69 files changed, 2359 insertions, 1166 deletions

diff --git a/externals/CMakeLists.txt b/externals/CMakeLists.txt
index df7a5e0a9..9be5b2780 100644
--- a/externals/CMakeLists.txt
+++ b/externals/CMakeLists.txt
@@ -4,6 +4,13 @@ list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/CMakeModules")
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/externals/find-modules")
 include(DownloadExternals)
 
+# xbyak
+if (ARCHITECTURE_x86 OR ARCHITECTURE_x86_64)
+    add_library(xbyak INTERFACE)
+    target_include_directories(xbyak SYSTEM INTERFACE ./xbyak/xbyak)
+    target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
+endif()
+
 # Catch
 add_library(catch-single-include INTERFACE)
 target_include_directories(catch-single-include INTERFACE catch/single_include)
@@ -75,11 +82,3 @@ if (ENABLE_WEB_SERVICE)
     target_compile_definitions(httplib INTERFACE -DCPPHTTPLIB_OPENSSL_SUPPORT)
     target_link_libraries(httplib INTERFACE OpenSSL::SSL OpenSSL::Crypto)
 endif()
-
-if (NOT TARGET xbyak)
-    if (ARCHITECTURE_x86 OR ARCHITECTURE_x86_64)
-        add_library(xbyak INTERFACE)
-        target_include_directories(xbyak SYSTEM INTERFACE ./xbyak/xbyak)
-        target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
-    endif()
-endif()
diff --git a/externals/sirit b/externals/sirit
-Subproject a62c5bbc100a5e5a31ea0ccc4a78d8fa6a4167c
+Subproject eefca56afd49379bdebc97ded8b480839f93088
diff --git a/src/core/hle/kernel/process.cpp b/src/core/hle/kernel/process.cpp
index 36724569f..c4c5199b1 100644
--- a/src/core/hle/kernel/process.cpp
+++ b/src/core/hle/kernel/process.cpp
@@ -132,7 +132,8 @@ std::shared_ptr<ResourceLimit> Process::GetResourceLimit() const {
 
 u64 Process::GetTotalPhysicalMemoryAvailable() const {
     const u64 capacity{resource_limit->GetCurrentResourceValue(ResourceType::PhysicalMemory) +
-                       page_table->GetTotalHeapSize() + image_size + main_thread_stack_size};
+                       page_table->GetTotalHeapSize() + GetSystemResourceSize() + image_size +
+                       main_thread_stack_size};
 
     if (capacity < memory_usage_capacity) {
         return capacity;
@@ -146,7 +147,8 @@ u64 Process::GetTotalPhysicalMemoryAvailableWithoutSystemResource() const {
 }
 
 u64 Process::GetTotalPhysicalMemoryUsed() const {
-    return image_size + main_thread_stack_size + page_table->GetTotalHeapSize();
+    return image_size + main_thread_stack_size + page_table->GetTotalHeapSize() +
+           GetSystemResourceSize();
 }
 
 u64 Process::GetTotalPhysicalMemoryUsedWithoutSystemResource() const {
diff --git a/src/core/hle/kernel/readable_event.cpp b/src/core/hle/kernel/readable_event.cpp
index 00860fcbd..ef5e19e63 100644
--- a/src/core/hle/kernel/readable_event.cpp
+++ b/src/core/hle/kernel/readable_event.cpp
@@ -38,7 +38,7 @@ void ReadableEvent::Clear() {
 
 ResultCode ReadableEvent::Reset() {
     if (!is_signaled) {
-        LOG_ERROR(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}",
+        LOG_TRACE(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}",
                   GetObjectId(), GetTypeName(), GetName());
         return ERR_INVALID_STATE;
     }
diff --git a/src/core/hle/kernel/resource_limit.cpp b/src/core/hle/kernel/resource_limit.cpp
index d9beaa3a4..212e442f4 100644
--- a/src/core/hle/kernel/resource_limit.cpp
+++ b/src/core/hle/kernel/resource_limit.cpp
@@ -24,13 +24,9 @@ bool ResourceLimit::Reserve(ResourceType resource, s64 amount, u64 timeout) {
     const std::size_t index{ResourceTypeToIndex(resource)};
 
     s64 new_value = current[index] + amount;
-    while (new_value > limit[index] && available[index] + amount <= limit[index]) {
+    if (new_value > limit[index] && available[index] + amount <= limit[index]) {
         // TODO(bunnei): This is wrong for multicore, we should wait the calling thread for timeout
         new_value = current[index] + amount;
-
-        if (timeout >= 0) {
-            break;
-        }
     }
 
     if (new_value <= limit[index]) {
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 cc2192e5c..0d913334e 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp
@@ -25,7 +25,7 @@ u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input,
     case IoctlCommand::IocGetCharacteristicsCommand:
         return GetCharacteristics(input, output, output2, version);
     case IoctlCommand::IocGetTPCMasksCommand:
-        return GetTPCMasks(input, output);
+        return GetTPCMasks(input, output, output2, version);
     case IoctlCommand::IocGetActiveSlotMaskCommand:
         return GetActiveSlotMask(input, output);
     case IoctlCommand::IocZcullGetCtxSizeCommand:
@@ -98,17 +98,22 @@ u32 nvhost_ctrl_gpu::GetCharacteristics(const std::vector<u8>& input, std::vecto
     return 0;
 }
 
-u32 nvhost_ctrl_gpu::GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl_gpu::GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output,
+                                 std::vector<u8>& output2, IoctlVersion version) {
     IoctlGpuGetTpcMasksArgs params{};
     std::memcpy(&params, input.data(), input.size());
-    LOG_INFO(Service_NVDRV, "called, mask=0x{:X}, mask_buf_addr=0x{:X}", params.mask_buf_size,
-             params.mask_buf_addr);
-    // TODO(ogniK): Confirm value on hardware
-    if (params.mask_buf_size)
-        params.tpc_mask_size = 4 * 1; // 4 * num_gpc
-    else
-        params.tpc_mask_size = 0;
-    std::memcpy(output.data(), &params, sizeof(params));
+    LOG_DEBUG(Service_NVDRV, "called, mask_buffer_size=0x{:X}", params.mask_buffer_size);
+    if (params.mask_buffer_size != 0) {
+        params.tcp_mask = 3;
+    }
+
+    if (version == IoctlVersion::Version3) {
+        std::memcpy(output.data(), input.data(), output.size());
+        std::memcpy(output2.data(), &params.tcp_mask, output2.size());
+    } else {
+        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 07b644ec5..ef60f72ce 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h
@@ -92,16 +92,11 @@ private:
                   "IoctlCharacteristics is incorrect size");
 
     struct IoctlGpuGetTpcMasksArgs {
-        /// [in]  TPC mask buffer size reserved by userspace. Should be at least
-        /// sizeof(__u32) * fls(gpc_mask) to receive TPC mask for each GPC.
-        /// [out] full kernel buffer size
-        u32_le mask_buf_size;
-        u32_le reserved;
-
-        /// [in]  pointer to TPC mask buffer. It will receive one 32-bit TPC mask per GPC or 0 if
-        /// GPC is not enabled or not present. This parameter is ignored if mask_buf_size is 0.
-        u64_le mask_buf_addr;
-        u64_le tpc_mask_size; // Nintendo add this?
+        u32_le mask_buffer_size{};
+        INSERT_PADDING_WORDS(1);
+        u64_le mask_buffer_address{};
+        u32_le tcp_mask{};
+        INSERT_PADDING_WORDS(1);
     };
     static_assert(sizeof(IoctlGpuGetTpcMasksArgs) == 24,
                   "IoctlGpuGetTpcMasksArgs is incorrect size");
@@ -166,7 +161,8 @@ private:
 
     u32 GetCharacteristics(const std::vector<u8>& input, std::vector<u8>& output,
                            std::vector<u8>& output2, IoctlVersion version);
-    u32 GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 GetTPCMasks(const std::vector<u8>& input, std::vector<u8>& output, std::vector<u8>& output2,
+                    IoctlVersion version);
     u32 GetActiveSlotMask(const std::vector<u8>& input, std::vector<u8>& output);
     u32 ZCullGetCtxSize(const std::vector<u8>& input, std::vector<u8>& output);
     u32 ZCullGetInfo(const std::vector<u8>& input, std::vector<u8>& output);
diff --git a/src/core/settings.h b/src/core/settings.h
index 9d916d5cb..33e1e06cd 100644
--- a/src/core/settings.h
+++ b/src/core/settings.h
@@ -474,6 +474,7 @@ struct Values {
     bool reporting_services;
     bool quest_flag;
     bool disable_cpu_opt;
+    bool disable_macro_jit;
 
     // BCAT
     std::string bcat_backend;
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index d6ee82836..39d5d8401 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -25,6 +25,12 @@ add_library(video_core STATIC
     engines/shader_bytecode.h
     engines/shader_header.h
     engines/shader_type.h
+    macro/macro.cpp
+    macro/macro.h
+    macro/macro_interpreter.cpp
+    macro/macro_interpreter.h
+    macro/macro_jit_x64.cpp
+    macro/macro_jit_x64.h
     fence_manager.h
     gpu.cpp
     gpu.h
@@ -36,8 +42,6 @@ add_library(video_core STATIC
     gpu_thread.h
     guest_driver.cpp
     guest_driver.h
-    macro_interpreter.cpp
-    macro_interpreter.h
     memory_manager.cpp
     memory_manager.h
     morton.cpp
@@ -45,8 +49,6 @@ add_library(video_core STATIC
     query_cache.h
     rasterizer_accelerated.cpp
     rasterizer_accelerated.h
-    rasterizer_cache.cpp
-    rasterizer_cache.h
     rasterizer_interface.h
     renderer_base.cpp
     renderer_base.h
@@ -89,6 +91,7 @@ add_library(video_core STATIC
     renderer_opengl/utils.h
     sampler_cache.cpp
     sampler_cache.h
+    shader_cache.h
     shader/decode/arithmetic.cpp
     shader/decode/arithmetic_immediate.cpp
     shader/decode/bfe.cpp
diff --git a/src/video_core/buffer_cache/buffer_block.h b/src/video_core/buffer_cache/buffer_block.h
index e35ee0b67..e64170e66 100644
--- a/src/video_core/buffer_cache/buffer_block.h
+++ b/src/video_core/buffer_cache/buffer_block.h
@@ -15,48 +15,47 @@ namespace VideoCommon {
 
 class BufferBlock {
 public:
-    bool Overlaps(const VAddr start, const VAddr end) const {
+    bool Overlaps(VAddr start, VAddr end) const {
         return (cpu_addr < end) && (cpu_addr_end > start);
     }
 
-    bool IsInside(const VAddr other_start, const VAddr other_end) const {
+    bool IsInside(VAddr other_start, VAddr other_end) const {
         return cpu_addr <= other_start && other_end <= cpu_addr_end;
     }
 
-    std::size_t GetOffset(const VAddr in_addr) {
+    std::size_t Offset(VAddr in_addr) const {
         return static_cast<std::size_t>(in_addr - cpu_addr);
     }
 
-    VAddr GetCpuAddr() const {
+    VAddr CpuAddr() const {
         return cpu_addr;
     }
 
-    VAddr GetCpuAddrEnd() const {
+    VAddr CpuAddrEnd() const {
         return cpu_addr_end;
     }
 
-    void SetCpuAddr(const VAddr new_addr) {
+    void SetCpuAddr(VAddr new_addr) {
         cpu_addr = new_addr;
         cpu_addr_end = new_addr + size;
     }
 
-    std::size_t GetSize() const {
+    std::size_t Size() const {
         return size;
     }
 
-    void SetEpoch(u64 new_epoch) {
-        epoch = new_epoch;
+    u64 Epoch() const {
+        return epoch;
     }
 
-    u64 GetEpoch() {
-        return epoch;
+    void SetEpoch(u64 new_epoch) {
+        epoch = new_epoch;
     }
 
 protected:
-    explicit BufferBlock(VAddr cpu_addr, const std::size_t size) : size{size} {
-        SetCpuAddr(cpu_addr);
+    explicit BufferBlock(VAddr cpu_addr_, std::size_t size_) : size{size_} {
+        SetCpuAddr(cpu_addr_);
     }
-    ~BufferBlock() = default;
 
 private:
     VAddr cpu_addr{};
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index b88fce2cd..308d8b55f 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -30,12 +30,16 @@
 
 namespace VideoCommon {
 
-template <typename OwnerBuffer, typename BufferType, typename StreamBuffer>
+template <typename Buffer, typename BufferType, typename StreamBuffer>
 class BufferCache {
     using IntervalSet = boost::icl::interval_set<VAddr>;
     using IntervalType = typename IntervalSet::interval_type;
     using VectorMapInterval = boost::container::small_vector<MapInterval*, 1>;
 
+    static constexpr u64 WRITE_PAGE_BIT = 11;
+    static constexpr u64 BLOCK_PAGE_BITS = 21;
+    static constexpr u64 BLOCK_PAGE_SIZE = 1ULL << BLOCK_PAGE_BITS;
+
 public:
     using BufferInfo = std::pair<BufferType, u64>;
 
@@ -82,7 +86,7 @@ public:
             }
         }
 
-        OwnerBuffer block = GetBlock(cpu_addr, size);
+        Buffer* const block = GetBlock(cpu_addr, size);
         MapInterval* const map = MapAddress(block, gpu_addr, cpu_addr, size);
         if (!map) {
             return {GetEmptyBuffer(size), 0};
@@ -98,7 +102,7 @@ public:
             }
         }
 
-        return {ToHandle(block), static_cast<u64>(block->GetOffset(cpu_addr))};
+        return {block->Handle(), static_cast<u64>(block->Offset(cpu_addr))};
     }
 
     /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
@@ -110,31 +114,37 @@ public:
         });
     }
 
-    void Map(std::size_t max_size) {
+    /// Prepares the buffer cache for data uploading
+    /// @param max_size Maximum number of bytes that will be uploaded
+    /// @return True when a stream buffer invalidation was required, false otherwise
+    bool Map(std::size_t max_size) {
         std::lock_guard lock{mutex};
 
+        bool invalidated;
         std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
         buffer_offset = buffer_offset_base;
+
+        return invalidated;
     }
 
-    /// Finishes the upload stream, returns true on bindings invalidation.
-    bool Unmap() {
+    /// Finishes the upload stream
+    void Unmap() {
         std::lock_guard lock{mutex};
-
         stream_buffer->Unmap(buffer_offset - buffer_offset_base);
-        return std::exchange(invalidated, false);
     }
 
+    /// Function called at the end of each frame, inteded for deferred operations
     void TickFrame() {
         ++epoch;
+
         while (!pending_destruction.empty()) {
             // Delay at least 4 frames before destruction.
             // This is due to triple buffering happening on some drivers.
             static constexpr u64 epochs_to_destroy = 5;
-            if (pending_destruction.front()->GetEpoch() + epochs_to_destroy > epoch) {
+            if (pending_destruction.front()->Epoch() + epochs_to_destroy > epoch) {
                 break;
             }
-            pending_destruction.pop_front();
+            pending_destruction.pop();
         }
     }
 
@@ -249,23 +259,21 @@ public:
 
 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()} {}
+                         std::unique_ptr<StreamBuffer> stream_buffer_)
+        : rasterizer{rasterizer}, system{system}, stream_buffer{std::move(stream_buffer_)},
+          stream_buffer_handle{stream_buffer->Handle()} {}
 
     ~BufferCache() = default;
 
-    virtual BufferType ToHandle(const OwnerBuffer& storage) = 0;
+    virtual std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) = 0;
 
-    virtual OwnerBuffer CreateBlock(VAddr cpu_addr, std::size_t size) = 0;
-
-    virtual void UploadBlockData(const OwnerBuffer& buffer, std::size_t offset, std::size_t size,
+    virtual void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                                  const u8* data) = 0;
 
-    virtual void DownloadBlockData(const OwnerBuffer& buffer, std::size_t offset, std::size_t size,
+    virtual void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                                    u8* data) = 0;
 
-    virtual void CopyBlock(const OwnerBuffer& src, const OwnerBuffer& dst, std::size_t src_offset,
+    virtual void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
                            std::size_t dst_offset, std::size_t size) = 0;
 
     virtual BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) {
@@ -321,7 +329,7 @@ protected:
     }
 
 private:
-    MapInterval* MapAddress(const OwnerBuffer& block, GPUVAddr gpu_addr, VAddr cpu_addr,
+    MapInterval* MapAddress(const Buffer* block, GPUVAddr gpu_addr, VAddr cpu_addr,
                             std::size_t size) {
         const VectorMapInterval overlaps = GetMapsInRange(cpu_addr, size);
         if (overlaps.empty()) {
@@ -329,11 +337,11 @@ private:
             const VAddr cpu_addr_end = cpu_addr + size;
             if (memory_manager.IsGranularRange(gpu_addr, size)) {
                 u8* host_ptr = memory_manager.GetPointer(gpu_addr);
-                UploadBlockData(block, block->GetOffset(cpu_addr), size, host_ptr);
+                UploadBlockData(*block, block->Offset(cpu_addr), size, host_ptr);
             } else {
                 staging_buffer.resize(size);
                 memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size);
-                UploadBlockData(block, block->GetOffset(cpu_addr), size, staging_buffer.data());
+                UploadBlockData(*block, block->Offset(cpu_addr), size, staging_buffer.data());
             }
             return Register(MapInterval(cpu_addr, cpu_addr_end, gpu_addr));
         }
@@ -376,7 +384,7 @@ private:
         return map;
     }
 
-    void UpdateBlock(const OwnerBuffer& block, VAddr start, VAddr end,
+    void UpdateBlock(const Buffer* block, VAddr start, VAddr end,
                      const VectorMapInterval& overlaps) {
         const IntervalType base_interval{start, end};
         IntervalSet interval_set{};
@@ -386,13 +394,13 @@ private:
             interval_set.subtract(subtract);
         }
         for (auto& interval : interval_set) {
-            std::size_t size = interval.upper() - interval.lower();
-            if (size > 0) {
-                staging_buffer.resize(size);
-                system.Memory().ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size);
-                UploadBlockData(block, block->GetOffset(interval.lower()), size,
-                                staging_buffer.data());
+            const std::size_t size = interval.upper() - interval.lower();
+            if (size == 0) {
+                continue;
             }
+            staging_buffer.resize(size);
+            system.Memory().ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size);
+            UploadBlockData(*block, block->Offset(interval.lower()), size, staging_buffer.data());
         }
     }
 
@@ -422,10 +430,14 @@ private:
     }
 
     void FlushMap(MapInterval* map) {
+        const auto it = blocks.find(map->start >> BLOCK_PAGE_BITS);
+        ASSERT_OR_EXECUTE(it != blocks.end(), return;);
+
+        std::shared_ptr<Buffer> block = it->second;
+
         const std::size_t size = map->end - map->start;
-        OwnerBuffer block = blocks[map->start >> block_page_bits];
         staging_buffer.resize(size);
-        DownloadBlockData(block, block->GetOffset(map->start), size, staging_buffer.data());
+        DownloadBlockData(*block, block->Offset(map->start), size, staging_buffer.data());
         system.Memory().WriteBlockUnsafe(map->start, staging_buffer.data(), size);
         map->MarkAsModified(false, 0);
     }
@@ -448,97 +460,89 @@ private:
         buffer_offset = offset_aligned;
     }
 
-    OwnerBuffer EnlargeBlock(OwnerBuffer buffer) {
-        const std::size_t old_size = buffer->GetSize();
-        const std::size_t new_size = old_size + block_page_size;
-        const VAddr cpu_addr = buffer->GetCpuAddr();
-        OwnerBuffer new_buffer = CreateBlock(cpu_addr, new_size);
-        CopyBlock(buffer, new_buffer, 0, 0, old_size);
-        buffer->SetEpoch(epoch);
-        pending_destruction.push_back(buffer);
+    std::shared_ptr<Buffer> EnlargeBlock(std::shared_ptr<Buffer> buffer) {
+        const std::size_t old_size = buffer->Size();
+        const std::size_t new_size = old_size + BLOCK_PAGE_SIZE;
+        const VAddr cpu_addr = buffer->CpuAddr();
+        std::shared_ptr<Buffer> new_buffer = CreateBlock(cpu_addr, new_size);
+        CopyBlock(*buffer, *new_buffer, 0, 0, old_size);
+        QueueDestruction(std::move(buffer));
+
         const VAddr cpu_addr_end = cpu_addr + new_size - 1;
-        u64 page_start = cpu_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
-            blocks[page_start] = new_buffer;
-            ++page_start;
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
+            blocks.insert_or_assign(page_start, new_buffer);
         }
+
         return new_buffer;
     }
 
-    OwnerBuffer MergeBlocks(OwnerBuffer first, OwnerBuffer second) {
-        const std::size_t size_1 = first->GetSize();
-        const std::size_t size_2 = second->GetSize();
-        const VAddr first_addr = first->GetCpuAddr();
-        const VAddr second_addr = second->GetCpuAddr();
+    std::shared_ptr<Buffer> MergeBlocks(std::shared_ptr<Buffer> first,
+                                        std::shared_ptr<Buffer> second) {
+        const std::size_t size_1 = first->Size();
+        const std::size_t size_2 = second->Size();
+        const VAddr first_addr = first->CpuAddr();
+        const VAddr second_addr = second->CpuAddr();
         const VAddr new_addr = std::min(first_addr, second_addr);
         const std::size_t new_size = size_1 + size_2;
-        OwnerBuffer 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);
+
+        std::shared_ptr<Buffer> new_buffer = CreateBlock(new_addr, new_size);
+        CopyBlock(*first, *new_buffer, 0, new_buffer->Offset(first_addr), size_1);
+        CopyBlock(*second, *new_buffer, 0, new_buffer->Offset(second_addr), size_2);
+        QueueDestruction(std::move(first));
+        QueueDestruction(std::move(second));
+
         const VAddr cpu_addr_end = new_addr + new_size - 1;
-        u64 page_start = new_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
-            blocks[page_start] = new_buffer;
-            ++page_start;
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = new_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
+            blocks.insert_or_assign(page_start, new_buffer);
         }
         return new_buffer;
     }
 
-    OwnerBuffer GetBlock(const VAddr cpu_addr, const std::size_t size) {
-        OwnerBuffer found;
+    Buffer* GetBlock(VAddr cpu_addr, std::size_t size) {
+        std::shared_ptr<Buffer> found;
+
         const VAddr cpu_addr_end = cpu_addr + size - 1;
-        u64 page_start = cpu_addr >> block_page_bits;
-        const u64 page_end = cpu_addr_end >> block_page_bits;
-        while (page_start <= page_end) {
+        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
+        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
             auto it = blocks.find(page_start);
             if (it == blocks.end()) {
                 if (found) {
                     found = EnlargeBlock(found);
-                } else {
-                    const VAddr 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;
+                    continue;
                 }
+                const VAddr start_addr = page_start << BLOCK_PAGE_BITS;
+                found = CreateBlock(start_addr, BLOCK_PAGE_SIZE);
+                blocks.insert_or_assign(page_start, found);
+                continue;
+            }
+            if (!found) {
+                found = it->second;
+                continue;
+            }
+            if (found != it->second) {
+                found = MergeBlocks(std::move(found), it->second);
             }
-            ++page_start;
         }
-        return found;
+        return found.get();
     }
 
-    void MarkRegionAsWritten(const VAddr start, const VAddr end) {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    void MarkRegionAsWritten(VAddr start, VAddr end) {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             auto it = written_pages.find(page_start);
             if (it != written_pages.end()) {
                 it->second = it->second + 1;
             } else {
-                written_pages[page_start] = 1;
+                written_pages.insert_or_assign(page_start, 1);
             }
-            ++page_start;
         }
     }
 
-    void UnmarkRegionAsWritten(const VAddr start, const VAddr end) {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    void UnmarkRegionAsWritten(VAddr start, VAddr end) {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             auto it = written_pages.find(page_start);
             if (it != written_pages.end()) {
                 if (it->second > 1) {
@@ -547,22 +551,24 @@ private:
                     written_pages.erase(it);
                 }
             }
-            ++page_start;
         }
     }
 
-    bool IsRegionWritten(const VAddr start, const VAddr end) const {
-        u64 page_start = start >> write_page_bit;
-        const u64 page_end = end >> write_page_bit;
-        while (page_start <= page_end) {
+    bool IsRegionWritten(VAddr start, VAddr end) const {
+        const u64 page_end = end >> WRITE_PAGE_BIT;
+        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
             if (written_pages.count(page_start) > 0) {
                 return true;
             }
-            ++page_start;
         }
         return false;
     }
 
+    void QueueDestruction(std::shared_ptr<Buffer> buffer) {
+        buffer->SetEpoch(epoch);
+        pending_destruction.push(std::move(buffer));
+    }
+
     void MarkForAsyncFlush(MapInterval* map) {
         if (!uncommitted_flushes) {
             uncommitted_flushes = std::make_shared<std::unordered_set<MapInterval*>>();
@@ -574,9 +580,7 @@ private:
     Core::System& system;
 
     std::unique_ptr<StreamBuffer> stream_buffer;
-    BufferType stream_buffer_handle{};
-
-    bool invalidated = false;
+    BufferType stream_buffer_handle;
 
     u8* buffer_ptr = nullptr;
     u64 buffer_offset = 0;
@@ -586,18 +590,15 @@ private:
     boost::intrusive::set<MapInterval, boost::intrusive::compare<MapIntervalCompare>>
         mapped_addresses;
 
-    static constexpr u64 write_page_bit = 11;
     std::unordered_map<u64, u32> written_pages;
+    std::unordered_map<u64, std::shared_ptr<Buffer>> blocks;
 
-    static constexpr u64 block_page_bits = 21;
-    static constexpr u64 block_page_size = 1ULL << block_page_bits;
-    std::unordered_map<u64, OwnerBuffer> blocks;
-
-    std::list<OwnerBuffer> pending_destruction;
+    std::queue<std::shared_ptr<Buffer>> pending_destruction;
     u64 epoch = 0;
     u64 modified_ticks = 0;
 
     std::vector<u8> staging_buffer;
+
     std::list<MapInterval*> marked_for_unregister;
 
     std::shared_ptr<std::unordered_set<MapInterval*>> uncommitted_flushes;
diff --git a/src/video_core/engines/const_buffer_engine_interface.h b/src/video_core/engines/const_buffer_engine_interface.h
index ebe139504..f46e81bb7 100644
--- a/src/video_core/engines/const_buffer_engine_interface.h
+++ b/src/video_core/engines/const_buffer_engine_interface.h
@@ -93,6 +93,7 @@ public:
     virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
     virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                                     u64 offset) const = 0;
+    virtual SamplerDescriptor AccessSampler(u32 handle) const = 0;
     virtual u32 GetBoundBuffer() const = 0;
 
     virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0;
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index f6237fc6a..a82b06a38 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -92,8 +92,11 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con
     ASSERT(stage == ShaderType::Compute);
     const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
     const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset;
+    return AccessSampler(memory_manager.Read<u32>(tex_info_address));
+}
 
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const {
+    const Texture::TextureHandle tex_handle{handle};
     const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
     SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
     result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h
index 18ceedfaf..b7f668d88 100644
--- a/src/video_core/engines/kepler_compute.h
+++ b/src/video_core/engines/kepler_compute.h
@@ -219,6 +219,8 @@ public:
     SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                             u64 offset) const override;
 
+    SamplerDescriptor AccessSampler(u32 handle) const override;
+
     u32 GetBoundBuffer() const override {
         return regs.tex_cb_index;
     }
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 13ef2e42d..ea3c8a963 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -25,9 +25,8 @@ constexpr u32 MacroRegistersStart = 0xE00;
 Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                      MemoryManager& memory_manager)
     : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager},
-      macro_interpreter{*this}, upload_state{memory_manager, regs.upload} {
+      macro_engine{GetMacroEngine(*this)}, upload_state{memory_manager, regs.upload} {
     dirty.flags.flip();
-
     InitializeRegisterDefaults();
 }
 
@@ -120,7 +119,7 @@ void Maxwell3D::InitializeRegisterDefaults() {
     mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true;
 }
 
-void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters) {
+void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters) {
     // Reset the current macro.
     executing_macro = 0;
 
@@ -129,7 +128,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3
         ((method - MacroRegistersStart) >> 1) % static_cast<u32>(macro_positions.size());
 
     // Execute the current macro.
-    macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters);
+    macro_engine->Execute(macro_positions[entry], parameters);
     if (mme_draw.current_mode != MMEDrawMode::Undefined) {
         FlushMMEInlineDraw();
     }
@@ -165,7 +164,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
 
         // Call the macro when there are no more parameters in the command buffer
         if (is_last_call) {
-            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            CallMacroMethod(executing_macro, macro_params);
             macro_params.clear();
         }
         return;
@@ -201,7 +200,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
         break;
     }
     case MAXWELL3D_REG_INDEX(macros.data): {
-        ProcessMacroUpload(arg);
+        macro_engine->AddCode(regs.macros.upload_address, arg);
         break;
     }
     case MAXWELL3D_REG_INDEX(macros.bind): {
@@ -310,7 +309,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
 
         // Call the macro when there are no more parameters in the command buffer
         if (amount == methods_pending) {
-            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            CallMacroMethod(executing_macro, macro_params);
             macro_params.clear();
         }
         return;
@@ -424,9 +423,7 @@ void Maxwell3D::FlushMMEInlineDraw() {
 }
 
 void Maxwell3D::ProcessMacroUpload(u32 data) {
-    ASSERT_MSG(regs.macros.upload_address < macro_memory.size(),
-               "upload_address exceeded macro_memory size!");
-    macro_memory[regs.macros.upload_address++] = data;
+    macro_engine->AddCode(regs.macros.upload_address++, data);
 }
 
 void Maxwell3D::ProcessMacroBind(u32 data) {
@@ -743,8 +740,11 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b
     const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
     const auto& tex_info_buffer = shader.const_buffers[const_buffer];
     const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
+    return AccessSampler(memory_manager.Read<u32>(tex_info_address));
+}
 
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
+SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const {
+    const Texture::TextureHandle tex_handle{handle};
     const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
     SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
     result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 05dd6b39b..d5fe25065 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -23,7 +23,7 @@
 #include "video_core/engines/engine_upload.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/gpu.h"
-#include "video_core/macro_interpreter.h"
+#include "video_core/macro/macro.h"
 #include "video_core/textures/texture.h"
 
 namespace Core {
@@ -598,6 +598,7 @@ public:
                 BitField<4, 3, u32> block_height;
                 BitField<8, 3, u32> block_depth;
                 BitField<12, 1, InvMemoryLayout> type;
+                BitField<16, 1, u32> is_3d;
             } memory_layout;
             union {
                 BitField<0, 16, u32> layers;
@@ -1403,6 +1404,8 @@ public:
     SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                             u64 offset) const override;
 
+    SamplerDescriptor AccessSampler(u32 handle) const override;
+
     u32 GetBoundBuffer() const override {
         return regs.tex_cb_index;
     }
@@ -1411,15 +1414,6 @@ public:
 
     const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
 
-    /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than
-    /// we've seen used.
-    using MacroMemory = std::array<u32, 0x40000>;
-
-    /// Gets a reference to macro memory.
-    const MacroMemory& GetMacroMemory() const {
-        return macro_memory;
-    }
-
     bool ShouldExecute() const {
         return execute_on;
     }
@@ -1468,16 +1462,13 @@ private:
 
     std::array<bool, Regs::NUM_REGS> mme_inline{};
 
-    /// Memory for macro code
-    MacroMemory macro_memory;
-
     /// Macro method that is currently being executed / being fed parameters.
     u32 executing_macro = 0;
     /// Parameters that have been submitted to the macro call so far.
     std::vector<u32> macro_params;
 
     /// Interpreter for the macro codes uploaded to the GPU.
-    MacroInterpreter macro_interpreter;
+    std::unique_ptr<MacroEngine> macro_engine;
 
     static constexpr u32 null_cb_data = 0xFFFFFFFF;
     struct {
@@ -1506,7 +1497,7 @@ private:
      * @param num_parameters Number of arguments
      * @param parameters Arguments to the method call
      */
-    void CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters);
+    void CallMacroMethod(u32 method, const std::vector<u32>& parameters);
 
     /// Handles writes to the macro uploading register.
     void ProcessMacroUpload(u32 data);
diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp
new file mode 100644
index 000000000..89077a2d8
--- /dev/null
+++ b/src/video_core/macro/macro.cpp
@@ -0,0 +1,45 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "core/settings.h"
+#include "video_core/macro/macro.h"
+#include "video_core/macro/macro_interpreter.h"
+#include "video_core/macro/macro_jit_x64.h"
+
+namespace Tegra {
+
+void MacroEngine::AddCode(u32 method, u32 data) {
+    uploaded_macro_code[method].push_back(data);
+}
+
+void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
+    auto compiled_macro = macro_cache.find(method);
+    if (compiled_macro != macro_cache.end()) {
+        compiled_macro->second->Execute(parameters, method);
+    } else {
+        // Macro not compiled, check if it's uploaded and if so, compile it
+        auto macro_code = uploaded_macro_code.find(method);
+        if (macro_code == uploaded_macro_code.end()) {
+            UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method);
+            return;
+        }
+        macro_cache[method] = Compile(macro_code->second);
+        macro_cache[method]->Execute(parameters, method);
+    }
+}
+
+std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d) {
+    if (Settings::values.disable_macro_jit) {
+        return std::make_unique<MacroInterpreter>(maxwell3d);
+    }
+#ifdef ARCHITECTURE_x86_64
+    return std::make_unique<MacroJITx64>(maxwell3d);
+#else
+    return std::make_unique<MacroInterpreter>(maxwell3d);
+#endif
+}
+
+} // namespace Tegra
diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h
new file mode 100644
index 000000000..b76ed891f
--- /dev/null
+++ b/src/video_core/macro/macro.h
@@ -0,0 +1,128 @@
+// Copyright 2020 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 <vector>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+
+namespace Tegra {
+namespace Engines {
+class Maxwell3D;
+}
+namespace Macro {
+constexpr std::size_t NUM_MACRO_REGISTERS = 8;
+enum class Operation : u32 {
+    ALU = 0,
+    AddImmediate = 1,
+    ExtractInsert = 2,
+    ExtractShiftLeftImmediate = 3,
+    ExtractShiftLeftRegister = 4,
+    Read = 5,
+    Unused = 6, // This operation doesn't seem to be a valid encoding.
+    Branch = 7,
+};
+
+enum class ALUOperation : u32 {
+    Add = 0,
+    AddWithCarry = 1,
+    Subtract = 2,
+    SubtractWithBorrow = 3,
+    // Operations 4-7 don't seem to be valid encodings.
+    Xor = 8,
+    Or = 9,
+    And = 10,
+    AndNot = 11,
+    Nand = 12
+};
+
+enum class ResultOperation : u32 {
+    IgnoreAndFetch = 0,
+    Move = 1,
+    MoveAndSetMethod = 2,
+    FetchAndSend = 3,
+    MoveAndSend = 4,
+    FetchAndSetMethod = 5,
+    MoveAndSetMethodFetchAndSend = 6,
+    MoveAndSetMethodSend = 7
+};
+
+enum class BranchCondition : u32 {
+    Zero = 0,
+    NotZero = 1,
+};
+
+union Opcode {
+    u32 raw;
+    BitField<0, 3, Operation> operation;
+    BitField<4, 3, ResultOperation> result_operation;
+    BitField<4, 1, BranchCondition> branch_condition;
+    // If set on a branch, then the branch doesn't have a delay slot.
+    BitField<5, 1, u32> branch_annul;
+    BitField<7, 1, u32> is_exit;
+    BitField<8, 3, u32> dst;
+    BitField<11, 3, u32> src_a;
+    BitField<14, 3, u32> src_b;
+    // The signed immediate overlaps the second source operand and the alu operation.
+    BitField<14, 18, s32> immediate;
+
+    BitField<17, 5, ALUOperation> alu_operation;
+
+    // Bitfield instructions data
+    BitField<17, 5, u32> bf_src_bit;
+    BitField<22, 5, u32> bf_size;
+    BitField<27, 5, u32> bf_dst_bit;
+
+    u32 GetBitfieldMask() const {
+        return (1 << bf_size) - 1;
+    }
+
+    s32 GetBranchTarget() const {
+        return static_cast<s32>(immediate * sizeof(u32));
+    }
+};
+
+union MethodAddress {
+    u32 raw;
+    BitField<0, 12, u32> address;
+    BitField<12, 6, u32> increment;
+};
+
+} // namespace Macro
+
+class CachedMacro {
+public:
+    virtual ~CachedMacro() = default;
+    /**
+     * Executes the macro code with the specified input parameters.
+     * @param code The macro byte code to execute
+     * @param parameters The parameters of the macro
+     */
+    virtual void Execute(const std::vector<u32>& parameters, u32 method) = 0;
+};
+
+class MacroEngine {
+public:
+    virtual ~MacroEngine() = default;
+
+    // Store the uploaded macro code to compile them when they're called.
+    void AddCode(u32 method, u32 data);
+
+    // Compiles the macro if its not in the cache, and executes the compiled macro
+    void Execute(u32 method, const std::vector<u32>& parameters);
+
+protected:
+    virtual std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) = 0;
+
+private:
+    std::unordered_map<u32, std::unique_ptr<CachedMacro>> macro_cache;
+    std::unordered_map<u32, std::vector<u32>> uploaded_macro_code;
+};
+
+std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d);
+
+} // namespace Tegra
diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp
index 947364928..5edff27aa 100644
--- a/src/video_core/macro_interpreter.cpp
+++ b/src/video_core/macro/macro_interpreter.cpp
@@ -1,4 +1,4 @@
-// Copyright 2018 yuzu Emulator Project
+// Copyright 2020 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
@@ -6,109 +6,46 @@
 #include "common/logging/log.h"
 #include "common/microprofile.h"
 #include "video_core/engines/maxwell_3d.h"
-#include "video_core/macro_interpreter.h"
+#include "video_core/macro/macro_interpreter.h"
 
 MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
 
 namespace Tegra {
-namespace {
-enum class Operation : u32 {
-    ALU = 0,
-    AddImmediate = 1,
-    ExtractInsert = 2,
-    ExtractShiftLeftImmediate = 3,
-    ExtractShiftLeftRegister = 4,
-    Read = 5,
-    Unused = 6, // This operation doesn't seem to be a valid encoding.
-    Branch = 7,
-};
-} // Anonymous namespace
-
-enum class MacroInterpreter::ALUOperation : u32 {
-    Add = 0,
-    AddWithCarry = 1,
-    Subtract = 2,
-    SubtractWithBorrow = 3,
-    // Operations 4-7 don't seem to be valid encodings.
-    Xor = 8,
-    Or = 9,
-    And = 10,
-    AndNot = 11,
-    Nand = 12
-};
-
-enum class MacroInterpreter::ResultOperation : u32 {
-    IgnoreAndFetch = 0,
-    Move = 1,
-    MoveAndSetMethod = 2,
-    FetchAndSend = 3,
-    MoveAndSend = 4,
-    FetchAndSetMethod = 5,
-    MoveAndSetMethodFetchAndSend = 6,
-    MoveAndSetMethodSend = 7
-};
-
-enum class MacroInterpreter::BranchCondition : u32 {
-    Zero = 0,
-    NotZero = 1,
-};
-
-union MacroInterpreter::Opcode {
-    u32 raw;
-    BitField<0, 3, Operation> operation;
-    BitField<4, 3, ResultOperation> result_operation;
-    BitField<4, 1, BranchCondition> branch_condition;
-    // If set on a branch, then the branch doesn't have a delay slot.
-    BitField<5, 1, u32> branch_annul;
-    BitField<7, 1, u32> is_exit;
-    BitField<8, 3, u32> dst;
-    BitField<11, 3, u32> src_a;
-    BitField<14, 3, u32> src_b;
-    // The signed immediate overlaps the second source operand and the alu operation.
-    BitField<14, 18, s32> immediate;
-
-    BitField<17, 5, ALUOperation> alu_operation;
-
-    // Bitfield instructions data
-    BitField<17, 5, u32> bf_src_bit;
-    BitField<22, 5, u32> bf_size;
-    BitField<27, 5, u32> bf_dst_bit;
-
-    u32 GetBitfieldMask() const {
-        return (1 << bf_size) - 1;
-    }
+MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
 
-    s32 GetBranchTarget() const {
-        return static_cast<s32>(immediate * sizeof(u32));
-    }
-};
+std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroInterpreterImpl>(maxwell3d, code);
+}
 
-MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+MacroInterpreterImpl::MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d,
+                                           const std::vector<u32>& code)
+    : maxwell3d(maxwell3d), code(code) {}
 
-void MacroInterpreter::Execute(u32 offset, std::size_t num_parameters, const u32* parameters) {
+void MacroInterpreterImpl::Execute(const std::vector<u32>& parameters, u32 method) {
     MICROPROFILE_SCOPE(MacroInterp);
     Reset();
 
     registers[1] = parameters[0];
+    num_parameters = parameters.size();
 
     if (num_parameters > parameters_capacity) {
         parameters_capacity = num_parameters;
         this->parameters = std::make_unique<u32[]>(num_parameters);
     }
-    std::memcpy(this->parameters.get(), parameters, num_parameters * sizeof(u32));
+    std::memcpy(this->parameters.get(), parameters.data(), num_parameters * sizeof(u32));
     this->num_parameters = num_parameters;
 
     // Execute the code until we hit an exit condition.
     bool keep_executing = true;
     while (keep_executing) {
-        keep_executing = Step(offset, false);
+        keep_executing = Step(false);
     }
 
     // Assert the the macro used all the input parameters
     ASSERT(next_parameter_index == num_parameters);
 }
 
-void MacroInterpreter::Reset() {
+void MacroInterpreterImpl::Reset() {
     registers = {};
     pc = 0;
     delayed_pc = {};
@@ -120,10 +57,10 @@ void MacroInterpreter::Reset() {
     carry_flag = false;
 }
 
-bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
+bool MacroInterpreterImpl::Step(bool is_delay_slot) {
     u32 base_address = pc;
 
-    Opcode opcode = GetOpcode(offset);
+    Macro::Opcode opcode = GetOpcode();
     pc += 4;
 
     // Update the program counter if we were delayed
@@ -134,18 +71,18 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
     }
 
     switch (opcode.operation) {
-    case Operation::ALU: {
+    case Macro::Operation::ALU: {
         u32 result = GetALUResult(opcode.alu_operation, GetRegister(opcode.src_a),
                                   GetRegister(opcode.src_b));
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::AddImmediate: {
+    case Macro::Operation::AddImmediate: {
         ProcessResult(opcode.result_operation, opcode.dst,
                       GetRegister(opcode.src_a) + opcode.immediate);
         break;
     }
-    case Operation::ExtractInsert: {
+    case Macro::Operation::ExtractInsert: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -155,7 +92,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, dst);
         break;
     }
-    case Operation::ExtractShiftLeftImmediate: {
+    case Macro::Operation::ExtractShiftLeftImmediate: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -164,7 +101,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::ExtractShiftLeftRegister: {
+    case Macro::Operation::ExtractShiftLeftRegister: {
         u32 dst = GetRegister(opcode.src_a);
         u32 src = GetRegister(opcode.src_b);
 
@@ -173,12 +110,12 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::Read: {
+    case Macro::Operation::Read: {
         u32 result = Read(GetRegister(opcode.src_a) + opcode.immediate);
         ProcessResult(opcode.result_operation, opcode.dst, result);
         break;
     }
-    case Operation::Branch: {
+    case Macro::Operation::Branch: {
         ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
         u32 value = GetRegister(opcode.src_a);
         bool taken = EvaluateBranchCondition(opcode.branch_condition, value);
@@ -191,7 +128,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
 
             delayed_pc = base_address + opcode.GetBranchTarget();
             // Execute one more instruction due to the delay slot.
-            return Step(offset, true);
+            return Step(true);
         }
         break;
     }
@@ -204,51 +141,44 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
     // cause an exit if it's executed inside a delay slot.
     if (opcode.is_exit && !is_delay_slot) {
         // Exit has a delay slot, execute the next instruction
-        Step(offset, true);
+        Step(true);
         return false;
     }
 
     return true;
 }
 
-MacroInterpreter::Opcode MacroInterpreter::GetOpcode(u32 offset) const {
-    const auto& macro_memory{maxwell3d.GetMacroMemory()};
-    ASSERT((pc % sizeof(u32)) == 0);
-    ASSERT((pc + offset) < macro_memory.size() * sizeof(u32));
-    return {macro_memory[offset + pc / sizeof(u32)]};
-}
-
-u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) {
+u32 MacroInterpreterImpl::GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b) {
     switch (operation) {
-    case ALUOperation::Add: {
+    case Macro::ALUOperation::Add: {
         const u64 result{static_cast<u64>(src_a) + src_b};
         carry_flag = result > 0xffffffff;
         return static_cast<u32>(result);
     }
-    case ALUOperation::AddWithCarry: {
+    case Macro::ALUOperation::AddWithCarry: {
         const u64 result{static_cast<u64>(src_a) + src_b + (carry_flag ? 1ULL : 0ULL)};
         carry_flag = result > 0xffffffff;
         return static_cast<u32>(result);
     }
-    case ALUOperation::Subtract: {
+    case Macro::ALUOperation::Subtract: {
         const u64 result{static_cast<u64>(src_a) - src_b};
         carry_flag = result < 0x100000000;
         return static_cast<u32>(result);
     }
-    case ALUOperation::SubtractWithBorrow: {
+    case Macro::ALUOperation::SubtractWithBorrow: {
         const u64 result{static_cast<u64>(src_a) - src_b - (carry_flag ? 0ULL : 1ULL)};
         carry_flag = result < 0x100000000;
         return static_cast<u32>(result);
     }
-    case ALUOperation::Xor:
+    case Macro::ALUOperation::Xor:
         return src_a ^ src_b;
-    case ALUOperation::Or:
+    case Macro::ALUOperation::Or:
         return src_a | src_b;
-    case ALUOperation::And:
+    case Macro::ALUOperation::And:
         return src_a & src_b;
-    case ALUOperation::AndNot:
+    case Macro::ALUOperation::AndNot:
         return src_a & ~src_b;
-    case ALUOperation::Nand:
+    case Macro::ALUOperation::Nand:
         return ~(src_a & src_b);
 
     default:
@@ -257,43 +187,43 @@ u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b)
     }
 }
 
-void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 result) {
+void MacroInterpreterImpl::ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result) {
     switch (operation) {
-    case ResultOperation::IgnoreAndFetch:
+    case Macro::ResultOperation::IgnoreAndFetch:
         // Fetch parameter and ignore result.
         SetRegister(reg, FetchParameter());
         break;
-    case ResultOperation::Move:
+    case Macro::ResultOperation::Move:
         // Move result.
         SetRegister(reg, result);
         break;
-    case ResultOperation::MoveAndSetMethod:
+    case Macro::ResultOperation::MoveAndSetMethod:
         // Move result and use as Method Address.
         SetRegister(reg, result);
         SetMethodAddress(result);
         break;
-    case ResultOperation::FetchAndSend:
+    case Macro::ResultOperation::FetchAndSend:
         // Fetch parameter and send result.
         SetRegister(reg, FetchParameter());
         Send(result);
         break;
-    case ResultOperation::MoveAndSend:
+    case Macro::ResultOperation::MoveAndSend:
         // Move and send result.
         SetRegister(reg, result);
         Send(result);
         break;
-    case ResultOperation::FetchAndSetMethod:
+    case Macro::ResultOperation::FetchAndSetMethod:
         // Fetch parameter and use result as Method Address.
         SetRegister(reg, FetchParameter());
         SetMethodAddress(result);
         break;
-    case ResultOperation::MoveAndSetMethodFetchAndSend:
+    case Macro::ResultOperation::MoveAndSetMethodFetchAndSend:
         // Move result and use as Method Address, then fetch and send parameter.
         SetRegister(reg, result);
         SetMethodAddress(result);
         Send(FetchParameter());
         break;
-    case ResultOperation::MoveAndSetMethodSend:
+    case Macro::ResultOperation::MoveAndSetMethodSend:
         // Move result and use as Method Address, then send bits 12:17 of result.
         SetRegister(reg, result);
         SetMethodAddress(result);
@@ -304,16 +234,28 @@ void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 res
     }
 }
 
-u32 MacroInterpreter::FetchParameter() {
-    ASSERT(next_parameter_index < num_parameters);
-    return parameters[next_parameter_index++];
+bool MacroInterpreterImpl::EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const {
+    switch (cond) {
+    case Macro::BranchCondition::Zero:
+        return value == 0;
+    case Macro::BranchCondition::NotZero:
+        return value != 0;
+    }
+    UNREACHABLE();
+    return true;
 }
 
-u32 MacroInterpreter::GetRegister(u32 register_id) const {
+Macro::Opcode MacroInterpreterImpl::GetOpcode() const {
+    ASSERT((pc % sizeof(u32)) == 0);
+    ASSERT(pc < code.size() * sizeof(u32));
+    return {code[pc / sizeof(u32)]};
+}
+
+u32 MacroInterpreterImpl::GetRegister(u32 register_id) const {
     return registers.at(register_id);
 }
 
-void MacroInterpreter::SetRegister(u32 register_id, u32 value) {
+void MacroInterpreterImpl::SetRegister(u32 register_id, u32 value) {
     // Register 0 is hardwired as the zero register.
     // Ensure no writes to it actually occur.
     if (register_id == 0) {
@@ -323,30 +265,24 @@ void MacroInterpreter::SetRegister(u32 register_id, u32 value) {
     registers.at(register_id) = value;
 }
 
-void MacroInterpreter::SetMethodAddress(u32 address) {
+void MacroInterpreterImpl::SetMethodAddress(u32 address) {
     method_address.raw = address;
 }
 
-void MacroInterpreter::Send(u32 value) {
+void MacroInterpreterImpl::Send(u32 value) {
     maxwell3d.CallMethodFromMME(method_address.address, value);
     // Increment the method address by the method increment.
     method_address.address.Assign(method_address.address.Value() +
                                   method_address.increment.Value());
 }
 
-u32 MacroInterpreter::Read(u32 method) const {
+u32 MacroInterpreterImpl::Read(u32 method) const {
     return maxwell3d.GetRegisterValue(method);
 }
 
-bool MacroInterpreter::EvaluateBranchCondition(BranchCondition cond, u32 value) const {
-    switch (cond) {
-    case BranchCondition::Zero:
-        return value == 0;
-    case BranchCondition::NotZero:
-        return value != 0;
-    }
-    UNREACHABLE();
-    return true;
+u32 MacroInterpreterImpl::FetchParameter() {
+    ASSERT(next_parameter_index < num_parameters);
+    return parameters[next_parameter_index++];
 }
 
 } // namespace Tegra
diff --git a/src/video_core/macro_interpreter.h b/src/video_core/macro/macro_interpreter.h
index 631146d89..90217fc89 100644
--- a/src/video_core/macro_interpreter.h
+++ b/src/video_core/macro/macro_interpreter.h
@@ -1,44 +1,37 @@
-// Copyright 2018 yuzu Emulator Project
+// Copyright 2020 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
-
 #include <array>
 #include <optional>
-
+#include <vector>
 #include "common/bit_field.h"
 #include "common/common_types.h"
+#include "video_core/macro/macro.h"
 
 namespace Tegra {
 namespace Engines {
 class Maxwell3D;
 }
 
-class MacroInterpreter final {
+class MacroInterpreter final : public MacroEngine {
 public:
     explicit MacroInterpreter(Engines::Maxwell3D& maxwell3d);
 
-    /**
-     * Executes the macro code with the specified input parameters.
-     * @param offset Offset to start execution at.
-     * @param parameters The parameters of the macro.
-     */
-    void Execute(u32 offset, std::size_t num_parameters, const u32* parameters);
+protected:
+    std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override;
 
 private:
-    enum class ALUOperation : u32;
-    enum class BranchCondition : u32;
-    enum class ResultOperation : u32;
-
-    union Opcode;
+    Engines::Maxwell3D& maxwell3d;
+};
 
-    union MethodAddress {
-        u32 raw;
-        BitField<0, 12, u32> address;
-        BitField<12, 6, u32> increment;
-    };
+class MacroInterpreterImpl : public CachedMacro {
+public:
+    MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code);
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
 
+private:
     /// Resets the execution engine state, zeroing registers, etc.
     void Reset();
 
@@ -49,20 +42,20 @@ private:
      * @param is_delay_slot Whether the current step is being executed due to a delay slot in a
      * previous instruction.
      */
-    bool Step(u32 offset, bool is_delay_slot);
+    bool Step(bool is_delay_slot);
 
     /// Calculates the result of an ALU operation. src_a OP src_b;
-    u32 GetALUResult(ALUOperation operation, u32 src_a, u32 src_b);
+    u32 GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b);
 
     /// Performs the result operation on the input result and stores it in the specified register
     /// (if necessary).
-    void ProcessResult(ResultOperation operation, u32 reg, u32 result);
+    void ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result);
 
     /// Evaluates the branch condition and returns whether the branch should be taken or not.
-    bool EvaluateBranchCondition(BranchCondition cond, u32 value) const;
+    bool EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const;
 
     /// Reads an opcode at the current program counter location.
-    Opcode GetOpcode(u32 offset) const;
+    Macro::Opcode GetOpcode() const;
 
     /// Returns the specified register's value. Register 0 is hardcoded to always return 0.
     u32 GetRegister(u32 register_id) const;
@@ -89,13 +82,11 @@ private:
     /// Program counter to execute at after the delay slot is executed.
     std::optional<u32> delayed_pc;
 
-    static constexpr std::size_t NumMacroRegisters = 8;
-
     /// General purpose macro registers.
-    std::array<u32, NumMacroRegisters> registers = {};
+    std::array<u32, Macro::NUM_MACRO_REGISTERS> registers = {};
 
     /// Method address to use for the next Send instruction.
-    MethodAddress method_address = {};
+    Macro::MethodAddress method_address = {};
 
     /// Input parameters of the current macro.
     std::unique_ptr<u32[]> parameters;
@@ -105,5 +96,7 @@ private:
     u32 next_parameter_index = 0;
 
     bool carry_flag = false;
+    const std::vector<u32>& code;
 };
+
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp
new file mode 100644
index 000000000..11c1cc3be
--- /dev/null
+++ b/src/video_core/macro/macro_jit_x64.cpp
@@ -0,0 +1,640 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/microprofile.h"
+#include "common/x64/xbyak_util.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro_interpreter.h"
+#include "video_core/macro/macro_jit_x64.h"
+
+MICROPROFILE_DEFINE(MacroJitCompile, "GPU", "Compile macro JIT", MP_RGB(173, 255, 47));
+MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255, 0));
+
+namespace Tegra {
+static const Xbyak::Reg64 PARAMETERS = Xbyak::util::r9;
+static const Xbyak::Reg64 REGISTERS = Xbyak::util::r10;
+static const Xbyak::Reg64 STATE = Xbyak::util::r11;
+static const Xbyak::Reg64 NEXT_PARAMETER = Xbyak::util::r12;
+static const Xbyak::Reg32 RESULT = Xbyak::util::r13d;
+static const Xbyak::Reg64 RESULT_64 = Xbyak::util::r13;
+static const Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d;
+static const Xbyak::Reg64 METHOD_ADDRESS_64 = Xbyak::util::r14;
+static const Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
+
+static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
+    PARAMETERS,
+    REGISTERS,
+    STATE,
+    NEXT_PARAMETER,
+    RESULT,
+    METHOD_ADDRESS,
+    BRANCH_HOLDER,
+});
+
+MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+
+std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
+}
+
+MacroJITx64Impl::MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code)
+    : Xbyak::CodeGenerator(MAX_CODE_SIZE), code(code), maxwell3d(maxwell3d) {
+    Compile();
+}
+
+MacroJITx64Impl::~MacroJITx64Impl() = default;
+
+void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) {
+    MICROPROFILE_SCOPE(MacroJitExecute);
+    ASSERT_OR_EXECUTE(program != nullptr, { return; });
+    JITState state{};
+    state.maxwell3d = &maxwell3d;
+    state.registers = {};
+    state.parameters = parameters.data();
+    program(&state);
+}
+
+void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
+    const bool is_a_zero = opcode.src_a == 0;
+    const bool is_b_zero = opcode.src_b == 0;
+    const bool valid_operation = !is_a_zero && !is_b_zero;
+    const bool is_move_operation = !is_a_zero && is_b_zero;
+    const bool has_zero_register = is_a_zero || is_b_zero;
+
+    Xbyak::Reg64 src_a;
+    Xbyak::Reg32 src_b;
+
+    if (!optimizer.zero_reg_skip) {
+        src_a = Compile_GetRegister(opcode.src_a, RESULT_64);
+        src_b = Compile_GetRegister(opcode.src_b, ebx);
+    } else {
+        if (!is_a_zero) {
+            src_a = Compile_GetRegister(opcode.src_a, RESULT_64);
+        }
+        if (!is_b_zero) {
+            src_b = Compile_GetRegister(opcode.src_b, ebx);
+        }
+    }
+    Xbyak::Label skip_carry{};
+
+    bool has_emitted = false;
+
+    switch (opcode.alu_operation) {
+    case Macro::ALUOperation::Add:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                add(src_a, src_b);
+            }
+        } else {
+            add(src_a, src_b);
+        }
+
+        if (!optimizer.can_skip_carry) {
+            setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        }
+        break;
+    case Macro::ALUOperation::AddWithCarry:
+        bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
+        adc(src_a, src_b);
+        setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        break;
+    case Macro::ALUOperation::Subtract:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                sub(src_a, src_b);
+                has_emitted = true;
+            }
+        } else {
+            sub(src_a, src_b);
+            has_emitted = true;
+        }
+        if (!optimizer.can_skip_carry && has_emitted) {
+            setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        }
+        break;
+    case Macro::ALUOperation::SubtractWithBorrow:
+        bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
+        sbb(src_a, src_b);
+        setc(byte[STATE + offsetof(JITState, carry_flag)]);
+        break;
+    case Macro::ALUOperation::Xor:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                xor_(src_a, src_b);
+            }
+        } else {
+            xor_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::Or:
+        if (optimizer.zero_reg_skip) {
+            if (valid_operation) {
+                or_(src_a, src_b);
+            }
+        } else {
+            or_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::And:
+        if (optimizer.zero_reg_skip) {
+            if (!has_zero_register) {
+                and_(src_a, src_b);
+            }
+        } else {
+            and_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::AndNot:
+        if (optimizer.zero_reg_skip) {
+            if (!is_a_zero) {
+                not_(src_b);
+                and_(src_a, src_b);
+            }
+        } else {
+            not_(src_b);
+            and_(src_a, src_b);
+        }
+        break;
+    case Macro::ALUOperation::Nand:
+        if (optimizer.zero_reg_skip) {
+            if (!is_a_zero) {
+                and_(src_a, src_b);
+                not_(src_a);
+            }
+        } else {
+            and_(src_a, src_b);
+            not_(src_a);
+        }
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented ALU operation {}",
+                          static_cast<std::size_t>(opcode.alu_operation.Value()));
+        break;
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_AddImmediate(Macro::Opcode opcode) {
+    if (optimizer.skip_dummy_addimmediate) {
+        // Games tend to use this as an exit instruction placeholder. It's to encode an instruction
+        // without doing anything. In our case we can just not emit anything.
+        if (opcode.result_operation == Macro::ResultOperation::Move && opcode.dst == 0) {
+            return;
+        }
+    }
+    // Check for redundant moves
+    if (optimizer.optimize_for_method_move &&
+        opcode.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
+        if (next_opcode.has_value()) {
+            const auto next = *next_opcode;
+            if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
+                return;
+            }
+        }
+    }
+    if (optimizer.zero_reg_skip && opcode.src_a == 0) {
+        if (opcode.immediate == 0) {
+            xor_(RESULT, RESULT);
+        } else {
+            mov(RESULT, opcode.immediate);
+        }
+    } else {
+        auto result = Compile_GetRegister(opcode.src_a, RESULT);
+        if (opcode.immediate > 2) {
+            add(result, opcode.immediate);
+        } else if (opcode.immediate == 1) {
+            inc(result);
+        } else if (opcode.immediate < 0) {
+            sub(result, opcode.immediate * -1);
+        }
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) {
+    auto dst = Compile_GetRegister(opcode.src_a, RESULT);
+    auto src = Compile_GetRegister(opcode.src_b, eax);
+
+    if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) {
+        shr(src, opcode.bf_src_bit);
+    } else if (opcode.bf_src_bit == 31) {
+        xor_(src, src);
+    }
+    // Don't bother masking the whole register since we're using a 32 bit register
+    if (opcode.bf_size != 31 && opcode.bf_size != 0) {
+        and_(src, opcode.GetBitfieldMask());
+    } else if (opcode.bf_size == 0) {
+        xor_(src, src);
+    }
+    if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) {
+        shl(src, opcode.bf_dst_bit);
+    } else if (opcode.bf_dst_bit == 31) {
+        xor_(src, src);
+    }
+
+    const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit);
+    if (mask != 0xffffffff) {
+        and_(dst, mask);
+    }
+    or_(dst, src);
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) {
+    auto dst = Compile_GetRegister(opcode.src_a, eax);
+    auto src = Compile_GetRegister(opcode.src_b, RESULT);
+
+    shr(src, al);
+    if (opcode.bf_size != 0 && opcode.bf_size != 31) {
+        and_(src, opcode.GetBitfieldMask());
+    } else if (opcode.bf_size == 0) {
+        xor_(src, src);
+    }
+
+    if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) {
+        shl(src, opcode.bf_dst_bit);
+    } else if (opcode.bf_dst_bit == 31) {
+        xor_(src, src);
+    }
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) {
+    auto dst = Compile_GetRegister(opcode.src_a, eax);
+    auto src = Compile_GetRegister(opcode.src_b, RESULT);
+
+    if (opcode.bf_src_bit != 0) {
+        shr(src, opcode.bf_src_bit);
+    }
+
+    if (opcode.bf_size != 31) {
+        and_(src, opcode.GetBitfieldMask());
+    }
+    shl(src, al);
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+static u32 Read(Engines::Maxwell3D* maxwell3d, u32 method) {
+    return maxwell3d->GetRegisterValue(method);
+}
+
+static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
+    maxwell3d->CallMethodFromMME(method_address.address, value);
+}
+
+void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
+    if (optimizer.zero_reg_skip && opcode.src_a == 0) {
+        if (opcode.immediate == 0) {
+            xor_(RESULT, RESULT);
+        } else {
+            mov(RESULT, opcode.immediate);
+        }
+    } else {
+        auto result = Compile_GetRegister(opcode.src_a, RESULT);
+        if (opcode.immediate > 2) {
+            add(result, opcode.immediate);
+        } else if (opcode.immediate == 1) {
+            inc(result);
+        } else if (opcode.immediate < 0) {
+            sub(result, opcode.immediate * -1);
+        }
+    }
+    Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
+    mov(Common::X64::ABI_PARAM1, qword[STATE]);
+    mov(Common::X64::ABI_PARAM2, RESULT);
+    Common::X64::CallFarFunction(*this, &Read);
+    Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
+    mov(RESULT, Common::X64::ABI_RETURN.cvt32());
+    Compile_ProcessResult(opcode.result_operation, opcode.dst);
+}
+
+void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
+    Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
+    mov(Common::X64::ABI_PARAM1, qword[STATE]);
+    mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS);
+    mov(Common::X64::ABI_PARAM3, value);
+    Common::X64::CallFarFunction(*this, &Send);
+    Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
+
+    Xbyak::Label dont_process{};
+    // Get increment
+    test(METHOD_ADDRESS, 0x3f000);
+    // If zero, method address doesn't update
+    je(dont_process);
+
+    mov(ecx, METHOD_ADDRESS);
+    and_(METHOD_ADDRESS, 0xfff);
+    shr(ecx, 12);
+    and_(ecx, 0x3f);
+    lea(eax, ptr[rcx + METHOD_ADDRESS_64]);
+    sal(ecx, 12);
+    or_(eax, ecx);
+
+    mov(METHOD_ADDRESS, eax);
+
+    L(dont_process);
+}
+
+void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
+    ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
+    const s32 jump_address =
+        static_cast<s32>(pc) + static_cast<s32>(opcode.GetBranchTarget() / sizeof(s32));
+
+    Xbyak::Label end;
+    auto value = Compile_GetRegister(opcode.src_a, eax);
+    test(value, value);
+    if (optimizer.has_delayed_pc) {
+        switch (opcode.branch_condition) {
+        case Macro::BranchCondition::Zero:
+            jne(end, T_NEAR);
+            break;
+        case Macro::BranchCondition::NotZero:
+            je(end, T_NEAR);
+            break;
+        }
+
+        if (opcode.branch_annul) {
+            xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+            jmp(labels[jump_address], T_NEAR);
+        } else {
+            Xbyak::Label handle_post_exit{};
+            Xbyak::Label skip{};
+            jmp(skip, T_NEAR);
+            if (opcode.is_exit) {
+                L(handle_post_exit);
+                // Execute 1 instruction
+                mov(BRANCH_HOLDER, end_of_code);
+                // Jump to next instruction to skip delay slot check
+                jmp(labels[jump_address], T_NEAR);
+            } else {
+                L(handle_post_exit);
+                xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+                jmp(labels[jump_address], T_NEAR);
+            }
+            L(skip);
+            mov(BRANCH_HOLDER, handle_post_exit);
+            jmp(delay_skip[pc], T_NEAR);
+        }
+    } else {
+        switch (opcode.branch_condition) {
+        case Macro::BranchCondition::Zero:
+            je(labels[jump_address], T_NEAR);
+            break;
+        case Macro::BranchCondition::NotZero:
+            jne(labels[jump_address], T_NEAR);
+            break;
+        }
+    }
+
+    L(end);
+}
+
+void Tegra::MacroJITx64Impl::Optimizer_ScanFlags() {
+    optimizer.can_skip_carry = true;
+    optimizer.has_delayed_pc = false;
+    for (auto raw_op : code) {
+        Macro::Opcode op{};
+        op.raw = raw_op;
+
+        if (op.operation == Macro::Operation::ALU) {
+            // Scan for any ALU operations which actually use the carry flag, if they don't exist in
+            // our current code we can skip emitting the carry flag handling operations
+            if (op.alu_operation == Macro::ALUOperation::AddWithCarry ||
+                op.alu_operation == Macro::ALUOperation::SubtractWithBorrow) {
+                optimizer.can_skip_carry = false;
+            }
+        }
+
+        if (op.operation == Macro::Operation::Branch) {
+            if (!op.branch_annul) {
+                optimizer.has_delayed_pc = true;
+            }
+        }
+    }
+}
+
+void MacroJITx64Impl::Compile() {
+    MICROPROFILE_SCOPE(MacroJitCompile);
+    bool keep_executing = true;
+    labels.fill(Xbyak::Label());
+
+    Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
+    // JIT state
+    mov(STATE, Common::X64::ABI_PARAM1);
+    mov(PARAMETERS, qword[Common::X64::ABI_PARAM1 +
+                          static_cast<Xbyak::uint32>(offsetof(JITState, parameters))]);
+    mov(REGISTERS, Common::X64::ABI_PARAM1);
+    add(REGISTERS, static_cast<Xbyak::uint32>(offsetof(JITState, registers)));
+    xor_(RESULT, RESULT);
+    xor_(METHOD_ADDRESS, METHOD_ADDRESS);
+    xor_(NEXT_PARAMETER, NEXT_PARAMETER);
+    xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+
+    mov(dword[REGISTERS + 4], Compile_FetchParameter());
+
+    // Track get register for zero registers and mark it as no-op
+    optimizer.zero_reg_skip = true;
+
+    // AddImmediate tends to be used as a NOP instruction, if we detect this we can
+    // completely skip the entire code path and no emit anything
+    optimizer.skip_dummy_addimmediate = true;
+
+    // SMO tends to emit a lot of unnecessary method moves, we can mitigate this by only emitting
+    // one if our register isn't "dirty"
+    optimizer.optimize_for_method_move = true;
+
+    // Check to see if we can skip emitting certain instructions
+    Optimizer_ScanFlags();
+
+    const u32 op_count = static_cast<u32>(code.size());
+    for (u32 i = 0; i < op_count; i++) {
+        if (i < op_count - 1) {
+            pc = i + 1;
+            next_opcode = GetOpCode();
+        } else {
+            next_opcode = {};
+        }
+        pc = i;
+        Compile_NextInstruction();
+    }
+
+    L(end_of_code);
+
+    Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
+    ret();
+    ready();
+    program = getCode<ProgramType>();
+}
+
+bool MacroJITx64Impl::Compile_NextInstruction() {
+    const auto opcode = GetOpCode();
+    if (labels[pc].getAddress()) {
+        return false;
+    }
+
+    L(labels[pc]);
+
+    switch (opcode.operation) {
+    case Macro::Operation::ALU:
+        Compile_ALU(opcode);
+        break;
+    case Macro::Operation::AddImmediate:
+        Compile_AddImmediate(opcode);
+        break;
+    case Macro::Operation::ExtractInsert:
+        Compile_ExtractInsert(opcode);
+        break;
+    case Macro::Operation::ExtractShiftLeftImmediate:
+        Compile_ExtractShiftLeftImmediate(opcode);
+        break;
+    case Macro::Operation::ExtractShiftLeftRegister:
+        Compile_ExtractShiftLeftRegister(opcode);
+        break;
+    case Macro::Operation::Read:
+        Compile_Read(opcode);
+        break;
+    case Macro::Operation::Branch:
+        Compile_Branch(opcode);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented opcode {}", opcode.operation.Value());
+        break;
+    }
+
+    if (optimizer.has_delayed_pc) {
+        if (opcode.is_exit) {
+            mov(rax, end_of_code);
+            test(BRANCH_HOLDER, BRANCH_HOLDER);
+            cmove(BRANCH_HOLDER, rax);
+            // Jump to next instruction to skip delay slot check
+            je(labels[pc + 1], T_NEAR);
+        } else {
+            // TODO(ogniK): Optimize delay slot branching
+            Xbyak::Label no_delay_slot{};
+            test(BRANCH_HOLDER, BRANCH_HOLDER);
+            je(no_delay_slot, T_NEAR);
+            mov(rax, BRANCH_HOLDER);
+            xor_(BRANCH_HOLDER, BRANCH_HOLDER);
+            jmp(rax);
+            L(no_delay_slot);
+        }
+        L(delay_skip[pc]);
+        if (opcode.is_exit) {
+            return false;
+        }
+    } else {
+        test(BRANCH_HOLDER, BRANCH_HOLDER);
+        jne(end_of_code, T_NEAR);
+        if (opcode.is_exit) {
+            inc(BRANCH_HOLDER);
+            return false;
+        }
+    }
+    return true;
+}
+
+Xbyak::Reg32 Tegra::MacroJITx64Impl::Compile_FetchParameter() {
+    mov(eax, dword[PARAMETERS + NEXT_PARAMETER * sizeof(u32)]);
+    inc(NEXT_PARAMETER);
+    return eax;
+}
+
+Xbyak::Reg32 MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg32 dst) {
+    if (index == 0) {
+        // Register 0 is always zero
+        xor_(dst, dst);
+    } else {
+        mov(dst, dword[REGISTERS + index * sizeof(u32)]);
+    }
+
+    return dst;
+}
+
+Xbyak::Reg64 Tegra::MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg64 dst) {
+    if (index == 0) {
+        // Register 0 is always zero
+        xor_(dst, dst);
+    } else {
+        mov(dst, dword[REGISTERS + index * sizeof(u32)]);
+    }
+
+    return dst;
+}
+
+void Tegra::MacroJITx64Impl::Compile_WriteCarry(Xbyak::Reg64 dst) {
+    Xbyak::Label zero{}, end{};
+    xor_(ecx, ecx);
+    shr(dst, 32);
+    setne(cl);
+    mov(dword[STATE + offsetof(JITState, carry_flag)], ecx);
+}
+
+void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u32 reg) {
+    auto SetRegister = [=](u32 reg, Xbyak::Reg32 result) {
+        // Register 0 is supposed to always return 0. NOP is implemented as a store to the zero
+        // register.
+        if (reg == 0) {
+            return;
+        }
+        mov(dword[REGISTERS + reg * sizeof(u32)], result);
+    };
+    auto SetMethodAddress = [=](Xbyak::Reg32 reg) { mov(METHOD_ADDRESS, reg); };
+
+    switch (operation) {
+    case Macro::ResultOperation::IgnoreAndFetch:
+        SetRegister(reg, Compile_FetchParameter());
+        break;
+    case Macro::ResultOperation::Move:
+        SetRegister(reg, RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSetMethod:
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        break;
+    case Macro::ResultOperation::FetchAndSend:
+        // Fetch parameter and send result.
+        SetRegister(reg, Compile_FetchParameter());
+        Compile_Send(RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSend:
+        // Move and send result.
+        SetRegister(reg, RESULT);
+        Compile_Send(RESULT);
+        break;
+    case Macro::ResultOperation::FetchAndSetMethod:
+        // Fetch parameter and use result as Method Address.
+        SetRegister(reg, Compile_FetchParameter());
+        SetMethodAddress(RESULT);
+        break;
+    case Macro::ResultOperation::MoveAndSetMethodFetchAndSend:
+        // Move result and use as Method Address, then fetch and send parameter.
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        Compile_Send(Compile_FetchParameter());
+        break;
+    case Macro::ResultOperation::MoveAndSetMethodSend:
+        // Move result and use as Method Address, then send bits 12:17 of result.
+        SetRegister(reg, RESULT);
+        SetMethodAddress(RESULT);
+        shr(RESULT, 12);
+        and_(RESULT, 0b111111);
+        Compile_Send(RESULT);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented macro operation {}", static_cast<std::size_t>(operation));
+    }
+}
+
+Macro::Opcode MacroJITx64Impl::GetOpCode() const {
+    ASSERT(pc < code.size());
+    return {code[pc]};
+}
+
+std::bitset<32> MacroJITx64Impl::PersistentCallerSavedRegs() const {
+    return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED;
+}
+
+} // namespace Tegra
diff --git a/src/video_core/macro/macro_jit_x64.h b/src/video_core/macro/macro_jit_x64.h
new file mode 100644
index 000000000..71f738b9a
--- /dev/null
+++ b/src/video_core/macro/macro_jit_x64.h
@@ -0,0 +1,100 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <bitset>
+#include <xbyak.h>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "common/x64/xbyak_abi.h"
+#include "video_core/macro/macro.h"
+
+namespace Tegra {
+
+namespace Engines {
+class Maxwell3D;
+}
+
+/// MAX_CODE_SIZE is arbitrarily chosen based on current booting games
+constexpr size_t MAX_CODE_SIZE = 0x10000;
+
+class MacroJITx64 final : public MacroEngine {
+public:
+    explicit MacroJITx64(Engines::Maxwell3D& maxwell3d);
+
+protected:
+    std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override;
+
+private:
+    Engines::Maxwell3D& maxwell3d;
+};
+
+class MacroJITx64Impl : public Xbyak::CodeGenerator, public CachedMacro {
+public:
+    MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code);
+    ~MacroJITx64Impl();
+
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
+
+    void Compile_ALU(Macro::Opcode opcode);
+    void Compile_AddImmediate(Macro::Opcode opcode);
+    void Compile_ExtractInsert(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftRegister(Macro::Opcode opcode);
+    void Compile_Read(Macro::Opcode opcode);
+    void Compile_Branch(Macro::Opcode opcode);
+
+private:
+    void Optimizer_ScanFlags();
+
+    void Compile();
+    bool Compile_NextInstruction();
+
+    Xbyak::Reg32 Compile_FetchParameter();
+    Xbyak::Reg32 Compile_GetRegister(u32 index, Xbyak::Reg32 dst);
+    Xbyak::Reg64 Compile_GetRegister(u32 index, Xbyak::Reg64 dst);
+    void Compile_WriteCarry(Xbyak::Reg64 dst);
+
+    void Compile_ProcessResult(Macro::ResultOperation operation, u32 reg);
+    void Compile_Send(Xbyak::Reg32 value);
+
+    Macro::Opcode GetOpCode() const;
+    std::bitset<32> PersistentCallerSavedRegs() const;
+
+    struct JITState {
+        Engines::Maxwell3D* maxwell3d{};
+        std::array<u32, Macro::NUM_MACRO_REGISTERS> registers{};
+        const u32* parameters{};
+        u32 carry_flag{};
+    };
+    static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0");
+    using ProgramType = void (*)(JITState*);
+
+    struct OptimizerState {
+        bool can_skip_carry{};
+        bool has_delayed_pc{};
+        bool zero_reg_skip{};
+        bool skip_dummy_addimmediate{};
+        bool optimize_for_method_move{};
+    };
+    OptimizerState optimizer{};
+
+    std::optional<Macro::Opcode> next_opcode{};
+    ProgramType program{nullptr};
+
+    std::array<Xbyak::Label, MAX_CODE_SIZE> labels;
+    std::array<Xbyak::Label, MAX_CODE_SIZE> delay_skip;
+    Xbyak::Label end_of_code{};
+
+    bool is_delay_slot{};
+    u32 pc{};
+    std::optional<u32> delayed_pc;
+
+    const std::vector<u32>& code;
+    Engines::Maxwell3D& maxwell3d;
+};
+
+} // namespace Tegra
diff --git a/src/video_core/rasterizer_cache.cpp b/src/video_core/rasterizer_cache.cpp
deleted file mode 100644
index 093b2cdf4..000000000
--- a/src/video_core/rasterizer_cache.cpp
+++ /dev/null
@@ -1,7 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "video_core/rasterizer_cache.h"
-
-RasterizerCacheObject::~RasterizerCacheObject() = default;
diff --git a/src/video_core/rasterizer_cache.h b/src/video_core/rasterizer_cache.h
deleted file mode 100644
index 096ee337c..000000000
--- a/src/video_core/rasterizer_cache.h
+++ /dev/null
@@ -1,253 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <mutex>
-#include <set>
-#include <unordered_map>
-
-#include <boost/icl/interval_map.hpp>
-#include <boost/range/iterator_range_core.hpp>
-
-#include "common/common_types.h"
-#include "core/settings.h"
-#include "video_core/gpu.h"
-#include "video_core/rasterizer_interface.h"
-
-class RasterizerCacheObject {
-public:
-    explicit RasterizerCacheObject(const VAddr cpu_addr) : cpu_addr{cpu_addr} {}
-
-    virtual ~RasterizerCacheObject();
-
-    VAddr GetCpuAddr() const {
-        return cpu_addr;
-    }
-
-    /// Gets the size of the shader in guest memory, required for cache management
-    virtual std::size_t GetSizeInBytes() const = 0;
-
-    /// Sets whether the cached object should be considered registered
-    void SetIsRegistered(bool registered) {
-        is_registered = registered;
-    }
-
-    /// Returns true if the cached object is registered
-    bool IsRegistered() const {
-        return is_registered;
-    }
-
-    /// Returns true if the cached object is dirty
-    bool IsDirty() const {
-        return is_dirty;
-    }
-
-    /// Returns ticks from when this cached object was last modified
-    u64 GetLastModifiedTicks() const {
-        return last_modified_ticks;
-    }
-
-    /// Marks an object as recently modified, used to specify whether it is clean or dirty
-    template <class T>
-    void MarkAsModified(bool dirty, T& cache) {
-        is_dirty = dirty;
-        last_modified_ticks = cache.GetModifiedTicks();
-    }
-
-    void SetMemoryMarked(bool is_memory_marked_) {
-        is_memory_marked = is_memory_marked_;
-    }
-
-    bool IsMemoryMarked() const {
-        return is_memory_marked;
-    }
-
-    void SetSyncPending(bool is_sync_pending_) {
-        is_sync_pending = is_sync_pending_;
-    }
-
-    bool IsSyncPending() const {
-        return is_sync_pending;
-    }
-
-private:
-    bool is_registered{};      ///< Whether the object is currently registered with the cache
-    bool is_dirty{};           ///< Whether the object is dirty (out of sync with guest memory)
-    bool is_memory_marked{};   ///< Whether the object is marking rasterizer memory.
-    bool is_sync_pending{};    ///< Whether the object is pending deletion.
-    u64 last_modified_ticks{}; ///< When the object was last modified, used for in-order flushing
-    VAddr cpu_addr{};          ///< Cpu address memory, unique from emulated virtual address space
-};
-
-template <class T>
-class RasterizerCache : NonCopyable {
-    friend class RasterizerCacheObject;
-
-public:
-    explicit RasterizerCache(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {}
-
-    /// Write any cached resources overlapping the specified region back to memory
-    void FlushRegion(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
-
-        const auto& objects{GetSortedObjectsFromRegion(addr, size)};
-        for (auto& object : objects) {
-            FlushObject(object);
-        }
-    }
-
-    /// Mark the specified region as being invalidated
-    void InvalidateRegion(VAddr addr, u64 size) {
-        std::lock_guard lock{mutex};
-
-        const auto& objects{GetSortedObjectsFromRegion(addr, size)};
-        for (auto& object : objects) {
-            if (!object->IsRegistered()) {
-                // Skip duplicates
-                continue;
-            }
-            Unregister(object);
-        }
-    }
-
-    void OnCPUWrite(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
-
-        for (const auto& object : GetSortedObjectsFromRegion(addr, size)) {
-            if (object->IsRegistered()) {
-                UnmarkMemory(object);
-                object->SetSyncPending(true);
-                marked_for_unregister.emplace_back(object);
-            }
-        }
-    }
-
-    void SyncGuestHost() {
-        std::lock_guard lock{mutex};
-
-        for (const auto& object : marked_for_unregister) {
-            if (object->IsRegistered()) {
-                object->SetSyncPending(false);
-                Unregister(object);
-            }
-        }
-        marked_for_unregister.clear();
-    }
-
-    /// Invalidates everything in the cache
-    void InvalidateAll() {
-        std::lock_guard lock{mutex};
-
-        while (interval_cache.begin() != interval_cache.end()) {
-            Unregister(*interval_cache.begin()->second.begin());
-        }
-    }
-
-protected:
-    /// Tries to get an object from the cache with the specified cache address
-    T TryGet(VAddr addr) const {
-        const auto iter = map_cache.find(addr);
-        if (iter != map_cache.end())
-            return iter->second;
-        return nullptr;
-    }
-
-    /// Register an object into the cache
-    virtual void Register(const T& object) {
-        std::lock_guard lock{mutex};
-
-        object->SetIsRegistered(true);
-        interval_cache.add({GetInterval(object), ObjectSet{object}});
-        map_cache.insert({object->GetCpuAddr(), object});
-        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), 1);
-        object->SetMemoryMarked(true);
-    }
-
-    /// Unregisters an object from the cache
-    virtual void Unregister(const T& object) {
-        std::lock_guard lock{mutex};
-
-        UnmarkMemory(object);
-        object->SetIsRegistered(false);
-        if (object->IsSyncPending()) {
-            marked_for_unregister.remove(object);
-            object->SetSyncPending(false);
-        }
-        const VAddr addr = object->GetCpuAddr();
-        interval_cache.subtract({GetInterval(object), ObjectSet{object}});
-        map_cache.erase(addr);
-    }
-
-    void UnmarkMemory(const T& object) {
-        if (!object->IsMemoryMarked()) {
-            return;
-        }
-        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), -1);
-        object->SetMemoryMarked(false);
-    }
-
-    /// Returns a ticks counter used for tracking when cached objects were last modified
-    u64 GetModifiedTicks() {
-        std::lock_guard lock{mutex};
-
-        return ++modified_ticks;
-    }
-
-    virtual void FlushObjectInner(const T& object) = 0;
-
-    /// Flushes the specified object, updating appropriate cache state as needed
-    void FlushObject(const T& object) {
-        std::lock_guard lock{mutex};
-
-        if (!object->IsDirty()) {
-            return;
-        }
-        FlushObjectInner(object);
-        object->MarkAsModified(false, *this);
-    }
-
-    std::recursive_mutex mutex;
-
-private:
-    /// Returns a list of cached objects from the specified memory region, ordered by access time
-    std::vector<T> GetSortedObjectsFromRegion(VAddr addr, u64 size) {
-        if (size == 0) {
-            return {};
-        }
-
-        std::vector<T> objects;
-        const ObjectInterval interval{addr, addr + size};
-        for (auto& pair : boost::make_iterator_range(interval_cache.equal_range(interval))) {
-            for (auto& cached_object : pair.second) {
-                if (!cached_object) {
-                    continue;
-                }
-                objects.push_back(cached_object);
-            }
-        }
-
-        std::sort(objects.begin(), objects.end(), [](const T& a, const T& b) -> bool {
-            return a->GetLastModifiedTicks() < b->GetLastModifiedTicks();
-        });
-
-        return objects;
-    }
-
-    using ObjectSet = std::set<T>;
-    using ObjectCache = std::unordered_map<VAddr, T>;
-    using IntervalCache = boost::icl::interval_map<VAddr, ObjectSet>;
-    using ObjectInterval = typename IntervalCache::interval_type;
-
-    static auto GetInterval(const T& object) {
-        return ObjectInterval::right_open(object->GetCpuAddr(),
-                                          object->GetCpuAddr() + object->GetSizeInBytes());
-    }
-
-    ObjectCache map_cache;
-    IntervalCache interval_cache; ///< Cache of objects
-    u64 modified_ticks{};         ///< Counter of cache state ticks, used for in-order flushing
-    VideoCore::RasterizerInterface& rasterizer;
-    std::list<T> marked_for_unregister;
-};
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
index 9964ea894..ad0577a4f 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
@@ -22,13 +22,12 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128));
 
-CachedBufferBlock::CachedBufferBlock(VAddr cpu_addr, const std::size_t size)
-    : VideoCommon::BufferBlock{cpu_addr, size} {
+Buffer::Buffer(VAddr cpu_addr, const std::size_t size) : VideoCommon::BufferBlock{cpu_addr, size} {
     gl_buffer.Create();
     glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
 }
 
-CachedBufferBlock::~CachedBufferBlock() = default;
+Buffer::~Buffer() = default;
 
 OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
                                const Device& device, std::size_t stream_size)
@@ -48,12 +47,8 @@ OGLBufferCache::~OGLBufferCache() {
     glDeleteBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs));
 }
 
-Buffer OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<CachedBufferBlock>(cpu_addr, size);
-}
-
-GLuint OGLBufferCache::ToHandle(const Buffer& buffer) {
-    return buffer->GetHandle();
+std::shared_ptr<Buffer> OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
+    return std::make_shared<Buffer>(cpu_addr, size);
 }
 
 GLuint OGLBufferCache::GetEmptyBuffer(std::size_t) {
@@ -62,7 +57,7 @@ GLuint OGLBufferCache::GetEmptyBuffer(std::size_t) {
 
 void OGLBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                                      const u8* data) {
-    glNamedBufferSubData(buffer->GetHandle(), static_cast<GLintptr>(offset),
+    glNamedBufferSubData(buffer.Handle(), static_cast<GLintptr>(offset),
                          static_cast<GLsizeiptr>(size), data);
 }
 
@@ -70,20 +65,20 @@ void OGLBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
                                        u8* data) {
     MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
     glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
-    glGetNamedBufferSubData(buffer->GetHandle(), static_cast<GLintptr>(offset),
+    glGetNamedBufferSubData(buffer.Handle(), static_cast<GLintptr>(offset),
                             static_cast<GLsizeiptr>(size), data);
 }
 
 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),
+    glCopyNamedBufferSubData(src.Handle(), dst.Handle(), static_cast<GLintptr>(src_offset),
                              static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size));
 }
 
 OGLBufferCache::BufferInfo OGLBufferCache::ConstBufferUpload(const void* raw_pointer,
                                                              std::size_t size) {
     DEBUG_ASSERT(cbuf_cursor < std::size(cbufs));
-    const GLuint& cbuf = cbufs[cbuf_cursor++];
+    const GLuint cbuf = cbufs[cbuf_cursor++];
     glNamedBufferSubData(cbuf, 0, static_cast<GLsizeiptr>(size), raw_pointer);
     return {cbuf, 0};
 }
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index a9e86cfc7..a49aaf9c4 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -10,7 +10,6 @@
 #include "common/common_types.h"
 #include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/engines/maxwell_3d.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"
 
@@ -24,17 +23,12 @@ class Device;
 class OGLStreamBuffer;
 class RasterizerOpenGL;
 
-class CachedBufferBlock;
-
-using Buffer = std::shared_ptr<CachedBufferBlock>;
-using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
-
-class CachedBufferBlock : public VideoCommon::BufferBlock {
+class Buffer : public VideoCommon::BufferBlock {
 public:
-    explicit CachedBufferBlock(VAddr cpu_addr, const std::size_t size);
-    ~CachedBufferBlock();
+    explicit Buffer(VAddr cpu_addr, const std::size_t size);
+    ~Buffer();
 
-    GLuint GetHandle() const {
+    GLuint Handle() const {
         return gl_buffer.handle;
     }
 
@@ -42,6 +36,7 @@ private:
     OGLBuffer gl_buffer;
 };
 
+using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
 class OGLBufferCache final : public GenericBufferCache {
 public:
     explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
@@ -55,9 +50,7 @@ public:
     }
 
 protected:
-    Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
-
-    GLuint ToHandle(const Buffer& buffer) override;
+    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
 
     void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                          const u8* data) override;
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index b772c37d9..890fc6c63 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -185,12 +185,20 @@ bool IsASTCSupported() {
 Device::Device()
     : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} {
     const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR));
-    const auto renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER));
+    const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION));
     const std::vector extensions = GetExtensions();
 
     const bool is_nvidia = vendor == "NVIDIA Corporation";
     const bool is_amd = vendor == "ATI Technologies Inc.";
 
+    bool disable_fast_buffer_sub_data = false;
+    if (is_nvidia && version == "4.6.0 NVIDIA 443.24") {
+        LOG_WARNING(
+            Render_OpenGL,
+            "Beta driver 443.24 is known to have issues. There might be performance issues.");
+        disable_fast_buffer_sub_data = true;
+    }
+
     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);
@@ -204,9 +212,10 @@ Device::Device()
     has_variable_aoffi = TestVariableAoffi();
     has_component_indexing_bug = is_amd;
     has_precise_bug = TestPreciseBug();
-    has_fast_buffer_sub_data = is_nvidia;
+    has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data;
     use_assembly_shaders = Settings::values.use_assembly_shaders && GLAD_GL_NV_gpu_program5 &&
-                           GLAD_GL_NV_compute_program5;
+                           GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback &&
+                           GLAD_GL_NV_transform_feedback2;
 
     LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
     LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug);
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 55e79aaf6..2d6c11320 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -30,6 +30,7 @@
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/maxwell_to_gl.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
+#include "video_core/shader_cache.h"
 
 namespace OpenGL {
 
@@ -65,10 +66,22 @@ constexpr std::size_t NumSupportedVertexAttributes = 16;
 template <typename Engine, typename Entry>
 Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
                                                ShaderType shader_type, std::size_t index = 0) {
+    if constexpr (std::is_same_v<Entry, SamplerEntry>) {
+        if (entry.is_separated) {
+            const u32 buffer_1 = entry.buffer;
+            const u32 buffer_2 = entry.secondary_buffer;
+            const u32 offset_1 = entry.offset;
+            const u32 offset_2 = entry.secondary_offset;
+            const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
+            const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
+            return engine.GetTextureInfo(handle_1 | handle_2);
+        }
+    }
     if (entry.is_bindless) {
-        const auto tex_handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
-        return engine.GetTextureInfo(tex_handle);
+        const u32 handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
+        return engine.GetTextureInfo(handle);
     }
+
     const auto& gpu_profile = engine.AccessGuestDriverProfile();
     const u32 offset = entry.offset + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
     if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
@@ -93,6 +106,34 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
     return buffer.size;
 }
 
+/// Translates hardware transform feedback indices
+/// @param location Hardware location
+/// @return Pair of ARB_transform_feedback3 token stream first and third arguments
+/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt
+std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) {
+    const u8 index = location / 4;
+    if (index >= 8 && index <= 39) {
+        return {GL_GENERIC_ATTRIB_NV, index - 8};
+    }
+    if (index >= 48 && index <= 55) {
+        return {GL_TEXTURE_COORD_NV, index - 48};
+    }
+    switch (index) {
+    case 7:
+        return {GL_POSITION, 0};
+    case 40:
+        return {GL_PRIMARY_COLOR_NV, 0};
+    case 41:
+        return {GL_SECONDARY_COLOR_NV, 0};
+    case 42:
+        return {GL_BACK_PRIMARY_COLOR_NV, 0};
+    case 43:
+        return {GL_BACK_SECONDARY_COLOR_NV, 0};
+    }
+    UNIMPLEMENTED_MSG("index={}", static_cast<int>(index));
+    return {GL_POSITION, 0};
+}
+
 void oglEnable(GLenum cap, bool state) {
     (state ? glEnable : glDisable)(cap);
 }
@@ -282,7 +323,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
             continue;
         }
 
-        Shader shader{shader_cache.GetStageProgram(program)};
+        Shader* const shader = shader_cache.GetStageProgram(program);
 
         if (device.UseAssemblyShaders()) {
             // Check for ARB limitation. We only have 16 SSBOs per context state. To workaround this
@@ -576,7 +617,16 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
                    (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
 
     // Prepare the vertex array.
-    buffer_cache.Map(buffer_size);
+    const bool invalidated = buffer_cache.Map(buffer_size);
+
+    if (invalidated) {
+        // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty
+        auto& dirty = gpu.dirty.flags;
+        dirty[Dirty::VertexBuffers] = true;
+        for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) {
+            dirty[index] = true;
+        }
+    }
 
     // Prepare vertex array format.
     SetupVertexFormat();
@@ -842,7 +892,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
     return true;
 }
 
-void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) {
     static constexpr std::array PARAMETER_LUT = {
         GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
         GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV,
@@ -872,7 +922,7 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad
     }
 }
 
-void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
     const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
     const auto& entries = kernel->GetEntries();
@@ -941,7 +991,7 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding,
     }
 }
 
-void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) {
     auto& gpu{system.GPU()};
     auto& memory_manager{gpu.MemoryManager()};
     const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]};
@@ -956,7 +1006,7 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shad
     }
 }
 
-void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) {
     auto& gpu{system.GPU()};
     auto& memory_manager{gpu.MemoryManager()};
     const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
@@ -979,7 +1029,7 @@ void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& e
                       static_cast<GLsizeiptr>(size));
 }
 
-void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& maxwell3d = system.GPU().Maxwell3D();
     u32 binding = device.GetBaseBindings(stage_index).sampler;
@@ -992,7 +1042,7 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader&
     }
 }
 
-void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& compute = system.GPU().KeplerCompute();
     u32 binding = 0;
@@ -1021,7 +1071,7 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu
     }
 }
 
-void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& shader) {
+void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) {
     const auto& maxwell3d = system.GPU().Maxwell3D();
     u32 binding = device.GetBaseBindings(stage_index).image;
     for (const auto& entry : shader->GetEntries().images) {
@@ -1031,7 +1081,7 @@ void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& sh
     }
 }
 
-void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
+void RasterizerOpenGL::SetupComputeImages(Shader* shader) {
     const auto& compute = system.GPU().KeplerCompute();
     u32 binding = 0;
     for (const auto& entry : shader->GetEntries().images) {
@@ -1547,12 +1597,70 @@ void RasterizerOpenGL::SyncFramebufferSRGB() {
     oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb);
 }
 
+void RasterizerOpenGL::SyncTransformFeedback() {
+    // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal
+    // when this is required.
+    const auto& regs = system.GPU().Maxwell3D().regs;
+
+    static constexpr std::size_t STRIDE = 3;
+    std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs;
+    std::array<GLint, Maxwell::NumTransformFeedbackBuffers> streams;
+
+    GLint* cursor = attribs.data();
+    GLint* current_stream = streams.data();
+
+    for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) {
+        const auto& layout = regs.tfb_layouts[feedback];
+        UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding");
+        if (layout.varying_count == 0) {
+            continue;
+        }
+
+        *current_stream = static_cast<GLint>(feedback);
+        if (current_stream != streams.data()) {
+            // When stepping one stream, push the expected token
+            cursor[0] = GL_NEXT_BUFFER_NV;
+            cursor[1] = 0;
+            cursor[2] = 0;
+            cursor += STRIDE;
+        }
+        ++current_stream;
+
+        const auto& locations = regs.tfb_varying_locs[feedback];
+        std::optional<u8> current_index;
+        for (u32 offset = 0; offset < layout.varying_count; ++offset) {
+            const u8 location = locations[offset];
+            const u8 index = location / 4;
+
+            if (current_index == index) {
+                // Increase number of components of the previous attachment
+                ++cursor[-2];
+                continue;
+            }
+            current_index = index;
+
+            std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location);
+            cursor[1] = 1;
+            cursor += STRIDE;
+        }
+    }
+
+    const GLsizei num_attribs = static_cast<GLsizei>((cursor - attribs.data()) / STRIDE);
+    const GLsizei num_strides = static_cast<GLsizei>(current_stream - streams.data());
+    glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(),
+                                       GL_INTERLEAVED_ATTRIBS);
+}
+
 void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
     const auto& regs = system.GPU().Maxwell3D().regs;
     if (regs.tfb_enabled == 0) {
         return;
     }
 
+    if (device.UseAssemblyShaders()) {
+        SyncTransformFeedback();
+    }
+
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
@@ -1579,6 +1687,10 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
                           static_cast<GLsizeiptr>(size));
     }
 
+    // We may have to call BeginTransformFeedbackNV here since they seem to call different
+    // implementations on Nvidia's driver (the pointer is different) but we are using
+    // ARB_transform_feedback3 features with NV_transform_feedback interactions and the ARB
+    // extension doesn't define BeginTransformFeedback (without NV) interactions. It just works.
     glBeginTransformFeedback(GL_POINTS);
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index f5dc56a0e..4f082592f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -19,7 +19,6 @@
 #include "video_core/engines/const_buffer_info.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/rasterizer_accelerated.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
@@ -100,10 +99,10 @@ private:
     void ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil);
 
     /// Configures the current constbuffers to use for the draw command.
-    void SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader);
+    void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader);
 
     /// Configures the current constbuffers to use for the kernel invocation.
-    void SetupComputeConstBuffers(const Shader& kernel);
+    void SetupComputeConstBuffers(Shader* kernel);
 
     /// Configures a constant buffer.
     void SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer,
@@ -111,30 +110,30 @@ private:
                           std::size_t unified_offset);
 
     /// Configures the current global memory entries to use for the draw command.
-    void SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader);
+    void SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader);
 
     /// Configures the current global memory entries to use for the kernel invocation.
-    void SetupComputeGlobalMemory(const Shader& kernel);
+    void SetupComputeGlobalMemory(Shader* kernel);
 
     /// Configures a constant buffer.
     void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
                            std::size_t size);
 
     /// Configures the current textures to use for the draw command.
-    void SetupDrawTextures(std::size_t stage_index, const Shader& shader);
+    void SetupDrawTextures(std::size_t stage_index, Shader* shader);
 
     /// Configures the textures used in a compute shader.
-    void SetupComputeTextures(const Shader& kernel);
+    void SetupComputeTextures(Shader* kernel);
 
     /// Configures a texture.
     void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
                       const SamplerEntry& entry);
 
     /// Configures images in a graphics shader.
-    void SetupDrawImages(std::size_t stage_index, const Shader& shader);
+    void SetupDrawImages(std::size_t stage_index, Shader* shader);
 
     /// Configures images in a compute shader.
-    void SetupComputeImages(const Shader& shader);
+    void SetupComputeImages(Shader* shader);
 
     /// Configures an image.
     void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
@@ -202,6 +201,10 @@ private:
     /// Syncs the framebuffer sRGB state to match the guest state
     void SyncFramebufferSRGB();
 
+    /// Syncs transform feedback state to match guest state
+    /// @note Only valid on assembly shaders
+    void SyncTransformFeedback();
+
     /// Begin a transform feedback
     void BeginTransformFeedback(GLenum primitive_mode);
 
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index a991ca64a..c28486b1d 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -29,6 +29,7 @@
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace OpenGL {
 
@@ -194,12 +195,9 @@ std::unordered_set<GLenum> GetSupportedFormats() {
 
 } // Anonymous namespace
 
-CachedShader::CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
-                           std::shared_ptr<VideoCommon::Shader::Registry> registry,
-                           ShaderEntries entries, ProgramSharedPtr program_)
-    : RasterizerCacheObject{cpu_addr}, registry{std::move(registry)}, entries{std::move(entries)},
-      size_in_bytes{size_in_bytes}, program{std::move(program_)} {
-    // Assign either the assembly program or source program. We can't have both.
+Shader::Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry_, ShaderEntries entries_,
+               ProgramSharedPtr program_)
+    : registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)} {
     handle = program->assembly_program.handle;
     if (handle == 0) {
         handle = program->source_program.handle;
@@ -207,16 +205,16 @@ CachedShader::CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
     ASSERT(handle != 0);
 }
 
-CachedShader::~CachedShader() = default;
+Shader::~Shader() = default;
 
-GLuint CachedShader::GetHandle() const {
+GLuint Shader::GetHandle() const {
     DEBUG_ASSERT(registry->IsConsistent());
     return handle;
 }
 
-Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
-                                           Maxwell::ShaderProgram program_type, ProgramCode code,
-                                           ProgramCode code_b) {
+std::unique_ptr<Shader> Shader::CreateStageFromMemory(const ShaderParameters& params,
+                                                      Maxwell::ShaderProgram program_type,
+                                                      ProgramCode code, ProgramCode code_b) {
     const auto shader_type = GetShaderType(program_type);
     const std::size_t size_in_bytes = code.size() * sizeof(u64);
 
@@ -241,12 +239,12 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
     entry.bindless_samplers = registry->GetBindlessSamplers();
     params.disk_cache.SaveEntry(std::move(entry));
 
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params.cpu_addr, size_in_bytes, std::move(registry),
-                         MakeEntries(params.device, ir, shader_type), std::move(program)));
+    return std::unique_ptr<Shader>(new Shader(
+        std::move(registry), MakeEntries(params.device, ir, shader_type), std::move(program)));
 }
 
-Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) {
+std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& params,
+                                                       ProgramCode code) {
     const std::size_t size_in_bytes = code.size() * sizeof(u64);
 
     auto& engine = params.system.GPU().KeplerCompute();
@@ -266,23 +264,23 @@ Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, Prog
     entry.bindless_samplers = registry->GetBindlessSamplers();
     params.disk_cache.SaveEntry(std::move(entry));
 
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params.cpu_addr, size_in_bytes, std::move(registry),
-                         MakeEntries(params.device, ir, ShaderType::Compute), std::move(program)));
+    return std::unique_ptr<Shader>(new Shader(std::move(registry),
+                                              MakeEntries(params.device, ir, ShaderType::Compute),
+                                              std::move(program)));
 }
 
-Shader CachedShader::CreateFromCache(const ShaderParameters& params,
-                                     const PrecompiledShader& precompiled_shader,
-                                     std::size_t size_in_bytes) {
-    return std::shared_ptr<CachedShader>(
-        new CachedShader(params.cpu_addr, size_in_bytes, precompiled_shader.registry,
-                         precompiled_shader.entries, precompiled_shader.program));
+std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params,
+                                                const PrecompiledShader& precompiled_shader) {
+    return std::unique_ptr<Shader>(new Shader(
+        precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program));
 }
 
 ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
                                      Core::Frontend::EmuWindow& emu_window, const Device& device)
-    : RasterizerCache{rasterizer}, system{system}, emu_window{emu_window}, device{device},
-      disk_cache{system} {}
+    : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system},
+      emu_window{emu_window}, device{device}, disk_cache{system} {}
+
+ShaderCacheOpenGL::~ShaderCacheOpenGL() = default;
 
 void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
                                       const VideoCore::DiskResourceLoadCallback& callback) {
@@ -436,7 +434,7 @@ ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram(
     return program;
 }
 
-Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
+Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     if (!system.GPU().Maxwell3D().dirty.flags[Dirty::Shaders]) {
         return last_shaders[static_cast<std::size_t>(program)];
     }
@@ -446,8 +444,7 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
 
     // Look up shader in the cache based on address
     const auto cpu_addr{memory_manager.GpuToCpuAddress(address)};
-    Shader shader{cpu_addr ? TryGet(*cpu_addr) : null_shader};
-    if (shader) {
+    if (Shader* const shader{cpu_addr ? TryGet(*cpu_addr) : null_shader.get()}) {
         return last_shaders[static_cast<std::size_t>(program)] = shader;
     }
 
@@ -468,30 +465,29 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     const ShaderParameters params{system,    disk_cache, device,
                                   *cpu_addr, host_ptr,   unique_identifier};
 
+    std::unique_ptr<Shader> shader;
     const auto found = runtime_cache.find(unique_identifier);
     if (found == runtime_cache.end()) {
-        shader = CachedShader::CreateStageFromMemory(params, program, std::move(code),
-                                                     std::move(code_b));
+        shader = Shader::CreateStageFromMemory(params, program, std::move(code), std::move(code_b));
     } else {
-        const std::size_t size_in_bytes = code.size() * sizeof(u64);
-        shader = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
+        shader = Shader::CreateFromCache(params, found->second);
     }
 
+    Shader* const result = shader.get();
     if (cpu_addr) {
-        Register(shader);
+        Register(std::move(shader), *cpu_addr, code.size() * sizeof(u64));
     } else {
-        null_shader = shader;
+        null_shader = std::move(shader);
     }
 
-    return last_shaders[static_cast<std::size_t>(program)] = shader;
+    return last_shaders[static_cast<std::size_t>(program)] = result;
 }
 
-Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
+Shader* ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
     auto& memory_manager{system.GPU().MemoryManager()};
     const auto cpu_addr{memory_manager.GpuToCpuAddress(code_addr)};
 
-    auto kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
-    if (kernel) {
+    if (Shader* const kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get()) {
         return kernel;
     }
 
@@ -503,20 +499,21 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
     const ShaderParameters params{system,    disk_cache, device,
                                   *cpu_addr, host_ptr,   unique_identifier};
 
+    std::unique_ptr<Shader> kernel;
     const auto found = runtime_cache.find(unique_identifier);
     if (found == runtime_cache.end()) {
-        kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
+        kernel = Shader::CreateKernelFromMemory(params, std::move(code));
     } else {
-        const std::size_t size_in_bytes = code.size() * sizeof(u64);
-        kernel = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
+        kernel = Shader::CreateFromCache(params, found->second);
     }
 
+    Shader* const result = kernel.get();
     if (cpu_addr) {
-        Register(kernel);
+        Register(std::move(kernel), *cpu_addr, code.size() * sizeof(u64));
     } else {
-        null_kernel = kernel;
+        null_kernel = std::move(kernel);
     }
-    return kernel;
+    return result;
 }
 
 } // 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 b2ae8d7f9..6848f1388 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -18,12 +18,12 @@
 
 #include "common/common_types.h"
 #include "video_core/engines/shader_type.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_disk_cache.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace Core {
 class System;
@@ -35,12 +35,10 @@ class EmuWindow;
 
 namespace OpenGL {
 
-class CachedShader;
 class Device;
 class RasterizerOpenGL;
 struct UnspecializedShader;
 
-using Shader = std::shared_ptr<CachedShader>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct ProgramHandle {
@@ -64,62 +62,53 @@ struct ShaderParameters {
     u64 unique_identifier;
 };
 
-class CachedShader final : public RasterizerCacheObject {
+class Shader final {
 public:
-    ~CachedShader();
+    ~Shader();
 
     /// Gets the GL program handle for the shader
     GLuint GetHandle() const;
 
-    /// Returns the size in bytes of the shader
-    std::size_t GetSizeInBytes() const override {
-        return size_in_bytes;
-    }
-
     /// Gets the shader entries for the shader
     const ShaderEntries& GetEntries() const {
         return entries;
     }
 
-    static Shader CreateStageFromMemory(const ShaderParameters& params,
-                                        Maxwell::ShaderProgram program_type,
-                                        ProgramCode program_code, ProgramCode program_code_b);
-    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code);
+    static std::unique_ptr<Shader> CreateStageFromMemory(const ShaderParameters& params,
+                                                         Maxwell::ShaderProgram program_type,
+                                                         ProgramCode program_code,
+                                                         ProgramCode program_code_b);
+    static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params,
+                                                          ProgramCode code);
 
-    static Shader CreateFromCache(const ShaderParameters& params,
-                                  const PrecompiledShader& precompiled_shader,
-                                  std::size_t size_in_bytes);
+    static std::unique_ptr<Shader> CreateFromCache(const ShaderParameters& params,
+                                                   const PrecompiledShader& precompiled_shader);
 
 private:
-    explicit CachedShader(VAddr cpu_addr, std::size_t size_in_bytes,
-                          std::shared_ptr<VideoCommon::Shader::Registry> registry,
-                          ShaderEntries entries, ProgramSharedPtr program);
+    explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries,
+                    ProgramSharedPtr program);
 
     std::shared_ptr<VideoCommon::Shader::Registry> registry;
     ShaderEntries entries;
-    std::size_t size_in_bytes = 0;
     ProgramSharedPtr program;
     GLuint handle = 0;
 };
 
-class ShaderCacheOpenGL final : public RasterizerCache<Shader> {
+class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> {
 public:
     explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
                                Core::Frontend::EmuWindow& emu_window, const Device& device);
+    ~ShaderCacheOpenGL() override;
 
     /// Loads disk cache for the current game
     void LoadDiskCache(const std::atomic_bool& stop_loading,
                        const VideoCore::DiskResourceLoadCallback& callback);
 
     /// Gets the current specified shader stage program
-    Shader GetStageProgram(Maxwell::ShaderProgram 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 {}
+    Shader* GetComputeKernel(GPUVAddr code_addr);
 
 private:
     ProgramSharedPtr GeneratePrecompiledProgram(
@@ -132,10 +121,10 @@ private:
     ShaderDiskCacheOpenGL disk_cache;
     std::unordered_map<u64, PrecompiledShader> runtime_cache;
 
-    Shader null_shader{};
-    Shader null_kernel{};
+    std::unique_ptr<Shader> null_shader;
+    std::unique_ptr<Shader> null_kernel;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 502b95973..d6e30b321 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -626,7 +626,9 @@ private:
                 break;
             }
         }
-        if (stage != ShaderType::Vertex || device.HasVertexViewportLayer()) {
+
+        if (stage != ShaderType::Geometry &&
+            (stage != ShaderType::Vertex || device.HasVertexViewportLayer())) {
             if (ir.UsesLayer()) {
                 code.AddLine("int gl_Layer;");
             }
@@ -655,6 +657,16 @@ private:
         --code.scope;
         code.AddLine("}};");
         code.AddNewLine();
+
+        if (stage == ShaderType::Geometry) {
+            if (ir.UsesLayer()) {
+                code.AddLine("out int gl_Layer;");
+            }
+            if (ir.UsesViewportIndex()) {
+                code.AddLine("out int gl_ViewportIndex;");
+            }
+        }
+        code.AddNewLine();
     }
 
     void DeclareRegisters() {
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 9e95a122b..653c3f2f9 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
@@ -29,6 +29,8 @@ using VideoCommon::Shader::KeyMap;
 
 namespace {
 
+using VideoCommon::Shader::SeparateSamplerKey;
+
 using ShaderCacheVersionHash = std::array<u8, 64>;
 
 struct ConstBufferKey {
@@ -37,18 +39,26 @@ struct ConstBufferKey {
     u32 value = 0;
 };
 
-struct BoundSamplerKey {
+struct BoundSamplerEntry {
     u32 offset = 0;
     Tegra::Engines::SamplerDescriptor sampler;
 };
 
-struct BindlessSamplerKey {
+struct SeparateSamplerEntry {
+    u32 cbuf1 = 0;
+    u32 cbuf2 = 0;
+    u32 offset1 = 0;
+    u32 offset2 = 0;
+    Tegra::Engines::SamplerDescriptor sampler;
+};
+
+struct BindlessSamplerEntry {
     u32 cbuf = 0;
     u32 offset = 0;
     Tegra::Engines::SamplerDescriptor sampler;
 };
 
-constexpr u32 NativeVersion = 20;
+constexpr u32 NativeVersion = 21;
 
 ShaderCacheVersionHash GetShaderCacheVersionHash() {
     ShaderCacheVersionHash hash{};
@@ -87,12 +97,14 @@ bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) {
     u32 texture_handler_size_value;
     u32 num_keys;
     u32 num_bound_samplers;
+    u32 num_separate_samplers;
     u32 num_bindless_samplers;
     if (file.ReadArray(&unique_identifier, 1) != 1 || file.ReadArray(&bound_buffer, 1) != 1 ||
         file.ReadArray(&is_texture_handler_size_known, 1) != 1 ||
         file.ReadArray(&texture_handler_size_value, 1) != 1 ||
         file.ReadArray(&graphics_info, 1) != 1 || file.ReadArray(&compute_info, 1) != 1 ||
         file.ReadArray(&num_keys, 1) != 1 || file.ReadArray(&num_bound_samplers, 1) != 1 ||
+        file.ReadArray(&num_separate_samplers, 1) != 1 ||
         file.ReadArray(&num_bindless_samplers, 1) != 1) {
         return false;
     }
@@ -101,23 +113,32 @@ bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) {
     }
 
     std::vector<ConstBufferKey> flat_keys(num_keys);
-    std::vector<BoundSamplerKey> flat_bound_samplers(num_bound_samplers);
-    std::vector<BindlessSamplerKey> flat_bindless_samplers(num_bindless_samplers);
+    std::vector<BoundSamplerEntry> flat_bound_samplers(num_bound_samplers);
+    std::vector<SeparateSamplerEntry> flat_separate_samplers(num_separate_samplers);
+    std::vector<BindlessSamplerEntry> flat_bindless_samplers(num_bindless_samplers);
     if (file.ReadArray(flat_keys.data(), flat_keys.size()) != flat_keys.size() ||
         file.ReadArray(flat_bound_samplers.data(), flat_bound_samplers.size()) !=
             flat_bound_samplers.size() ||
+        file.ReadArray(flat_separate_samplers.data(), flat_separate_samplers.size()) !=
+            flat_separate_samplers.size() ||
         file.ReadArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) !=
             flat_bindless_samplers.size()) {
         return false;
     }
-    for (const auto& key : flat_keys) {
-        keys.insert({{key.cbuf, key.offset}, key.value});
+    for (const auto& entry : flat_keys) {
+        keys.insert({{entry.cbuf, entry.offset}, entry.value});
     }
-    for (const auto& key : flat_bound_samplers) {
-        bound_samplers.emplace(key.offset, key.sampler);
+    for (const auto& entry : flat_bound_samplers) {
+        bound_samplers.emplace(entry.offset, entry.sampler);
     }
-    for (const auto& key : flat_bindless_samplers) {
-        bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
+    for (const auto& entry : flat_separate_samplers) {
+        SeparateSamplerKey key;
+        key.buffers = {entry.cbuf1, entry.cbuf2};
+        key.offsets = {entry.offset1, entry.offset2};
+        separate_samplers.emplace(key, entry.sampler);
+    }
+    for (const auto& entry : flat_bindless_samplers) {
+        bindless_samplers.insert({{entry.cbuf, entry.offset}, entry.sampler});
     }
 
     return true;
@@ -142,6 +163,7 @@ bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const {
         file.WriteObject(graphics_info) != 1 || file.WriteObject(compute_info) != 1 ||
         file.WriteObject(static_cast<u32>(keys.size())) != 1 ||
         file.WriteObject(static_cast<u32>(bound_samplers.size())) != 1 ||
+        file.WriteObject(static_cast<u32>(separate_samplers.size())) != 1 ||
         file.WriteObject(static_cast<u32>(bindless_samplers.size())) != 1) {
         return false;
     }
@@ -152,22 +174,34 @@ bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const {
         flat_keys.push_back(ConstBufferKey{address.first, address.second, value});
     }
 
-    std::vector<BoundSamplerKey> flat_bound_samplers;
+    std::vector<BoundSamplerEntry> flat_bound_samplers;
     flat_bound_samplers.reserve(bound_samplers.size());
     for (const auto& [address, sampler] : bound_samplers) {
-        flat_bound_samplers.push_back(BoundSamplerKey{address, sampler});
+        flat_bound_samplers.push_back(BoundSamplerEntry{address, sampler});
+    }
+
+    std::vector<SeparateSamplerEntry> flat_separate_samplers;
+    flat_separate_samplers.reserve(separate_samplers.size());
+    for (const auto& [key, sampler] : separate_samplers) {
+        SeparateSamplerEntry entry;
+        std::tie(entry.cbuf1, entry.cbuf2) = key.buffers;
+        std::tie(entry.offset1, entry.offset2) = key.offsets;
+        entry.sampler = sampler;
+        flat_separate_samplers.push_back(entry);
     }
 
-    std::vector<BindlessSamplerKey> flat_bindless_samplers;
+    std::vector<BindlessSamplerEntry> flat_bindless_samplers;
     flat_bindless_samplers.reserve(bindless_samplers.size());
     for (const auto& [address, sampler] : bindless_samplers) {
         flat_bindless_samplers.push_back(
-            BindlessSamplerKey{address.first, address.second, sampler});
+            BindlessSamplerEntry{address.first, address.second, sampler});
     }
 
     return file.WriteArray(flat_keys.data(), flat_keys.size()) == flat_keys.size() &&
            file.WriteArray(flat_bound_samplers.data(), flat_bound_samplers.size()) ==
                flat_bound_samplers.size() &&
+           file.WriteArray(flat_separate_samplers.data(), flat_separate_samplers.size()) ==
+               flat_separate_samplers.size() &&
            file.WriteArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) ==
                flat_bindless_samplers.size();
 }
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 d5be52e40..a79cef0e9 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
@@ -57,6 +57,7 @@ struct ShaderDiskCacheEntry {
     VideoCommon::Shader::ComputeInfo compute_info;
     VideoCommon::Shader::KeyMap keys;
     VideoCommon::Shader::BoundSamplerMap bound_samplers;
+    VideoCommon::Shader::SeparateSamplerMap separate_samplers;
     VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
 };
 
diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp
index 6ec328c53..932a2f69e 100644
--- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp
+++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp
@@ -49,14 +49,6 @@ OGLStreamBuffer::~OGLStreamBuffer() {
     gl_buffer.Release();
 }
 
-GLuint OGLStreamBuffer::GetHandle() const {
-    return gl_buffer.handle;
-}
-
-GLsizeiptr OGLStreamBuffer::GetSize() const {
-    return buffer_size;
-}
-
 std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
     ASSERT(size <= buffer_size);
     ASSERT(alignment <= buffer_size);
diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h
index f8383cbd4..866da3594 100644
--- a/src/video_core/renderer_opengl/gl_stream_buffer.h
+++ b/src/video_core/renderer_opengl/gl_stream_buffer.h
@@ -17,9 +17,6 @@ public:
                              bool use_persistent = true);
     ~OGLStreamBuffer();
 
-    GLuint GetHandle() const;
-    GLsizeiptr GetSize() const;
-
     /*
      * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes
      * and the optional alignment requirement.
@@ -32,6 +29,14 @@ public:
 
     void Unmap(GLsizeiptr size);
 
+    GLuint Handle() const {
+        return gl_buffer.handle;
+    }
+
+    GLsizeiptr Size() const {
+        return buffer_size;
+    }
+
 private:
     OGLBuffer gl_buffer;
 
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 57db5a08b..61505879b 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -263,9 +263,14 @@ CachedSurface::CachedSurface(const GPUVAddr gpu_addr, const SurfaceParams& param
     target = GetTextureTarget(params.target);
     texture = CreateTexture(params, target, internal_format, texture_buffer);
     DecorateSurfaceName();
-    main_view = CreateViewInner(
-        ViewParams(params.target, 0, params.is_layered ? params.depth : 1, 0, params.num_levels),
-        true);
+
+    u32 num_layers = 1;
+    if (params.is_layered || params.target == SurfaceTarget::Texture3D) {
+        num_layers = params.depth;
+    }
+
+    main_view =
+        CreateViewInner(ViewParams(params.target, 0, num_layers, 0, params.num_levels), true);
 }
 
 CachedSurface::~CachedSurface() = default;
@@ -413,20 +418,23 @@ CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& p
 
 CachedSurfaceView::~CachedSurfaceView() = default;
 
-void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
+void CachedSurfaceView::Attach(GLenum attachment, GLenum fb_target) const {
     ASSERT(params.num_levels == 1);
 
+    if (params.target == SurfaceTarget::Texture3D) {
+        if (params.num_layers > 1) {
+            ASSERT(params.base_layer == 0);
+            glFramebufferTexture(fb_target, attachment, surface.texture.handle, params.base_level);
+        } else {
+            glFramebufferTexture3D(fb_target, attachment, target, surface.texture.handle,
+                                   params.base_level, params.base_layer);
+        }
+        return;
+    }
+
     if (params.num_layers > 1) {
-        // Layered framebuffer attachments
         UNIMPLEMENTED_IF(params.base_layer != 0);
-
-        switch (params.target) {
-        case SurfaceTarget::Texture2DArray:
-            glFramebufferTexture(target, attachment, GetTexture(), 0);
-            break;
-        default:
-            UNIMPLEMENTED();
-        }
+        glFramebufferTexture(fb_target, attachment, GetTexture(), 0);
         return;
     }
 
@@ -434,16 +442,16 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
     const GLuint texture = surface.GetTexture();
     switch (surface.GetSurfaceParams().target) {
     case SurfaceTarget::Texture1D:
-        glFramebufferTexture1D(target, attachment, view_target, texture, params.base_level);
+        glFramebufferTexture1D(fb_target, attachment, view_target, texture, params.base_level);
         break;
     case SurfaceTarget::Texture2D:
-        glFramebufferTexture2D(target, attachment, view_target, texture, params.base_level);
+        glFramebufferTexture2D(fb_target, attachment, view_target, texture, params.base_level);
         break;
     case SurfaceTarget::Texture1DArray:
     case SurfaceTarget::Texture2DArray:
     case SurfaceTarget::TextureCubemap:
     case SurfaceTarget::TextureCubeArray:
-        glFramebufferTextureLayer(target, attachment, texture, params.base_level,
+        glFramebufferTextureLayer(fb_target, attachment, texture, params.base_level,
                                   params.base_layer);
         break;
     default:
@@ -500,8 +508,13 @@ OGLTextureView CachedSurfaceView::CreateTextureView() const {
     OGLTextureView texture_view;
     texture_view.Create();
 
-    glTextureView(texture_view.handle, target, surface.texture.handle, format, params.base_level,
-                  params.num_levels, params.base_layer, params.num_layers);
+    if (target == GL_TEXTURE_3D) {
+        glTextureView(texture_view.handle, target, surface.texture.handle, format,
+                      params.base_level, params.num_levels, 0, 1);
+    } else {
+        glTextureView(texture_view.handle, target, surface.texture.handle, format,
+                      params.base_level, params.num_levels, params.base_layer, params.num_layers);
+    }
     ApplyTextureDefaults(surface.GetSurfaceParams(), texture_view.handle);
 
     return texture_view;
@@ -544,8 +557,8 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
                                    const Tegra::Engines::Fermi2D::Config& copy_config) {
     const auto& src_params{src_view->GetSurfaceParams()};
     const auto& dst_params{dst_view->GetSurfaceParams()};
-    UNIMPLEMENTED_IF(src_params.target == SurfaceTarget::Texture3D);
-    UNIMPLEMENTED_IF(dst_params.target == SurfaceTarget::Texture3D);
+    UNIMPLEMENTED_IF(src_params.depth != 1);
+    UNIMPLEMENTED_IF(dst_params.depth != 1);
 
     state_tracker.NotifyScissor0();
     state_tracker.NotifyFramebuffer();
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index 8a2ac8603..bfc4ddf5d 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -80,8 +80,10 @@ public:
     explicit CachedSurfaceView(CachedSurface& surface, const ViewParams& params, bool is_proxy);
     ~CachedSurfaceView();
 
-    /// Attaches this texture view to the current bound GL_DRAW_FRAMEBUFFER
-    void Attach(GLenum attachment, GLenum target) const;
+    /// @brief Attaches this texture view to the currently bound fb_target framebuffer
+    /// @param attachment   Attachment to bind textures to
+    /// @param fb_target    Framebuffer target to attach to (e.g. DRAW_FRAMEBUFFER)
+    void Attach(GLenum attachment, GLenum fb_target) const;
 
     GLuint GetTexture(Tegra::Texture::SwizzleSource x_source,
                       Tegra::Texture::SwizzleSource y_source,
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index e7952924a..6214fcbc3 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -751,11 +751,9 @@ void RendererOpenGL::RenderScreenshot() {
 }
 
 bool RendererOpenGL::Init() {
-    if (GLAD_GL_KHR_debug) {
+    if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) {
         glEnable(GL_DEBUG_OUTPUT);
-        if (Settings::values.renderer_debug) {
-            glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
-        }
+        glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
         glDebugMessageCallback(DebugHandler, nullptr);
     }
 
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 5f33d9e40..1fde38328 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -37,8 +37,8 @@ std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKSch
 
 } // Anonymous namespace
 
-CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
-                                     VAddr cpu_addr, std::size_t size)
+Buffer::Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VAddr cpu_addr,
+               std::size_t size)
     : VideoCommon::BufferBlock{cpu_addr, size} {
     VkBufferCreateInfo ci;
     ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
@@ -54,7 +54,7 @@ CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& me
     buffer.commit = memory_manager.Commit(buffer.handle, false);
 }
 
-CachedBufferBlock::~CachedBufferBlock() = default;
+Buffer::~Buffer() = default;
 
 VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
                              const VKDevice& device, VKMemoryManager& memory_manager,
@@ -67,12 +67,8 @@ VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::S
 
 VKBufferCache::~VKBufferCache() = default;
 
-Buffer VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<CachedBufferBlock>(device, memory_manager, cpu_addr, size);
-}
-
-VkBuffer VKBufferCache::ToHandle(const Buffer& buffer) {
-    return buffer->GetHandle();
+std::shared_ptr<Buffer> VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
+    return std::make_shared<Buffer>(device, memory_manager, cpu_addr, size);
 }
 
 VkBuffer VKBufferCache::GetEmptyBuffer(std::size_t size) {
@@ -91,7 +87,7 @@ void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, st
     std::memcpy(staging.commit->Map(size), data, size);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
+    scheduler.Record([staging = *staging.handle, buffer = buffer.Handle(), offset,
                       size](vk::CommandBuffer cmdbuf) {
         cmdbuf.CopyBuffer(staging, buffer, VkBufferCopy{0, offset, size});
 
@@ -114,7 +110,7 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
                                       u8* data) {
     const auto& staging = staging_pool.GetUnusedBuffer(size, true);
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
+    scheduler.Record([staging = *staging.handle, buffer = buffer.Handle(), offset,
                       size](vk::CommandBuffer cmdbuf) {
         VkBufferMemoryBarrier barrier;
         barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
@@ -141,8 +137,8 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
 void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
                               std::size_t dst_offset, std::size_t size) {
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([src_buffer = src->GetHandle(), dst_buffer = dst->GetHandle(), src_offset,
-                      dst_offset, size](vk::CommandBuffer cmdbuf) {
+    scheduler.Record([src_buffer = src.Handle(), dst_buffer = dst.Handle(), src_offset, dst_offset,
+                      size](vk::CommandBuffer cmdbuf) {
         cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size});
 
         std::array<VkBufferMemoryBarrier, 2> barriers;
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index a54583e7d..9ebbef835 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -8,7 +8,6 @@
 
 #include "common/common_types.h"
 #include "video_core/buffer_cache/buffer_cache.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/renderer_vulkan/vk_stream_buffer.h"
@@ -24,13 +23,13 @@ class VKDevice;
 class VKMemoryManager;
 class VKScheduler;
 
-class CachedBufferBlock final : public VideoCommon::BufferBlock {
+class Buffer final : public VideoCommon::BufferBlock {
 public:
-    explicit CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
-                               VAddr cpu_addr, std::size_t size);
-    ~CachedBufferBlock();
+    explicit Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VAddr cpu_addr,
+                    std::size_t size);
+    ~Buffer();
 
-    VkBuffer GetHandle() const {
+    VkBuffer Handle() const {
         return *buffer.handle;
     }
 
@@ -38,8 +37,6 @@ private:
     VKBuffer buffer;
 };
 
-using Buffer = std::shared_ptr<CachedBufferBlock>;
-
 class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
 public:
     explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
@@ -50,9 +47,7 @@ public:
     VkBuffer GetEmptyBuffer(std::size_t size) override;
 
 protected:
-    VkBuffer ToHandle(const Buffer& buffer) override;
-
-    Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
+    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
 
     void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                          const u8* data) override;
diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
index 8e1b46277..281bf9ac3 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
@@ -53,8 +53,9 @@ vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
     };
     add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
     add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
-    add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.texel_buffers.size());
+    add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size());
     add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
+    add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size());
     add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
 
     VkDescriptorSetLayoutCreateInfo ci;
diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
index 890fd52cf..9259b618d 100644
--- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
@@ -42,6 +42,7 @@ vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() {
         {VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60},
         {VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64},
         {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64},
+        {VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64},
         {VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}};
 
     VkDescriptorPoolCreateInfo ci;
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 65a1c6245..ea66e621e 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -27,6 +27,7 @@
 #include "video_core/renderer_vulkan/wrapper.h"
 #include "video_core/shader/compiler_settings.h"
 #include "video_core/shader/memory_util.h"
+#include "video_core/shader_cache.h"
 
 namespace Vulkan {
 
@@ -45,6 +46,7 @@ constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
 constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
 constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
+constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
 constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
 
 constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
@@ -104,8 +106,9 @@ u32 FillDescriptorLayout(const ShaderEntries& entries,
     u32 binding = base_binding;
     AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
     AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
-    AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers);
+    AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels);
     AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
+    AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels);
     AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
     return binding;
 }
@@ -130,19 +133,18 @@ bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) con
     return std::memcmp(&rhs, this, sizeof *this) == 0;
 }
 
-CachedShader::CachedShader(Core::System& system, Tegra::Engines::ShaderType stage,
-                           GPUVAddr gpu_addr, VAddr cpu_addr, ProgramCode program_code,
-                           u32 main_offset)
-    : RasterizerCacheObject{cpu_addr}, gpu_addr{gpu_addr}, program_code{std::move(program_code)},
+Shader::Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
+               VideoCommon::Shader::ProgramCode program_code, u32 main_offset)
+    : gpu_addr{gpu_addr}, program_code{std::move(program_code)},
       registry{stage, GetEngine(system, stage)}, shader_ir{this->program_code, main_offset,
                                                            compiler_settings, registry},
       entries{GenerateShaderEntries(shader_ir)} {}
 
-CachedShader::~CachedShader() = default;
+Shader::~Shader() = default;
 
-Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine(
-    Core::System& system, Tegra::Engines::ShaderType stage) {
-    if (stage == Tegra::Engines::ShaderType::Compute) {
+Tegra::Engines::ConstBufferEngineInterface& Shader::GetEngine(Core::System& system,
+                                                              Tegra::Engines::ShaderType stage) {
+    if (stage == ShaderType::Compute) {
         return system.GPU().KeplerCompute();
     } else {
         return system.GPU().Maxwell3D();
@@ -154,16 +156,16 @@ VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasteri
                                  VKDescriptorPool& descriptor_pool,
                                  VKUpdateDescriptorQueue& update_descriptor_queue,
                                  VKRenderPassCache& renderpass_cache)
-    : RasterizerCache{rasterizer}, system{system}, device{device}, scheduler{scheduler},
-      descriptor_pool{descriptor_pool}, update_descriptor_queue{update_descriptor_queue},
-      renderpass_cache{renderpass_cache} {}
+    : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system}, device{device},
+      scheduler{scheduler}, descriptor_pool{descriptor_pool},
+      update_descriptor_queue{update_descriptor_queue}, renderpass_cache{renderpass_cache} {}
 
 VKPipelineCache::~VKPipelineCache() = default;
 
-std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
+std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
     const auto& gpu = system.GPU().Maxwell3D();
 
-    std::array<Shader, Maxwell::MaxShaderProgram> shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> shaders{};
     for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         const auto program{static_cast<Maxwell::ShaderProgram>(index)};
 
@@ -176,24 +178,28 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
         const GPUVAddr program_addr{GetShaderAddress(system, program)};
         const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
         ASSERT(cpu_addr);
-        auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
-        if (!shader) {
+
+        Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
+        if (!result) {
             const auto host_ptr{memory_manager.GetPointer(program_addr)};
 
             // No shader found - create a new one
             constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
-            const auto stage = static_cast<Tegra::Engines::ShaderType>(index == 0 ? 0 : index - 1);
+            const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1);
             ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
+            const std::size_t size_in_bytes = code.size() * sizeof(u64);
+
+            auto shader = std::make_unique<Shader>(system, stage, program_addr, std::move(code),
+                                                   stage_offset);
+            result = shader.get();
 
-            shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
-                                                    std::move(code), stage_offset);
             if (cpu_addr) {
-                Register(shader);
+                Register(std::move(shader), *cpu_addr, size_in_bytes);
             } else {
-                null_shader = shader;
+                null_shader = std::move(shader);
             }
         }
-        shaders[index] = std::move(shader);
+        shaders[index] = result;
     }
     return last_shaders = shaders;
 }
@@ -234,19 +240,22 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
     const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
     ASSERT(cpu_addr);
 
-    auto shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
+    Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get();
     if (!shader) {
         // No shader found - create a new one
         const auto host_ptr = memory_manager.GetPointer(program_addr);
 
         ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
-        shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
-                                                program_addr, *cpu_addr, std::move(code),
-                                                KERNEL_MAIN_OFFSET);
+        const std::size_t size_in_bytes = code.size() * sizeof(u64);
+
+        auto shader_info = std::make_unique<Shader>(system, ShaderType::Compute, program_addr,
+                                                    std::move(code), KERNEL_MAIN_OFFSET);
+        shader = shader_info.get();
+
         if (cpu_addr) {
-            Register(shader);
+            Register(std::move(shader_info), *cpu_addr, size_in_bytes);
         } else {
-            null_kernel = shader;
+            null_kernel = std::move(shader_info);
         }
     }
 
@@ -262,7 +271,7 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
     return *entry;
 }
 
-void VKPipelineCache::Unregister(const Shader& shader) {
+void VKPipelineCache::OnShaderRemoval(Shader* shader) {
     bool finished = false;
     const auto Finish = [&] {
         // TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
@@ -294,8 +303,6 @@ void VKPipelineCache::Unregister(const Shader& shader) {
         Finish();
         it = compute_cache.erase(it);
     }
-
-    RasterizerCache::Unregister(shader);
 }
 
 std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
@@ -330,12 +337,11 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
         }
 
         const GPUVAddr gpu_addr = GetShaderAddress(system, program_enum);
-        const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
-        const auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
-        ASSERT(shader);
+        const std::optional<VAddr> cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
+        Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
 
         const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
-        const auto program_type = GetShaderType(program_enum);
+        const ShaderType program_type = GetShaderType(program_enum);
         const auto& entries = shader->GetEntries();
         program[stage] = {
             Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
@@ -377,16 +383,17 @@ void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u3
         return;
     }
 
-    if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) {
-        // Nvidia has a bug where updating multiple uniform texels at once causes the driver to
-        // crash.
+    if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER ||
+                  descriptor_type == STORAGE_TEXEL_BUFFER) {
+        // Nvidia has a bug where updating multiple texels at once causes the driver to crash.
+        // Note: Fixed in driver Windows 443.24, Linux 440.66.15
         for (u32 i = 0; i < count; ++i) {
             VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
             entry.dstBinding = binding + i;
             entry.dstArrayElement = 0;
             entry.descriptorCount = 1;
             entry.descriptorType = descriptor_type;
-            entry.offset = offset + i * entry_size;
+            entry.offset = static_cast<std::size_t>(offset + i * entry_size);
             entry.stride = entry_size;
         }
     } else if (count > 0) {
@@ -407,8 +414,9 @@ void FillDescriptorUpdateTemplateEntries(
     std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
     AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
     AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
-    AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers);
+    AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels);
     AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
+    AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels);
     AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
index 0b5796fef..0a36e5112 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
@@ -17,7 +17,6 @@
 #include "common/common_types.h"
 #include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_renderpass_cache.h"
@@ -26,6 +25,7 @@
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader_cache.h"
 
 namespace Core {
 class System;
@@ -41,8 +41,6 @@ class VKFence;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
 
-class CachedShader;
-using Shader = std::shared_ptr<CachedShader>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct GraphicsPipelineCacheKey {
@@ -102,21 +100,16 @@ struct hash<Vulkan::ComputePipelineCacheKey> {
 
 namespace Vulkan {
 
-class CachedShader final : public RasterizerCacheObject {
+class Shader {
 public:
-    explicit CachedShader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
-                          VAddr cpu_addr, VideoCommon::Shader::ProgramCode program_code,
-                          u32 main_offset);
-    ~CachedShader();
+    explicit Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
+                    VideoCommon::Shader::ProgramCode program_code, u32 main_offset);
+    ~Shader();
 
     GPUVAddr GetGpuAddr() const {
         return gpu_addr;
     }
 
-    std::size_t GetSizeInBytes() const override {
-        return program_code.size() * sizeof(u64);
-    }
-
     VideoCommon::Shader::ShaderIR& GetIR() {
         return shader_ir;
     }
@@ -144,25 +137,23 @@ private:
     ShaderEntries entries;
 };
 
-class VKPipelineCache final : public RasterizerCache<Shader> {
+class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> {
 public:
     explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
                              const VKDevice& device, VKScheduler& scheduler,
                              VKDescriptorPool& descriptor_pool,
                              VKUpdateDescriptorQueue& update_descriptor_queue,
                              VKRenderPassCache& renderpass_cache);
-    ~VKPipelineCache();
+    ~VKPipelineCache() override;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> GetShaders();
+    std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders();
 
     VKGraphicsPipeline& GetGraphicsPipeline(const GraphicsPipelineCacheKey& key);
 
     VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
 
 protected:
-    void Unregister(const Shader& shader) override;
-
-    void FlushObjectInner(const Shader& object) override {}
+    void OnShaderRemoval(Shader* shader) final;
 
 private:
     std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders(
@@ -175,10 +166,10 @@ private:
     VKUpdateDescriptorQueue& update_descriptor_queue;
     VKRenderPassCache& renderpass_cache;
 
-    Shader null_shader{};
-    Shader null_kernel{};
+    std::unique_ptr<Shader> null_shader;
+    std::unique_ptr<Shader> null_kernel;
 
-    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+    std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
 
     GraphicsPipelineCacheKey last_graphics_key;
     VKGraphicsPipeline* last_graphics_pipeline = nullptr;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index a3d992ed3..184b2238a 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -38,6 +38,7 @@
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/renderer_vulkan/wrapper.h"
+#include "video_core/shader_cache.h"
 
 namespace Vulkan {
 
@@ -98,7 +99,7 @@ VkRect2D GetScissorState(const Maxwell& regs, std::size_t index) {
 }
 
 std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
-    const std::array<Shader, Maxwell::MaxShaderProgram>& shaders) {
+    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
     std::array<GPUVAddr, Maxwell::MaxShaderProgram> addresses;
     for (std::size_t i = 0; i < std::size(addresses); ++i) {
         addresses[i] = shaders[i] ? shaders[i]->GetGpuAddr() : 0;
@@ -117,6 +118,17 @@ template <typename Engine, typename Entry>
 Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
                                                std::size_t stage, std::size_t index = 0) {
     const auto stage_type = static_cast<Tegra::Engines::ShaderType>(stage);
+    if constexpr (std::is_same_v<Entry, SamplerEntry>) {
+        if (entry.is_separated) {
+            const u32 buffer_1 = entry.buffer;
+            const u32 buffer_2 = entry.secondary_buffer;
+            const u32 offset_1 = entry.offset;
+            const u32 offset_2 = entry.secondary_offset;
+            const u32 handle_1 = engine.AccessConstBuffer32(stage_type, buffer_1, offset_1);
+            const u32 handle_2 = engine.AccessConstBuffer32(stage_type, buffer_2, offset_2);
+            return engine.GetTextureInfo(handle_1 | handle_2);
+        }
+    }
     if (entry.is_bindless) {
         const auto tex_handle = engine.AccessConstBuffer32(stage_type, entry.buffer, entry.offset);
         return engine.GetTextureInfo(tex_handle);
@@ -468,8 +480,9 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     const auto& entries = pipeline.GetEntries();
     SetupComputeConstBuffers(entries);
     SetupComputeGlobalBuffers(entries);
-    SetupComputeTexelBuffers(entries);
+    SetupComputeUniformTexels(entries);
     SetupComputeTextures(entries);
+    SetupComputeStorageTexels(entries);
     SetupComputeImages(entries);
 
     buffer_cache.Unmap();
@@ -715,7 +728,7 @@ std::tuple<VkFramebuffer, VkExtent2D> RasterizerVulkan::ConfigureFramebuffers(
         if (!view) {
             return false;
         }
-        key.views.push_back(view->GetHandle());
+        key.views.push_back(view->GetAttachment());
         key.width = std::min(key.width, view->GetWidth());
         key.height = std::min(key.height, view->GetHeight());
         key.layers = std::min(key.layers, view->GetNumLayers());
@@ -775,20 +788,21 @@ RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineSt
 }
 
 void RasterizerVulkan::SetupShaderDescriptors(
-    const std::array<Shader, Maxwell::MaxShaderProgram>& shaders) {
+    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
     texture_cache.GuardSamplers(true);
 
     for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
         // Skip VertexA stage
-        const auto& shader = shaders[stage + 1];
+        Shader* const shader = shaders[stage + 1];
         if (!shader) {
             continue;
         }
         const auto& entries = shader->GetEntries();
         SetupGraphicsConstBuffers(entries, stage);
         SetupGraphicsGlobalBuffers(entries, stage);
-        SetupGraphicsTexelBuffers(entries, stage);
+        SetupGraphicsUniformTexels(entries, stage);
         SetupGraphicsTextures(entries, stage);
+        SetupGraphicsStorageTexels(entries, stage);
         SetupGraphicsImages(entries, stage);
     }
     texture_cache.GuardSamplers(false);
@@ -838,6 +852,10 @@ void RasterizerVulkan::BeginTransformFeedback() {
     if (regs.tfb_enabled == 0) {
         return;
     }
+    if (!device.IsExtTransformFeedbackSupported()) {
+        LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
+        return;
+    }
 
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
@@ -866,6 +884,9 @@ void RasterizerVulkan::EndTransformFeedback() {
     if (regs.tfb_enabled == 0) {
         return;
     }
+    if (!device.IsExtTransformFeedbackSupported()) {
+        return;
+    }
 
     scheduler.Record(
         [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
@@ -976,12 +997,12 @@ void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries,
     }
 }
 
-void RasterizerVulkan::SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage) {
+void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage) {
     MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& gpu = system.GPU().Maxwell3D();
-    for (const auto& entry : entries.texel_buffers) {
+    for (const auto& entry : entries.uniform_texels) {
         const auto image = GetTextureInfo(gpu, entry, stage).tic;
-        SetupTexelBuffer(image, entry);
+        SetupUniformTexels(image, entry);
     }
 }
 
@@ -996,6 +1017,15 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std::
     }
 }
 
+void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage) {
+    MICROPROFILE_SCOPE(Vulkan_Textures);
+    const auto& gpu = system.GPU().Maxwell3D();
+    for (const auto& entry : entries.storage_texels) {
+        const auto image = GetTextureInfo(gpu, entry, stage).tic;
+        SetupStorageTexel(image, entry);
+    }
+}
+
 void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage) {
     MICROPROFILE_SCOPE(Vulkan_Images);
     const auto& gpu = system.GPU().Maxwell3D();
@@ -1028,12 +1058,12 @@ void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) {
     }
 }
 
-void RasterizerVulkan::SetupComputeTexelBuffers(const ShaderEntries& entries) {
+void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
     MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& gpu = system.GPU().KeplerCompute();
-    for (const auto& entry : entries.texel_buffers) {
+    for (const auto& entry : entries.uniform_texels) {
         const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
-        SetupTexelBuffer(image, entry);
+        SetupUniformTexels(image, entry);
     }
 }
 
@@ -1048,6 +1078,15 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
     }
 }
 
+void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
+    MICROPROFILE_SCOPE(Vulkan_Textures);
+    const auto& gpu = system.GPU().KeplerCompute();
+    for (const auto& entry : entries.storage_texels) {
+        const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
+        SetupStorageTexel(image, entry);
+    }
+}
+
 void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
     MICROPROFILE_SCOPE(Vulkan_Images);
     const auto& gpu = system.GPU().KeplerCompute();
@@ -1097,8 +1136,8 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd
     update_descriptor_queue.AddBuffer(buffer, offset, size);
 }
 
-void RasterizerVulkan::SetupTexelBuffer(const Tegra::Texture::TICEntry& tic,
-                                        const TexelBufferEntry& entry) {
+void RasterizerVulkan::SetupUniformTexels(const Tegra::Texture::TICEntry& tic,
+                                          const UniformTexelEntry& entry) {
     const auto view = texture_cache.GetTextureSurface(tic, entry);
     ASSERT(view->IsBufferView());
 
@@ -1110,8 +1149,8 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
     auto view = texture_cache.GetTextureSurface(texture.tic, entry);
     ASSERT(!view->IsBufferView());
 
-    const auto image_view = view->GetHandle(texture.tic.x_source, texture.tic.y_source,
-                                            texture.tic.z_source, texture.tic.w_source);
+    const VkImageView image_view = view->GetImageView(texture.tic.x_source, texture.tic.y_source,
+                                                      texture.tic.z_source, texture.tic.w_source);
     const auto sampler = sampler_cache.GetSampler(texture.tsc);
     update_descriptor_queue.AddSampledImage(sampler, image_view);
 
@@ -1120,6 +1159,14 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
     sampled_views.push_back(ImageView{std::move(view), image_layout});
 }
 
+void RasterizerVulkan::SetupStorageTexel(const Tegra::Texture::TICEntry& tic,
+                                         const StorageTexelEntry& entry) {
+    const auto view = texture_cache.GetImageSurface(tic, entry);
+    ASSERT(view->IsBufferView());
+
+    update_descriptor_queue.AddTexelBuffer(view->GetBufferView());
+}
+
 void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry) {
     auto view = texture_cache.GetImageSurface(tic, entry);
 
@@ -1129,7 +1176,8 @@ void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const Ima
 
     UNIMPLEMENTED_IF(tic.IsBuffer());
 
-    const auto image_view = view->GetHandle(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
+    const VkImageView image_view =
+        view->GetImageView(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
     update_descriptor_queue.AddImage(image_view);
 
     const auto image_layout = update_descriptor_queue.GetLastImageLayout();
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 0ed0e48c6..c8c187606 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -168,7 +168,7 @@ private:
                                  bool is_indexed, bool is_instanced);
 
     /// Setup descriptors in the graphics pipeline.
-    void SetupShaderDescriptors(const std::array<Shader, Maxwell::MaxShaderProgram>& shaders);
+    void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders);
 
     void SetupImageTransitions(Texceptions texceptions,
                                const std::array<View, Maxwell::NumRenderTargets>& color_attachments,
@@ -193,12 +193,15 @@ private:
     /// Setup global buffers in the graphics pipeline.
     void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage);
 
-    /// Setup texel buffers in the graphics pipeline.
-    void SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage);
+    /// Setup uniform texels in the graphics pipeline.
+    void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
 
     /// Setup textures in the graphics pipeline.
     void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
 
+    /// Setup storage texels in the graphics pipeline.
+    void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage);
+
     /// Setup images in the graphics pipeline.
     void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
 
@@ -209,11 +212,14 @@ private:
     void SetupComputeGlobalBuffers(const ShaderEntries& entries);
 
     /// Setup texel buffers in the compute pipeline.
-    void SetupComputeTexelBuffers(const ShaderEntries& entries);
+    void SetupComputeUniformTexels(const ShaderEntries& entries);
 
     /// Setup textures in the compute pipeline.
     void SetupComputeTextures(const ShaderEntries& entries);
 
+    /// Setup storage texels in the compute pipeline.
+    void SetupComputeStorageTexels(const ShaderEntries& entries);
+
     /// Setup images in the compute pipeline.
     void SetupComputeImages(const ShaderEntries& entries);
 
@@ -222,10 +228,12 @@ private:
 
     void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
 
-    void SetupTexelBuffer(const Tegra::Texture::TICEntry& image, const TexelBufferEntry& entry);
+    void SetupUniformTexels(const Tegra::Texture::TICEntry& image, const UniformTexelEntry& entry);
 
     void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry);
 
+    void SetupStorageTexel(const Tegra::Texture::TICEntry& tic, const StorageTexelEntry& entry);
+
     void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
 
     void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index a13e8baa7..97429cc59 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -400,8 +400,9 @@ private:
         u32 binding = specialization.base_binding;
         binding = DeclareConstantBuffers(binding);
         binding = DeclareGlobalBuffers(binding);
-        binding = DeclareTexelBuffers(binding);
+        binding = DeclareUniformTexels(binding);
         binding = DeclareSamplers(binding);
+        binding = DeclareStorageTexels(binding);
         binding = DeclareImages(binding);
 
         const Id main = OpFunction(t_void, {}, TypeFunction(t_void));
@@ -889,7 +890,7 @@ private:
         return binding;
     }
 
-    u32 DeclareTexelBuffers(u32 binding) {
+    u32 DeclareUniformTexels(u32 binding) {
         for (const auto& sampler : ir.GetSamplers()) {
             if (!sampler.is_buffer) {
                 continue;
@@ -910,7 +911,7 @@ private:
             Decorate(id, spv::Decoration::Binding, binding++);
             Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
 
-            texel_buffers.emplace(sampler.index, TexelBuffer{image_type, id});
+            uniform_texels.emplace(sampler.index, TexelBuffer{image_type, id});
         }
         return binding;
     }
@@ -945,31 +946,48 @@ private:
         return binding;
     }
 
-    u32 DeclareImages(u32 binding) {
+    u32 DeclareStorageTexels(u32 binding) {
         for (const auto& image : ir.GetImages()) {
-            const auto [dim, arrayed] = GetImageDim(image);
-            constexpr int depth = 0;
-            constexpr bool ms = false;
-            constexpr int sampled = 2; // This won't be accessed with a sampler
-            constexpr auto format = spv::ImageFormat::Unknown;
-            const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {});
-            const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
-            const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
-            AddGlobalVariable(Name(id, fmt::format("image_{}", image.index)));
-
-            Decorate(id, spv::Decoration::Binding, binding++);
-            Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
-            if (image.is_read && !image.is_written) {
-                Decorate(id, spv::Decoration::NonWritable);
-            } else if (image.is_written && !image.is_read) {
-                Decorate(id, spv::Decoration::NonReadable);
+            if (image.type != Tegra::Shader::ImageType::TextureBuffer) {
+                continue;
             }
+            DeclareImage(image, binding);
+        }
+        return binding;
+    }
 
-            images.emplace(image.index, StorageImage{image_type, id});
+    u32 DeclareImages(u32 binding) {
+        for (const auto& image : ir.GetImages()) {
+            if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
+                continue;
+            }
+            DeclareImage(image, binding);
         }
         return binding;
     }
 
+    void DeclareImage(const Image& image, u32& binding) {
+        const auto [dim, arrayed] = GetImageDim(image);
+        constexpr int depth = 0;
+        constexpr bool ms = false;
+        constexpr int sampled = 2; // This won't be accessed with a sampler
+        const auto format = image.is_atomic ? spv::ImageFormat::R32ui : spv::ImageFormat::Unknown;
+        const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {});
+        const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
+        const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
+        AddGlobalVariable(Name(id, fmt::format("image_{}", image.index)));
+
+        Decorate(id, spv::Decoration::Binding, binding++);
+        Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
+        if (image.is_read && !image.is_written) {
+            Decorate(id, spv::Decoration::NonWritable);
+        } else if (image.is_written && !image.is_read) {
+            Decorate(id, spv::Decoration::NonReadable);
+        }
+
+        images.emplace(image.index, StorageImage{image_type, id});
+    }
+
     bool IsRenderTargetEnabled(u32 rt) const {
         for (u32 component = 0; component < 4; ++component) {
             if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
@@ -1256,7 +1274,7 @@ private:
                 } else {
                     UNREACHABLE_MSG("Unmanaged offset node type");
                 }
-                pointer = OpAccessChain(t_cbuf_float, buffer_id, Constant(t_uint, 0), buffer_index,
+                pointer = OpAccessChain(t_cbuf_float, buffer_id, v_uint_zero, buffer_index,
                                         buffer_element);
             }
             return {OpLoad(t_float, pointer), Type::Float};
@@ -1611,7 +1629,7 @@ private:
 
         const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b);
         const Id carry = OpCompositeExtract(t_uint, result, 1);
-        return {OpINotEqual(t_bool, carry, Constant(t_uint, 0)), Type::Bool};
+        return {OpINotEqual(t_bool, carry, v_uint_zero), Type::Bool};
     }
 
     Expression LogicalAssign(Operation operation) {
@@ -1674,7 +1692,7 @@ private:
         const auto& meta = std::get<MetaTexture>(operation.GetMeta());
         const u32 index = meta.sampler.index;
         if (meta.sampler.is_buffer) {
-            const auto& entry = texel_buffers.at(index);
+            const auto& entry = uniform_texels.at(index);
             return OpLoad(entry.image_type, entry.image);
         } else {
             const auto& entry = sampled_images.at(index);
@@ -1951,39 +1969,20 @@ private:
         return {};
     }
 
-    Expression AtomicImageAdd(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageMin(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageMax(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageAnd(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Expression AtomicImageOr(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
+    template <Id (Module::*func)(Id, Id, Id, Id, Id)>
+    Expression AtomicImage(Operation operation) {
+        const auto& meta{std::get<MetaImage>(operation.GetMeta())};
+        ASSERT(meta.values.size() == 1);
 
-    Expression AtomicImageXor(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
+        const Id coordinate = GetCoordinates(operation, Type::Int);
+        const Id image = images.at(meta.image.index).image;
+        const Id sample = v_uint_zero;
+        const Id pointer = OpImageTexelPointer(t_image_uint, image, coordinate, sample);
 
-    Expression AtomicImageExchange(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
+        const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
+        const Id semantics = v_uint_zero;
+        const Id value = AsUint(Visit(meta.values[0]));
+        return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
     }
 
     template <Id (Module::*func)(Id, Id, Id, Id, Id)>
@@ -1998,7 +1997,7 @@ private:
             return {v_float_zero, Type::Float};
         }
         const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
-        const Id semantics = Constant(t_uint, 0);
+        const Id semantics = v_uint_zero;
         const Id value = AsUint(Visit(operation[1]));
 
         return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
@@ -2622,11 +2621,11 @@ private:
 
         &SPIRVDecompiler::ImageLoad,
         &SPIRVDecompiler::ImageStore,
-        &SPIRVDecompiler::AtomicImageAdd,
-        &SPIRVDecompiler::AtomicImageAnd,
-        &SPIRVDecompiler::AtomicImageOr,
-        &SPIRVDecompiler::AtomicImageXor,
-        &SPIRVDecompiler::AtomicImageExchange,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicXor>,
+        &SPIRVDecompiler::AtomicImage<&Module::OpAtomicExchange>,
 
         &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
         &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
@@ -2768,8 +2767,11 @@ private:
         Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0);
     const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct);
 
+    const Id t_image_uint = TypePointer(spv::StorageClass::Image, t_uint);
+
     const Id v_float_zero = Constant(t_float, 0.0f);
     const Id v_float_one = Constant(t_float, 1.0f);
+    const Id v_uint_zero = Constant(t_uint, 0);
 
     // Nvidia uses these defaults for varyings (e.g. position and generic attributes)
     const Id v_varying_default =
@@ -2794,15 +2796,16 @@ private:
     std::unordered_map<u8, GenericVaryingDescription> output_attributes;
     std::map<u32, Id> constant_buffers;
     std::map<GlobalMemoryBase, Id> global_buffers;
-    std::map<u32, TexelBuffer> texel_buffers;
+    std::map<u32, TexelBuffer> uniform_texels;
     std::map<u32, SampledImage> sampled_images;
+    std::map<u32, TexelBuffer> storage_texels;
     std::map<u32, StorageImage> images;
 
+    std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
     Id instance_index{};
     Id vertex_index{};
     Id base_instance{};
     Id base_vertex{};
-    std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
     Id frag_depth{};
     Id frag_coord{};
     Id front_facing{};
@@ -3058,13 +3061,17 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
     }
     for (const auto& sampler : ir.GetSamplers()) {
         if (sampler.is_buffer) {
-            entries.texel_buffers.emplace_back(sampler);
+            entries.uniform_texels.emplace_back(sampler);
         } else {
             entries.samplers.emplace_back(sampler);
         }
     }
     for (const auto& image : ir.GetImages()) {
-        entries.images.emplace_back(image);
+        if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
+            entries.storage_texels.emplace_back(image);
+        } else {
+            entries.images.emplace_back(image);
+        }
     }
     for (const auto& attribute : ir.GetInputAttributes()) {
         if (IsGenericAttribute(attribute)) {
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index b7af26388..2b0e90396 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -21,8 +21,9 @@ class VKDevice;
 namespace Vulkan {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using TexelBufferEntry = VideoCommon::Shader::Sampler;
+using UniformTexelEntry = VideoCommon::Shader::Sampler;
 using SamplerEntry = VideoCommon::Shader::Sampler;
+using StorageTexelEntry = VideoCommon::Shader::Image;
 using ImageEntry = VideoCommon::Shader::Image;
 
 constexpr u32 DESCRIPTOR_SET = 0;
@@ -66,13 +67,15 @@ private:
 struct ShaderEntries {
     u32 NumBindings() const {
         return static_cast<u32>(const_buffers.size() + global_buffers.size() +
-                                texel_buffers.size() + samplers.size() + images.size());
+                                uniform_texels.size() + samplers.size() + storage_texels.size() +
+                                images.size());
     }
 
     std::vector<ConstBufferEntry> const_buffers;
     std::vector<GlobalBufferEntry> global_buffers;
-    std::vector<TexelBufferEntry> texel_buffers;
+    std::vector<UniformTexelEntry> uniform_texels;
     std::vector<SamplerEntry> samplers;
+    std::vector<StorageTexelEntry> storage_texels;
     std::vector<ImageEntry> images;
     std::set<u32> attributes;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
diff --git a/src/video_core/renderer_vulkan/vk_stream_buffer.h b/src/video_core/renderer_vulkan/vk_stream_buffer.h
index dfddf7ad6..c765c60a0 100644
--- a/src/video_core/renderer_vulkan/vk_stream_buffer.h
+++ b/src/video_core/renderer_vulkan/vk_stream_buffer.h
@@ -35,7 +35,7 @@ public:
     /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
     void Unmap(u64 size);
 
-    VkBuffer GetHandle() const {
+    VkBuffer Handle() const {
         return *buffer;
     }
 
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 2f1d5021d..430031665 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -100,8 +100,8 @@ vk::Buffer CreateBuffer(const VKDevice& device, const SurfaceParams& params,
     ci.pNext = nullptr;
     ci.flags = 0;
     ci.size = static_cast<VkDeviceSize>(host_memory_size);
-    ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
-               VK_BUFFER_USAGE_TRANSFER_DST_BIT;
+    ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT |
+               VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
     ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
     ci.queueFamilyIndexCount = 0;
     ci.pQueueFamilyIndices = nullptr;
@@ -167,6 +167,7 @@ VkImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceP
         ci.extent = {params.width, params.height, 1};
         break;
     case SurfaceTarget::Texture3D:
+        ci.flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT;
         ci.extent = {params.width, params.height, params.depth};
         break;
     case SurfaceTarget::TextureBuffer:
@@ -176,6 +177,12 @@ VkImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceP
     return ci;
 }
 
+u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, Tegra::Texture::SwizzleSource y_source,
+                  Tegra::Texture::SwizzleSource z_source, Tegra::Texture::SwizzleSource w_source) {
+    return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) |
+           (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source);
+}
+
 } // Anonymous namespace
 
 CachedSurface::CachedSurface(Core::System& system, const VKDevice& device,
@@ -203,9 +210,11 @@ CachedSurface::CachedSurface(Core::System& system, const VKDevice& device,
     }
 
     // TODO(Rodrigo): Move this to a virtual function.
-    main_view = CreateViewInner(
-        ViewParams(params.target, 0, static_cast<u32>(params.GetNumLayers()), 0, params.num_levels),
-        true);
+    u32 num_layers = 1;
+    if (params.is_layered || params.target == SurfaceTarget::Texture3D) {
+        num_layers = params.depth;
+    }
+    main_view = CreateView(ViewParams(params.target, 0, num_layers, 0, params.num_levels));
 }
 
 CachedSurface::~CachedSurface() = default;
@@ -253,12 +262,8 @@ void CachedSurface::DecorateSurfaceName() {
 }
 
 View CachedSurface::CreateView(const ViewParams& params) {
-    return CreateViewInner(params, false);
-}
-
-View CachedSurface::CreateViewInner(const ViewParams& params, bool is_proxy) {
     // TODO(Rodrigo): Add name decorations
-    return views[params] = std::make_shared<CachedSurfaceView>(device, *this, params, is_proxy);
+    return views[params] = std::make_shared<CachedSurfaceView>(device, *this, params);
 }
 
 void CachedSurface::UploadBuffer(const std::vector<u8>& staging_buffer) {
@@ -342,18 +347,27 @@ VkImageSubresourceRange CachedSurface::GetImageSubresourceRange() const {
 }
 
 CachedSurfaceView::CachedSurfaceView(const VKDevice& device, CachedSurface& surface,
-                                     const ViewParams& params, bool is_proxy)
+                                     const ViewParams& params)
     : VideoCommon::ViewBase{params}, params{surface.GetSurfaceParams()},
       image{surface.GetImageHandle()}, buffer_view{surface.GetBufferViewHandle()},
       aspect_mask{surface.GetAspectMask()}, device{device}, surface{surface},
-      base_layer{params.base_layer}, num_layers{params.num_layers}, base_level{params.base_level},
-      num_levels{params.num_levels}, image_view_type{image ? GetImageViewType(params.target)
-                                                           : VK_IMAGE_VIEW_TYPE_1D} {}
+      base_level{params.base_level}, num_levels{params.num_levels},
+      image_view_type{image ? GetImageViewType(params.target) : VK_IMAGE_VIEW_TYPE_1D} {
+    if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) {
+        base_layer = 0;
+        num_layers = 1;
+        base_slice = params.base_layer;
+        num_slices = params.num_layers;
+    } else {
+        base_layer = params.base_layer;
+        num_layers = params.num_layers;
+    }
+}
 
 CachedSurfaceView::~CachedSurfaceView() = default;
 
-VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y_source,
-                                         SwizzleSource z_source, SwizzleSource w_source) {
+VkImageView CachedSurfaceView::GetImageView(SwizzleSource x_source, SwizzleSource y_source,
+                                            SwizzleSource z_source, SwizzleSource w_source) {
     const u32 new_swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source);
     if (last_image_view && last_swizzle == new_swizzle) {
         return last_image_view;
@@ -399,6 +413,11 @@ VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y
             });
     }
 
+    if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) {
+        ASSERT(base_slice == 0);
+        ASSERT(num_slices == params.depth);
+    }
+
     VkImageViewCreateInfo ci;
     ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
     ci.pNext = nullptr;
@@ -417,6 +436,35 @@ VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y
     return last_image_view = *image_view;
 }
 
+VkImageView CachedSurfaceView::GetAttachment() {
+    if (render_target) {
+        return *render_target;
+    }
+
+    VkImageViewCreateInfo ci;
+    ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
+    ci.pNext = nullptr;
+    ci.flags = 0;
+    ci.image = surface.GetImageHandle();
+    ci.format = surface.GetImage().GetFormat();
+    ci.components = {VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
+                     VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY};
+    ci.subresourceRange.aspectMask = aspect_mask;
+    ci.subresourceRange.baseMipLevel = base_level;
+    ci.subresourceRange.levelCount = num_levels;
+    if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) {
+        ci.viewType = num_slices > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D;
+        ci.subresourceRange.baseArrayLayer = base_slice;
+        ci.subresourceRange.layerCount = num_slices;
+    } else {
+        ci.viewType = image_view_type;
+        ci.subresourceRange.baseArrayLayer = base_layer;
+        ci.subresourceRange.layerCount = num_layers;
+    }
+    render_target = device.GetLogical().CreateImageView(ci);
+    return *render_target;
+}
+
 VKTextureCache::VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                                const VKDevice& device, VKResourceManager& resource_manager,
                                VKMemoryManager& memory_manager, VKScheduler& scheduler,
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index f211ccb1e..807e26c8a 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -91,7 +91,6 @@ protected:
     void DecorateSurfaceName();
 
     View CreateView(const ViewParams& params) override;
-    View CreateViewInner(const ViewParams& params, bool is_proxy);
 
 private:
     void UploadBuffer(const std::vector<u8>& staging_buffer);
@@ -120,23 +119,20 @@ private:
 class CachedSurfaceView final : public VideoCommon::ViewBase {
 public:
     explicit CachedSurfaceView(const VKDevice& device, CachedSurface& surface,
-                               const ViewParams& params, bool is_proxy);
+                               const ViewParams& params);
     ~CachedSurfaceView();
 
-    VkImageView GetHandle(Tegra::Texture::SwizzleSource x_source,
-                          Tegra::Texture::SwizzleSource y_source,
-                          Tegra::Texture::SwizzleSource z_source,
-                          Tegra::Texture::SwizzleSource w_source);
+    VkImageView GetImageView(Tegra::Texture::SwizzleSource x_source,
+                             Tegra::Texture::SwizzleSource y_source,
+                             Tegra::Texture::SwizzleSource z_source,
+                             Tegra::Texture::SwizzleSource w_source);
+
+    VkImageView GetAttachment();
 
     bool IsSameSurface(const CachedSurfaceView& rhs) const {
         return &surface == &rhs.surface;
     }
 
-    VkImageView GetHandle() {
-        return GetHandle(Tegra::Texture::SwizzleSource::R, Tegra::Texture::SwizzleSource::G,
-                         Tegra::Texture::SwizzleSource::B, Tegra::Texture::SwizzleSource::A);
-    }
-
     u32 GetWidth() const {
         return params.GetMipWidth(base_level);
     }
@@ -180,14 +176,6 @@ public:
     }
 
 private:
-    static u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source,
-                             Tegra::Texture::SwizzleSource y_source,
-                             Tegra::Texture::SwizzleSource z_source,
-                             Tegra::Texture::SwizzleSource w_source) {
-        return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) |
-               (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source);
-    }
-
     // Store a copy of these values to avoid double dereference when reading them
     const SurfaceParams params;
     const VkImage image;
@@ -196,15 +184,18 @@ private:
 
     const VKDevice& device;
     CachedSurface& surface;
-    const u32 base_layer;
-    const u32 num_layers;
     const u32 base_level;
     const u32 num_levels;
     const VkImageViewType image_view_type;
+    u32 base_layer = 0;
+    u32 num_layers = 0;
+    u32 base_slice = 0;
+    u32 num_slices = 0;
 
     VkImageView last_image_view = nullptr;
     u32 last_swizzle = 0;
 
+    vk::ImageView render_target;
     std::unordered_map<u32, vk::ImageView> view_cache;
 };
 
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index 8f0bb996e..29ebf65ba 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -357,13 +357,11 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(SamplerInfo info, u32 offset,
-                                               std::optional<u32> buffer) {
+ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(
+    SamplerInfo info, std::optional<Tegra::Engines::SamplerDescriptor> sampler) {
     if (info.IsComplete()) {
         return info;
     }
-    const auto sampler = buffer ? registry.ObtainBindlessSampler(*buffer, offset)
-                                : registry.ObtainBoundSampler(offset);
     if (!sampler) {
         LOG_WARNING(HW_GPU, "Unknown sampler info");
         info.type = info.type.value_or(Tegra::Shader::TextureType::Texture2D);
@@ -381,8 +379,8 @@ ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(SamplerInfo info, u32 offset,
 
 std::optional<Sampler> ShaderIR::GetSampler(Tegra::Shader::Sampler sampler,
                                             SamplerInfo sampler_info) {
-    const auto offset = static_cast<u32>(sampler.index.Value());
-    const auto info = GetSamplerInfo(sampler_info, offset);
+    const u32 offset = static_cast<u32>(sampler.index.Value());
+    const auto info = GetSamplerInfo(sampler_info, registry.ObtainBoundSampler(offset));
 
     // If this sampler has already been used, return the existing mapping.
     const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
@@ -404,20 +402,19 @@ std::optional<Sampler> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
     const Node sampler_register = GetRegister(reg);
     const auto [base_node, tracked_sampler_info] =
         TrackBindlessSampler(sampler_register, global_code, static_cast<s64>(global_code.size()));
-    ASSERT(base_node != nullptr);
-    if (base_node == nullptr) {
+    if (!base_node) {
+        UNREACHABLE();
         return std::nullopt;
     }
 
-    if (const auto bindless_sampler_info =
-            std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
-        const u32 buffer = bindless_sampler_info->GetIndex();
-        const u32 offset = bindless_sampler_info->GetOffset();
-        info = GetSamplerInfo(info, offset, buffer);
+    if (const auto sampler_info = std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
+        const u32 buffer = sampler_info->index;
+        const u32 offset = sampler_info->offset;
+        info = GetSamplerInfo(info, registry.ObtainBindlessSampler(buffer, offset));
 
         // If this sampler has already been used, return the existing mapping.
         const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
-                                     [buffer = buffer, offset = offset](const Sampler& entry) {
+                                     [buffer, offset](const Sampler& entry) {
                                          return entry.buffer == buffer && entry.offset == offset;
                                      });
         if (it != used_samplers.end()) {
@@ -431,10 +428,32 @@ std::optional<Sampler> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
         return used_samplers.emplace_back(next_index, offset, buffer, *info.type, *info.is_array,
                                           *info.is_shadow, *info.is_buffer, false);
     }
-    if (const auto array_sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
-        const u32 base_offset = array_sampler_info->GetBaseOffset() / 4;
-        index_var = GetCustomVariable(array_sampler_info->GetIndexVar());
-        info = GetSamplerInfo(info, base_offset);
+    if (const auto sampler_info = std::get_if<SeparateSamplerNode>(&*tracked_sampler_info)) {
+        const std::pair indices = sampler_info->indices;
+        const std::pair offsets = sampler_info->offsets;
+        info = GetSamplerInfo(info, registry.ObtainSeparateSampler(indices, offsets));
+
+        // Try to use an already created sampler if it exists
+        const auto it = std::find_if(
+            used_samplers.begin(), used_samplers.end(), [indices, offsets](const Sampler& entry) {
+                return offsets == std::pair{entry.offset, entry.secondary_offset} &&
+                       indices == std::pair{entry.buffer, entry.secondary_buffer};
+            });
+        if (it != used_samplers.end()) {
+            ASSERT(it->is_separated && it->type == info.type && it->is_array == info.is_array &&
+                   it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer);
+            return *it;
+        }
+
+        // Otherwise create a new mapping for this sampler
+        const u32 next_index = static_cast<u32>(used_samplers.size());
+        return used_samplers.emplace_back(next_index, offsets, indices, *info.type, *info.is_array,
+                                          *info.is_shadow, *info.is_buffer);
+    }
+    if (const auto sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
+        const u32 base_offset = sampler_info->base_offset / 4;
+        index_var = GetCustomVariable(sampler_info->bindless_var);
+        info = GetSamplerInfo(info, registry.ObtainBoundSampler(base_offset));
 
         // If this sampler has already been used, return the existing mapping.
         const auto it = std::find_if(
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index c5e5165ff..8f230d57a 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -275,10 +275,11 @@ using Node = std::shared_ptr<NodeData>;
 using Node4 = std::array<Node, 4>;
 using NodeBlock = std::vector<Node>;
 
-class BindlessSamplerNode;
-class ArraySamplerNode;
+struct ArraySamplerNode;
+struct BindlessSamplerNode;
+struct SeparateSamplerNode;
 
-using TrackSamplerData = std::variant<BindlessSamplerNode, ArraySamplerNode>;
+using TrackSamplerData = std::variant<BindlessSamplerNode, SeparateSamplerNode, ArraySamplerNode>;
 using TrackSampler = std::shared_ptr<TrackSamplerData>;
 
 struct Sampler {
@@ -288,63 +289,51 @@ struct Sampler {
         : index{index}, offset{offset}, type{type}, is_array{is_array}, is_shadow{is_shadow},
           is_buffer{is_buffer}, is_indexed{is_indexed} {}
 
+    /// Separate sampler constructor
+    constexpr explicit Sampler(u32 index, std::pair<u32, u32> offsets, std::pair<u32, u32> buffers,
+                               Tegra::Shader::TextureType type, bool is_array, bool is_shadow,
+                               bool is_buffer)
+        : index{index}, offset{offsets.first}, secondary_offset{offsets.second},
+          buffer{buffers.first}, secondary_buffer{buffers.second}, type{type}, is_array{is_array},
+          is_shadow{is_shadow}, is_buffer{is_buffer}, is_separated{true} {}
+
     /// Bindless samplers constructor
     constexpr explicit Sampler(u32 index, u32 offset, u32 buffer, Tegra::Shader::TextureType type,
                                bool is_array, bool is_shadow, bool is_buffer, bool is_indexed)
         : index{index}, offset{offset}, buffer{buffer}, type{type}, is_array{is_array},
           is_shadow{is_shadow}, is_buffer{is_buffer}, is_bindless{true}, is_indexed{is_indexed} {}
 
-    u32 index = 0;  ///< Emulated index given for the this sampler.
-    u32 offset = 0; ///< Offset in the const buffer from where the sampler is being read.
-    u32 buffer = 0; ///< Buffer where the bindless sampler is being read (unused on bound samplers).
-    u32 size = 1;   ///< Size of the sampler.
+    u32 index = 0;            ///< Emulated index given for the this sampler.
+    u32 offset = 0;           ///< Offset in the const buffer from where the sampler is being read.
+    u32 secondary_offset = 0; ///< Secondary offset in the const buffer.
+    u32 buffer = 0;           ///< Buffer where the bindless sampler is read.
+    u32 secondary_buffer = 0; ///< Secondary buffer where the bindless sampler is read.
+    u32 size = 1;             ///< Size of the sampler.
 
     Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc)
-    bool is_array = false;    ///< Whether the texture is being sampled as an array texture or not.
-    bool is_shadow = false;   ///< Whether the texture is being sampled as a depth texture or not.
-    bool is_buffer = false;   ///< Whether the texture is a texture buffer without sampler.
-    bool is_bindless = false; ///< Whether this sampler belongs to a bindless texture or not.
-    bool is_indexed = false;  ///< Whether this sampler is an indexed array of textures.
+    bool is_array = false;     ///< Whether the texture is being sampled as an array texture or not.
+    bool is_shadow = false;    ///< Whether the texture is being sampled as a depth texture or not.
+    bool is_buffer = false;    ///< Whether the texture is a texture buffer without sampler.
+    bool is_bindless = false;  ///< Whether this sampler belongs to a bindless texture or not.
+    bool is_indexed = false;   ///< Whether this sampler is an indexed array of textures.
+    bool is_separated = false; ///< Whether the image and sampler is separated or not.
 };
 
 /// Represents a tracked bindless sampler into a direct const buffer
-class ArraySamplerNode final {
-public:
-    explicit ArraySamplerNode(u32 index, u32 base_offset, u32 bindless_var)
-        : index{index}, base_offset{base_offset}, bindless_var{bindless_var} {}
-
-    constexpr u32 GetIndex() const {
-        return index;
-    }
-
-    constexpr u32 GetBaseOffset() const {
-        return base_offset;
-    }
-
-    constexpr u32 GetIndexVar() const {
-        return bindless_var;
-    }
-
-private:
+struct ArraySamplerNode {
     u32 index;
     u32 base_offset;
     u32 bindless_var;
 };
 
-/// Represents a tracked bindless sampler into a direct const buffer
-class BindlessSamplerNode final {
-public:
-    explicit BindlessSamplerNode(u32 index, u32 offset) : index{index}, offset{offset} {}
-
-    constexpr u32 GetIndex() const {
-        return index;
-    }
-
-    constexpr u32 GetOffset() const {
-        return offset;
-    }
+/// Represents a tracked separate sampler image pair that was folded statically
+struct SeparateSamplerNode {
+    std::pair<u32, u32> indices;
+    std::pair<u32, u32> offsets;
+};
 
-private:
+/// Represents a tracked bindless sampler into a direct const buffer
+struct BindlessSamplerNode {
     u32 index;
     u32 offset;
 };
diff --git a/src/video_core/shader/node_helper.h b/src/video_core/shader/node_helper.h
index 11231bbea..1e0886185 100644
--- a/src/video_core/shader/node_helper.h
+++ b/src/video_core/shader/node_helper.h
@@ -48,7 +48,7 @@ Node MakeNode(Args&&... args) {
 template <typename T, typename... Args>
 TrackSampler MakeTrackSampler(Args&&... args) {
     static_assert(std::is_convertible_v<T, TrackSamplerData>);
-    return std::make_shared<TrackSamplerData>(T(std::forward<Args>(args)...));
+    return std::make_shared<TrackSamplerData>(T{std::forward<Args>(args)...});
 }
 
 template <typename... Args>
diff --git a/src/video_core/shader/registry.cpp b/src/video_core/shader/registry.cpp
index af70b3f35..cdf274e54 100644
--- a/src/video_core/shader/registry.cpp
+++ b/src/video_core/shader/registry.cpp
@@ -93,6 +93,26 @@ std::optional<SamplerDescriptor> Registry::ObtainBoundSampler(u32 offset) {
     return value;
 }
 
+std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainSeparateSampler(
+    std::pair<u32, u32> buffers, std::pair<u32, u32> offsets) {
+    SeparateSamplerKey key;
+    key.buffers = buffers;
+    key.offsets = offsets;
+    const auto iter = separate_samplers.find(key);
+    if (iter != separate_samplers.end()) {
+        return iter->second;
+    }
+    if (!engine) {
+        return std::nullopt;
+    }
+
+    const u32 handle_1 = engine->AccessConstBuffer32(stage, key.buffers.first, key.offsets.first);
+    const u32 handle_2 = engine->AccessConstBuffer32(stage, key.buffers.second, key.offsets.second);
+    const SamplerDescriptor value = engine->AccessSampler(handle_1 | handle_2);
+    separate_samplers.emplace(key, value);
+    return value;
+}
+
 std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainBindlessSampler(u32 buffer,
                                                                                  u32 offset) {
     const std::pair key = {buffer, offset};
diff --git a/src/video_core/shader/registry.h b/src/video_core/shader/registry.h
index 0c80d35fd..231206765 100644
--- a/src/video_core/shader/registry.h
+++ b/src/video_core/shader/registry.h
@@ -19,8 +19,39 @@
 
 namespace VideoCommon::Shader {
 
+struct SeparateSamplerKey {
+    std::pair<u32, u32> buffers;
+    std::pair<u32, u32> offsets;
+};
+
+} // namespace VideoCommon::Shader
+
+namespace std {
+
+template <>
+struct hash<VideoCommon::Shader::SeparateSamplerKey> {
+    std::size_t operator()(const VideoCommon::Shader::SeparateSamplerKey& key) const noexcept {
+        return std::hash<u32>{}(key.buffers.first ^ key.buffers.second ^ key.offsets.first ^
+                                key.offsets.second);
+    }
+};
+
+template <>
+struct equal_to<VideoCommon::Shader::SeparateSamplerKey> {
+    bool operator()(const VideoCommon::Shader::SeparateSamplerKey& lhs,
+                    const VideoCommon::Shader::SeparateSamplerKey& rhs) const noexcept {
+        return lhs.buffers == rhs.buffers && lhs.offsets == rhs.offsets;
+    }
+};
+
+} // namespace std
+
+namespace VideoCommon::Shader {
+
 using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
 using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
+using SeparateSamplerMap =
+    std::unordered_map<SeparateSamplerKey, Tegra::Engines::SamplerDescriptor>;
 using BindlessSamplerMap =
     std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
 
@@ -73,6 +104,9 @@ public:
 
     std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
 
+    std::optional<Tegra::Engines::SamplerDescriptor> ObtainSeparateSampler(
+        std::pair<u32, u32> buffers, std::pair<u32, u32> offsets);
+
     std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
 
     /// Inserts a key.
@@ -128,6 +162,7 @@ private:
     Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
     KeyMap keys;
     BoundSamplerMap bound_samplers;
+    SeparateSamplerMap separate_samplers;
     BindlessSamplerMap bindless_samplers;
     u32 bound_buffer;
     GraphicsInfo graphics_info;
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 15ae152f2..3a98b2104 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -330,8 +330,8 @@ private:
     OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation);
 
     /// Queries the missing sampler info from the execution context.
-    SamplerInfo GetSamplerInfo(SamplerInfo info, u32 offset,
-                               std::optional<u32> buffer = std::nullopt);
+    SamplerInfo GetSamplerInfo(SamplerInfo info,
+                               std::optional<Tegra::Engines::SamplerDescriptor> sampler);
 
     /// Accesses a texture sampler.
     std::optional<Sampler> GetSampler(Tegra::Shader::Sampler sampler, SamplerInfo info);
@@ -409,8 +409,14 @@ private:
 
     std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
 
-    std::tuple<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code,
-                                                        s64 cursor);
+    std::pair<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code,
+                                                       s64 cursor);
+
+    std::pair<Node, TrackSampler> HandleBindlessIndirectRead(const CbufNode& cbuf,
+                                                             const OperationNode& operation,
+                                                             Node gpr, Node base_offset,
+                                                             Node tracked, const NodeBlock& code,
+                                                             s64 cursor);
 
     std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const;
 
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
index eb97bfd41..d5ed81442 100644
--- a/src/video_core/shader/track.cpp
+++ b/src/video_core/shader/track.cpp
@@ -14,6 +14,7 @@
 namespace VideoCommon::Shader {
 
 namespace {
+
 std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor,
                                    OperationCode operation_code) {
     for (; cursor >= 0; --cursor) {
@@ -63,7 +64,8 @@ bool AmendNodeCv(std::size_t amend_index, Node node) {
     if (const auto operation = std::get_if<OperationNode>(&*node)) {
         operation->SetAmendIndex(amend_index);
         return true;
-    } else if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
+    }
+    if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
         conditional->SetAmendIndex(amend_index);
         return true;
     }
@@ -72,40 +74,27 @@ bool AmendNodeCv(std::size_t amend_index, Node node) {
 
 } // Anonymous namespace
 
-std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code,
-                                                              s64 cursor) {
+std::pair<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code,
+                                                             s64 cursor) {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
+        const u32 cbuf_index = cbuf->GetIndex();
+
         // Constant buffer found, test if it's an immediate
         const auto& offset = cbuf->GetOffset();
         if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
-            auto track =
-                MakeTrackSampler<BindlessSamplerNode>(cbuf->GetIndex(), immediate->GetValue());
+            auto track = MakeTrackSampler<BindlessSamplerNode>(cbuf_index, immediate->GetValue());
             return {tracked, track};
         }
         if (const auto operation = std::get_if<OperationNode>(&*offset)) {
             const u32 bound_buffer = registry.GetBoundBuffer();
-            if (bound_buffer != cbuf->GetIndex()) {
+            if (bound_buffer != cbuf_index) {
                 return {};
             }
-            const auto pair = DecoupleIndirectRead(*operation);
-            if (!pair) {
-                return {};
+            if (const std::optional pair = DecoupleIndirectRead(*operation)) {
+                auto [gpr, base_offset] = *pair;
+                return HandleBindlessIndirectRead(*cbuf, *operation, gpr, base_offset, tracked,
+                                                  code, cursor);
             }
-            auto [gpr, base_offset] = *pair;
-            const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset);
-            const auto& gpu_driver = registry.AccessGuestDriverProfile();
-            const u32 bindless_cv = NewCustomVariable();
-            Node op =
-                Operation(OperationCode::UDiv, gpr, Immediate(gpu_driver.GetTextureHandlerSize()));
-
-            const Node cv_node = GetCustomVariable(bindless_cv);
-            Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op));
-            const std::size_t amend_index = DeclareAmend(std::move(amend_op));
-            AmendNodeCv(amend_index, code[cursor]);
-            // TODO Implement Bindless Index custom variable
-            auto track = MakeTrackSampler<ArraySamplerNode>(cbuf->GetIndex(),
-                                                            offset_inm->GetValue(), bindless_cv);
-            return {tracked, track};
         }
         return {};
     }
@@ -122,10 +111,23 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
         return TrackBindlessSampler(source, code, new_cursor);
     }
     if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
-        for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) {
-            if (auto found = TrackBindlessSampler((*operation)[i - 1], code, cursor);
-                std::get<0>(found)) {
-                // Cbuf found in operand.
+        const OperationNode& op = *operation;
+
+        const OperationCode opcode = operation->GetCode();
+        if (opcode == OperationCode::IBitwiseOr || opcode == OperationCode::UBitwiseOr) {
+            ASSERT(op.GetOperandsCount() == 2);
+            auto [node_a, index_a, offset_a] = TrackCbuf(op[0], code, cursor);
+            auto [node_b, index_b, offset_b] = TrackCbuf(op[1], code, cursor);
+            if (node_a && node_b) {
+                auto track = MakeTrackSampler<SeparateSamplerNode>(std::pair{index_a, index_b},
+                                                                   std::pair{offset_a, offset_b});
+                return {tracked, std::move(track)};
+            }
+        }
+        std::size_t i = op.GetOperandsCount();
+        while (i--) {
+            if (auto found = TrackBindlessSampler(op[i - 1], code, cursor); std::get<0>(found)) {
+                // Constant buffer found in operand.
                 return found;
             }
         }
@@ -139,6 +141,26 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
     return {};
 }
 
+std::pair<Node, TrackSampler> ShaderIR::HandleBindlessIndirectRead(
+    const CbufNode& cbuf, const OperationNode& operation, Node gpr, Node base_offset, Node tracked,
+    const NodeBlock& code, s64 cursor) {
+    const auto offset_imm = std::get<ImmediateNode>(*base_offset);
+    const auto& gpu_driver = registry.AccessGuestDriverProfile();
+    const u32 bindless_cv = NewCustomVariable();
+    const u32 texture_handler_size = gpu_driver.GetTextureHandlerSize();
+    Node op = Operation(OperationCode::UDiv, gpr, Immediate(texture_handler_size));
+
+    Node cv_node = GetCustomVariable(bindless_cv);
+    Node amend_op = Operation(OperationCode::Assign, std::move(cv_node), std::move(op));
+    const std::size_t amend_index = DeclareAmend(std::move(amend_op));
+    AmendNodeCv(amend_index, code[cursor]);
+
+    // TODO: Implement bindless index custom variable
+    auto track =
+        MakeTrackSampler<ArraySamplerNode>(cbuf.GetIndex(), offset_imm.GetValue(), bindless_cv);
+    return {tracked, track};
+}
+
 std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code,
                                                s64 cursor) const {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h
new file mode 100644
index 000000000..a23c23886
--- /dev/null
+++ b/src/video_core/shader_cache.h
@@ -0,0 +1,228 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <memory>
+#include <mutex>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace VideoCommon {
+
+template <class T>
+class ShaderCache {
+    static constexpr u64 PAGE_SHIFT = 14;
+
+    struct Entry {
+        VAddr addr_start;
+        VAddr addr_end;
+        T* data;
+
+        bool is_memory_marked = true;
+
+        constexpr bool Overlaps(VAddr start, VAddr end) const noexcept {
+            return start < addr_end && addr_start < end;
+        }
+    };
+
+public:
+    virtual ~ShaderCache() = default;
+
+    /// @brief Removes shaders inside a given region
+    /// @note Checks for ranges
+    /// @param addr Start address of the invalidation
+    /// @param size Number of bytes of the invalidation
+    void InvalidateRegion(VAddr addr, std::size_t size) {
+        std::scoped_lock lock{invalidation_mutex};
+        InvalidatePagesInRegion(addr, size);
+        RemovePendingShaders();
+    }
+
+    /// @brief Unmarks a memory region as cached and marks it for removal
+    /// @param addr Start address of the CPU write operation
+    /// @param size Number of bytes of the CPU write operation
+    void OnCPUWrite(VAddr addr, std::size_t size) {
+        std::lock_guard lock{invalidation_mutex};
+        InvalidatePagesInRegion(addr, size);
+    }
+
+    /// @brief Flushes delayed removal operations
+    void SyncGuestHost() {
+        std::scoped_lock lock{invalidation_mutex};
+        RemovePendingShaders();
+    }
+
+    /// @brief Tries to obtain a cached shader starting in a given address
+    /// @note Doesn't check for ranges, the given address has to be the start of the shader
+    /// @param addr Start address of the shader, this doesn't cache for region
+    /// @return Pointer to a valid shader, nullptr when nothing is found
+    T* TryGet(VAddr addr) const {
+        std::scoped_lock lock{lookup_mutex};
+
+        const auto it = lookup_cache.find(addr);
+        if (it == lookup_cache.end()) {
+            return nullptr;
+        }
+        return it->second->data;
+    }
+
+protected:
+    explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_) : rasterizer{rasterizer_} {}
+
+    /// @brief Register in the cache a given entry
+    /// @param data Shader to store in the cache
+    /// @param addr Start address of the shader that will be registered
+    /// @param size Size in bytes of the shader
+    void Register(std::unique_ptr<T> data, VAddr addr, std::size_t size) {
+        std::scoped_lock lock{invalidation_mutex, lookup_mutex};
+
+        const VAddr addr_end = addr + size;
+        Entry* const entry = NewEntry(addr, addr_end, data.get());
+
+        const u64 page_end = addr_end >> PAGE_SHIFT;
+        for (u64 page = addr >> PAGE_SHIFT; page <= page_end; ++page) {
+            invalidation_cache[page].push_back(entry);
+        }
+
+        storage.push_back(std::move(data));
+
+        rasterizer.UpdatePagesCachedCount(addr, size, 1);
+    }
+
+    /// @brief Called when a shader is going to be removed
+    /// @param shader Shader that will be removed
+    /// @pre invalidation_cache is locked
+    /// @pre lookup_mutex is locked
+    virtual void OnShaderRemoval([[maybe_unused]] T* shader) {}
+
+private:
+    /// @brief Invalidate pages in a given region
+    /// @pre invalidation_mutex is locked
+    void InvalidatePagesInRegion(VAddr addr, std::size_t size) {
+        const VAddr addr_end = addr + size;
+        const u64 page_end = addr_end >> PAGE_SHIFT;
+        for (u64 page = addr >> PAGE_SHIFT; page <= page_end; ++page) {
+            const auto it = invalidation_cache.find(page);
+            if (it == invalidation_cache.end()) {
+                continue;
+            }
+
+            std::vector<Entry*>& entries = it->second;
+            InvalidatePageEntries(entries, addr, addr_end);
+
+            // If there's nothing else in this page, remove it to avoid overpopulating the hash map.
+            if (entries.empty()) {
+                invalidation_cache.erase(it);
+            }
+        }
+    }
+
+    /// @brief Remove shaders marked for deletion
+    /// @pre invalidation_mutex is locked
+    void RemovePendingShaders() {
+        if (marked_for_removal.empty()) {
+            return;
+        }
+        std::scoped_lock lock{lookup_mutex};
+
+        std::vector<T*> removed_shaders;
+        removed_shaders.reserve(marked_for_removal.size());
+
+        for (Entry* const entry : marked_for_removal) {
+            if (lookup_cache.erase(entry->addr_start) > 0) {
+                removed_shaders.push_back(entry->data);
+            }
+        }
+        marked_for_removal.clear();
+
+        if (!removed_shaders.empty()) {
+            RemoveShadersFromStorage(std::move(removed_shaders));
+        }
+    }
+
+    /// @brief Invalidates entries in a given range for the passed page
+    /// @param entries         Vector of entries in the page, it will be modified on overlaps
+    /// @param addr            Start address of the invalidation
+    /// @param addr_end        Non-inclusive end address of the invalidation
+    /// @pre invalidation_mutex is locked
+    void InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr, VAddr addr_end) {
+        auto it = entries.begin();
+        while (it != entries.end()) {
+            Entry* const entry = *it;
+            if (!entry->Overlaps(addr, addr_end)) {
+                ++it;
+                continue;
+            }
+            UnmarkMemory(entry);
+            marked_for_removal.push_back(entry);
+
+            it = entries.erase(it);
+        }
+    }
+
+    /// @brief Unmarks an entry from the rasterizer cache
+    /// @param entry Entry to unmark from memory
+    void UnmarkMemory(Entry* entry) {
+        if (!entry->is_memory_marked) {
+            return;
+        }
+        entry->is_memory_marked = false;
+
+        const VAddr addr = entry->addr_start;
+        const std::size_t size = entry->addr_end - addr;
+        rasterizer.UpdatePagesCachedCount(addr, size, -1);
+    }
+
+    /// @brief Removes a vector of shaders from a list
+    /// @param removed_shaders Shaders to be removed from the storage, it can contain duplicates
+    /// @pre invalidation_mutex is locked
+    /// @pre lookup_mutex is locked
+    void RemoveShadersFromStorage(std::vector<T*> removed_shaders) {
+        // Remove duplicates
+        std::sort(removed_shaders.begin(), removed_shaders.end());
+        removed_shaders.erase(std::unique(removed_shaders.begin(), removed_shaders.end()),
+                              removed_shaders.end());
+
+        // Now that there are no duplicates, we can notify removals
+        for (T* const shader : removed_shaders) {
+            OnShaderRemoval(shader);
+        }
+
+        // Remove them from the cache
+        const auto is_removed = [&removed_shaders](std::unique_ptr<T>& shader) {
+            return std::find(removed_shaders.begin(), removed_shaders.end(), shader.get()) !=
+                   removed_shaders.end();
+        };
+        storage.erase(std::remove_if(storage.begin(), storage.end(), is_removed), storage.end());
+    }
+
+    /// @brief Creates a new entry in the lookup cache and returns its pointer
+    /// @pre lookup_mutex is locked
+    Entry* NewEntry(VAddr addr, VAddr addr_end, T* data) {
+        auto entry = std::make_unique<Entry>(Entry{addr, addr_end, data});
+        Entry* const entry_pointer = entry.get();
+
+        lookup_cache.emplace(addr, std::move(entry));
+        return entry_pointer;
+    }
+
+    VideoCore::RasterizerInterface& rasterizer;
+
+    mutable std::mutex lookup_mutex;
+    std::mutex invalidation_mutex;
+
+    std::unordered_map<u64, std::unique_ptr<Entry>> lookup_cache;
+    std::unordered_map<u64, std::vector<Entry*>> invalidation_cache;
+    std::vector<std::unique_ptr<T>> storage;
+    std::vector<Entry*> marked_for_removal;
+};
+
+} // namespace VideoCommon
diff --git a/src/video_core/texture_cache/surface_base.cpp b/src/video_core/texture_cache/surface_base.cpp
index 715f39d0d..94d3a6ae5 100644
--- a/src/video_core/texture_cache/surface_base.cpp
+++ b/src/video_core/texture_cache/surface_base.cpp
@@ -248,12 +248,11 @@ void SurfaceBaseImpl::FlushBuffer(Tegra::MemoryManager& memory_manager,
 
     // Use an extra temporal buffer
     auto& tmp_buffer = staging_cache.GetBuffer(1);
-    // Special case for 3D Texture Segments
-    const bool must_read_current_data =
-        params.block_depth > 0 && params.target == VideoCore::Surface::SurfaceTarget::Texture2D;
     tmp_buffer.resize(guest_memory_size);
     host_ptr = tmp_buffer.data();
-    if (must_read_current_data) {
+
+    if (params.target == SurfaceTarget::Texture3D) {
+        // Special case for 3D texture segments
         memory_manager.ReadBlockUnsafe(gpu_addr, host_ptr, guest_memory_size);
     }
 
diff --git a/src/video_core/texture_cache/surface_base.h b/src/video_core/texture_cache/surface_base.h
index 79e10ffbb..173f2edba 100644
--- a/src/video_core/texture_cache/surface_base.h
+++ b/src/video_core/texture_cache/surface_base.h
@@ -217,8 +217,8 @@ public:
     }
 
     bool IsProtected() const {
-        // Only 3D Slices are to be protected
-        return is_target && params.block_depth > 0;
+        // Only 3D slices are to be protected
+        return is_target && params.target == SurfaceTarget::Texture3D;
     }
 
     bool IsRenderTarget() const {
@@ -250,6 +250,11 @@ public:
         return GetView(ViewParams(overview_params.target, 0, num_layers, 0, params.num_levels));
     }
 
+    TView Emplace3DView(u32 slice, u32 depth, u32 base_level, u32 num_levels) {
+        return GetView(ViewParams(VideoCore::Surface::SurfaceTarget::Texture3D, slice, depth,
+                                  base_level, num_levels));
+    }
+
     std::optional<TView> EmplaceIrregularView(const SurfaceParams& view_params,
                                               const GPUVAddr view_addr,
                                               const std::size_t candidate_size, const u32 mipmap,
@@ -272,8 +277,8 @@ public:
     std::optional<TView> EmplaceView(const SurfaceParams& view_params, const GPUVAddr view_addr,
                                      const std::size_t candidate_size) {
         if (params.target == SurfaceTarget::Texture3D ||
-            (params.num_levels == 1 && !params.is_layered) ||
-            view_params.target == SurfaceTarget::Texture3D) {
+            view_params.target == SurfaceTarget::Texture3D ||
+            (params.num_levels == 1 && !params.is_layered)) {
             return {};
         }
         const auto layer_mipmap{GetLayerMipmap(view_addr)};
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index 884fabffe..0b2b2b8c4 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -215,10 +215,19 @@ SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::siz
     params.num_levels = 1;
     params.emulated_levels = 1;
 
-    const bool is_layered = config.layers > 1 && params.block_depth == 0;
-    params.is_layered = is_layered;
-    params.depth = is_layered ? config.layers.Value() : 1;
-    params.target = is_layered ? SurfaceTarget::Texture2DArray : SurfaceTarget::Texture2D;
+    if (config.memory_layout.is_3d != 0) {
+        params.depth = config.layers.Value();
+        params.is_layered = false;
+        params.target = SurfaceTarget::Texture3D;
+    } else if (config.layers > 1) {
+        params.depth = config.layers.Value();
+        params.is_layered = true;
+        params.target = SurfaceTarget::Texture2DArray;
+    } else {
+        params.depth = 1;
+        params.is_layered = false;
+        params.target = SurfaceTarget::Texture2D;
+    }
     return params;
 }
 
@@ -237,7 +246,7 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
     params.width = config.width;
     params.height = config.height;
     params.pitch = config.pitch;
-    // TODO(Rodrigo): Try to guess the surface target from depth and layer parameters
+    // TODO(Rodrigo): Try to guess texture arrays from parameters
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
     params.num_levels = 1;
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 6f63217a2..b543fc8c0 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -298,15 +298,13 @@ public:
         const GPUVAddr src_gpu_addr = src_config.Address();
         const GPUVAddr dst_gpu_addr = dst_config.Address();
         DeduceBestBlit(src_params, dst_params, src_gpu_addr, dst_gpu_addr);
-        const std::optional<VAddr> dst_cpu_addr =
-            system.GPU().MemoryManager().GpuToCpuAddress(dst_gpu_addr);
-        const std::optional<VAddr> src_cpu_addr =
-            system.GPU().MemoryManager().GpuToCpuAddress(src_gpu_addr);
-        std::pair<TSurface, TView> dst_surface =
-            GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false);
-        std::pair<TSurface, TView> src_surface =
-            GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false);
-        ImageBlit(src_surface.second, dst_surface.second, copy_config);
+
+        const auto& memory_manager = system.GPU().MemoryManager();
+        const std::optional<VAddr> dst_cpu_addr = memory_manager.GpuToCpuAddress(dst_gpu_addr);
+        const std::optional<VAddr> src_cpu_addr = memory_manager.GpuToCpuAddress(src_gpu_addr);
+        std::pair dst_surface = GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false);
+        TView src_surface = GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false).second;
+        ImageBlit(src_surface, dst_surface.second, copy_config);
         dst_surface.first->MarkAsModified(true, Tick());
     }
 
@@ -508,12 +506,12 @@ private:
             return RecycleStrategy::Flush;
         }
         // 3D Textures decision
-        if (params.block_depth > 1 || params.target == SurfaceTarget::Texture3D) {
+        if (params.target == SurfaceTarget::Texture3D) {
             return RecycleStrategy::Flush;
         }
         for (const auto& s : overlaps) {
             const auto& s_params = s->GetSurfaceParams();
-            if (s_params.block_depth > 1 || s_params.target == SurfaceTarget::Texture3D) {
+            if (s_params.target == SurfaceTarget::Texture3D) {
                 return RecycleStrategy::Flush;
             }
         }
@@ -731,51 +729,9 @@ private:
      */
     std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(VectorSurface& overlaps,
                                                                const SurfaceParams& params,
-                                                               const GPUVAddr gpu_addr,
-                                                               const VAddr cpu_addr,
+                                                               GPUVAddr gpu_addr, VAddr cpu_addr,
                                                                bool preserve_contents) {
-        if (params.target == SurfaceTarget::Texture3D) {
-            bool failed = false;
-            if (params.num_levels > 1) {
-                // We can't handle mipmaps in 3D textures yet, better fallback to LLE approach
-                return std::nullopt;
-            }
-            TSurface new_surface = GetUncachedSurface(gpu_addr, params);
-            bool modified = false;
-            for (auto& surface : overlaps) {
-                const SurfaceParams& src_params = surface->GetSurfaceParams();
-                if (src_params.target != SurfaceTarget::Texture2D) {
-                    failed = true;
-                    break;
-                }
-                if (src_params.height != params.height) {
-                    failed = true;
-                    break;
-                }
-                if (src_params.block_depth != params.block_depth ||
-                    src_params.block_height != params.block_height) {
-                    failed = true;
-                    break;
-                }
-                const u32 offset = static_cast<u32>(surface->GetCpuAddr() - cpu_addr);
-                const auto offsets = params.GetBlockOffsetXYZ(offset);
-                const auto z = std::get<2>(offsets);
-                modified |= surface->IsModified();
-                const CopyParams copy_params(0, 0, 0, 0, 0, z, 0, 0, params.width, params.height,
-                                             1);
-                ImageCopy(surface, new_surface, copy_params);
-            }
-            if (failed) {
-                return std::nullopt;
-            }
-            for (const auto& surface : overlaps) {
-                Unregister(surface);
-            }
-            new_surface->MarkAsModified(modified, Tick());
-            Register(new_surface);
-            auto view = new_surface->GetMainView();
-            return {{std::move(new_surface), view}};
-        } else {
+        if (params.target != SurfaceTarget::Texture3D) {
             for (const auto& surface : overlaps) {
                 if (!surface->MatchTarget(params.target)) {
                     if (overlaps.size() == 1 && surface->GetCpuAddr() == cpu_addr) {
@@ -791,11 +747,60 @@ private:
                     continue;
                 }
                 if (surface->MatchesStructure(params) == MatchStructureResult::FullMatch) {
-                    return {{surface, surface->GetMainView()}};
+                    return std::make_pair(surface, surface->GetMainView());
                 }
             }
             return InitializeSurface(gpu_addr, params, preserve_contents);
         }
+
+        if (params.num_levels > 1) {
+            // We can't handle mipmaps in 3D textures yet, better fallback to LLE approach
+            return std::nullopt;
+        }
+
+        if (overlaps.size() == 1) {
+            const auto& surface = overlaps[0];
+            const SurfaceParams& overlap_params = surface->GetSurfaceParams();
+            // Don't attempt to render to textures with more than one level for now
+            // The texture has to be to the right or the sample address if we want to render to it
+            if (overlap_params.num_levels == 1 && cpu_addr >= surface->GetCpuAddr()) {
+                const u32 offset = static_cast<u32>(cpu_addr - surface->GetCpuAddr());
+                const u32 slice = std::get<2>(params.GetBlockOffsetXYZ(offset));
+                if (slice < overlap_params.depth) {
+                    auto view = surface->Emplace3DView(slice, params.depth, 0, 1);
+                    return std::make_pair(std::move(surface), std::move(view));
+                }
+            }
+        }
+
+        TSurface new_surface = GetUncachedSurface(gpu_addr, params);
+        bool modified = false;
+
+        for (auto& surface : overlaps) {
+            const SurfaceParams& src_params = surface->GetSurfaceParams();
+            if (src_params.target != SurfaceTarget::Texture2D ||
+                src_params.height != params.height ||
+                src_params.block_depth != params.block_depth ||
+                src_params.block_height != params.block_height) {
+                return std::nullopt;
+            }
+            modified |= surface->IsModified();
+
+            const u32 offset = static_cast<u32>(surface->GetCpuAddr() - cpu_addr);
+            const u32 slice = std::get<2>(params.GetBlockOffsetXYZ(offset));
+            const u32 width = params.width;
+            const u32 height = params.height;
+            const CopyParams copy_params(0, 0, 0, 0, 0, slice, 0, 0, width, height, 1);
+            ImageCopy(surface, new_surface, copy_params);
+        }
+        for (const auto& surface : overlaps) {
+            Unregister(surface);
+        }
+        new_surface->MarkAsModified(modified, Tick());
+        Register(new_surface);
+
+        TView view = new_surface->GetMainView();
+        return std::make_pair(std::move(new_surface), std::move(view));
     }
 
     /**
@@ -873,7 +878,7 @@ private:
             }
         }
 
-        // Check if it's a 3D texture
+        // Manage 3D textures
         if (params.block_depth > 0) {
             auto surface =
                 Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr, preserve_contents);
diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp
index d6c9e5013..32c81dc70 100644
--- a/src/yuzu/configuration/config.cpp
+++ b/src/yuzu/configuration/config.cpp
@@ -533,6 +533,8 @@ void Config::ReadDebuggingValues() {
     Settings::values.quest_flag = ReadSetting(QStringLiteral("quest_flag"), false).toBool();
     Settings::values.disable_cpu_opt =
         ReadSetting(QStringLiteral("disable_cpu_opt"), false).toBool();
+    Settings::values.disable_macro_jit =
+        ReadSetting(QStringLiteral("disable_macro_jit"), false).toBool();
 
     qt_config->endGroup();
 }
@@ -1007,6 +1009,7 @@ void Config::SaveDebuggingValues() {
     WriteSetting(QStringLiteral("dump_nso"), Settings::values.dump_nso, false);
     WriteSetting(QStringLiteral("quest_flag"), Settings::values.quest_flag, false);
     WriteSetting(QStringLiteral("disable_cpu_opt"), Settings::values.disable_cpu_opt, false);
+    WriteSetting(QStringLiteral("disable_macro_jit"), Settings::values.disable_macro_jit, false);
 
     qt_config->endGroup();
 }
diff --git a/src/yuzu/configuration/configure_debug.cpp b/src/yuzu/configuration/configure_debug.cpp
index c2026763e..2c77441fd 100644
--- a/src/yuzu/configuration/configure_debug.cpp
+++ b/src/yuzu/configuration/configure_debug.cpp
@@ -39,6 +39,8 @@ void ConfigureDebug::SetConfiguration() {
     ui->disable_cpu_opt->setChecked(Settings::values.disable_cpu_opt);
     ui->enable_graphics_debugging->setEnabled(!Core::System::GetInstance().IsPoweredOn());
     ui->enable_graphics_debugging->setChecked(Settings::values.renderer_debug);
+    ui->disable_macro_jit->setEnabled(!Core::System::GetInstance().IsPoweredOn());
+    ui->disable_macro_jit->setChecked(Settings::values.disable_macro_jit);
 }
 
 void ConfigureDebug::ApplyConfiguration() {
@@ -51,6 +53,7 @@ void ConfigureDebug::ApplyConfiguration() {
     Settings::values.quest_flag = ui->quest_flag->isChecked();
     Settings::values.disable_cpu_opt = ui->disable_cpu_opt->isChecked();
     Settings::values.renderer_debug = ui->enable_graphics_debugging->isChecked();
+    Settings::values.disable_macro_jit = ui->disable_macro_jit->isChecked();
     Debugger::ToggleConsole();
     Log::Filter filter;
     filter.ParseFilterString(Settings::values.log_filter);
diff --git a/src/yuzu/configuration/configure_debug.ui b/src/yuzu/configuration/configure_debug.ui
index e0d4c4a44..46f0208c6 100644
--- a/src/yuzu/configuration/configure_debug.ui
+++ b/src/yuzu/configuration/configure_debug.ui
@@ -148,6 +148,19 @@
         </property>
        </widget>
       </item>
+      <item>
+       <widget class="QCheckBox" name="disable_macro_jit">
+        <property name="enabled">
+         <bool>true</bool>
+        </property>
+        <property name="whatsThis">
+         <string>When checked, it disables the macro Just In Time compiler. Enabled this makes games run slower</string>
+        </property>
+        <property name="text">
+         <string>Disable Macro JIT</string>
+        </property>
+       </widget>
+      </item>
      </layout>
     </widget>
    </item>
diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp
index 5f9cc158e..659b9f701 100644
--- a/src/yuzu_cmd/config.cpp
+++ b/src/yuzu_cmd/config.cpp
@@ -430,6 +430,8 @@ void Config::ReadValues() {
     Settings::values.quest_flag = sdl2_config->GetBoolean("Debugging", "quest_flag", false);
     Settings::values.disable_cpu_opt =
         sdl2_config->GetBoolean("Debugging", "disable_cpu_opt", false);
+    Settings::values.disable_macro_jit =
+        sdl2_config->GetBoolean("Debugging", "disable_macro_jit", false);
 
     const auto title_list = sdl2_config->Get("AddOns", "title_ids", "");
     std::stringstream ss(title_list);
diff --git a/src/yuzu_cmd/default_ini.h b/src/yuzu_cmd/default_ini.h
index 102502084..45c07ed5d 100644
--- a/src/yuzu_cmd/default_ini.h
+++ b/src/yuzu_cmd/default_ini.h
@@ -286,6 +286,8 @@ quest_flag =
 # Determines whether or not JIT CPU optimizations are enabled
 # false: Optimizations Enabled, true: Optimizations Disabled
 disable_cpu_opt =
+# Enables/Disables the macro JIT compiler
+disable_macro_jit=false
 
 [WebService]
 # Whether or not to enable telemetry