target) : target{target} {}
 
 BindBuffersRangePushBuffer::~BindBuffersRangePushBuffer() = default;
 
 void BindBuffersRangePushBuffer::Setup(GLuint first_) {
     first = first_;
-    buffers.clear();
+    buffer_pointers.clear();
     offsets.clear();
     sizes.clear();
 }
 
-void BindBuffersRangePushBuffer::Push(GLuint buffer, GLintptr offset, GLsizeiptr size) {
-    buffers.push_back(buffer);
+void BindBuffersRangePushBuffer::Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size) {
+    buffer_pointers.push_back(buffer);
     offsets.push_back(offset);
     sizes.push_back(size);
 }
 
-void BindBuffersRangePushBuffer::Bind() const {
-    const std::size_t count{buffers.size()};
+void BindBuffersRangePushBuffer::Bind() {
+    // Ensure sizes are valid.
+    const std::size_t count{buffer_pointers.size()};
     DEBUG_ASSERT(count == offsets.size() && count == sizes.size());
     if (count == 0) {
         return;
     }
+
+    // Dereference buffers.
+    buffers.resize(count);
+    std::transform(buffer_pointers.begin(), buffer_pointers.end(), buffers.begin(),
+                   [](const GLuint* pointer) { return *pointer; });
+
     glBindBuffersRange(target, first, static_cast<GLsizei>(count), buffers.data(), offsets.data(),
                        sizes.data());
 }
diff --git a/src/video_core/renderer_opengl/utils.h b/src/video_core/renderer_opengl/utils.h
index 4a752f3b4..6c2b45546 100644
--- a/src/video_core/renderer_opengl/utils.h
+++ b/src/video_core/renderer_opengl/utils.h
@@ -11,20 +11,49 @@
 
 namespace OpenGL {
 
-class BindBuffersRangePushBuffer {
+class VertexArrayPushBuffer final {
 public:
-    BindBuffersRangePushBuffer(GLenum target);
+    explicit VertexArrayPushBuffer();
+    ~VertexArrayPushBuffer();
+
+    void Setup(GLuint vao_);
+
+    void SetIndexBuffer(const GLuint* buffer);
+
+    void SetVertexBuffer(GLuint binding_index, const GLuint* buffer, GLintptr offset,
+                         GLsizei stride);
+
+    void Bind();
+
+private:
+    struct Entry {
+        GLuint binding_index{};
+        const GLuint* buffer{};
+        GLintptr offset{};
+        GLsizei stride{};
+    };
+
+    GLuint vao{};
+    const GLuint* index_buffer{};
+    std::vector<Entry> vertex_buffers;
+};
+
+class BindBuffersRangePushBuffer final {
+public:
+    explicit BindBuffersRangePushBuffer(GLenum target);
     ~BindBuffersRangePushBuffer();
 
     void Setup(GLuint first_);
 
-    void Push(GLuint buffer, GLintptr offset, GLsizeiptr size);
+    void Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size);
 
-    void Bind() const;
+    void Bind();
 
 private:
-    GLenum target;
-    GLuint first;
+    GLenum target{};
+    GLuint first{};
+    std::vector<const GLuint*> buffer_pointers;
+
     std::vector<GLuint> buffers;
     std::vector<GLintptr> offsets;
     std::vector<GLsizeiptr> sizes;
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 02a9f5ecb..d2e9f4031 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -109,8 +109,8 @@ void VKBufferCache::Reserve(std::size_t max_size) {
     }
 }
 
-VKExecutionContext VKBufferCache::Send(VKExecutionContext exctx) {
-    return stream_buffer->Send(exctx, buffer_offset - buffer_offset_base);
+void VKBufferCache::Send() {
+    stream_buffer->Send(buffer_offset - buffer_offset_base);
 }
 
 void VKBufferCache::AlignBuffer(std::size_t alignment) {
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index 3edf460df..49f13bcdc 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -77,7 +77,7 @@ public:
     void Reserve(std::size_t max_size);
 
     /// Ensures that the set data is sent to the device.
-    [[nodiscard]] VKExecutionContext Send(VKExecutionContext exctx);
+    void Send();
 
     /// Returns the buffer cache handle.
     vk::Buffer GetBuffer() const {
diff --git a/src/video_core/renderer_vulkan/vk_sampler_cache.h b/src/video_core/renderer_vulkan/vk_sampler_cache.h
index 771b05c73..1f73b716b 100644
--- a/src/video_core/renderer_vulkan/vk_sampler_cache.h
+++ b/src/video_core/renderer_vulkan/vk_sampler_cache.h
@@ -4,9 +4,6 @@
 
 #pragma once
 
-#include <unordered_map>
-
-#include "common/common_types.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/sampler_cache.h"
 #include "video_core/textures/texture.h"
@@ -21,9 +18,9 @@ public:
     ~VKSamplerCache();
 
 protected:
-    UniqueSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const;
+    UniqueSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
 
-    vk::Sampler ToSamplerType(const UniqueSampler& sampler) const;
+    vk::Sampler ToSamplerType(const UniqueSampler& sampler) const override;
 
 private:
     const VKDevice& device;
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index f1fea1871..0f8116458 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -19,23 +19,19 @@ VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_man
 
 VKScheduler::~VKScheduler() = default;
 
-VKExecutionContext VKScheduler::GetExecutionContext() const {
-    return VKExecutionContext(current_fence, current_cmdbuf);
-}
-
-VKExecutionContext VKScheduler::Flush(vk::Semaphore semaphore) {
+void VKScheduler::Flush(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
-    current_fence->Release();
+    if (release_fence)
+        current_fence->Release();
     AllocateNewContext();
-    return GetExecutionContext();
 }
 
-VKExecutionContext VKScheduler::Finish(vk::Semaphore semaphore) {
+void VKScheduler::Finish(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
     current_fence->Wait();
-    current_fence->Release();
+    if (release_fence)
+        current_fence->Release();
     AllocateNewContext();
-    return GetExecutionContext();
 }
 
 void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h
index cfaf5376f..0e5b49c7f 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.h
+++ b/src/video_core/renderer_vulkan/vk_scheduler.h
@@ -10,10 +10,43 @@
 namespace Vulkan {
 
 class VKDevice;
-class VKExecutionContext;
 class VKFence;
 class VKResourceManager;
 
+class VKFenceView {
+public:
+    VKFenceView() = default;
+    VKFenceView(VKFence* const& fence) : fence{fence} {}
+
+    VKFence* operator->() const noexcept {
+        return fence;
+    }
+
+    operator VKFence&() const noexcept {
+        return *fence;
+    }
+
+private:
+    VKFence* const& fence;
+};
+
+class VKCommandBufferView {
+public:
+    VKCommandBufferView() = default;
+    VKCommandBufferView(const vk::CommandBuffer& cmdbuf) : cmdbuf{cmdbuf} {}
+
+    const vk::CommandBuffer* operator->() const noexcept {
+        return &cmdbuf;
+    }
+
+    operator vk::CommandBuffer() const noexcept {
+        return cmdbuf;
+    }
+
+private:
+    const vk::CommandBuffer& cmdbuf;
+};
+
 /// The scheduler abstracts command buffer and fence management with an interface that's able to do
 /// OpenGL-like operations on Vulkan command buffers.
 class VKScheduler {
@@ -21,16 +54,21 @@ public:
     explicit VKScheduler(const VKDevice& device, VKResourceManager& resource_manager);
     ~VKScheduler();
 
-    /// Gets the current execution context.
-    [[nodiscard]] VKExecutionContext GetExecutionContext() const;
+    /// Gets a reference to the current fence.
+    VKFenceView GetFence() const {
+        return current_fence;
+    }
+
+    /// Gets a reference to the current command buffer.
+    VKCommandBufferView GetCommandBuffer() const {
+        return current_cmdbuf;
+    }
 
-    /// Sends the current execution context to the GPU. It invalidates the current execution context
-    /// and returns a new one.
-    VKExecutionContext Flush(vk::Semaphore semaphore = nullptr);
+    /// Sends the current execution context to the GPU.
+    void Flush(bool release_fence = true, vk::Semaphore semaphore = nullptr);
 
-    /// Sends the current execution context to the GPU and waits for it to complete. It invalidates
-    /// the current execution context and returns a new one.
-    VKExecutionContext Finish(vk::Semaphore semaphore = nullptr);
+    /// Sends the current execution context to the GPU and waits for it to complete.
+    void Finish(bool release_fence = true, vk::Semaphore semaphore = nullptr);
 
 private:
     void SubmitExecution(vk::Semaphore semaphore);
@@ -44,26 +82,4 @@ private:
     VKFence* next_fence = nullptr;
 };
 
-class VKExecutionContext {
-    friend class VKScheduler;
-
-public:
-    VKExecutionContext() = default;
-
-    VKFence& GetFence() const {
-        return *fence;
-    }
-
-    vk::CommandBuffer GetCommandBuffer() const {
-        return cmdbuf;
-    }
-
-private:
-    explicit VKExecutionContext(VKFence* fence, vk::CommandBuffer cmdbuf)
-        : fence{fence}, cmdbuf{cmdbuf} {}
-
-    VKFence* fence{};
-    vk::CommandBuffer cmdbuf;
-};
-
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index 97ce214b1..a35b45c9c 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -205,10 +205,6 @@ public:
     }
 
 private:
-    using OperationDecompilerFn = Id (SPIRVDecompiler::*)(Operation);
-    using OperationDecompilersArray =
-        std::array<OperationDecompilerFn, static_cast<std::size_t>(OperationCode::Amount)>;
-
     static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount);
 
     void AllocateBindings() {
@@ -430,20 +426,17 @@ private:
         instance_index = DeclareBuiltIn(spv::BuiltIn::InstanceIndex, spv::StorageClass::Input,
                                         t_in_uint, "instance_index");
 
-        bool is_point_size_declared = false;
         bool is_clip_distances_declared = false;
         for (const auto index : ir.GetOutputAttributes()) {
-            if (index == Attribute::Index::PointSize) {
-                is_point_size_declared = true;
-            } else if (index == Attribute::Index::ClipDistances0123 ||
-                       index == Attribute::Index::ClipDistances4567) {
+            if (index == Attribute::Index::ClipDistances0123 ||
+                index == Attribute::Index::ClipDistances4567) {
                 is_clip_distances_declared = true;
             }
         }
 
         std::vector<Id> members;
         members.push_back(t_float4);
-        if (is_point_size_declared) {
+        if (ir.UsesPointSize()) {
             members.push_back(t_float);
         }
         if (is_clip_distances_declared) {
@@ -466,7 +459,7 @@ private:
 
         position_index = MemberDecorateBuiltIn(spv::BuiltIn::Position, "position", true);
         point_size_index =
-            MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", is_point_size_declared);
+            MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", ir.UsesPointSize());
         clip_distances_index = MemberDecorateBuiltIn(spv::BuiltIn::ClipDistance, "clip_distances",
                                                      is_clip_distances_declared);
 
@@ -712,7 +705,8 @@ private:
                 case Attribute::Index::Position:
                     return AccessElement(t_out_float, per_vertex, position_index,
                                          abuf->GetElement());
-                case Attribute::Index::PointSize:
+                case Attribute::Index::LayerViewportPointSize:
+                    UNIMPLEMENTED_IF(abuf->GetElement() != 3);
                     return AccessElement(t_out_float, per_vertex, point_size_index);
                 case Attribute::Index::ClipDistances0123:
                     return AccessElement(t_out_float, per_vertex, clip_distances_index,
@@ -741,6 +735,16 @@ private:
         return {};
     }
 
+    Id FCastHalf0(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id FCastHalf1(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id HNegate(Operation operation) {
         UNIMPLEMENTED();
         return {};
@@ -751,6 +755,11 @@ private:
         return {};
     }
 
+    Id HCastFloat(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id HUnpack(Operation operation) {
         UNIMPLEMENTED();
         return {};
@@ -806,12 +815,7 @@ private:
         return {};
     }
 
-    Id LogicalAll2(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id LogicalAny2(Operation operation) {
+    Id LogicalAnd2(Operation operation) {
         UNIMPLEMENTED();
         return {};
     }
@@ -949,6 +953,14 @@ private:
         return {};
     }
 
+    Id BranchIndirect(Operation operation) {
+        const Id op_a = VisitOperand<Type::Uint>(operation, 0);
+
+        Emit(OpStore(jmp_to, op_a));
+        BranchingOp([&]() { Emit(OpBranch(continue_label)); });
+        return {};
+    }
+
     Id PushFlowStack(Operation operation) {
         const auto target = std::get_if<ImmediateNode>(&*operation[0]);
         ASSERT(target);
@@ -1060,6 +1072,26 @@ private:
         return {};
     }
 
+    Id BallotThread(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteAll(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteAny(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteEqual(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id DeclareBuiltIn(spv::BuiltIn builtin, spv::StorageClass storage, Id type,
                       const std::string& name) {
         const Id id = OpVariable(type, storage);
@@ -1200,7 +1232,7 @@ private:
         return {};
     }
 
-    static constexpr OperationDecompilersArray operation_decompilers = {
+    static constexpr std::array operation_decompilers = {
         &SPIRVDecompiler::Assign,
 
         &SPIRVDecompiler::Ternary<&Module::OpSelect, Type::Float, Type::Bool, Type::Float,
@@ -1213,6 +1245,8 @@ private:
         &SPIRVDecompiler::Unary<&Module::OpFNegate, Type::Float>,
         &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::Float>,
         &SPIRVDecompiler::Ternary<&Module::OpFClamp, Type::Float>,
+        &SPIRVDecompiler::FCastHalf0,
+        &SPIRVDecompiler::FCastHalf1,
         &SPIRVDecompiler::Binary<&Module::OpFMin, Type::Float>,
         &SPIRVDecompiler::Binary<&Module::OpFMax, Type::Float>,
         &SPIRVDecompiler::Unary<&Module::OpCos, Type::Float>,
@@ -1273,6 +1307,7 @@ private:
         &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>,
         &SPIRVDecompiler::HNegate,
         &SPIRVDecompiler::HClamp,
+        &SPIRVDecompiler::HCastFloat,
         &SPIRVDecompiler::HUnpack,
         &SPIRVDecompiler::HMergeF32,
         &SPIRVDecompiler::HMergeH0,
@@ -1285,8 +1320,7 @@ private:
         &SPIRVDecompiler::Binary<&Module::OpLogicalNotEqual, Type::Bool>,
         &SPIRVDecompiler::Unary<&Module::OpLogicalNot, Type::Bool>,
         &SPIRVDecompiler::LogicalPick2,
-        &SPIRVDecompiler::LogicalAll2,
-        &SPIRVDecompiler::LogicalAny2,
+        &SPIRVDecompiler::LogicalAnd2,
 
         &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::Float>,
         &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::Float>,
@@ -1334,6 +1368,7 @@ private:
         &SPIRVDecompiler::ImageStore,
 
         &SPIRVDecompiler::Branch,
+        &SPIRVDecompiler::BranchIndirect,
         &SPIRVDecompiler::PushFlowStack,
         &SPIRVDecompiler::PopFlowStack,
         &SPIRVDecompiler::Exit,
@@ -1349,7 +1384,13 @@ private:
         &SPIRVDecompiler::WorkGroupId<0>,
         &SPIRVDecompiler::WorkGroupId<1>,
         &SPIRVDecompiler::WorkGroupId<2>,
+
+        &SPIRVDecompiler::BallotThread,
+        &SPIRVDecompiler::VoteAll,
+        &SPIRVDecompiler::VoteAny,
+        &SPIRVDecompiler::VoteEqual,
     };
+    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
     const VKDevice& device;
     const ShaderIR& ir;
diff --git a/src/video_core/renderer_vulkan/vk_stream_buffer.cpp b/src/video_core/renderer_vulkan/vk_stream_buffer.cpp
index 58ffa42f2..62f1427f5 100644
--- a/src/video_core/renderer_vulkan/vk_stream_buffer.cpp
+++ b/src/video_core/renderer_vulkan/vk_stream_buffer.cpp
@@ -46,12 +46,12 @@ std::tuple<u8*, u64, bool> VKStreamBuffer::Reserve(u64 size) {
     return {mapped_pointer + offset, offset, invalidation_mark.has_value()};
 }
 
-VKExecutionContext VKStreamBuffer::Send(VKExecutionContext exctx, u64 size) {
+void VKStreamBuffer::Send(u64 size) {
     ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
 
     if (invalidation_mark) {
         // TODO(Rodrigo): Find a better way to invalidate than waiting for all watches to finish.
-        exctx = scheduler.Flush();
+        scheduler.Flush();
         std::for_each(watches.begin(), watches.begin() + *invalidation_mark,
                       [&](auto& resource) { resource->Wait(); });
         invalidation_mark = std::nullopt;
@@ -62,11 +62,9 @@ VKExecutionContext VKStreamBuffer::Send(VKExecutionContext exctx, u64 size) {
         ReserveWatches(WATCHES_RESERVE_CHUNK);
     }
     // Add a watch for this allocation.
-    watches[used_watches++]->Watch(exctx.GetFence());
+    watches[used_watches++]->Watch(scheduler.GetFence());
 
     offset += size;
-
-    return exctx;
 }
 
 void VKStreamBuffer::CreateBuffers(VKMemoryManager& memory_manager, vk::BufferUsageFlags usage) {
diff --git a/src/video_core/renderer_vulkan/vk_stream_buffer.h b/src/video_core/renderer_vulkan/vk_stream_buffer.h
index 69d036ccd..842e54162 100644
--- a/src/video_core/renderer_vulkan/vk_stream_buffer.h
+++ b/src/video_core/renderer_vulkan/vk_stream_buffer.h
@@ -37,7 +37,7 @@ public:
     std::tuple<u8*, u64, bool> Reserve(u64 size);
 
     /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
-    [[nodiscard]] VKExecutionContext Send(VKExecutionContext exctx, u64 size);
+    void Send(u64 size);
 
     vk::Buffer GetBuffer() const {
         return *buffer;
diff --git a/src/video_core/shader/control_flow.cpp b/src/video_core/shader/control_flow.cpp
new file mode 100644
index 000000000..ec3a76690
--- /dev/null
+++ b/src/video_core/shader/control_flow.cpp
@@ -0,0 +1,481 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <list>
+#include <map>
+#include <stack>
+#include <unordered_map>
+#include <unordered_set>
+#include <vector>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/shader/control_flow.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+namespace {
+using Tegra::Shader::Instruction;
+using Tegra::Shader::OpCode;
+
+constexpr s32 unassigned_branch = -2;
+
+struct Query {
+    u32 address{};
+    std::stack<u32> ssy_stack{};
+    std::stack<u32> pbk_stack{};
+};
+
+struct BlockStack {
+    BlockStack() = default;
+    explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
+    std::stack<u32> ssy_stack{};
+    std::stack<u32> pbk_stack{};
+};
+
+struct BlockBranchInfo {
+    Condition condition{};
+    s32 address{exit_branch};
+    bool kill{};
+    bool is_sync{};
+    bool is_brk{};
+    bool ignore{};
+};
+
+struct BlockInfo {
+    u32 start{};
+    u32 end{};
+    bool visited{};
+    BlockBranchInfo branch{};
+
+    bool IsInside(const u32 address) const {
+        return start <= address && address <= end;
+    }
+};
+
+struct CFGRebuildState {
+    explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size,
+                             const u32 start)
+        : start{start}, program_code{program_code}, program_size{program_size} {}
+
+    u32 start{};
+    std::vector<BlockInfo> block_info{};
+    std::list<u32> inspect_queries{};
+    std::list<Query> queries{};
+    std::unordered_map<u32, u32> registered{};
+    std::unordered_set<u32> labels{};
+    std::map<u32, u32> ssy_labels{};
+    std::map<u32, u32> pbk_labels{};
+    std::unordered_map<u32, BlockStack> stacks{};
+    const ProgramCode& program_code;
+    const std::size_t program_size;
+};
+
+enum class BlockCollision : u32 { None, Found, Inside };
+
+std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address) {
+    const auto& blocks = state.block_info;
+    for (u32 index = 0; index < blocks.size(); index++) {
+        if (blocks[index].start == address) {
+            return {BlockCollision::Found, index};
+        }
+        if (blocks[index].IsInside(address)) {
+            return {BlockCollision::Inside, index};
+        }
+    }
+    return {BlockCollision::None, 0xFFFFFFFF};
+}
+
+struct ParseInfo {
+    BlockBranchInfo branch_info{};
+    u32 end_address{};
+};
+
+BlockInfo& CreateBlockInfo(CFGRebuildState& state, u32 start, u32 end) {
+    auto& it = state.block_info.emplace_back();
+    it.start = start;
+    it.end = end;
+    const u32 index = static_cast<u32>(state.block_info.size() - 1);
+    state.registered.insert({start, index});
+    return it;
+}
+
+Pred GetPredicate(u32 index, bool negated) {
+    return static_cast<Pred>(index + (negated ? 8 : 0));
+}
+
+/**
+ * Returns whether the instruction at the specified offset is a 'sched' instruction.
+ * Sched instructions always appear before a sequence of 3 instructions.
+ */
+constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
+    constexpr u32 SchedPeriod = 4;
+    u32 absolute_offset = offset - main_offset;
+
+    return (absolute_offset % SchedPeriod) == 0;
+}
+
+enum class ParseResult : u32 {
+    ControlCaught,
+    BlockEnd,
+    AbnormalFlow,
+};
+
+std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) {
+    u32 offset = static_cast<u32>(address);
+    const u32 end_address = static_cast<u32>(state.program_size / sizeof(Instruction));
+    ParseInfo parse_info{};
+
+    const auto insert_label = [](CFGRebuildState& state, u32 address) {
+        const auto pair = state.labels.emplace(address);
+        if (pair.second) {
+            state.inspect_queries.push_back(address);
+        }
+    };
+
+    while (true) {
+        if (offset >= end_address) {
+            // ASSERT_OR_EXECUTE can't be used, as it ignores the break
+            ASSERT_MSG(false, "Shader passed the current limit!");
+            parse_info.branch_info.address = exit_branch;
+            parse_info.branch_info.ignore = false;
+            break;
+        }
+        if (state.registered.count(offset) != 0) {
+            parse_info.branch_info.address = offset;
+            parse_info.branch_info.ignore = true;
+            break;
+        }
+        if (IsSchedInstruction(offset, state.start)) {
+            offset++;
+            continue;
+        }
+        const Instruction instr = {state.program_code[offset]};
+        const auto opcode = OpCode::Decode(instr);
+        if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) {
+            offset++;
+            continue;
+        }
+
+        switch (opcode->get().GetId()) {
+        case OpCode::Id::EXIT: {
+            const auto pred_index = static_cast<u32>(instr.pred.pred_index);
+            parse_info.branch_info.condition.predicate =
+                GetPredicate(pred_index, instr.negate_pred != 0);
+            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+                offset++;
+                continue;
+            }
+            const ConditionCode cc = instr.flow_condition_code;
+            parse_info.branch_info.condition.cc = cc;
+            if (cc == ConditionCode::F) {
+                offset++;
+                continue;
+            }
+            parse_info.branch_info.address = exit_branch;
+            parse_info.branch_info.kill = false;
+            parse_info.branch_info.is_sync = false;
+            parse_info.branch_info.is_brk = false;
+            parse_info.branch_info.ignore = false;
+            parse_info.end_address = offset;
+
+            return {ParseResult::ControlCaught, parse_info};
+        }
+        case OpCode::Id::BRA: {
+            if (instr.bra.constant_buffer != 0) {
+                return {ParseResult::AbnormalFlow, parse_info};
+            }
+            const auto pred_index = static_cast<u32>(instr.pred.pred_index);
+            parse_info.branch_info.condition.predicate =
+                GetPredicate(pred_index, instr.negate_pred != 0);
+            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+                offset++;
+                continue;
+            }
+            const ConditionCode cc = instr.flow_condition_code;
+            parse_info.branch_info.condition.cc = cc;
+            if (cc == ConditionCode::F) {
+                offset++;
+                continue;
+            }
+            const u32 branch_offset = offset + instr.bra.GetBranchTarget();
+            if (branch_offset == 0) {
+                parse_info.branch_info.address = exit_branch;
+            } else {
+                parse_info.branch_info.address = branch_offset;
+            }
+            insert_label(state, branch_offset);
+            parse_info.branch_info.kill = false;
+            parse_info.branch_info.is_sync = false;
+            parse_info.branch_info.is_brk = false;
+            parse_info.branch_info.ignore = false;
+            parse_info.end_address = offset;
+
+            return {ParseResult::ControlCaught, parse_info};
+        }
+        case OpCode::Id::SYNC: {
+            const auto pred_index = static_cast<u32>(instr.pred.pred_index);
+            parse_info.branch_info.condition.predicate =
+                GetPredicate(pred_index, instr.negate_pred != 0);
+            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+                offset++;
+                continue;
+            }
+            const ConditionCode cc = instr.flow_condition_code;
+            parse_info.branch_info.condition.cc = cc;
+            if (cc == ConditionCode::F) {
+                offset++;
+                continue;
+            }
+            parse_info.branch_info.address = unassigned_branch;
+            parse_info.branch_info.kill = false;
+            parse_info.branch_info.is_sync = true;
+            parse_info.branch_info.is_brk = false;
+            parse_info.branch_info.ignore = false;
+            parse_info.end_address = offset;
+
+            return {ParseResult::ControlCaught, parse_info};
+        }
+        case OpCode::Id::BRK: {
+            const auto pred_index = static_cast<u32>(instr.pred.pred_index);
+            parse_info.branch_info.condition.predicate =
+                GetPredicate(pred_index, instr.negate_pred != 0);
+            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+                offset++;
+                continue;
+            }
+            const ConditionCode cc = instr.flow_condition_code;
+            parse_info.branch_info.condition.cc = cc;
+            if (cc == ConditionCode::F) {
+                offset++;
+                continue;
+            }
+            parse_info.branch_info.address = unassigned_branch;
+            parse_info.branch_info.kill = false;
+            parse_info.branch_info.is_sync = false;
+            parse_info.branch_info.is_brk = true;
+            parse_info.branch_info.ignore = false;
+            parse_info.end_address = offset;
+
+            return {ParseResult::ControlCaught, parse_info};
+        }
+        case OpCode::Id::KIL: {
+            const auto pred_index = static_cast<u32>(instr.pred.pred_index);
+            parse_info.branch_info.condition.predicate =
+                GetPredicate(pred_index, instr.negate_pred != 0);
+            if (parse_info.branch_info.condition.predicate == Pred::NeverExecute) {
+                offset++;
+                continue;
+            }
+            const ConditionCode cc = instr.flow_condition_code;
+            parse_info.branch_info.condition.cc = cc;
+            if (cc == ConditionCode::F) {
+                offset++;
+                continue;
+            }
+            parse_info.branch_info.address = exit_branch;
+            parse_info.branch_info.kill = true;
+            parse_info.branch_info.is_sync = false;
+            parse_info.branch_info.is_brk = false;
+            parse_info.branch_info.ignore = false;
+            parse_info.end_address = offset;
+
+            return {ParseResult::ControlCaught, parse_info};
+        }
+        case OpCode::Id::SSY: {
+            const u32 target = offset + instr.bra.GetBranchTarget();
+            insert_label(state, target);
+            state.ssy_labels.emplace(offset, target);
+            break;
+        }
+        case OpCode::Id::PBK: {
+            const u32 target = offset + instr.bra.GetBranchTarget();
+            insert_label(state, target);
+            state.pbk_labels.emplace(offset, target);
+            break;
+        }
+        case OpCode::Id::BRX: {
+            return {ParseResult::AbnormalFlow, parse_info};
+        }
+        default:
+            break;
+        }
+
+        offset++;
+    }
+    parse_info.branch_info.kill = false;
+    parse_info.branch_info.is_sync = false;
+    parse_info.branch_info.is_brk = false;
+    parse_info.end_address = offset - 1;
+    return {ParseResult::BlockEnd, parse_info};
+}
+
+bool TryInspectAddress(CFGRebuildState& state) {
+    if (state.inspect_queries.empty()) {
+        return false;
+    }
+
+    const u32 address = state.inspect_queries.front();
+    state.inspect_queries.pop_front();
+    const auto [result, block_index] = TryGetBlock(state, address);
+    switch (result) {
+    case BlockCollision::Found: {
+        return true;
+    }
+    case BlockCollision::Inside: {
+        // This case is the tricky one:
+        // We need to Split the block in 2 sepparate blocks
+        const u32 end = state.block_info[block_index].end;
+        BlockInfo& new_block = CreateBlockInfo(state, address, end);
+        BlockInfo& current_block = state.block_info[block_index];
+        current_block.end = address - 1;
+        new_block.branch = current_block.branch;
+        BlockBranchInfo forward_branch{};
+        forward_branch.address = address;
+        forward_branch.ignore = true;
+        current_block.branch = forward_branch;
+        return true;
+    }
+    default:
+        break;
+    }
+    const auto [parse_result, parse_info] = ParseCode(state, address);
+    if (parse_result == ParseResult::AbnormalFlow) {
+        // if it's AbnormalFlow, we end it as false, ending the CFG reconstruction
+        return false;
+    }
+
+    BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address);
+    block_info.branch = parse_info.branch_info;
+    if (parse_info.branch_info.condition.IsUnconditional()) {
+        return true;
+    }
+
+    const u32 fallthrough_address = parse_info.end_address + 1;
+    state.inspect_queries.push_front(fallthrough_address);
+    return true;
+}
+
+bool TryQuery(CFGRebuildState& state) {
+    const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels,
+                                  BlockInfo& block) {
+        auto gather_start = labels.lower_bound(block.start);
+        const auto gather_end = labels.upper_bound(block.end);
+        while (gather_start != gather_end) {
+            cc.push(gather_start->second);
+            ++gather_start;
+        }
+    };
+    if (state.queries.empty()) {
+        return false;
+    }
+
+    Query& q = state.queries.front();
+    const u32 block_index = state.registered[q.address];
+    BlockInfo& block = state.block_info[block_index];
+    // If the block is visited, check if the stacks match, else gather the ssy/pbk
+    // labels into the current stack and look if the branch at the end of the block
+    // consumes a label. Schedule new queries accordingly
+    if (block.visited) {
+        BlockStack& stack = state.stacks[q.address];
+        const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
+                              (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
+        state.queries.pop_front();
+        return all_okay;
+    }
+    block.visited = true;
+    state.stacks.insert_or_assign(q.address, BlockStack{q});
+
+    Query q2(q);
+    state.queries.pop_front();
+    gather_labels(q2.ssy_stack, state.ssy_labels, block);
+    gather_labels(q2.pbk_stack, state.pbk_labels, block);
+    if (!block.branch.condition.IsUnconditional()) {
+        q2.address = block.end + 1;
+        state.queries.push_back(q2);
+    }
+
+    Query conditional_query{q2};
+    if (block.branch.is_sync) {
+        if (block.branch.address == unassigned_branch) {
+            block.branch.address = conditional_query.ssy_stack.top();
+        }
+        conditional_query.ssy_stack.pop();
+    }
+    if (block.branch.is_brk) {
+        if (block.branch.address == unassigned_branch) {
+            block.branch.address = conditional_query.pbk_stack.top();
+        }
+        conditional_query.pbk_stack.pop();
+    }
+    conditional_query.address = block.branch.address;
+    state.queries.push_back(std::move(conditional_query));
+    return true;
+}
+} // Anonymous namespace
+
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+                                              std::size_t program_size, u32 start_address) {
+    CFGRebuildState state{program_code, program_size, start_address};
+
+    // Inspect Code and generate blocks
+    state.labels.clear();
+    state.labels.emplace(start_address);
+    state.inspect_queries.push_back(state.start);
+    while (!state.inspect_queries.empty()) {
+        if (!TryInspectAddress(state)) {
+            return {};
+        }
+    }
+
+    // Decompile Stacks
+    state.queries.push_back(Query{state.start, {}, {}});
+    bool decompiled = true;
+    while (!state.queries.empty()) {
+        if (!TryQuery(state)) {
+            decompiled = false;
+            break;
+        }
+    }
+
+    // Sort and organize results
+    std::sort(state.block_info.begin(), state.block_info.end(),
+              [](const BlockInfo& a, const BlockInfo& b) { return a.start < b.start; });
+    ShaderCharacteristics result_out{};
+    result_out.decompilable = decompiled;
+    result_out.start = start_address;
+    result_out.end = start_address;
+    for (const auto& block : state.block_info) {
+        ShaderBlock new_block{};
+        new_block.start = block.start;
+        new_block.end = block.end;
+        new_block.ignore_branch = block.branch.ignore;
+        if (!new_block.ignore_branch) {
+            new_block.branch.cond = block.branch.condition;
+            new_block.branch.kills = block.branch.kill;
+            new_block.branch.address = block.branch.address;
+        }
+        result_out.end = std::max(result_out.end, block.end);
+        result_out.blocks.push_back(new_block);
+    }
+    if (result_out.decompilable) {
+        result_out.labels = std::move(state.labels);
+        return {std::move(result_out)};
+    }
+
+    // If it's not decompilable, merge the unlabelled blocks together
+    auto back = result_out.blocks.begin();
+    auto next = std::next(back);
+    while (next != result_out.blocks.end()) {
+        if (state.labels.count(next->start) == 0 && next->start == back->end + 1) {
+            back->end = next->end;
+            next = result_out.blocks.erase(next);
+            continue;
+        }
+        back = next;
+        ++next;
+    }
+    return {std::move(result_out)};
+}
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/control_flow.h b/src/video_core/shader/control_flow.h
new file mode 100644
index 000000000..b0a5e4f8c
--- /dev/null
+++ b/src/video_core/shader/control_flow.h
@@ -0,0 +1,79 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <list>
+#include <optional>
+#include <unordered_set>
+
+#include "video_core/engines/shader_bytecode.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+
+using Tegra::Shader::ConditionCode;
+using Tegra::Shader::Pred;
+
+constexpr s32 exit_branch = -1;
+
+struct Condition {
+    Pred predicate{Pred::UnusedIndex};
+    ConditionCode cc{ConditionCode::T};
+
+    bool IsUnconditional() const {
+        return predicate == Pred::UnusedIndex && cc == ConditionCode::T;
+    }
+
+    bool operator==(const Condition& other) const {
+        return std::tie(predicate, cc) == std::tie(other.predicate, other.cc);
+    }
+
+    bool operator!=(const Condition& other) const {
+        return !operator==(other);
+    }
+};
+
+struct ShaderBlock {
+    struct Branch {
+        Condition cond{};
+        bool kills{};
+        s32 address{};
+
+        bool operator==(const Branch& b) const {
+            return std::tie(cond, kills, address) == std::tie(b.cond, b.kills, b.address);
+        }
+
+        bool operator!=(const Branch& b) const {
+            return !operator==(b);
+        }
+    };
+
+    u32 start{};
+    u32 end{};
+    bool ignore_branch{};
+    Branch branch{};
+
+    bool operator==(const ShaderBlock& sb) const {
+        return std::tie(start, end, ignore_branch, branch) ==
+               std::tie(sb.start, sb.end, sb.ignore_branch, sb.branch);
+    }
+
+    bool operator!=(const ShaderBlock& sb) const {
+        return !operator==(sb);
+    }
+};
+
+struct ShaderCharacteristics {
+    std::list<ShaderBlock> blocks{};
+    bool decompilable{};
+    u32 start{};
+    u32 end{};
+    std::unordered_set<u32> labels{};
+};
+
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+                                              std::size_t program_size, u32 start_address);
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp
index 2c9ff28f2..47a9fd961 100644
--- a/src/video_core/shader/decode.cpp
+++ b/src/video_core/shader/decode.cpp
@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "video_core/engines/shader_bytecode.h"
 #include "video_core/engines/shader_header.h"
+#include "video_core/shader/control_flow.h"
 #include "video_core/shader/node_helper.h"
 #include "video_core/shader/shader_ir.h"
 
@@ -21,20 +22,6 @@ using Tegra::Shader::OpCode;
 
 namespace {
 
-/// Merges exit method of two parallel branches.
-constexpr ExitMethod ParallelExit(ExitMethod a, ExitMethod b) {
-    if (a == ExitMethod::Undetermined) {
-        return b;
-    }
-    if (b == ExitMethod::Undetermined) {
-        return a;
-    }
-    if (a == b) {
-        return a;
-    }
-    return ExitMethod::Conditional;
-}
-
 /**
  * Returns whether the instruction at the specified offset is a 'sched' instruction.
  * Sched instructions always appear before a sequence of 3 instructions.
@@ -51,85 +38,104 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
 void ShaderIR::Decode() {
     std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
 
-    std::set<u32> labels;
-    const ExitMethod exit_method = Scan(main_offset, MAX_PROGRAM_LENGTH, labels);
-    if (exit_method != ExitMethod::AlwaysEnd) {
-        UNREACHABLE_MSG("Program does not always end");
-    }
-
-    if (labels.empty()) {
-        basic_blocks.insert({main_offset, DecodeRange(main_offset, MAX_PROGRAM_LENGTH)});
+    disable_flow_stack = false;
+    const auto info = ScanFlow(program_code, program_size, main_offset);
+    if (info) {
+        const auto& shader_info = *info;
+        coverage_begin = shader_info.start;
+        coverage_end = shader_info.end;
+        if (shader_info.decompilable) {
+            disable_flow_stack = true;
+            const auto insert_block = [this](NodeBlock& nodes, u32 label) {
+                if (label == static_cast<u32>(exit_branch)) {
+                    return;
+                }
+                basic_blocks.insert({label, nodes});
+            };
+            const auto& blocks = shader_info.blocks;
+            NodeBlock current_block;
+            u32 current_label = static_cast<u32>(exit_branch);
+            for (auto& block : blocks) {
+                if (shader_info.labels.count(block.start) != 0) {
+                    insert_block(current_block, current_label);
+                    current_block.clear();
+                    current_label = block.start;
+                }
+                if (!block.ignore_branch) {
+                    DecodeRangeInner(current_block, block.start, block.end);
+                    InsertControlFlow(current_block, block);
+                } else {
+                    DecodeRangeInner(current_block, block.start, block.end + 1);
+                }
+            }
+            insert_block(current_block, current_label);
+            return;
+        }
+        LOG_WARNING(HW_GPU, "Flow Stack Removing Failed! Falling back to old method");
+        // we can't decompile it, fallback to standard method
+        for (const auto& block : shader_info.blocks) {
+            basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
+        }
         return;
     }
+    LOG_WARNING(HW_GPU, "Flow Analysis Failed! Falling back to brute force compiling");
+
+    // Now we need to deal with an undecompilable shader. We need to brute force
+    // a shader that captures every position.
+    coverage_begin = main_offset;
+    const u32 shader_end = static_cast<u32>(program_size / sizeof(u64));
+    coverage_end = shader_end;
+    for (u32 label = main_offset; label < shader_end; label++) {
+        basic_blocks.insert({label, DecodeRange(label, label + 1)});
+    }
+}
 
-    labels.insert(main_offset);
-
-    for (const u32 label : labels) {
-        const auto next_it = labels.lower_bound(label + 1);
-        const u32 next_label = next_it == labels.end() ? MAX_PROGRAM_LENGTH : *next_it;
+NodeBlock ShaderIR::DecodeRange(u32 begin, u32 end) {
+    NodeBlock basic_block;
+    DecodeRangeInner(basic_block, begin, end);
+    return basic_block;
+}
 
-        basic_blocks.insert({label, DecodeRange(label, next_label)});
+void ShaderIR::DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end) {
+    for (u32 pc = begin; pc < (begin > end ? MAX_PROGRAM_LENGTH : end);) {
+        pc = DecodeInstr(bb, pc);
     }
 }
 
-ExitMethod ShaderIR::Scan(u32 begin, u32 end, std::set<u32>& labels) {
-    const auto [iter, inserted] =
-        exit_method_map.emplace(std::make_pair(begin, end), ExitMethod::Undetermined);
-    ExitMethod& exit_method = iter->second;
-    if (!inserted)
-        return exit_method;
-
-    for (u32 offset = begin; offset != end && offset != MAX_PROGRAM_LENGTH; ++offset) {
-        coverage_begin = std::min(coverage_begin, offset);
-        coverage_end = std::max(coverage_end, offset + 1);
-
-        const Instruction instr = {program_code[offset]};
-        const auto opcode = OpCode::Decode(instr);
-        if (!opcode)
-            continue;
-        switch (opcode->get().GetId()) {
-        case OpCode::Id::EXIT: {
-            // The EXIT instruction can be predicated, which means that the shader can conditionally
-            // end on this instruction. We have to consider the case where the condition is not met
-            // and check the exit method of that other basic block.
-            using Tegra::Shader::Pred;
-            if (instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex)) {
-                return exit_method = ExitMethod::AlwaysEnd;
-            } else {
-                const ExitMethod not_met = Scan(offset + 1, end, labels);
-                return exit_method = ParallelExit(ExitMethod::AlwaysEnd, not_met);
-            }
+void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) {
+    const auto apply_conditions = [&](const Condition& cond, Node n) -> Node {
+        Node result = n;
+        if (cond.cc != ConditionCode::T) {
+            result = Conditional(GetConditionCode(cond.cc), {result});
         }
-        case OpCode::Id::BRA: {
-            const u32 target = offset + instr.bra.GetBranchTarget();
-            labels.insert(target);
-            const ExitMethod no_jmp = Scan(offset + 1, end, labels);
-            const ExitMethod jmp = Scan(target, end, labels);
-            return exit_method = ParallelExit(no_jmp, jmp);
-        }
-        case OpCode::Id::SSY:
-        case OpCode::Id::PBK: {
-            // The SSY and PBK use a similar encoding as the BRA instruction.
-            UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
-                                 "Constant buffer branching is not supported");
-            const u32 target = offset + instr.bra.GetBranchTarget();
-            labels.insert(target);
-            // Continue scanning for an exit method.
-            break;
+        if (cond.predicate != Pred::UnusedIndex) {
+            u32 pred = static_cast<u32>(cond.predicate);
+            const bool is_neg = pred > 7;
+            if (is_neg) {
+                pred -= 8;
+            }
+            result = Conditional(GetPredicate(pred, is_neg), {result});
         }
-        default:
-            break;
+        return result;
+    };
+    if (block.branch.address < 0) {
+        if (block.branch.kills) {
+            Node n = Operation(OperationCode::Discard);
+            n = apply_conditions(block.branch.cond, n);
+            bb.push_back(n);
+            global_code.push_back(n);
+            return;
         }
+        Node n = Operation(OperationCode::Exit);
+        n = apply_conditions(block.branch.cond, n);
+        bb.push_back(n);
+        global_code.push_back(n);
+        return;
     }
-    return exit_method = ExitMethod::AlwaysReturn;
-}
-
-NodeBlock ShaderIR::DecodeRange(u32 begin, u32 end) {
-    NodeBlock basic_block;
-    for (u32 pc = begin; pc < (begin > end ? MAX_PROGRAM_LENGTH : end);) {
-        pc = DecodeInstr(basic_block, pc);
-    }
-    return basic_block;
+    Node n = Operation(OperationCode::Branch, Immediate(block.branch.address));
+    n = apply_conditions(block.branch.cond, n);
+    bb.push_back(n);
+    global_code.push_back(n);
 }
 
 u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
@@ -140,15 +146,18 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
 
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
+    const u32 nv_address = ConvertAddressToNvidiaSpace(pc);
 
     // Decoding failure
     if (!opcode) {
         UNIMPLEMENTED_MSG("Unhandled instruction: {0:x}", instr.value);
+        bb.push_back(Comment(fmt::format("{:05x} Unimplemented Shader instruction (0x{:016x})",
+                                         nv_address, instr.value)));
         return pc + 1;
     }
 
-    bb.push_back(
-        Comment(fmt::format("{}: {} (0x{:016x})", pc, opcode->get().GetName(), instr.value)));
+    bb.push_back(Comment(
+        fmt::format("{:05x} {} (0x{:016x})", nv_address, opcode->get().GetName(), instr.value)));
 
     using Tegra::Shader::Pred;
     UNIMPLEMENTED_IF_MSG(instr.pred.full_pred == Pred::NeverExecute,
@@ -167,6 +176,7 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
         {OpCode::Type::Ffma, &ShaderIR::DecodeFfma},
         {OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2},
         {OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
+        {OpCode::Type::Warp, &ShaderIR::DecodeWarp},
         {OpCode::Type::Memory, &ShaderIR::DecodeMemory},
         {OpCode::Type::Texture, &ShaderIR::DecodeTexture},
         {OpCode::Type::Image, &ShaderIR::DecodeImage},
diff --git a/src/video_core/shader/decode/arithmetic.cpp b/src/video_core/shader/decode/arithmetic.cpp
index 87d8fecaa..1473c282a 100644
--- a/src/video_core/shader/decode/arithmetic.cpp
+++ b/src/video_core/shader/decode/arithmetic.cpp
@@ -42,11 +42,14 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
     case OpCode::Id::FMUL_R:
     case OpCode::Id::FMUL_IMM: {
         // FMUL does not have 'abs' bits and only the second operand has a 'neg' bit.
-        UNIMPLEMENTED_IF_MSG(instr.fmul.tab5cb8_2 != 0, "FMUL tab5cb8_2({}) is not implemented",
-                             instr.fmul.tab5cb8_2.Value());
-        UNIMPLEMENTED_IF_MSG(
-            instr.fmul.tab5c68_0 != 1, "FMUL tab5cb8_0({}) is not implemented",
-            instr.fmul.tab5c68_0.Value()); // SMO typical sends 1 here which seems to be the default
+        if (instr.fmul.tab5cb8_2 != 0) {
+            LOG_WARNING(HW_GPU, "FMUL tab5cb8_2({}) is not implemented",
+                        instr.fmul.tab5cb8_2.Value());
+        }
+        if (instr.fmul.tab5c68_0 != 1) {
+            LOG_WARNING(HW_GPU, "FMUL tab5cb8_0({}) is not implemented",
+                        instr.fmul.tab5c68_0.Value());
+        }
 
         op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b);
 
diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
index 7bcf38f23..6466fc011 100644
--- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp
+++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
@@ -23,7 +23,9 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
             LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName());
         }
     } else {
-        UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None);
+        if (instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None) {
+            LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName());
+        }
     }
 
     Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a);
diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp
index 4221f0c58..32facd6ba 100644
--- a/src/video_core/shader/decode/conversion.cpp
+++ b/src/video_core/shader/decode/conversion.cpp
@@ -14,6 +14,12 @@ using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
 
+namespace {
+constexpr OperationCode GetFloatSelector(u64 selector) {
+    return selector == 0 ? OperationCode::FCastHalf0 : OperationCode::FCastHalf1;
+}
+} // Anonymous namespace
+
 u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
@@ -22,7 +28,7 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     case OpCode::Id::I2I_R:
     case OpCode::Id::I2I_C:
     case OpCode::Id::I2I_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.selector);
+        UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
         UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
         UNIMPLEMENTED_IF(instr.alu.saturate_d);
 
@@ -57,8 +63,8 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     case OpCode::Id::I2F_R:
     case OpCode::Id::I2F_C:
     case OpCode::Id::I2F_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
-        UNIMPLEMENTED_IF(instr.conversion.selector);
+        UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
+        UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in I2F is not implemented");
 
@@ -82,14 +88,19 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
         value = GetOperandAbsNegFloat(value, false, instr.conversion.negate_a);
 
         SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
+
+        if (instr.conversion.dst_size == Register::Size::Short) {
+            value = Operation(OperationCode::HCastFloat, PRECISE, value);
+        }
+
         SetRegister(bb, instr.gpr0, value);
         break;
     }
     case OpCode::Id::F2F_R:
     case OpCode::Id::F2F_C:
     case OpCode::Id::F2F_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.f2f.dst_size != Register::Size::Word);
-        UNIMPLEMENTED_IF(instr.conversion.f2f.src_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
+        UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in F2F is not implemented");
 
@@ -107,6 +118,13 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
             }
         }();
 
+        if (instr.conversion.src_size == Register::Size::Short) {
+            value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
+                              std::move(value));
+        } else {
+            ASSERT(instr.conversion.float_src.selector == 0);
+        }
+
         value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
 
         value = [&]() {
@@ -124,19 +142,24 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
             default:
                 UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}",
                                   static_cast<u32>(instr.conversion.f2f.rounding.Value()));
-                return Immediate(0);
+                return value;
             }
         }();
         value = GetSaturatedFloat(value, instr.alu.saturate_d);
 
         SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
+
+        if (instr.conversion.dst_size == Register::Size::Short) {
+            value = Operation(OperationCode::HCastFloat, PRECISE, value);
+        }
+
         SetRegister(bb, instr.gpr0, value);
         break;
     }
     case OpCode::Id::F2I_R:
     case OpCode::Id::F2I_C:
     case OpCode::Id::F2I_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in F2I is not implemented");
         Node value = [&]() {
@@ -153,6 +176,13 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
             }
         }();
 
+        if (instr.conversion.src_size == Register::Size::Short) {
+            value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
+                              std::move(value));
+        } else {
+            ASSERT(instr.conversion.float_src.selector == 0);
+        }
+
         value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
 
         value = [&]() {
diff --git a/src/video_core/shader/decode/ffma.cpp b/src/video_core/shader/decode/ffma.cpp
index 29be25ca3..ca2f39e8d 100644
--- a/src/video_core/shader/decode/ffma.cpp
+++ b/src/video_core/shader/decode/ffma.cpp
@@ -18,10 +18,12 @@ u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) {
     const auto opcode = OpCode::Decode(instr);
 
     UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented");
-    UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_0 != 1, "FFMA tab5980_0({}) not implemented",
-                         instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO
-    UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_1 != 0, "FFMA tab5980_1({}) not implemented",
-                         instr.ffma.tab5980_1.Value());
+    if (instr.ffma.tab5980_0 != 1) {
+        LOG_WARNING(HW_GPU, "FFMA tab5980_0({}) not implemented", instr.ffma.tab5980_0.Value());
+    }
+    if (instr.ffma.tab5980_1 != 0) {
+        LOG_WARNING(HW_GPU, "FFMA tab5980_1({}) not implemented", instr.ffma.tab5980_1.Value());
+    }
 
     const Node op_a = GetRegister(instr.gpr8);
 
diff --git a/src/video_core/shader/decode/float_set.cpp b/src/video_core/shader/decode/float_set.cpp
index f5013e44a..5614e8a0d 100644
--- a/src/video_core/shader/decode/float_set.cpp
+++ b/src/video_core/shader/decode/float_set.cpp
@@ -15,7 +15,6 @@ using Tegra::Shader::OpCode;
 
 u32 ShaderIR::DecodeFloatSet(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
-    const auto opcode = OpCode::Decode(instr);
 
     const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fset.abs_a != 0,
                                             instr.fset.neg_a != 0);
diff --git a/src/video_core/shader/decode/float_set_predicate.cpp b/src/video_core/shader/decode/float_set_predicate.cpp
index 2323052b0..200c2c983 100644
--- a/src/video_core/shader/decode/float_set_predicate.cpp
+++ b/src/video_core/shader/decode/float_set_predicate.cpp
@@ -16,10 +16,9 @@ using Tegra::Shader::Pred;
 
 u32 ShaderIR::DecodeFloatSetPredicate(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
-    const auto opcode = OpCode::Decode(instr);
 
-    const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fsetp.abs_a != 0,
-                                            instr.fsetp.neg_a != 0);
+    Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fsetp.abs_a != 0,
+                                      instr.fsetp.neg_a != 0);
     Node op_b = [&]() {
         if (instr.is_b_imm) {
             return GetImmediate19(instr);
@@ -29,12 +28,13 @@ u32 ShaderIR::DecodeFloatSetPredicate(NodeBlock& bb, u32 pc) {
             return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
         }
     }();
-    op_b = GetOperandAbsNegFloat(op_b, instr.fsetp.abs_b, false);
+    op_b = GetOperandAbsNegFloat(std::move(op_b), instr.fsetp.abs_b, instr.fsetp.neg_b);
 
     // We can't use the constant predicate as destination.
     ASSERT(instr.fsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
 
-    const Node predicate = GetPredicateComparisonFloat(instr.fsetp.cond, op_a, op_b);
+    const Node predicate =
+        GetPredicateComparisonFloat(instr.fsetp.cond, std::move(op_a), std::move(op_b));
     const Node second_pred = GetPredicate(instr.fsetp.pred39, instr.fsetp.neg_pred != 0);
 
     const OperationCode combiner = GetPredicateCombiner(instr.fsetp.op);
diff --git a/src/video_core/shader/decode/half_set_predicate.cpp b/src/video_core/shader/decode/half_set_predicate.cpp
index d59d15bd8..afea33e5f 100644
--- a/src/video_core/shader/decode/half_set_predicate.cpp
+++ b/src/video_core/shader/decode/half_set_predicate.cpp
@@ -18,43 +18,56 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    UNIMPLEMENTED_IF(instr.hsetp2.ftz != 0);
+    DEBUG_ASSERT(instr.hsetp2.ftz == 0);
 
     Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a);
     op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a);
 
-    Node op_b = [&]() {
-        switch (opcode->get().GetId()) {
-        case OpCode::Id::HSETP2_R:
-            return GetOperandAbsNegHalf(GetRegister(instr.gpr20), instr.hsetp2.abs_a,
-                                        instr.hsetp2.negate_b);
-        default:
-            UNREACHABLE();
-            return Immediate(0);
-        }
-    }();
-    op_b = UnpackHalfFloat(op_b, instr.hsetp2.type_b);
-
-    // We can't use the constant predicate as destination.
-    ASSERT(instr.hsetp2.pred3 != static_cast<u64>(Pred::UnusedIndex));
-
-    const Node second_pred = GetPredicate(instr.hsetp2.pred39, instr.hsetp2.neg_pred != 0);
+    Tegra::Shader::PredCondition cond{};
+    bool h_and{};
+    Node op_b{};
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::HSETP2_C:
+        cond = instr.hsetp2.cbuf_and_imm.cond;
+        h_and = instr.hsetp2.cbuf_and_imm.h_and;
+        op_b = GetOperandAbsNegHalf(GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
+                                    instr.hsetp2.cbuf.abs_b, instr.hsetp2.cbuf.negate_b);
+        break;
+    case OpCode::Id::HSETP2_IMM:
+        cond = instr.hsetp2.cbuf_and_imm.cond;
+        h_and = instr.hsetp2.cbuf_and_imm.h_and;
+        op_b = UnpackHalfImmediate(instr, true);
+        break;
+    case OpCode::Id::HSETP2_R:
+        cond = instr.hsetp2.reg.cond;
+        h_and = instr.hsetp2.reg.h_and;
+        op_b =
+            UnpackHalfFloat(GetOperandAbsNegHalf(GetRegister(instr.gpr20), instr.hsetp2.reg.abs_b,
+                                                 instr.hsetp2.reg.negate_b),
+                            instr.hsetp2.reg.type_b);
+        break;
+    default:
+        UNREACHABLE();
+        op_b = Immediate(0);
+    }
 
     const OperationCode combiner = GetPredicateCombiner(instr.hsetp2.op);
-    const OperationCode pair_combiner =
-        instr.hsetp2.h_and ? OperationCode::LogicalAll2 : OperationCode::LogicalAny2;
-
-    const Node comparison = GetPredicateComparisonHalf(instr.hsetp2.cond, op_a, op_b);
-    const Node first_pred = Operation(pair_combiner, comparison);
+    const Node combined_pred = GetPredicate(instr.hsetp2.pred3, instr.hsetp2.neg_pred);
 
-    // Set the primary predicate to the result of Predicate OP SecondPredicate
-    const Node value = Operation(combiner, first_pred, second_pred);
-    SetPredicate(bb, instr.hsetp2.pred3, value);
+    const auto Write = [&](u64 dest, Node src) {
+        SetPredicate(bb, dest, Operation(combiner, std::move(src), combined_pred));
+    };
 
-    if (instr.hsetp2.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
-        // Set the secondary predicate to the result of !Predicate OP SecondPredicate, if enabled
-        const Node negated_pred = Operation(OperationCode::LogicalNegate, first_pred);
-        SetPredicate(bb, instr.hsetp2.pred0, Operation(combiner, negated_pred, second_pred));
+    const Node comparison = GetPredicateComparisonHalf(cond, op_a, op_b);
+    const u64 first = instr.hsetp2.pred0;
+    const u64 second = instr.hsetp2.pred39;
+    if (h_and) {
+        const Node joined = Operation(OperationCode::LogicalAnd2, comparison);
+        Write(first, joined);
+        Write(second, Operation(OperationCode::LogicalNegate, joined));
+    } else {
+        Write(first, Operation(OperationCode::LogicalPick2, comparison, Immediate(0u)));
+        Write(second, Operation(OperationCode::LogicalPick2, comparison, Immediate(1u)));
     }
 
     return pc;
diff --git a/src/video_core/shader/decode/hfma2.cpp b/src/video_core/shader/decode/hfma2.cpp
index c3bcf1ae9..5b44cb79c 100644
--- a/src/video_core/shader/decode/hfma2.cpp
+++ b/src/video_core/shader/decode/hfma2.cpp
@@ -22,9 +22,9 @@ u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) {
     const auto opcode = OpCode::Decode(instr);
 
     if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) {
-        UNIMPLEMENTED_IF(instr.hfma2.rr.precision != HalfPrecision::None);
+        DEBUG_ASSERT(instr.hfma2.rr.precision == HalfPrecision::None);
     } else {
-        UNIMPLEMENTED_IF(instr.hfma2.precision != HalfPrecision::None);
+        DEBUG_ASSERT(instr.hfma2.precision == HalfPrecision::None);
     }
 
     constexpr auto identity = HalfType::H0_H1;
diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp
index 24f022cc0..77151a24b 100644
--- a/src/video_core/shader/decode/image.cpp
+++ b/src/video_core/shader/decode/image.cpp
@@ -95,12 +95,8 @@ const Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::Image
 const Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg,
                                         Tegra::Shader::ImageType type) {
     const Node image_register{GetRegister(reg)};
-    const Node base_image{
+    const auto [base_image, cbuf_index, cbuf_offset]{
         TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()))};
-    const auto cbuf{std::get_if<CbufNode>(&*base_image)};
-    const auto cbuf_offset_imm{std::get_if<ImmediateNode>(&*cbuf->GetOffset())};
-    const auto cbuf_offset{cbuf_offset_imm->GetValue()};
-    const auto cbuf_index{cbuf->GetIndex()};
     const auto cbuf_key{(static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset)};
 
     // If this image has already been used, return the existing mapping.
diff --git a/src/video_core/shader/decode/integer_set.cpp b/src/video_core/shader/decode/integer_set.cpp
index 46e3d5905..59809bcd8 100644
--- a/src/video_core/shader/decode/integer_set.cpp
+++ b/src/video_core/shader/decode/integer_set.cpp
@@ -14,7 +14,6 @@ using Tegra::Shader::OpCode;
 
 u32 ShaderIR::DecodeIntegerSet(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
-    const auto opcode = OpCode::Decode(instr);
 
     const Node op_a = GetRegister(instr.gpr8);
     const Node op_b = [&]() {
diff --git a/src/video_core/shader/decode/integer_set_predicate.cpp b/src/video_core/shader/decode/integer_set_predicate.cpp
index dd20775d7..25e48fef8 100644
--- a/src/video_core/shader/decode/integer_set_predicate.cpp
+++ b/src/video_core/shader/decode/integer_set_predicate.cpp
@@ -16,7 +16,6 @@ using Tegra::Shader::Pred;
 
 u32 ShaderIR::DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
-    const auto opcode = OpCode::Decode(instr);
 
     const Node op_a = GetRegister(instr.gpr8);
 
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 80fc0ccfc..ed108bea8 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -95,10 +95,10 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             const Node op_b =
                 GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 4, index);
 
-            SetTemporal(bb, 0, op_a);
-            SetTemporal(bb, 1, op_b);
-            SetRegister(bb, instr.gpr0, GetTemporal(0));
-            SetRegister(bb, instr.gpr0.Value() + 1, GetTemporal(1));
+            SetTemporary(bb, 0, op_a);
+            SetTemporary(bb, 1, op_b);
+            SetRegister(bb, instr.gpr0, GetTemporary(0));
+            SetRegister(bb, instr.gpr0.Value() + 1, GetTemporary(1));
             break;
         }
         default:
@@ -136,9 +136,9 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
                 }
             }();
             for (u32 i = 0; i < count; ++i)
-                SetTemporal(bb, i, GetLmem(i * 4));
+                SetTemporary(bb, i, GetLmem(i * 4));
             for (u32 i = 0; i < count; ++i)
-                SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+                SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
             break;
         }
         default:
@@ -172,10 +172,10 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
                 Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
             const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
 
-            SetTemporal(bb, i, gmem);
+            SetTemporary(bb, i, gmem);
         }
         for (u32 i = 0; i < count; ++i) {
-            SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+            SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
         }
         break;
     }
@@ -253,11 +253,11 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             TrackAndGetGlobalMemory(bb, instr, true);
 
         // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
-        SetTemporal(bb, 0, real_address_base);
+        SetTemporary(bb, 0, real_address_base);
 
         const u32 count = GetUniformTypeElementsCount(type);
         for (u32 i = 0; i < count; ++i) {
-            SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
+            SetTemporary(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
         }
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
@@ -265,7 +265,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
                 Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
             const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
 
-            bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1)));
+            bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporary(i + 1)));
         }
         break;
     }
@@ -297,18 +297,13 @@ std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeB
     const auto addr_register{GetRegister(instr.gmem.gpr)};
     const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
 
-    const Node base_address{
-        TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()))};
-    const auto cbuf = std::get_if<CbufNode>(&*base_address);
-    ASSERT(cbuf != nullptr);
-    const auto cbuf_offset_imm = std::get_if<ImmediateNode>(&*cbuf->GetOffset());
-    ASSERT(cbuf_offset_imm != nullptr);
-    const auto cbuf_offset = cbuf_offset_imm->GetValue();
+    const auto [base_address, index, offset] =
+        TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()));
+    ASSERT(base_address != nullptr);
 
-    bb.push_back(
-        Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", cbuf->GetIndex(), cbuf_offset)));
+    bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset)));
 
-    const GlobalMemoryBase descriptor{cbuf->GetIndex(), cbuf_offset};
+    const GlobalMemoryBase descriptor{index, offset};
     const auto& [entry, is_new] = used_global_memory.try_emplace(descriptor);
     auto& usage = entry->second;
     if (is_write) {
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index d46a8ab82..d46e0f823 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -22,6 +22,12 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
     const auto opcode = OpCode::Decode(instr);
 
     switch (opcode->get().GetId()) {
+    case OpCode::Id::NOP: {
+        UNIMPLEMENTED_IF(instr.nop.cc != Tegra::Shader::ConditionCode::T);
+        UNIMPLEMENTED_IF(instr.nop.trigger != 0);
+        // With the previous preconditions, this instruction is a no-operation.
+        break;
+    }
     case OpCode::Id::EXIT: {
         const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
         UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "EXIT condition code used: {}",
@@ -68,6 +74,13 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
             case SystemVariable::InvocationInfo:
                 LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete");
                 return Immediate(0u);
+            case SystemVariable::Tid: {
+                Node value = Immediate(0);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdX), 0, 9);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdY), 16, 9);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdZ), 26, 5);
+                return value;
+            }
             case SystemVariable::TidX:
                 return Operation(OperationCode::LocalInvocationIdX);
             case SystemVariable::TidY:
@@ -91,11 +104,46 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::BRA: {
-        UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
-                             "BRA with constant buffers are not implemented");
+        Node branch;
+        if (instr.bra.constant_buffer == 0) {
+            const u32 target = pc + instr.bra.GetBranchTarget();
+            branch = Operation(OperationCode::Branch, Immediate(target));
+        } else {
+            const u32 target = pc + 1;
+            const Node op_a = GetConstBuffer(instr.cbuf36.index, instr.cbuf36.GetOffset());
+            const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
+                                                 PRECISE, op_a, Immediate(3));
+            const Node operand =
+                Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
+            branch = Operation(OperationCode::BranchIndirect, operand);
+        }
 
-        const u32 target = pc + instr.bra.GetBranchTarget();
-        const Node branch = Operation(OperationCode::Branch, Immediate(target));
+        const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
+        if (cc != Tegra::Shader::ConditionCode::T) {
+            bb.push_back(Conditional(GetConditionCode(cc), {branch}));
+        } else {
+            bb.push_back(branch);
+        }
+        break;
+    }
+    case OpCode::Id::BRX: {
+        Node operand;
+        if (instr.brx.constant_buffer != 0) {
+            const s32 target = pc + 1;
+            const Node index = GetRegister(instr.gpr8);
+            const Node op_a =
+                GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index);
+            const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
+                                                 PRECISE, op_a, Immediate(3));
+            operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
+        } else {
+            const s32 target = pc + instr.brx.GetBranchExtend();
+            const Node op_a = GetRegister(instr.gpr8);
+            const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
+                                                 PRECISE, op_a, Immediate(3));
+            operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
+        }
+        const Node branch = Operation(OperationCode::BranchIndirect, operand);
 
         const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
         if (cc != Tegra::Shader::ConditionCode::T) {
@@ -109,6 +157,10 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
                              "Constant buffer flow is not supported");
 
+        if (disable_flow_stack) {
+            break;
+        }
+
         // The SSY opcode tells the GPU where to re-converge divergent execution paths with SYNC.
         const u32 target = pc + instr.bra.GetBranchTarget();
         bb.push_back(
@@ -119,6 +171,10 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
                              "Constant buffer PBK is not supported");
 
+        if (disable_flow_stack) {
+            break;
+        }
+
         // PBK pushes to a stack the address where BRK will jump to.
         const u32 target = pc + instr.bra.GetBranchTarget();
         bb.push_back(
@@ -130,6 +186,10 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "SYNC condition code used: {}",
                              static_cast<u32>(cc));
 
+        if (disable_flow_stack) {
+            break;
+        }
+
         // The SYNC opcode jumps to the address previously set by the SSY opcode
         bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Ssy));
         break;
@@ -138,6 +198,9 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
         UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "BRK condition code used: {}",
                              static_cast<u32>(cc));
+        if (disable_flow_stack) {
+            break;
+        }
 
         // The BRK opcode jumps to the address previously set by the PBK opcode
         bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Pbk));
diff --git a/src/video_core/shader/decode/predicate_set_register.cpp b/src/video_core/shader/decode/predicate_set_register.cpp
index febbfeb50..84dbc50fe 100644
--- a/src/video_core/shader/decode/predicate_set_register.cpp
+++ b/src/video_core/shader/decode/predicate_set_register.cpp
@@ -15,7 +15,6 @@ using Tegra::Shader::OpCode;
 
 u32 ShaderIR::DecodePredicateSetRegister(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
-    const auto opcode = OpCode::Decode(instr);
 
     UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                          "Condition codes generation in PSET is not implemented");
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index cb480be9b..0b934a069 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -181,10 +181,10 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
                 const Node value =
                     Operation(OperationCode::TextureQueryDimensions, meta,
                               GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0)));
-                SetTemporal(bb, indexer++, value);
+                SetTemporary(bb, indexer++, value);
             }
             for (u32 i = 0; i < indexer; ++i) {
-                SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+                SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
             }
             break;
         }
@@ -238,10 +238,10 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
             auto params = coords;
             MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, element};
             const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params));
-            SetTemporal(bb, indexer++, value);
+            SetTemporary(bb, indexer++, value);
         }
         for (u32 i = 0; i < indexer; ++i) {
-            SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+            SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
         }
         break;
     }
@@ -269,7 +269,13 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
             LOG_WARNING(HW_GPU, "TLDS.NODEP implementation is incomplete");
         }
 
-        WriteTexsInstructionFloat(bb, instr, GetTldsCode(instr, texture_type, is_array));
+        const Node4 components = GetTldsCode(instr, texture_type, is_array);
+
+        if (instr.tlds.fp32_flag) {
+            WriteTexsInstructionFloat(bb, instr, components);
+        } else {
+            WriteTexsInstructionHalfFloat(bb, instr, components);
+        }
         break;
     }
     default:
@@ -302,13 +308,9 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, Textu
 const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg, TextureType type,
                                             bool is_array, bool is_shadow) {
     const Node sampler_register = GetRegister(reg);
-    const Node base_sampler =
+    const auto [base_sampler, cbuf_index, cbuf_offset] =
         TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
-    const auto cbuf = std::get_if<CbufNode>(&*base_sampler);
-    const auto cbuf_offset_imm = std::get_if<ImmediateNode>(&*cbuf->GetOffset());
-    ASSERT(cbuf_offset_imm != nullptr);
-    const auto cbuf_offset = cbuf_offset_imm->GetValue();
-    const auto cbuf_index = cbuf->GetIndex();
+    ASSERT(base_sampler != nullptr);
     const auto cbuf_key = (static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset);
 
     // If this sampler has already been used, return the existing mapping.
@@ -334,11 +336,11 @@ void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const
             // Skip disabled components
             continue;
         }
-        SetTemporal(bb, dest_elem++, components[elem]);
+        SetTemporary(bb, dest_elem++, components[elem]);
     }
     // After writing values in temporals, move them to the real registers
     for (u32 i = 0; i < dest_elem; ++i) {
-        SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+        SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
     }
 }
 
@@ -351,17 +353,17 @@ void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr,
     for (u32 component = 0; component < 4; ++component) {
         if (!instr.texs.IsComponentEnabled(component))
             continue;
-        SetTemporal(bb, dest_elem++, components[component]);
+        SetTemporary(bb, dest_elem++, components[component]);
     }
 
     for (u32 i = 0; i < dest_elem; ++i) {
         if (i < 2) {
             // Write the first two swizzle components to gpr0 and gpr0+1
-            SetRegister(bb, instr.gpr0.Value() + i % 2, GetTemporal(i));
+            SetRegister(bb, instr.gpr0.Value() + i % 2, GetTemporary(i));
         } else {
             ASSERT(instr.texs.HasTwoDestinations());
             // Write the rest of the swizzle components to gpr28 and gpr28+1
-            SetRegister(bb, instr.gpr28.Value() + i % 2, GetTemporal(i));
+            SetRegister(bb, instr.gpr28.Value() + i % 2, GetTemporary(i));
         }
     }
 }
@@ -389,11 +391,11 @@ void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr,
         return;
     }
 
-    SetTemporal(bb, 0, first_value);
-    SetTemporal(bb, 1, Operation(OperationCode::HPack2, values[2], values[3]));
+    SetTemporary(bb, 0, first_value);
+    SetTemporary(bb, 1, Operation(OperationCode::HPack2, values[2], values[3]));
 
-    SetRegister(bb, instr.gpr0, GetTemporal(0));
-    SetRegister(bb, instr.gpr28, GetTemporal(1));
+    SetRegister(bb, instr.gpr0, GetTemporary(0));
+    SetRegister(bb, instr.gpr28, GetTemporary(1));
 }
 
 Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
diff --git a/src/video_core/shader/decode/warp.cpp b/src/video_core/shader/decode/warp.cpp
new file mode 100644
index 000000000..04ca74f46
--- /dev/null
+++ b/src/video_core/shader/decode/warp.cpp
@@ -0,0 +1,55 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/engines/shader_bytecode.h"
+#include "video_core/shader/node_helper.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+
+using Tegra::Shader::Instruction;
+using Tegra::Shader::OpCode;
+using Tegra::Shader::Pred;
+using Tegra::Shader::VoteOperation;
+
+namespace {
+OperationCode GetOperationCode(VoteOperation vote_op) {
+    switch (vote_op) {
+    case VoteOperation::All:
+        return OperationCode::VoteAll;
+    case VoteOperation::Any:
+        return OperationCode::VoteAny;
+    case VoteOperation::Eq:
+        return OperationCode::VoteEqual;
+    default:
+        UNREACHABLE_MSG("Invalid vote operation={}", static_cast<u64>(vote_op));
+        return OperationCode::VoteAll;
+    }
+}
+} // Anonymous namespace
+
+u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
+    const Instruction instr = {program_code[pc]};
+    const auto opcode = OpCode::Decode(instr);
+
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::VOTE: {
+        const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0);
+        const Node active = Operation(OperationCode::BallotThread, value);
+        const Node vote = Operation(GetOperationCode(instr.vote.operation), value);
+        SetRegister(bb, instr.gpr0, active);
+        SetPredicate(bb, instr.vote.dest_pred, vote);
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName());
+        break;
+    }
+
+    return pc;
+}
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/xmad.cpp b/src/video_core/shader/decode/xmad.cpp
index 93dee77d1..206961909 100644
--- a/src/video_core/shader/decode/xmad.cpp
+++ b/src/video_core/shader/decode/xmad.cpp
@@ -73,8 +73,8 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) {
     if (is_psl) {
         product = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, product, Immediate(16));
     }
-    SetTemporal(bb, 0, product);
-    product = GetTemporal(0);
+    SetTemporary(bb, 0, product);
+    product = GetTemporary(0);
 
     const Node original_c = op_c;
     const Tegra::Shader::XmadMode set_mode = mode; // Workaround to clang compile error
@@ -98,13 +98,13 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) {
         }
     }();
 
-    SetTemporal(bb, 1, op_c);
-    op_c = GetTemporal(1);
+    SetTemporary(bb, 1, op_c);
+    op_c = GetTemporary(1);
 
     // TODO(Rodrigo): Use an appropiate sign for this operation
     Node sum = Operation(OperationCode::IAdd, product, op_c);
-    SetTemporal(bb, 2, sum);
-    sum = GetTemporal(2);
+    SetTemporary(bb, 2, sum);
+    sum = GetTemporary(2);
     if (is_merge) {
         const Node a = BitfieldExtract(sum, 0, 16);
         const Node b =
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 0ac83fcf0..5db9313c4 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -30,6 +30,8 @@ enum class OperationCode {
     FNegate,       /// (MetaArithmetic, float a) -> float
     FAbsolute,     /// (MetaArithmetic, float a) -> float
     FClamp,        /// (MetaArithmetic, float value, float min, float max) -> float
+    FCastHalf0,    /// (MetaArithmetic, f16vec2 a) -> float
+    FCastHalf1,    /// (MetaArithmetic, f16vec2 a) -> float
     FMin,          /// (MetaArithmetic, float a, float b) -> float
     FMax,          /// (MetaArithmetic, float a, float b) -> float
     FCos,          /// (MetaArithmetic, float a) -> float
@@ -83,17 +85,18 @@ enum class OperationCode {
     UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint
     UBitCount,        /// (MetaArithmetic, uint) -> uint
 
-    HAdd,      /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
-    HMul,      /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
-    HFma,      /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
-    HAbsolute, /// (f16vec2 a) -> f16vec2
-    HNegate,   /// (f16vec2 a, bool first, bool second) -> f16vec2
-    HClamp,    /// (f16vec2 src, float min, float max) -> f16vec2
-    HUnpack,   /// (Tegra::Shader::HalfType, T value) -> f16vec2
-    HMergeF32, /// (f16vec2 src) -> float
-    HMergeH0,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
-    HMergeH1,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
-    HPack2,    /// (float a, float b) -> f16vec2
+    HAdd,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
+    HMul,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
+    HFma,       /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
+    HAbsolute,  /// (f16vec2 a) -> f16vec2
+    HNegate,    /// (f16vec2 a, bool first, bool second) -> f16vec2
+    HClamp,     /// (f16vec2 src, float min, float max) -> f16vec2
+    HCastFloat, /// (MetaArithmetic, float a) -> f16vec2
+    HUnpack,    /// (Tegra::Shader::HalfType, T value) -> f16vec2
+    HMergeF32,  /// (f16vec2 src) -> float
+    HMergeH0,   /// (f16vec2 dest, f16vec2 src) -> f16vec2
+    HMergeH1,   /// (f16vec2 dest, f16vec2 src) -> f16vec2
+    HPack2,     /// (float a, float b) -> f16vec2
 
     LogicalAssign, /// (bool& dst, bool src) -> void
     LogicalAnd,    /// (bool a, bool b) -> bool
@@ -101,8 +104,7 @@ enum class OperationCode {
     LogicalXor,    /// (bool a, bool b) -> bool
     LogicalNegate, /// (bool a) -> bool
     LogicalPick2,  /// (bool2 pair, uint index) -> bool
-    LogicalAll2,   /// (bool2 a) -> bool
-    LogicalAny2,   /// (bool2 a) -> bool
+    LogicalAnd2,   /// (bool2 a) -> bool
 
     LogicalFLessThan,     /// (float a, float b) -> bool
     LogicalFEqual,        /// (float a, float b) -> bool
@@ -148,11 +150,12 @@ enum class OperationCode {
 
     ImageStore, /// (MetaImage, float[N] coords) -> void
 
-    Branch,        /// (uint branch_target) -> void
-    PushFlowStack, /// (uint branch_target) -> void
-    PopFlowStack,  /// () -> void
-    Exit,          /// () -> void
-    Discard,       /// () -> void
+    Branch,         /// (uint branch_target) -> void
+    BranchIndirect, /// (uint branch_target) -> void
+    PushFlowStack,  /// (uint branch_target) -> void
+    PopFlowStack,   /// () -> void
+    Exit,           /// () -> void
+    Discard,        /// () -> void
 
     EmitVertex,   /// () -> void
     EndPrimitive, /// () -> void
@@ -165,6 +168,11 @@ enum class OperationCode {
     WorkGroupIdY,       /// () -> uint
     WorkGroupIdZ,       /// () -> uint
 
+    BallotThread, /// (bool) -> uint
+    VoteAll,      /// (bool) -> bool
+    VoteAny,      /// (bool) -> bool
+    VoteEqual,    /// (bool) -> bool
+
     Amount,
 };
 
diff --git a/src/video_core/shader/node_helper.cpp b/src/video_core/shader/node_helper.cpp
index 6fccbbba3..b3dcd291c 100644
--- a/src/video_core/shader/node_helper.cpp
+++ b/src/video_core/shader/node_helper.cpp
@@ -12,7 +12,7 @@
 namespace VideoCommon::Shader {
 
 Node Conditional(Node condition, std::vector<Node> code) {
-    return MakeNode<ConditionalNode>(condition, std::move(code));
+    return MakeNode<ConditionalNode>(std::move(condition), std::move(code));
 }
 
 Node Comment(std::string text) {
diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp
index 11b545cca..1e5c7f660 100644
--- a/src/video_core/shader/shader_ir.cpp
+++ b/src/video_core/shader/shader_ir.cpp
@@ -22,8 +22,8 @@ using Tegra::Shader::PredCondition;
 using Tegra::Shader::PredOperation;
 using Tegra::Shader::Register;
 
-ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset)
-    : program_code{program_code}, main_offset{main_offset} {
+ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset, const std::size_t size)
+    : program_code{program_code}, main_offset{main_offset}, program_size{size} {
     Decode();
 }
 
@@ -61,8 +61,17 @@ Node ShaderIR::GetConstBufferIndirect(u64 index_, u64 offset_, Node node) {
     const auto [entry, is_new] = used_cbufs.try_emplace(index);
     entry->second.MarkAsUsedIndirect();
 
-    const Node final_offset = Operation(OperationCode::UAdd, NO_PRECISE, node, Immediate(offset));
-    return MakeNode<CbufNode>(index, final_offset);
+    Node final_offset = [&] {
+        // Attempt to inline constant buffer without a variable offset. This is done to allow
+        // tracking LDC calls.
+        if (const auto gpr = std::get_if<GprNode>(&*node)) {
+            if (gpr->GetIndex() == Register::ZeroIndex) {
+                return Immediate(offset);
+            }
+        }
+        return Operation(OperationCode::UAdd, NO_PRECISE, std::move(node), Immediate(offset));
+    }();
+    return MakeNode<CbufNode>(index, std::move(final_offset));
 }
 
 Node ShaderIR::GetPredicate(u64 pred_, bool negated) {
@@ -80,7 +89,7 @@ Node ShaderIR::GetPredicate(bool immediate) {
 
 Node ShaderIR::GetInputAttribute(Attribute::Index index, u64 element, Node buffer) {
     used_input_attributes.emplace(index);
-    return MakeNode<AbufNode>(index, static_cast<u32>(element), buffer);
+    return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
 }
 
 Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_address, Node buffer) {
@@ -89,6 +98,22 @@ Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_addres
 }
 
 Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buffer) {
+    if (index == Attribute::Index::LayerViewportPointSize) {
+        switch (element) {
+        case 0:
+            UNIMPLEMENTED();
+            break;
+        case 1:
+            uses_layer = true;
+            break;
+        case 2:
+            uses_viewport_index = true;
+            break;
+        case 3:
+            uses_point_size = true;
+            break;
+        }
+    }
     if (index == Attribute::Index::ClipDistances0123 ||
         index == Attribute::Index::ClipDistances4567) {
         const auto clip_index =
@@ -97,7 +122,7 @@ Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buff
     }
     used_output_attributes.insert(index);
 
-    return MakeNode<AbufNode>(index, static_cast<u32>(element), buffer);
+    return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
 }
 
 Node ShaderIR::GetInternalFlag(InternalFlag flag, bool negated) {
@@ -109,19 +134,19 @@ Node ShaderIR::GetInternalFlag(InternalFlag flag, bool negated) {
 }
 
 Node ShaderIR::GetLocalMemory(Node address) {
-    return MakeNode<LmemNode>(address);
+    return MakeNode<LmemNode>(std::move(address));
 }
 
-Node ShaderIR::GetTemporal(u32 id) {
+Node ShaderIR::GetTemporary(u32 id) {
     return GetRegister(Register::ZeroIndex + 1 + id);
 }
 
 Node ShaderIR::GetOperandAbsNegFloat(Node value, bool absolute, bool negate) {
     if (absolute) {
-        value = Operation(OperationCode::FAbsolute, NO_PRECISE, value);
+        value = Operation(OperationCode::FAbsolute, NO_PRECISE, std::move(value));
     }
     if (negate) {
-        value = Operation(OperationCode::FNegate, NO_PRECISE, value);
+        value = Operation(OperationCode::FNegate, NO_PRECISE, std::move(value));
     }
     return value;
 }
@@ -130,24 +155,26 @@ Node ShaderIR::GetSaturatedFloat(Node value, bool saturate) {
     if (!saturate) {
         return value;
     }
-    const Node positive_zero = Immediate(std::copysignf(0, 1));
-    const Node positive_one = Immediate(1.0f);
-    return Operation(OperationCode::FClamp, NO_PRECISE, value, positive_zero, positive_one);
+
+    Node positive_zero = Immediate(std::copysignf(0, 1));
+    Node positive_one = Immediate(1.0f);
+    return Operation(OperationCode::FClamp, NO_PRECISE, std::move(value), std::move(positive_zero),
+                     std::move(positive_one));
 }
 
-Node ShaderIR::ConvertIntegerSize(Node value, Tegra::Shader::Register::Size size, bool is_signed) {
+Node ShaderIR::ConvertIntegerSize(Node value, Register::Size size, bool is_signed) {
     switch (size) {
     case Register::Size::Byte:
-        value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE, value,
-                                Immediate(24));
-        value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE, value,
-                                Immediate(24));
+        value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE,
+                                std::move(value), Immediate(24));
+        value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE,
+                                std::move(value), Immediate(24));
         return value;
     case Register::Size::Short:
-        value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE, value,
-                                Immediate(16));
-        value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE, value,
-                                Immediate(16));
+        value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE,
+                                std::move(value), Immediate(16));
+        value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE,
+                                std::move(value), Immediate(16));
     case Register::Size::Word:
         // Default - do nothing
         return value;
@@ -163,27 +190,29 @@ Node ShaderIR::GetOperandAbsNegInteger(Node value, bool absolute, bool negate, b
         return value;
     }
     if (absolute) {
-        value = Operation(OperationCode::IAbsolute, NO_PRECISE, value);
+        value = Operation(OperationCode::IAbsolute, NO_PRECISE, std::move(value));
     }
     if (negate) {
-        value = Operation(OperationCode::INegate, NO_PRECISE, value);
+        value = Operation(OperationCode::INegate, NO_PRECISE, std::move(value));
     }
     return value;
 }
 
 Node ShaderIR::UnpackHalfImmediate(Instruction instr, bool has_negation) {
-    const Node value = Immediate(instr.half_imm.PackImmediates());
+    Node value = Immediate(instr.half_imm.PackImmediates());
     if (!has_negation) {
         return value;
     }
-    const Node first_negate = GetPredicate(instr.half_imm.first_negate != 0);
-    const Node second_negate = GetPredicate(instr.half_imm.second_negate != 0);
 
-    return Operation(OperationCode::HNegate, NO_PRECISE, value, first_negate, second_negate);
+    Node first_negate = GetPredicate(instr.half_imm.first_negate != 0);
+    Node second_negate = GetPredicate(instr.half_imm.second_negate != 0);
+
+    return Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), std::move(first_negate),
+                     std::move(second_negate));
 }
 
 Node ShaderIR::UnpackHalfFloat(Node value, Tegra::Shader::HalfType type) {
-    return Operation(OperationCode::HUnpack, type, value);
+    return Operation(OperationCode::HUnpack, type, std::move(value));
 }
 
 Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) {
@@ -191,11 +220,11 @@ Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) {
     case Tegra::Shader::HalfMerge::H0_H1:
         return src;
     case Tegra::Shader::HalfMerge::F32:
-        return Operation(OperationCode::HMergeF32, src);
+        return Operation(OperationCode::HMergeF32, std::move(src));
     case Tegra::Shader::HalfMerge::Mrg_H0:
-        return Operation(OperationCode::HMergeH0, dest, src);
+        return Operation(OperationCode::HMergeH0, std::move(dest), std::move(src));
     case Tegra::Shader::HalfMerge::Mrg_H1:
-        return Operation(OperationCode::HMergeH1, dest, src);
+        return Operation(OperationCode::HMergeH1, std::move(dest), std::move(src));
     }
     UNREACHABLE();
     return src;
@@ -203,10 +232,10 @@ Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) {
 
 Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) {
     if (absolute) {
-        value = Operation(OperationCode::HAbsolute, NO_PRECISE, value);
+        value = Operation(OperationCode::HAbsolute, NO_PRECISE, std::move(value));
     }
     if (negate) {
-        value = Operation(OperationCode::HNegate, NO_PRECISE, value, GetPredicate(true),
+        value = Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), GetPredicate(true),
                           GetPredicate(true));
     }
     return value;
@@ -216,9 +245,11 @@ Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) {
     if (!saturate) {
         return value;
     }
-    const Node positive_zero = Immediate(std::copysignf(0, 1));
-    const Node positive_one = Immediate(1.0f);
-    return Operation(OperationCode::HClamp, NO_PRECISE, value, positive_zero, positive_one);
+
+    Node positive_zero = Immediate(std::copysignf(0, 1));
+    Node positive_one = Immediate(1.0f);
+    return Operation(OperationCode::HClamp, NO_PRECISE, std::move(value), std::move(positive_zero),
+                     std::move(positive_one));
 }
 
 Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) {
@@ -246,7 +277,6 @@ Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, N
         condition == PredCondition::LessEqualWithNan ||
         condition == PredCondition::GreaterThanWithNan ||
         condition == PredCondition::GreaterEqualWithNan) {
-
         predicate = Operation(OperationCode::LogicalOr, predicate,
                               Operation(OperationCode::LogicalFIsNan, op_a));
         predicate = Operation(OperationCode::LogicalOr, predicate,
@@ -275,7 +305,8 @@ Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_si
     UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(),
                          "Unknown predicate comparison operation");
 
-    Node predicate = SignedOperation(comparison->second, is_signed, NO_PRECISE, op_a, op_b);
+    Node predicate = SignedOperation(comparison->second, is_signed, NO_PRECISE, std::move(op_a),
+                                     std::move(op_b));
 
     UNIMPLEMENTED_IF_MSG(condition == PredCondition::LessThanWithNan ||
                              condition == PredCondition::NotEqualWithNan ||
@@ -305,9 +336,7 @@ Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition
     UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(),
                          "Unknown predicate comparison operation");
 
-    const Node predicate = Operation(comparison->second, NO_PRECISE, op_a, op_b);
-
-    return predicate;
+    return Operation(comparison->second, NO_PRECISE, std::move(op_a), std::move(op_b));
 }
 
 OperationCode ShaderIR::GetPredicateCombiner(PredOperation operation) {
@@ -333,31 +362,32 @@ Node ShaderIR::GetConditionCode(Tegra::Shader::ConditionCode cc) {
 }
 
 void ShaderIR::SetRegister(NodeBlock& bb, Register dest, Node src) {
-    bb.push_back(Operation(OperationCode::Assign, GetRegister(dest), src));
+    bb.push_back(Operation(OperationCode::Assign, GetRegister(dest), std::move(src)));
 }
 
 void ShaderIR::SetPredicate(NodeBlock& bb, u64 dest, Node src) {
-    bb.push_back(Operation(OperationCode::LogicalAssign, GetPredicate(dest), src));
+    bb.push_back(Operation(OperationCode::LogicalAssign, GetPredicate(dest), std::move(src)));
 }
 
 void ShaderIR::SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value) {
-    bb.push_back(Operation(OperationCode::LogicalAssign, GetInternalFlag(flag), value));
+    bb.push_back(Operation(OperationCode::LogicalAssign, GetInternalFlag(flag), std::move(value)));
 }
 
 void ShaderIR::SetLocalMemory(NodeBlock& bb, Node address, Node value) {
-    bb.push_back(Operation(OperationCode::Assign, GetLocalMemory(address), value));
+    bb.push_back(
+        Operation(OperationCode::Assign, GetLocalMemory(std::move(address)), std::move(value)));
 }
 
-void ShaderIR::SetTemporal(NodeBlock& bb, u32 id, Node value) {
-    SetRegister(bb, Register::ZeroIndex + 1 + id, value);
+void ShaderIR::SetTemporary(NodeBlock& bb, u32 id, Node value) {
+    SetRegister(bb, Register::ZeroIndex + 1 + id, std::move(value));
 }
 
 void ShaderIR::SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc) {
     if (!sets_cc) {
         return;
     }
-    const Node zerop = Operation(OperationCode::LogicalFEqual, value, Immediate(0.0f));
-    SetInternalFlag(bb, InternalFlag::Zero, zerop);
+    Node zerop = Operation(OperationCode::LogicalFEqual, std::move(value), Immediate(0.0f));
+    SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop));
     LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete");
 }
 
@@ -365,13 +395,18 @@ void ShaderIR::SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_
     if (!sets_cc) {
         return;
     }
-    const Node zerop = Operation(OperationCode::LogicalIEqual, value, Immediate(0));
-    SetInternalFlag(bb, InternalFlag::Zero, zerop);
+    Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), Immediate(0));
+    SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop));
     LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete");
 }
 
 Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) {
-    return Operation(OperationCode::UBitfieldExtract, NO_PRECISE, value, Immediate(offset),
+    return Operation(OperationCode::UBitfieldExtract, NO_PRECISE, std::move(value),
+                     Immediate(offset), Immediate(bits));
+}
+
+Node ShaderIR::BitfieldInsert(Node base, Node insert, u32 offset, u32 bits) {
+    return Operation(OperationCode::UBitfieldInsert, NO_PRECISE, base, insert, Immediate(offset),
                      Immediate(bits));
 }
 
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index e22548208..bcc9b79b6 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -5,13 +5,10 @@
 #pragma once
 
 #include <array>
-#include <cstring>
 #include <map>
 #include <optional>
 #include <set>
-#include <string>
 #include <tuple>
-#include <variant>
 #include <vector>
 
 #include "common/common_types.h"
@@ -22,18 +19,12 @@
 
 namespace VideoCommon::Shader {
 
+struct ShaderBlock;
+
 using ProgramCode = std::vector<u64>;
 
 constexpr u32 MAX_PROGRAM_LENGTH = 0x1000;
 
-/// Describes the behaviour of code path of a given entry point and a return point.
-enum class ExitMethod {
-    Undetermined, ///< Internal value. Only occur when analyzing JMP loop.
-    AlwaysReturn, ///< All code paths reach the return point.
-    Conditional,  ///< Code path reaches the return point or an END instruction conditionally.
-    AlwaysEnd,    ///< All code paths reach a END instruction.
-};
-
 class ConstBuffer {
 public:
     explicit ConstBuffer(u32 max_offset, bool is_indirect)
@@ -73,7 +64,7 @@ struct GlobalMemoryUsage {
 
 class ShaderIR final {
 public:
-    explicit ShaderIR(const ProgramCode& program_code, u32 main_offset);
+    explicit ShaderIR(const ProgramCode& program_code, u32 main_offset, std::size_t size);
     ~ShaderIR();
 
     const std::map<u32, NodeBlock>& GetBasicBlocks() const {
@@ -121,6 +112,18 @@ public:
         return static_cast<std::size_t>(coverage_end * sizeof(u64));
     }
 
+    bool UsesLayer() const {
+        return uses_layer;
+    }
+
+    bool UsesViewportIndex() const {
+        return uses_viewport_index;
+    }
+
+    bool UsesPointSize() const {
+        return uses_point_size;
+    }
+
     bool HasPhysicalAttributes() const {
         return uses_physical_attributes;
     }
@@ -129,12 +132,20 @@ public:
         return header;
     }
 
+    bool IsFlowStackDisabled() const {
+        return disable_flow_stack;
+    }
+
+    u32 ConvertAddressToNvidiaSpace(const u32 address) const {
+        return (address - main_offset) * sizeof(Tegra::Shader::Instruction);
+    }
+
 private:
     void Decode();
 
-    ExitMethod Scan(u32 begin, u32 end, std::set<u32>& labels);
-
     NodeBlock DecodeRange(u32 begin, u32 end);
+    void DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end);
+    void InsertControlFlow(NodeBlock& bb, const ShaderBlock& block);
 
     /**
      * Decodes a single instruction from Tegra to IR.
@@ -156,6 +167,7 @@ private:
     u32 DecodeFfma(NodeBlock& bb, u32 pc);
     u32 DecodeHfma2(NodeBlock& bb, u32 pc);
     u32 DecodeConversion(NodeBlock& bb, u32 pc);
+    u32 DecodeWarp(NodeBlock& bb, u32 pc);
     u32 DecodeMemory(NodeBlock& bb, u32 pc);
     u32 DecodeTexture(NodeBlock& bb, u32 pc);
     u32 DecodeImage(NodeBlock& bb, u32 pc);
@@ -196,8 +208,8 @@ private:
     Node GetInternalFlag(InternalFlag flag, bool negated = false);
     /// Generates a node representing a local memory address
     Node GetLocalMemory(Node address);
-    /// Generates a temporal, internally it uses a post-RZ register
-    Node GetTemporal(u32 id);
+    /// Generates a temporary, internally it uses a post-RZ register
+    Node GetTemporary(u32 id);
 
     /// Sets a register. src value must be a number-evaluated node.
     void SetRegister(NodeBlock& bb, Tegra::Shader::Register dest, Node src);
@@ -207,8 +219,8 @@ private:
     void SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value);
     /// Sets a local memory address. address and value must be a number-evaluated node
     void SetLocalMemory(NodeBlock& bb, Node address, Node value);
-    /// Sets a temporal. Internally it uses a post-RZ register
-    void SetTemporal(NodeBlock& bb, u32 id, Node value);
+    /// Sets a temporary. Internally it uses a post-RZ register
+    void SetTemporary(NodeBlock& bb, u32 id, Node value);
 
     /// Sets internal flags from a float
     void SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc = true);
@@ -268,6 +280,9 @@ private:
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);
 
+    /// Inserts a sequence of bits from a node
+    Node BitfieldInsert(Node base, Node insert, u32 offset, u32 bits);
+
     void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
                                   const Node4& components);
 
@@ -314,7 +329,7 @@ private:
     void WriteLop3Instruction(NodeBlock& bb, Tegra::Shader::Register dest, Node op_a, Node op_b,
                               Node op_c, Node imm_lut, bool sets_cc);
 
-    Node TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
+    std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
 
     std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const;
 
@@ -326,10 +341,11 @@ private:
 
     const ProgramCode& program_code;
     const u32 main_offset;
+    const std::size_t program_size;
+    bool disable_flow_stack{};
 
     u32 coverage_begin{};
     u32 coverage_end{};
-    std::map<std::pair<u32, u32>, ExitMethod> exit_method_map;
 
     std::map<u32, NodeBlock> basic_blocks;
     NodeBlock global_code;
@@ -343,6 +359,9 @@ private:
     std::set<Image> used_images;
     std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
     std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory;
+    bool uses_layer{};
+    bool uses_viewport_index{};
+    bool uses_point_size{};
     bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
 
     Tegra::Shader::Header header;
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
index fc957d980..55f5949e4 100644
--- a/src/video_core/shader/track.cpp
+++ b/src/video_core/shader/track.cpp
@@ -15,56 +15,63 @@ namespace {
 std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor,
                                    OperationCode operation_code) {
     for (; cursor >= 0; --cursor) {
-        const Node node = code.at(cursor);
+        Node node = code.at(cursor);
+
         if (const auto operation = std::get_if<OperationNode>(&*node)) {
             if (operation->GetCode() == operation_code) {
-                return {node, cursor};
+                return {std::move(node), cursor};
             }
         }
+
         if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
             const auto& conditional_code = conditional->GetCode();
-            const auto [found, internal_cursor] = FindOperation(
+            auto [found, internal_cursor] = FindOperation(
                 conditional_code, static_cast<s64>(conditional_code.size() - 1), operation_code);
             if (found) {
-                return {found, cursor};
+                return {std::move(found), cursor};
             }
         }
     }
     return {};
 }
-} // namespace
+} // Anonymous namespace
 
-Node ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const {
+std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code,
+                                               s64 cursor) const {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
-        // Cbuf found, but it has to be immediate
-        return std::holds_alternative<ImmediateNode>(*cbuf->GetOffset()) ? tracked : nullptr;
+        // Constant buffer found, test if it's an immediate
+        const auto offset = cbuf->GetOffset();
+        if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
+            return {tracked, cbuf->GetIndex(), immediate->GetValue()};
+        }
+        return {};
     }
     if (const auto gpr = std::get_if<GprNode>(&*tracked)) {
         if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
-            return nullptr;
+            return {};
         }
         // Reduce the cursor in one to avoid infinite loops when the instruction sets the same
         // register that it uses as operand
         const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1);
         if (!source) {
-            return nullptr;
+            return {};
         }
         return TrackCbuf(source, code, new_cursor);
     }
     if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
-        for (std::size_t i = 0; i < operation->GetOperandsCount(); ++i) {
-            if (const auto found = TrackCbuf((*operation)[i], code, cursor)) {
-                // Cbuf found in operand
+        for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) {
+            if (auto found = TrackCbuf((*operation)[i - 1], code, cursor); std::get<0>(found)) {
+                // Cbuf found in operand.
                 return found;
             }
         }
-        return nullptr;
+        return {};
     }
     if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) {
         const auto& conditional_code = conditional->GetCode();
         return TrackCbuf(tracked, conditional_code, static_cast<s64>(conditional_code.size()));
     }
-    return nullptr;
+    return {};
 }
 
 std::optional<u32> ShaderIR::TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const {
diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp
index c50f6354d..4ceb219be 100644
--- a/src/video_core/surface.cpp
+++ b/src/video_core/surface.cpp
@@ -445,11 +445,12 @@ PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat
     switch (format) {
     case Tegra::FramebufferConfig::PixelFormat::ABGR8:
         return PixelFormat::ABGR8U;
+    case Tegra::FramebufferConfig::PixelFormat::RGB565:
+        return PixelFormat::B5G6R5U;
     case Tegra::FramebufferConfig::PixelFormat::BGRA8:
         return PixelFormat::BGRA8;
     default:
-        LOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
-        UNREACHABLE();
+        UNIMPLEMENTED_MSG("Unimplemented format={}", static_cast<u32>(format));
         return PixelFormat::ABGR8U;
     }
 }
diff --git a/src/video_core/texture_cache/surface_base.cpp b/src/video_core/texture_cache/surface_base.cpp
index 7a0fdb19b..683c49207 100644
--- a/src/video_core/texture_cache/surface_base.cpp
+++ b/src/video_core/texture_cache/surface_base.cpp
@@ -24,9 +24,8 @@ StagingCache::StagingCache() = default;
 StagingCache::~StagingCache() = default;
 
 SurfaceBaseImpl::SurfaceBaseImpl(GPUVAddr gpu_addr, const SurfaceParams& params)
-    : params{params}, mipmap_sizes(params.num_levels),
-      mipmap_offsets(params.num_levels), gpu_addr{gpu_addr}, host_memory_size{
-                                                                 params.GetHostSizeInBytes()} {
+    : params{params}, host_memory_size{params.GetHostSizeInBytes()}, gpu_addr{gpu_addr},
+      mipmap_sizes(params.num_levels), mipmap_offsets(params.num_levels) {
     std::size_t offset = 0;
     for (u32 level = 0; level < params.num_levels; ++level) {
         const std::size_t mipmap_size{params.GetGuestMipmapSize(level)};
@@ -75,9 +74,12 @@ MatchStructureResult SurfaceBaseImpl::MatchesStructure(const SurfaceParams& rhs)
 
     // Linear Surface check
     if (!params.is_tiled) {
-        if (std::tie(params.width, params.height, params.pitch) ==
-            std::tie(rhs.width, rhs.height, rhs.pitch)) {
-            return MatchStructureResult::FullMatch;
+        if (std::tie(params.height, params.pitch) == std::tie(rhs.height, rhs.pitch)) {
+            if (params.width == rhs.width) {
+                return MatchStructureResult::FullMatch;
+            } else {
+                return MatchStructureResult::SemiMatch;
+            }
         }
         return MatchStructureResult::None;
     }
diff --git a/src/video_core/texture_cache/surface_base.h b/src/video_core/texture_cache/surface_base.h
index 8ba386a8a..bcce8d863 100644
--- a/src/video_core/texture_cache/surface_base.h
+++ b/src/video_core/texture_cache/surface_base.h
@@ -200,8 +200,9 @@ public:
         modification_tick = tick;
     }
 
-    void MarkAsRenderTarget(const bool is_target) {
+    void MarkAsRenderTarget(const bool is_target, const u32 index) {
         this->is_target = is_target;
+        this->index = index;
     }
 
     void MarkAsPicked(const bool is_picked) {
@@ -221,6 +222,10 @@ public:
         return is_target;
     }
 
+    u32 GetRenderTarget() const {
+        return index;
+    }
+
     bool IsRegistered() const {
         return is_registered;
     }
@@ -307,10 +312,13 @@ private:
         return view;
     }
 
+    static constexpr u32 NO_RT = 0xFFFFFFFF;
+
     bool is_modified{};
     bool is_target{};
     bool is_registered{};
     bool is_picked{};
+    u32 index{NO_RT};
     u64 modification_tick{};
 };
 
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index 9c56e2b4f..fd5472451 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -290,12 +290,19 @@ std::size_t SurfaceParams::GetLayerSize(bool as_host_size, bool uncompressed) co
 
 std::size_t SurfaceParams::GetInnerMipmapMemorySize(u32 level, bool as_host_size,
                                                     bool uncompressed) const {
-    const bool tiled{as_host_size ? false : is_tiled};
     const u32 width{GetMipmapSize(uncompressed, GetMipWidth(level), GetDefaultBlockWidth())};
     const u32 height{GetMipmapSize(uncompressed, GetMipHeight(level), GetDefaultBlockHeight())};
     const u32 depth{is_layered ? 1U : GetMipDepth(level)};
-    return Tegra::Texture::CalculateSize(tiled, GetBytesPerPixel(), width, height, depth,
-                                         GetMipBlockHeight(level), GetMipBlockDepth(level));
+    if (is_tiled) {
+        return Tegra::Texture::CalculateSize(!as_host_size, GetBytesPerPixel(), width, height,
+                                             depth, GetMipBlockHeight(level),
+                                             GetMipBlockDepth(level));
+    } else if (as_host_size || IsBuffer()) {
+        return GetBytesPerPixel() * width * height * depth;
+    } else {
+        // Linear Texture Case
+        return pitch * height * depth;
+    }
 }
 
 bool SurfaceParams::operator==(const SurfaceParams& rhs) const {
diff --git a/src/video_core/texture_cache/surface_params.h b/src/video_core/texture_cache/surface_params.h
index 358d6757c..e7ef66ee2 100644
--- a/src/video_core/texture_cache/surface_params.h
+++ b/src/video_core/texture_cache/surface_params.h
@@ -58,7 +58,6 @@ public:
     std::size_t GetHostSizeInBytes() const {
         std::size_t host_size_in_bytes;
         if (GetCompressionType() == SurfaceCompression::Converted) {
-            constexpr std::size_t rgb8_bpp = 4ULL;
             // ASTC is uncompressed in software, in emulated as RGBA8
             host_size_in_bytes = 0;
             for (u32 level = 0; level < num_levels; ++level) {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index c9e72531a..2ec0203d1 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -116,10 +116,10 @@ public:
         std::lock_guard lock{mutex};
         auto& maxwell3d = system.GPU().Maxwell3D();
 
-        if (!maxwell3d.dirty_flags.zeta_buffer) {
+        if (!maxwell3d.dirty.depth_buffer) {
             return depth_buffer.view;
         }
-        maxwell3d.dirty_flags.zeta_buffer = false;
+        maxwell3d.dirty.depth_buffer = false;
 
         const auto& regs{maxwell3d.regs};
         const auto gpu_addr{regs.zeta.Address()};
@@ -133,11 +133,11 @@ public:
             regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
         auto surface_view = GetSurface(gpu_addr, depth_params, preserve_contents, true);
         if (depth_buffer.target)
-            depth_buffer.target->MarkAsRenderTarget(false);
+            depth_buffer.target->MarkAsRenderTarget(false, NO_RT);
         depth_buffer.target = surface_view.first;
         depth_buffer.view = surface_view.second;
         if (depth_buffer.target)
-            depth_buffer.target->MarkAsRenderTarget(true);
+            depth_buffer.target->MarkAsRenderTarget(true, DEPTH_RT);
         return surface_view.second;
     }
 
@@ -145,10 +145,10 @@ public:
         std::lock_guard lock{mutex};
         ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
         auto& maxwell3d = system.GPU().Maxwell3D();
-        if (!maxwell3d.dirty_flags.color_buffer[index]) {
+        if (!maxwell3d.dirty.render_target[index]) {
             return render_targets[index].view;
         }
-        maxwell3d.dirty_flags.color_buffer.reset(index);
+        maxwell3d.dirty.render_target[index] = false;
 
         const auto& regs{maxwell3d.regs};
         if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 ||
@@ -167,11 +167,11 @@ public:
         auto surface_view = GetSurface(gpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
                                        preserve_contents, true);
         if (render_targets[index].target)
-            render_targets[index].target->MarkAsRenderTarget(false);
+            render_targets[index].target->MarkAsRenderTarget(false, NO_RT);
         render_targets[index].target = surface_view.first;
         render_targets[index].view = surface_view.second;
         if (render_targets[index].target)
-            render_targets[index].target->MarkAsRenderTarget(true);
+            render_targets[index].target->MarkAsRenderTarget(true, static_cast<u32>(index));
         return surface_view.second;
     }
 
@@ -191,7 +191,7 @@ public:
         if (depth_buffer.target == nullptr) {
             return;
         }
-        depth_buffer.target->MarkAsRenderTarget(false);
+        depth_buffer.target->MarkAsRenderTarget(false, NO_RT);
         depth_buffer.target = nullptr;
         depth_buffer.view = nullptr;
     }
@@ -200,7 +200,7 @@ public:
         if (render_targets[index].target == nullptr) {
             return;
         }
-        render_targets[index].target->MarkAsRenderTarget(false);
+        render_targets[index].target->MarkAsRenderTarget(false, NO_RT);
         render_targets[index].target = nullptr;
         render_targets[index].view = nullptr;
     }
@@ -270,6 +270,17 @@ protected:
     // and reading it from a sepparate buffer.
     virtual void BufferCopy(TSurface& src_surface, TSurface& dst_surface) = 0;
 
+    void ManageRenderTargetUnregister(TSurface& surface) {
+        auto& maxwell3d = system.GPU().Maxwell3D();
+        const u32 index = surface->GetRenderTarget();
+        if (index == DEPTH_RT) {
+            maxwell3d.dirty.depth_buffer = true;
+        } else {
+            maxwell3d.dirty.render_target[index] = true;
+        }
+        maxwell3d.dirty.render_settings = true;
+    }
+
     void Register(TSurface surface) {
         const GPUVAddr gpu_addr = surface->GetGpuAddr();
         const CacheAddr cache_ptr = ToCacheAddr(system.GPU().MemoryManager().GetPointer(gpu_addr));
@@ -294,8 +305,9 @@ protected:
         if (guard_render_targets && surface->IsProtected()) {
             return;
         }
-        const GPUVAddr gpu_addr = surface->GetGpuAddr();
-        const CacheAddr cache_ptr = surface->GetCacheAddr();
+        if (!guard_render_targets && surface->IsRenderTarget()) {
+            ManageRenderTargetUnregister(surface);
+        }
         const std::size_t size = surface->GetSizeInBytes();
         const VAddr cpu_addr = surface->GetCpuAddr();
         rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
@@ -649,15 +661,6 @@ private:
                 }
                 return {current_surface, *view};
             }
-            // The next case is unsafe, so if we r in accurate GPU, just skip it
-            if (Settings::values.use_accurate_gpu_emulation) {
-                return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
-                                      MatchTopologyResult::FullMatch);
-            }
-            // This is the case the texture is a part of the parent.
-            if (current_surface->MatchesSubTexture(params, gpu_addr)) {
-                return RebuildSurface(current_surface, params, is_render);
-            }
         } else {
             // If there are many overlaps, odds are they are subtextures of the candidate
             // surface. We try to construct a new surface based on the candidate parameters,
@@ -793,6 +796,9 @@ private:
     static constexpr u64 registry_page_size{1 << registry_page_bits};
     std::unordered_map<CacheAddr, std::vector<TSurface>> registry;
 
+    static constexpr u32 DEPTH_RT = 8;
+    static constexpr u32 NO_RT = 0xFFFFFFFF;
+
     // The L1 Cache is used for fast texture lookup before checking the overlaps
     // This avoids calculating size and other stuffs.
     std::unordered_map<CacheAddr, TSurface> l1_cache;
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 7e8295944..7df5f1452 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -257,19 +257,21 @@ std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y,
 
 void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
                     u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data,
-                    u32 block_height_bit) {
+                    u32 block_height_bit, u32 offset_x, u32 offset_y) {
     const u32 block_height = 1U << block_height_bit;
     const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) /
                                   gob_size_x};
     for (u32 line = 0; line < subrect_height; ++line) {
+        const u32 dst_y = line + offset_y;
         const u32 gob_address_y =
-            (line / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs +
-            ((line % (gob_size_y * block_height)) / gob_size_y) * gob_size;
-        const auto& table = legacy_swizzle_table[line % gob_size_y];
+            (dst_y / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs +
+            ((dst_y % (gob_size_y * block_height)) / gob_size_y) * gob_size;
+        const auto& table = legacy_swizzle_table[dst_y % gob_size_y];
         for (u32 x = 0; x < subrect_width; ++x) {
+            const u32 dst_x = x + offset_x;
             const u32 gob_address =
-                gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height;
-            const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x];
+                gob_address_y + (dst_x * bytes_per_pixel / gob_size_x) * gob_size * block_height;
+            const u32 swizzled_offset = gob_address + table[(dst_x * bytes_per_pixel) % gob_size_x];
             u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel;
             u8* dest_addr = swizzled_data + swizzled_offset;
 
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index eaec9b5a5..f1e3952bc 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -44,7 +44,8 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height
 
 /// Copies an untiled subrectangle into a tiled surface.
 void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
-                    u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height);
+                    u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
+                    u32 offset_x, u32 offset_y);
 
 /// Copies a tiled subrectangle into a linear surface.
 void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width,
diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h
index e3be018b9..e36bc2c04 100644
--- a/src/video_core/textures/texture.h
+++ b/src/video_core/textures/texture.h
@@ -213,7 +213,7 @@ struct TICEntry {
         if (header_version != TICHeaderVersion::OneDBuffer) {
             return width_minus_1 + 1;
         }
-        return (buffer_high_width_minus_one << 16) | buffer_low_width_minus_one;
+        return ((buffer_high_width_minus_one << 16) | buffer_low_width_minus_one) + 1;
     }
 
     u32 Height() const {
diff --git a/src/yuzu/CMakeLists.txt b/src/yuzu/CMakeLists.txt
index 3dc0e47d0..f051e17b4 100644
--- a/src/yuzu/CMakeLists.txt
+++ b/src/yuzu/CMakeLists.txt
@@ -1,5 +1,6 @@
 set(CMAKE_AUTOMOC ON)
 set(CMAKE_AUTORCC ON)
+set(CMAKE_AUTOUIC ON)
 set(CMAKE_INCLUDE_CURRENT_DIR ON)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${PROJECT_SOURCE_DIR}/CMakeModules)
 
@@ -7,6 +8,7 @@ add_executable(yuzu
     Info.plist
     about_dialog.cpp
     about_dialog.h
+    aboutdialog.ui
     applets/error.cpp
     applets/error.h
     applets/profile_select.cpp
@@ -17,42 +19,59 @@ add_executable(yuzu
     applets/web_browser.h
     bootmanager.cpp
     bootmanager.h
+    compatdb.ui
     compatibility_list.cpp
     compatibility_list.h
     configuration/config.cpp
     configuration/config.h
+    configuration/configure.ui
     configuration/configure_audio.cpp
     configuration/configure_audio.h
+    configuration/configure_audio.ui
     configuration/configure_debug.cpp
     configuration/configure_debug.h
+    configuration/configure_debug.ui
     configuration/configure_dialog.cpp
     configuration/configure_dialog.h
     configuration/configure_gamelist.cpp
     configuration/configure_gamelist.h
+    configuration/configure_gamelist.ui
     configuration/configure_general.cpp
     configuration/configure_general.h
+    configuration/configure_general.ui
     configuration/configure_graphics.cpp
     configuration/configure_graphics.h
+    configuration/configure_graphics.ui
     configuration/configure_hotkeys.cpp
     configuration/configure_hotkeys.h
+    configuration/configure_hotkeys.ui
     configuration/configure_input.cpp
     configuration/configure_input.h
+    configuration/configure_input.ui
     configuration/configure_input_player.cpp
     configuration/configure_input_player.h
+    configuration/configure_input_player.ui
     configuration/configure_input_simple.cpp
     configuration/configure_input_simple.h
+    configuration/configure_input_simple.ui
     configuration/configure_mouse_advanced.cpp
     configuration/configure_mouse_advanced.h
+    configuration/configure_mouse_advanced.ui
+    configuration/configure_per_general.cpp
+    configuration/configure_per_general.h
+    configuration/configure_per_general.ui
     configuration/configure_profile_manager.cpp
     configuration/configure_profile_manager.h
+    configuration/configure_profile_manager.ui
     configuration/configure_system.cpp
     configuration/configure_system.h
-    configuration/configure_per_general.cpp
-    configuration/configure_per_general.h
+    configuration/configure_system.ui
     configuration/configure_touchscreen_advanced.cpp
     configuration/configure_touchscreen_advanced.h
+    configuration/configure_touchscreen_advanced.ui
     configuration/configure_web.cpp
     configuration/configure_web.h
+    configuration/configure_web.ui
     debugger/graphics/graphics_breakpoint_observer.cpp
     debugger/graphics/graphics_breakpoint_observer.h
     debugger/graphics/graphics_breakpoints.cpp
@@ -72,12 +91,14 @@ add_executable(yuzu
     game_list_worker.h
     loading_screen.cpp
     loading_screen.h
+    loading_screen.ui
     hotkeys.cpp
     hotkeys.h
     main.cpp
     main.h
-    ui_settings.cpp
-    ui_settings.h
+    main.ui
+    uisettings.cpp
+    uisettings.h
     util/limitable_input_dialog.cpp
     util/limitable_input_dialog.h
     util/sequence_dialog/sequence_dialog.cpp
@@ -89,44 +110,18 @@ add_executable(yuzu
     yuzu.rc
 )
 
-set(UIS
-    aboutdialog.ui
-    configuration/configure.ui
-    configuration/configure_audio.ui
-    configuration/configure_debug.ui
-    configuration/configure_gamelist.ui
-    configuration/configure_general.ui
-    configuration/configure_graphics.ui
-    configuration/configure_hotkeys.ui
-    configuration/configure_input.ui
-    configuration/configure_input_player.ui
-    configuration/configure_input_simple.ui
-    configuration/configure_mouse_advanced.ui
-    configuration/configure_per_general.ui
-    configuration/configure_profile_manager.ui
-    configuration/configure_system.ui
-    configuration/configure_touchscreen_advanced.ui
-    configuration/configure_web.ui
-    compatdb.ui
-    loading_screen.ui
-    main.ui
-)
-
 file(GLOB COMPAT_LIST
      ${PROJECT_BINARY_DIR}/dist/compatibility_list/compatibility_list.qrc
      ${PROJECT_BINARY_DIR}/dist/compatibility_list/compatibility_list.json)
 file(GLOB_RECURSE ICONS ${PROJECT_SOURCE_DIR}/dist/icons/*)
 file(GLOB_RECURSE THEMES ${PROJECT_SOURCE_DIR}/dist/qt_themes/*)
 
-qt5_wrap_ui(UI_HDRS ${UIS})
 
 target_sources(yuzu
     PRIVATE
         ${COMPAT_LIST}
         ${ICONS}
         ${THEMES}
-        ${UI_HDRS}
-        ${UIS}
 )
 
 if (APPLE)
diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp
index 73978ff5b..0456248ac 100644
--- a/src/yuzu/configuration/config.cpp
+++ b/src/yuzu/configuration/config.cpp
@@ -11,7 +11,7 @@
 #include "core/hle/service/hid/controllers/npad.h"
 #include "input_common/main.h"
 #include "yuzu/configuration/config.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 Config::Config() {
     // TODO: Don't hardcode the path; let the frontend decide where to put the config files.
@@ -436,8 +436,6 @@ void Config::ReadControlValues() {
 void Config::ReadCoreValues() {
     qt_config->beginGroup(QStringLiteral("Core"));
 
-    Settings::values.cpu_jit_enabled =
-        ReadSetting(QStringLiteral("cpu_jit_enabled"), true).toBool();
     Settings::values.use_multi_core = ReadSetting(QStringLiteral("use_multi_core"), false).toBool();
 
     qt_config->endGroup();
@@ -518,6 +516,7 @@ void Config::ReadPathValues() {
 
     UISettings::values.roms_path = ReadSetting(QStringLiteral("romsPath")).toString();
     UISettings::values.symbols_path = ReadSetting(QStringLiteral("symbolsPath")).toString();
+    UISettings::values.screenshot_path = ReadSetting(QStringLiteral("screenshotPath")).toString();
     UISettings::values.game_directory_path =
         ReadSetting(QStringLiteral("gameListRootDir"), QStringLiteral(".")).toString();
     UISettings::values.game_directory_deepscan =
@@ -831,7 +830,6 @@ void Config::SaveControlValues() {
 void Config::SaveCoreValues() {
     qt_config->beginGroup(QStringLiteral("Core"));
 
-    WriteSetting(QStringLiteral("cpu_jit_enabled"), Settings::values.cpu_jit_enabled, true);
     WriteSetting(QStringLiteral("use_multi_core"), Settings::values.use_multi_core, false);
 
     qt_config->endGroup();
diff --git a/src/yuzu/configuration/configure_debug.cpp b/src/yuzu/configuration/configure_debug.cpp
index 9a13bb797..5b7e03056 100644
--- a/src/yuzu/configuration/configure_debug.cpp
+++ b/src/yuzu/configuration/configure_debug.cpp
@@ -12,13 +12,13 @@
 #include "ui_configure_debug.h"
 #include "yuzu/configuration/configure_debug.h"
 #include "yuzu/debugger/console.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 ConfigureDebug::ConfigureDebug(QWidget* parent) : QWidget(parent), ui(new Ui::ConfigureDebug) {
     ui->setupUi(this);
     SetConfiguration();
 
-    connect(ui->open_log_button, &QPushButton::pressed, []() {
+    connect(ui->open_log_button, &QPushButton::clicked, []() {
         QString path = QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::LogDir));
         QDesktopServices::openUrl(QUrl::fromLocalFile(path));
     });
diff --git a/src/yuzu/configuration/configure_gamelist.cpp b/src/yuzu/configuration/configure_gamelist.cpp
index d1724ba89..daedbc33e 100644
--- a/src/yuzu/configuration/configure_gamelist.cpp
+++ b/src/yuzu/configuration/configure_gamelist.cpp
@@ -9,7 +9,7 @@
 #include "core/settings.h"
 #include "ui_configure_gamelist.h"
 #include "yuzu/configuration/configure_gamelist.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace {
 constexpr std::array default_icon_sizes{
diff --git a/src/yuzu/configuration/configure_general.cpp b/src/yuzu/configuration/configure_general.cpp
index 7a6e921cd..75fcbfea3 100644
--- a/src/yuzu/configuration/configure_general.cpp
+++ b/src/yuzu/configuration/configure_general.cpp
@@ -6,7 +6,7 @@
 #include "core/settings.h"
 #include "ui_configure_general.h"
 #include "yuzu/configuration/configure_general.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 ConfigureGeneral::ConfigureGeneral(QWidget* parent)
     : QWidget(parent), ui(new Ui::ConfigureGeneral) {
diff --git a/src/yuzu/configuration/configure_input.cpp b/src/yuzu/configuration/configure_input.cpp
index 4dd775aab..7613197f2 100644
--- a/src/yuzu/configuration/configure_input.cpp
+++ b/src/yuzu/configuration/configure_input.cpp
@@ -79,7 +79,7 @@ ConfigureInput::ConfigureInput(QWidget* parent)
     LoadConfiguration();
     UpdateUIEnabled();
 
-    connect(ui->restore_defaults_button, &QPushButton::pressed, this,
+    connect(ui->restore_defaults_button, &QPushButton::clicked, this,
             &ConfigureInput::RestoreDefaults);
 
     for (auto* enabled : players_controller) {
@@ -96,20 +96,20 @@ ConfigureInput::ConfigureInput(QWidget* parent)
             &ConfigureInput::UpdateUIEnabled);
 
     for (std::size_t i = 0; i < players_configure.size(); ++i) {
-        connect(players_configure[i], &QPushButton::pressed, this,
+        connect(players_configure[i], &QPushButton::clicked, this,
                 [this, i] { CallConfigureDialog<ConfigureInputPlayer>(*this, i, false); });
     }
 
-    connect(ui->handheld_configure, &QPushButton::pressed, this,
+    connect(ui->handheld_configure, &QPushButton::clicked, this,
             [this] { CallConfigureDialog<ConfigureInputPlayer>(*this, 8, false); });
 
-    connect(ui->debug_configure, &QPushButton::pressed, this,
+    connect(ui->debug_configure, &QPushButton::clicked, this,
             [this] { CallConfigureDialog<ConfigureInputPlayer>(*this, 9, true); });
 
-    connect(ui->mouse_advanced, &QPushButton::pressed, this,
+    connect(ui->mouse_advanced, &QPushButton::clicked, this,
             [this] { CallConfigureDialog<ConfigureMouseAdvanced>(*this); });
 
-    connect(ui->touchscreen_advanced, &QPushButton::pressed, this,
+    connect(ui->touchscreen_advanced, &QPushButton::clicked, this,
             [this] { CallConfigureDialog<ConfigureTouchscreenAdvanced>(*this); });
 }
 
diff --git a/src/yuzu/configuration/configure_input_player.cpp b/src/yuzu/configuration/configure_input_player.cpp
index 916baccc1..7b70f307c 100644
--- a/src/yuzu/configuration/configure_input_player.cpp
+++ b/src/yuzu/configuration/configure_input_player.cpp
@@ -244,7 +244,7 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
         }
 
         button->setContextMenuPolicy(Qt::CustomContextMenu);
-        connect(button, &QPushButton::released, [=] {
+        connect(button, &QPushButton::clicked, [=] {
             HandleClick(
                 button_map[button_id],
                 [=](const Common::ParamPackage& params) { buttons_param[button_id] = params; },
@@ -273,7 +273,7 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
             }
 
             analog_button->setContextMenuPolicy(Qt::CustomContextMenu);
-            connect(analog_button, &QPushButton::released, [=]() {
+            connect(analog_button, &QPushButton::clicked, [=]() {
                 HandleClick(analog_map_buttons[analog_id][sub_button_id],
                             [=](const Common::ParamPackage& params) {
                                 SetAnalogButton(params, analogs_param[analog_id],
@@ -300,7 +300,7 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
                             menu_location));
                     });
         }
-        connect(analog_map_stick[analog_id], &QPushButton::released, [=] {
+        connect(analog_map_stick[analog_id], &QPushButton::clicked, [=] {
             QMessageBox::information(this, tr("Information"),
                                      tr("After pressing OK, first move your joystick horizontally, "
                                         "and then vertically."));
@@ -311,8 +311,8 @@ ConfigureInputPlayer::ConfigureInputPlayer(QWidget* parent, std::size_t player_i
         });
     }
 
-    connect(ui->buttonClearAll, &QPushButton::released, [this] { ClearAll(); });
-    connect(ui->buttonRestoreDefaults, &QPushButton::released, [this] { RestoreDefaults(); });
+    connect(ui->buttonClearAll, &QPushButton::clicked, [this] { ClearAll(); });
+    connect(ui->buttonRestoreDefaults, &QPushButton::clicked, [this] { RestoreDefaults(); });
 
     timeout_timer->setSingleShot(true);
     connect(timeout_timer.get(), &QTimer::timeout, [this] { SetPollingResult({}, true); });
diff --git a/src/yuzu/configuration/configure_input_simple.cpp b/src/yuzu/configuration/configure_input_simple.cpp
index 864803ea3..ab3a11d30 100644
--- a/src/yuzu/configuration/configure_input_simple.cpp
+++ b/src/yuzu/configuration/configure_input_simple.cpp
@@ -9,7 +9,7 @@
 #include "yuzu/configuration/configure_input.h"
 #include "yuzu/configuration/configure_input_player.h"
 #include "yuzu/configuration/configure_input_simple.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace {
 
@@ -101,7 +101,7 @@ ConfigureInputSimple::ConfigureInputSimple(QWidget* parent)
 
     connect(ui->profile_combobox, QOverload<int>::of(&QComboBox::currentIndexChanged), this,
             &ConfigureInputSimple::OnSelectProfile);
-    connect(ui->profile_configure, &QPushButton::pressed, this, &ConfigureInputSimple::OnConfigure);
+    connect(ui->profile_configure, &QPushButton::clicked, this, &ConfigureInputSimple::OnConfigure);
 
     LoadConfiguration();
 }
diff --git a/src/yuzu/configuration/configure_mouse_advanced.cpp b/src/yuzu/configuration/configure_mouse_advanced.cpp
index b7305e653..0a4abe34f 100644
--- a/src/yuzu/configuration/configure_mouse_advanced.cpp
+++ b/src/yuzu/configuration/configure_mouse_advanced.cpp
@@ -83,7 +83,7 @@ ConfigureMouseAdvanced::ConfigureMouseAdvanced(QWidget* parent)
         }
 
         button->setContextMenuPolicy(Qt::CustomContextMenu);
-        connect(button, &QPushButton::released, [=] {
+        connect(button, &QPushButton::clicked, [=] {
             HandleClick(
                 button_map[button_id],
                 [=](const Common::ParamPackage& params) { buttons_param[button_id] = params; },
@@ -104,8 +104,8 @@ ConfigureMouseAdvanced::ConfigureMouseAdvanced(QWidget* parent)
         });
     }
 
-    connect(ui->buttonClearAll, &QPushButton::released, [this] { ClearAll(); });
-    connect(ui->buttonRestoreDefaults, &QPushButton::released, [this] { RestoreDefaults(); });
+    connect(ui->buttonClearAll, &QPushButton::clicked, [this] { ClearAll(); });
+    connect(ui->buttonRestoreDefaults, &QPushButton::clicked, [this] { RestoreDefaults(); });
 
     timeout_timer->setSingleShot(true);
     connect(timeout_timer.get(), &QTimer::timeout, [this] { SetPollingResult({}, true); });
diff --git a/src/yuzu/configuration/configure_per_general.cpp b/src/yuzu/configuration/configure_per_general.cpp
index 90336e235..d7f259f12 100644
--- a/src/yuzu/configuration/configure_per_general.cpp
+++ b/src/yuzu/configuration/configure_per_general.cpp
@@ -23,7 +23,7 @@
 #include "yuzu/configuration/config.h"
 #include "yuzu/configuration/configure_input.h"
 #include "yuzu/configuration/configure_per_general.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 #include "yuzu/util/util.h"
 
 ConfigurePerGameGeneral::ConfigurePerGameGeneral(QWidget* parent, u64 title_id)
diff --git a/src/yuzu/configuration/configure_profile_manager.cpp b/src/yuzu/configuration/configure_profile_manager.cpp
index c90f4cdd8..f53423440 100644
--- a/src/yuzu/configuration/configure_profile_manager.cpp
+++ b/src/yuzu/configuration/configure_profile_manager.cpp
@@ -108,10 +108,10 @@ ConfigureProfileManager ::ConfigureProfileManager(QWidget* parent)
 
     connect(tree_view, &QTreeView::clicked, this, &ConfigureProfileManager::SelectUser);
 
-    connect(ui->pm_add, &QPushButton::pressed, this, &ConfigureProfileManager::AddUser);
-    connect(ui->pm_rename, &QPushButton::pressed, this, &ConfigureProfileManager::RenameUser);
-    connect(ui->pm_remove, &QPushButton::pressed, this, &ConfigureProfileManager::DeleteUser);
-    connect(ui->pm_set_image, &QPushButton::pressed, this, &ConfigureProfileManager::SetUserImage);
+    connect(ui->pm_add, &QPushButton::clicked, this, &ConfigureProfileManager::AddUser);
+    connect(ui->pm_rename, &QPushButton::clicked, this, &ConfigureProfileManager::RenameUser);
+    connect(ui->pm_remove, &QPushButton::clicked, this, &ConfigureProfileManager::DeleteUser);
+    connect(ui->pm_set_image, &QPushButton::clicked, this, &ConfigureProfileManager::SetUserImage);
 
     scene = new QGraphicsScene;
     ui->current_user_icon->setScene(scene);
diff --git a/src/yuzu/configuration/configure_touchscreen_advanced.cpp b/src/yuzu/configuration/configure_touchscreen_advanced.cpp
index 8ced28c75..7d7cc00b7 100644
--- a/src/yuzu/configuration/configure_touchscreen_advanced.cpp
+++ b/src/yuzu/configuration/configure_touchscreen_advanced.cpp
@@ -11,7 +11,7 @@ ConfigureTouchscreenAdvanced::ConfigureTouchscreenAdvanced(QWidget* parent)
     : QDialog(parent), ui(std::make_unique<Ui::ConfigureTouchscreenAdvanced>()) {
     ui->setupUi(this);
 
-    connect(ui->restore_defaults_button, &QPushButton::pressed, this,
+    connect(ui->restore_defaults_button, &QPushButton::clicked, this,
             &ConfigureTouchscreenAdvanced::RestoreDefaults);
 
     LoadConfiguration();
diff --git a/src/yuzu/configuration/configure_web.cpp b/src/yuzu/configuration/configure_web.cpp
index 5a70ef168..336b062b3 100644
--- a/src/yuzu/configuration/configure_web.cpp
+++ b/src/yuzu/configuration/configure_web.cpp
@@ -9,7 +9,7 @@
 #include "core/telemetry_session.h"
 #include "ui_configure_web.h"
 #include "yuzu/configuration/configure_web.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 ConfigureWeb::ConfigureWeb(QWidget* parent)
     : QWidget(parent), ui(std::make_unique<Ui::ConfigureWeb>()) {
diff --git a/src/yuzu/debugger/console.cpp b/src/yuzu/debugger/console.cpp
index 320898f6a..207ff4d58 100644
--- a/src/yuzu/debugger/console.cpp
+++ b/src/yuzu/debugger/console.cpp
@@ -10,7 +10,7 @@
 
 #include "common/logging/backend.h"
 #include "yuzu/debugger/console.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace Debugger {
 void ToggleConsole() {
diff --git a/src/yuzu/discord_impl.cpp b/src/yuzu/discord_impl.cpp
index 9d87a41eb..ea0079353 100644
--- a/src/yuzu/discord_impl.cpp
+++ b/src/yuzu/discord_impl.cpp
@@ -9,7 +9,7 @@
 #include "core/core.h"
 #include "core/loader/loader.h"
 #include "yuzu/discord_impl.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace DiscordRPC {
 
diff --git a/src/yuzu/game_list.cpp b/src/yuzu/game_list.cpp
index 1885587af..d18b96519 100644
--- a/src/yuzu/game_list.cpp
+++ b/src/yuzu/game_list.cpp
@@ -23,7 +23,7 @@
 #include "yuzu/game_list_p.h"
 #include "yuzu/game_list_worker.h"
 #include "yuzu/main.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 GameListSearchField::KeyReleaseEater::KeyReleaseEater(GameList* gamelist) : gamelist{gamelist} {}
 
diff --git a/src/yuzu/game_list_p.h b/src/yuzu/game_list_p.h
index 0b458ef48..ece534dd6 100644
--- a/src/yuzu/game_list_p.h
+++ b/src/yuzu/game_list_p.h
@@ -19,7 +19,7 @@
 #include "common/common_types.h"
 #include "common/logging/log.h"
 #include "common/string_util.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 #include "yuzu/util/util.h"
 
 /**
diff --git a/src/yuzu/game_list_worker.cpp b/src/yuzu/game_list_worker.cpp
index 4f30e9147..77f358630 100644
--- a/src/yuzu/game_list_worker.cpp
+++ b/src/yuzu/game_list_worker.cpp
@@ -29,7 +29,7 @@
 #include "yuzu/game_list.h"
 #include "yuzu/game_list_p.h"
 #include "yuzu/game_list_worker.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace {
 
diff --git a/src/yuzu/hotkeys.cpp b/src/yuzu/hotkeys.cpp
index 4582e7f21..d4e97fa16 100644
--- a/src/yuzu/hotkeys.cpp
+++ b/src/yuzu/hotkeys.cpp
@@ -7,7 +7,7 @@
 #include <QTreeWidgetItem>
 #include <QtGlobal>
 #include "yuzu/hotkeys.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 HotkeyRegistry::HotkeyRegistry() = default;
 HotkeyRegistry::~HotkeyRegistry() = default;
diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp
index ae21f4753..ac57229d5 100644
--- a/src/yuzu/main.cpp
+++ b/src/yuzu/main.cpp
@@ -100,7 +100,7 @@ static FileSys::VirtualFile VfsDirectoryCreateFileWrapper(const FileSys::Virtual
 #include "yuzu/hotkeys.h"
 #include "yuzu/loading_screen.h"
 #include "yuzu/main.h"
-#include "yuzu/ui_settings.h"
+#include "yuzu/uisettings.h"
 
 #ifdef USE_DISCORD_PRESENCE
 #include "yuzu/discord_impl.h"
@@ -119,6 +119,7 @@ Q_IMPORT_PLUGIN(QWindowsIntegrationPlugin);
 #endif
 
 #ifdef _WIN32
+#include <windows.h>
 extern "C" {
 // tells Nvidia and AMD drivers to use the dedicated GPU by default on laptops with switchable
 // graphics
@@ -747,6 +748,18 @@ void GMainWindow::OnDisplayTitleBars(bool show) {
     }
 }
 
+void GMainWindow::PreventOSSleep() {
+#ifdef _WIN32
+    SetThreadExecutionState(ES_CONTINUOUS | ES_SYSTEM_REQUIRED | ES_DISPLAY_REQUIRED);
+#endif
+}
+
+void GMainWindow::AllowOSSleep() {
+#ifdef _WIN32
+    SetThreadExecutionState(ES_CONTINUOUS);
+#endif
+}
+
 QStringList GMainWindow::GetUnsupportedGLExtensions() {
     QStringList unsupported_ext;
 
@@ -966,6 +979,8 @@ void GMainWindow::BootGame(const QString& filename) {
 }
 
 void GMainWindow::ShutdownGame() {
+    AllowOSSleep();
+
     discord_rpc->Pause();
     emu_thread->RequestStop();
 
@@ -1567,6 +1582,8 @@ void GMainWindow::OnMenuRecentFile() {
 }
 
 void GMainWindow::OnStartGame() {
+    PreventOSSleep();
+
     emu_thread->SetRunning(true);
 
     qRegisterMetaType<Core::Frontend::SoftwareKeyboardParameters>(
@@ -1598,6 +1615,8 @@ void GMainWindow::OnPauseGame() {
     ui.action_Pause->setEnabled(false);
     ui.action_Stop->setEnabled(true);
     ui.action_Capture_Screenshot->setEnabled(false);
+
+    AllowOSSleep();
 }
 
 void GMainWindow::OnStopGame() {
@@ -1843,13 +1862,14 @@ void GMainWindow::OnCoreError(Core::System::ResultStatus result, std::string det
            "data, or other bugs.");
     switch (result) {
     case Core::System::ResultStatus::ErrorSystemFiles: {
-        QString message = tr("yuzu was unable to locate a Switch system archive");
-        if (!details.empty()) {
-            message.append(tr(": %1. ").arg(QString::fromStdString(details)));
+        QString message;
+        if (details.empty()) {
+            message =
+                tr("yuzu was unable to locate a Switch system archive. %1").arg(common_message);
         } else {
-            message.append(tr(". "));
+            message = tr("yuzu was unable to locate a Switch system archive: %1. %2")
+                          .arg(QString::fromStdString(details), common_message);
         }
-        message.append(common_message);
 
         answer = QMessageBox::question(this, tr("System Archive Not Found"), message,
                                        QMessageBox::Yes | QMessageBox::No, QMessageBox::No);
@@ -1858,8 +1878,8 @@ void GMainWindow::OnCoreError(Core::System::ResultStatus result, std::string det
     }
 
     case Core::System::ResultStatus::ErrorSharedFont: {
-        QString message = tr("yuzu was unable to locate the Switch shared fonts. ");
-        message.append(common_message);
+        const QString message =
+            tr("yuzu was unable to locate the Switch shared fonts. %1").arg(common_message);
         answer = QMessageBox::question(this, tr("Shared Fonts Not Found"), message,
                                        QMessageBox::Yes | QMessageBox::No, QMessageBox::No);
         status_message = tr("Shared Font Missing");
diff --git a/src/yuzu/main.h b/src/yuzu/main.h
index 1137bbc7a..501608ddc 100644
--- a/src/yuzu/main.h
+++ b/src/yuzu/main.h
@@ -130,6 +130,9 @@ private:
     void ConnectWidgetEvents();
     void ConnectMenuEvents();
 
+    void PreventOSSleep();
+    void AllowOSSleep();
+
     QStringList GetUnsupportedGLExtensions();
     bool LoadROM(const QString& filename);
     void BootGame(const QString& filename);
diff --git a/src/yuzu/ui_settings.cpp b/src/yuzu/uisettings.cpp
index 4bdc302e0..7f7d247a3 100644
--- a/src/yuzu/ui_settings.cpp
+++ b/src/yuzu/uisettings.cpp
@@ -2,7 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include "ui_settings.h"
+#include "yuzu/uisettings.h"
 
 namespace UISettings {
 
diff --git a/src/yuzu/ui_settings.h b/src/yuzu/uisettings.h
index a62cd6911..a62cd6911 100644
--- a/src/yuzu/ui_settings.h
+++ b/src/yuzu/uisettings.h
diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp
index 30b22341b..067d58d80 100644
--- a/src/yuzu_cmd/config.cpp
+++ b/src/yuzu_cmd/config.cpp
@@ -340,7 +340,6 @@ void Config::ReadValues() {
     }
 
     // Core
-    Settings::values.cpu_jit_enabled = sdl2_config->GetBoolean("Core", "cpu_jit_enabled", true);
     Settings::values.use_multi_core = sdl2_config->GetBoolean("Core", "use_multi_core", false);
 
     // Renderer
diff --git a/src/yuzu_cmd/default_ini.h b/src/yuzu_cmd/default_ini.h
index 4f1add434..0cfc111a6 100644
--- a/src/yuzu_cmd/default_ini.h
+++ b/src/yuzu_cmd/default_ini.h
@@ -76,10 +76,6 @@ motion_device=
 touch_device=
 
 [Core]
-# Whether to use the Just-In-Time (JIT) compiler for CPU emulation
-# 0: Interpreter (slow), 1 (default): JIT (fast)
-cpu_jit_enabled =
-
 # Whether to use multi-core for CPU emulation
 # 0 (default): Disabled, 1: Enabled
 use_multi_core=
diff --git a/src/yuzu_tester/config.cpp b/src/yuzu_tester/config.cpp
index b96b7d279..9a11dc6c3 100644
--- a/src/yuzu_tester/config.cpp
+++ b/src/yuzu_tester/config.cpp
@@ -114,7 +114,6 @@ void Config::ReadValues() {
     }
 
     // Core
-    Settings::values.cpu_jit_enabled = sdl2_config->GetBoolean("Core", "cpu_jit_enabled", true);
     Settings::values.use_multi_core = sdl2_config->GetBoolean("Core", "use_multi_core", false);
 
     // Renderer
diff --git a/src/yuzu_tester/default_ini.h b/src/yuzu_tester/default_ini.h
index 0f880d8c7..9a3e86d68 100644
--- a/src/yuzu_tester/default_ini.h
+++ b/src/yuzu_tester/default_ini.h
@@ -8,10 +8,6 @@ namespace DefaultINI {
 
 const char* sdl2_config_file = R"(
 [Core]
-# Whether to use the Just-In-Time (JIT) compiler for CPU emulation
-# 0: Interpreter (slow), 1 (default): JIT (fast)
-cpu_jit_enabled =
-
 # Whether to use multi-core for CPU emulation
 # 0 (default): Disabled, 1: Enabled
 use_multi_core=
diff --git a/src/yuzu_tester/yuzu.cpp b/src/yuzu_tester/yuzu.cpp
index b589c3de3..0ee97aa54 100644
--- a/src/yuzu_tester/yuzu.cpp
+++ b/src/yuzu_tester/yuzu.cpp
@@ -92,7 +92,6 @@ int main(int argc, char** argv) {
 
     int option_index = 0;
 
-    char* endarg;
 #ifdef _WIN32
     int argc_w;
     auto argv_w = CommandLineToArgvW(GetCommandLineW(), &argc_w);
@@ -226,7 +225,7 @@ int main(int argc, char** argv) {
 
     switch (load_result) {
     case Core::System::ResultStatus::ErrorGetLoader:
-        LOG_CRITICAL(Frontend, "Failed to obtain loader for %s!", filepath.c_str());
+        LOG_CRITICAL(Frontend, "Failed to obtain loader for {}!", filepath);
         return -1;
     case Core::System::ResultStatus::ErrorLoader:
         LOG_CRITICAL(Frontend, "Failed to load ROM!");