16 files changed, 430 insertions, 69 deletions

diff --git a/externals/sirit b/externals/sirit
-Subproject 12f40a80324d7c154f19f25c448a5ce27d38cd1
+Subproject 9f4d057aa28c4e9509bdc767afb27b4aee303b7
diff --git a/src/video_core/renderer_vulkan/shaders/blit.frag b/src/video_core/renderer_vulkan/shaders/blit.frag
new file mode 100644
index 000000000..a06ecd24a
--- /dev/null
+++ b/src/video_core/renderer_vulkan/shaders/blit.frag
@@ -0,0 +1,24 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V $THIS_FILE -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#version 460 core
+
+layout (location = 0) in vec2 frag_tex_coord;
+
+layout (location = 0) out vec4 color;
+
+layout (binding = 1) uniform sampler2D color_texture;
+
+void main() {
+    color = texture(color_texture, frag_tex_coord);
+}
diff --git a/src/video_core/renderer_vulkan/shaders/blit.vert b/src/video_core/renderer_vulkan/shaders/blit.vert
new file mode 100644
index 000000000..c64d9235a
--- /dev/null
+++ b/src/video_core/renderer_vulkan/shaders/blit.vert
@@ -0,0 +1,28 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V $THIS_FILE -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#version 460 core
+
+layout (location = 0) in vec2 vert_position;
+layout (location = 1) in vec2 vert_tex_coord;
+
+layout (location = 0) out vec2 frag_tex_coord;
+
+layout (set = 0, binding = 0) uniform MatrixBlock {
+    mat4 modelview_matrix;
+};
+
+void main() {
+    gl_Position = modelview_matrix * vec4(vert_position, 0.0, 1.0);
+    frag_tex_coord = vert_tex_coord;
+}
diff --git a/src/video_core/renderer_vulkan/shaders/quad_array.comp b/src/video_core/renderer_vulkan/shaders/quad_array.comp
new file mode 100644
index 000000000..5a5703308
--- /dev/null
+++ b/src/video_core/renderer_vulkan/shaders/quad_array.comp
@@ -0,0 +1,37 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V $THIS_FILE -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#version 460 core
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) buffer OutputBuffer {
+    uint output_indexes[];
+};
+
+layout (push_constant) uniform PushConstants {
+    uint first;
+};
+
+void main() {
+    uint primitive = gl_GlobalInvocationID.x;
+    if (primitive * 6 >= output_indexes.length()) {
+        return;
+    }
+
+    const uint quad_map[6] = uint[](0, 1, 2, 0, 2, 3);
+    for (uint vertex = 0; vertex < 6; ++vertex) {
+        uint index = first + primitive * 4 + quad_map[vertex];
+        output_indexes[primitive * 6 + vertex] = index;
+    }
+}
diff --git a/src/video_core/renderer_vulkan/shaders/uint8.comp b/src/video_core/renderer_vulkan/shaders/uint8.comp
new file mode 100644
index 000000000..a320f3ae0
--- /dev/null
+++ b/src/video_core/renderer_vulkan/shaders/uint8.comp
@@ -0,0 +1,33 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V $THIS_FILE -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#version 460 core
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) readonly buffer InputBuffer {
+    uint8_t input_indexes[];
+};
+
+layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
+    uint16_t output_indexes[];
+};
+
+void main() {
+    uint id = gl_GlobalInvocationID.x;
+    if (id < input_indexes.length()) {
+        output_indexes[id] = uint16_t(input_indexes[id]);
+    }
+}
diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp
index 92854a4b3..939eebe83 100644
--- a/src/video_core/renderer_vulkan/vk_device.cpp
+++ b/src/video_core/renderer_vulkan/vk_device.cpp
@@ -3,12 +3,15 @@
 // Refer to the license.txt file included.
 
 #include <bitset>
+#include <chrono>
 #include <cstdlib>
 #include <optional>
 #include <set>
 #include <string_view>
+#include <thread>
 #include <vector>
 #include "common/assert.h"
+#include "core/settings.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_device.h"
 
@@ -201,6 +204,22 @@ vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format,
     return wanted_format;
 }
 
+void VKDevice::ReportLoss() const {
+    LOG_CRITICAL(Render_Vulkan, "Device loss occured!");
+
+    // Wait some time to let the log flush
+    std::this_thread::sleep_for(std::chrono::seconds{1});
+
+    if (!nv_device_diagnostic_checkpoints) {
+        return;
+    }
+
+    [[maybe_unused]] const std::vector data = graphics_queue.getCheckpointDataNV(dld);
+    // Catch here in debug builds (or with optimizations disabled) the last graphics pipeline to be
+    // executed. It can be done on a debugger by evaluating the expression:
+    // *(VKGraphicsPipeline*)data[0]
+}
+
 bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features,
                                       const vk::DispatchLoaderDynamic& dldi) const {
     // Disable for now to avoid converting ASTC twice.
@@ -381,6 +400,8 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
              VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true);
         Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME,
              false);
+        Test(extension, nv_device_diagnostic_checkpoints,
+             VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true);
     }
 
     if (khr_shader_float16_int8) {
@@ -464,6 +485,7 @@ std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() con
 std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
     const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical) {
     static constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
+                                        vk::Format::eA8B8G8R8UintPack32,
                                         vk::Format::eA8B8G8R8SnormPack32,
                                         vk::Format::eA8B8G8R8SrgbPack32,
                                         vk::Format::eB5G6R5UnormPack16,
diff --git a/src/video_core/renderer_vulkan/vk_device.h b/src/video_core/renderer_vulkan/vk_device.h
index a844c52df..72603f9f6 100644
--- a/src/video_core/renderer_vulkan/vk_device.h
+++ b/src/video_core/renderer_vulkan/vk_device.h
@@ -39,6 +39,9 @@ public:
     vk::Format GetSupportedFormat(vk::Format wanted_format, vk::FormatFeatureFlags wanted_usage,
                                   FormatType format_type) const;
 
+    /// Reports a device loss.
+    void ReportLoss() const;
+
     /// Returns the dispatch loader with direct function pointers of the device.
     const vk::DispatchLoaderDynamic& GetDispatchLoader() const {
         return dld;
@@ -159,6 +162,11 @@ public:
         return ext_shader_viewport_index_layer;
     }
 
+    /// Returns true if the device supports VK_NV_device_diagnostic_checkpoints.
+    bool IsNvDeviceDiagnosticCheckpoints() const {
+        return nv_device_diagnostic_checkpoints;
+    }
+
     /// Returns the vendor name reported from Vulkan.
     std::string_view GetVendorName() const {
         return vendor_name;
@@ -218,6 +226,7 @@ private:
     bool ext_index_type_uint8{};               ///< Support for VK_EXT_index_type_uint8.
     bool ext_depth_range_unrestricted{};       ///< Support for VK_EXT_depth_range_unrestricted.
     bool ext_shader_viewport_index_layer{};    ///< Support for VK_EXT_shader_viewport_index_layer.
+    bool nv_device_diagnostic_checkpoints{};   ///< Support for VK_NV_device_diagnostic_checkpoints.
 
     // Telemetry parameters
     std::string vendor_name;                      ///< Device's driver name.
diff --git a/src/video_core/renderer_vulkan/vk_resource_manager.cpp b/src/video_core/renderer_vulkan/vk_resource_manager.cpp
index 13c46e5b8..525b4bb46 100644
--- a/src/video_core/renderer_vulkan/vk_resource_manager.cpp
+++ b/src/video_core/renderer_vulkan/vk_resource_manager.cpp
@@ -72,12 +72,22 @@ VKFence::VKFence(const VKDevice& device, UniqueFence handle)
 VKFence::~VKFence() = default;
 
 void VKFence::Wait() {
+    static constexpr u64 timeout = std::numeric_limits<u64>::max();
     const auto dev = device.GetLogical();
     const auto& dld = device.GetDispatchLoader();
-    dev.waitForFences({*handle}, true, std::numeric_limits<u64>::max(), dld);
+    switch (const auto result = dev.waitForFences(1, &*handle, true, timeout, dld)) {
+    case vk::Result::eSuccess:
+        return;
+    case vk::Result::eErrorDeviceLost:
+        device.ReportLoss();
+        [[fallthrough]];
+    default:
+        vk::throwResultException(result, "vk::waitForFences");
+    }
 }
 
 void VKFence::Release() {
+    ASSERT(is_owned);
     is_owned = false;
 }
 
@@ -133,8 +143,32 @@ void VKFence::Unprotect(VKResource* resource) {
     protected_resources.erase(it);
 }
 
+void VKFence::RedirectProtection(VKResource* old_resource, VKResource* new_resource) noexcept {
+    std::replace(std::begin(protected_resources), std::end(protected_resources), old_resource,
+                 new_resource);
+}
+
 VKFenceWatch::VKFenceWatch() = default;
 
+VKFenceWatch::VKFenceWatch(VKFence& initial_fence) {
+    Watch(initial_fence);
+}
+
+VKFenceWatch::VKFenceWatch(VKFenceWatch&& rhs) noexcept {
+    fence = std::exchange(rhs.fence, nullptr);
+    if (fence) {
+        fence->RedirectProtection(&rhs, this);
+    }
+}
+
+VKFenceWatch& VKFenceWatch::operator=(VKFenceWatch&& rhs) noexcept {
+    fence = std::exchange(rhs.fence, nullptr);
+    if (fence) {
+        fence->RedirectProtection(&rhs, this);
+    }
+    return *this;
+}
+
 VKFenceWatch::~VKFenceWatch() {
     if (fence) {
         fence->Unprotect(this);
diff --git a/src/video_core/renderer_vulkan/vk_resource_manager.h b/src/video_core/renderer_vulkan/vk_resource_manager.h
index 08ee86fa6..d4cbc95a5 100644
--- a/src/video_core/renderer_vulkan/vk_resource_manager.h
+++ b/src/video_core/renderer_vulkan/vk_resource_manager.h
@@ -65,6 +65,9 @@ public:
     /// Removes protection for a resource.
     void Unprotect(VKResource* resource);
 
+    /// Redirects one protected resource to a new address.
+    void RedirectProtection(VKResource* old_resource, VKResource* new_resource) noexcept;
+
     /// Retreives the fence.
     operator vk::Fence() const {
         return *handle;
@@ -97,8 +100,13 @@ private:
 class VKFenceWatch final : public VKResource {
 public:
     explicit VKFenceWatch();
+    VKFenceWatch(VKFence& initial_fence);
+    VKFenceWatch(VKFenceWatch&&) noexcept;
+    VKFenceWatch(const VKFenceWatch&) = delete;
     ~VKFenceWatch() override;
 
+    VKFenceWatch& operator=(VKFenceWatch&&) noexcept;
+
     /// Waits for the fence to be released.
     void Wait();
 
@@ -116,6 +124,14 @@ public:
 
     void OnFenceRemoval(VKFence* signaling_fence) override;
 
+    /**
+     * Do not use it paired with Watch. Use TryWatch instead.
+     * Returns true when the watch is free.
+     */
+    bool IsUsed() const {
+        return fence != nullptr;
+    }
+
 private:
     VKFence* fence{}; ///< Fence watching this resource. nullptr when the watch is free.
 };
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index 6227bc70b..a8baf91de 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -543,7 +543,7 @@ private:
         }
 
         for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) {
-            if (!IsRenderTargetUsed(rt)) {
+            if (!specialization.enabled_rendertargets[rt]) {
                 continue;
             }
 
@@ -1555,40 +1555,48 @@ private:
 
     Expression Texture(Operation operation) {
         const auto& meta = std::get<MetaTexture>(operation.GetMeta());
-        UNIMPLEMENTED_IF(!meta.aoffi.empty());
 
         const bool can_implicit = stage == ShaderType::Fragment;
         const Id sampler = GetTextureSampler(operation);
         const Id coords = GetCoordinates(operation, Type::Float);
 
+        std::vector<Id> operands;
+        spv::ImageOperandsMask mask{};
+        if (meta.bias) {
+            mask = mask | spv::ImageOperandsMask::Bias;
+            operands.push_back(AsFloat(Visit(meta.bias)));
+        }
+
+        if (!can_implicit) {
+            mask = mask | spv::ImageOperandsMask::Lod;
+            operands.push_back(v_float_zero);
+        }
+
+        if (!meta.aoffi.empty()) {
+            mask = mask | spv::ImageOperandsMask::Offset;
+            operands.push_back(GetOffsetCoordinates(operation));
+        }
+
         if (meta.depth_compare) {
             // Depth sampling
             UNIMPLEMENTED_IF(meta.bias);
             const Id dref = AsFloat(Visit(meta.depth_compare));
             if (can_implicit) {
-                return {OpImageSampleDrefImplicitLod(t_float, sampler, coords, dref, {}),
-                        Type::Float};
+                return {
+                    OpImageSampleDrefImplicitLod(t_float, sampler, coords, dref, mask, operands),
+                    Type::Float};
             } else {
-                return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref,
-                                                     spv::ImageOperandsMask::Lod, v_float_zero),
-                        Type::Float};
+                return {
+                    OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands),
+                    Type::Float};
             }
         }
 
-        std::vector<Id> operands;
-        spv::ImageOperandsMask mask{};
-        if (meta.bias) {
-            mask = mask | spv::ImageOperandsMask::Bias;
-            operands.push_back(AsFloat(Visit(meta.bias)));
-        }
-
         Id texture;
         if (can_implicit) {
             texture = OpImageSampleImplicitLod(t_float4, sampler, coords, mask, operands);
         } else {
-            texture = OpImageSampleExplicitLod(t_float4, sampler, coords,
-                                               mask | spv::ImageOperandsMask::Lod, v_float_zero,
-                                               operands);
+            texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands);
         }
         return GetTextureElement(operation, texture, Type::Float);
     }
@@ -1601,7 +1609,8 @@ private:
         const Id lod = AsFloat(Visit(meta.lod));
 
         spv::ImageOperandsMask mask = spv::ImageOperandsMask::Lod;
-        std::vector<Id> operands;
+        std::vector<Id> operands{lod};
+
         if (!meta.aoffi.empty()) {
             mask = mask | spv::ImageOperandsMask::Offset;
             operands.push_back(GetOffsetCoordinates(operation));
@@ -1609,11 +1618,10 @@ private:
 
         if (meta.sampler.IsShadow()) {
             const Id dref = AsFloat(Visit(meta.depth_compare));
-            return {
-                OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, lod, operands),
-                Type::Float};
+            return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands),
+                    Type::Float};
         }
-        const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, lod, operands);
+        const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands);
         return GetTextureElement(operation, texture, Type::Float);
     }
 
@@ -1722,7 +1730,7 @@ private:
         const std::vector grad = {dx, dy};
 
         static constexpr auto mask = spv::ImageOperandsMask::Grad;
-        const Id texture = OpImageSampleImplicitLod(t_float4, sampler, coords, mask, grad);
+        const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, grad);
         return GetTextureElement(operation, texture, Type::Float);
     }
 
@@ -1833,7 +1841,7 @@ private:
     }
 
     void PreExit() {
-        if (stage == ShaderType::Vertex) {
+        if (stage == ShaderType::Vertex && specialization.ndc_minus_one_to_one) {
             const u32 position_index = out_indices.position.value();
             const Id z_pointer = AccessElement(t_out_float, out_vertex, position_index, 2U);
             const Id w_pointer = AccessElement(t_out_float, out_vertex, position_index, 3U);
@@ -1860,12 +1868,18 @@ private:
             // rendertargets/components are skipped in the register assignment.
             u32 current_reg = 0;
             for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
+                if (!specialization.enabled_rendertargets[rt]) {
+                    // Skip rendertargets that are not enabled
+                    continue;
+                }
                 // TODO(Subv): Figure out how dual-source blending is configured in the Switch.
                 for (u32 component = 0; component < 4; ++component) {
+                    const Id pointer = AccessElement(t_out_float, frag_colors.at(rt), component);
                     if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
-                        OpStore(AccessElement(t_out_float, frag_colors.at(rt), component),
-                                SafeGetRegister(current_reg));
+                        OpStore(pointer, SafeGetRegister(current_reg));
                         ++current_reg;
+                    } else {
+                        OpStore(pointer, component == 3 ? v_float_one : v_float_zero);
                     }
                 }
             }
@@ -1995,15 +2009,6 @@ private:
         return DeclareBuiltIn(builtin, spv::StorageClass::Input, type, std::move(name));
     }
 
-    bool IsRenderTargetUsed(u32 rt) const {
-        for (u32 component = 0; component < 4; ++component) {
-            if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
-                return true;
-            }
-        }
-        return false;
-    }
-
     template <typename... Args>
     Id AccessElement(Id pointer_type, Id composite, Args... elements_) {
         std::vector<Id> members;
@@ -2552,29 +2557,7 @@ public:
     }
 
     Id operator()(const ExprCondCode& expr) {
-        const Node cc = decomp.ir.GetConditionCode(expr.cc);
-        Id target;
-
-        if (const auto pred = std::get_if<PredicateNode>(&*cc)) {
-            const auto index = pred->GetIndex();
-            switch (index) {
-            case Tegra::Shader::Pred::NeverExecute:
-                target = decomp.v_false;
-                break;
-            case Tegra::Shader::Pred::UnusedIndex:
-                target = decomp.v_true;
-                break;
-            default:
-                target = decomp.predicates.at(index);
-                break;
-            }
-        } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) {
-            target = decomp.internal_flags.at(static_cast<u32>(flag->GetFlag()));
-        } else {
-            UNREACHABLE();
-        }
-
-        return decomp.OpLoad(decomp.t_bool, target);
+        return decomp.AsBool(decomp.Visit(decomp.ir.GetConditionCode(expr.cc)));
     }
 
     Id operator()(const ExprVar& expr) {
@@ -2589,7 +2572,7 @@ public:
         const Id target = decomp.Constant(decomp.t_uint, expr.value);
         Id gpr = decomp.OpLoad(decomp.t_float, decomp.registers.at(expr.gpr));
         gpr = decomp.OpBitcast(decomp.t_uint, gpr);
-        return decomp.OpLogicalEqual(decomp.t_uint, gpr, target);
+        return decomp.OpIEqual(decomp.t_bool, gpr, target);
     }
 
     Id Visit(const Expr& node) {
@@ -2659,11 +2642,11 @@ public:
         const Id loop_label = decomp.OpLabel();
         const Id endloop_label = decomp.OpLabel();
         const Id loop_start_block = decomp.OpLabel();
-        const Id loop_end_block = decomp.OpLabel();
+        const Id loop_continue_block = decomp.OpLabel();
         current_loop_exit = endloop_label;
         decomp.OpBranch(loop_label);
         decomp.AddLabel(loop_label);
-        decomp.OpLoopMerge(endloop_label, loop_end_block, spv::LoopControlMask::MaskNone);
+        decomp.OpLoopMerge(endloop_label, loop_continue_block, spv::LoopControlMask::MaskNone);
         decomp.OpBranch(loop_start_block);
         decomp.AddLabel(loop_start_block);
         ASTNode current = ast.nodes.GetFirst();
@@ -2671,6 +2654,8 @@ public:
             Visit(current);
             current = current->GetNext();
         }
+        decomp.OpBranch(loop_continue_block);
+        decomp.AddLabel(loop_continue_block);
         ExprDecompiler expr_parser{decomp};
         const Id condition = expr_parser.Visit(ast.condition);
         decomp.OpBranchConditional(condition, loop_label, endloop_label);
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index 2b01321b6..10794be1c 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -94,6 +94,7 @@ struct Specialization final {
     Maxwell::PrimitiveTopology primitive_topology{};
     std::optional<float> point_size{};
     std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{};
+    bool ndc_minus_one_to_one{};
 
     // Tessellation specific
     struct {
@@ -101,6 +102,9 @@ struct Specialization final {
         Maxwell::TessellationSpacing spacing{};
         bool clockwise{};
     } tessellation;
+
+    // Fragment specific
+    std::bitset<8> enabled_rendertargets;
 };
 // Old gcc versions don't consider this trivially copyable.
 // static_assert(std::is_trivially_copyable_v<Specialization>);
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 78e92f52e..c934d0719 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -22,7 +22,23 @@ using Tegra::Shader::Register;
 
 namespace {
 
-u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
+u32 GetLdgMemorySize(Tegra::Shader::UniformType uniform_type) {
+    switch (uniform_type) {
+    case Tegra::Shader::UniformType::UnsignedByte:
+    case Tegra::Shader::UniformType::Single:
+        return 1;
+    case Tegra::Shader::UniformType::Double:
+        return 2;
+    case Tegra::Shader::UniformType::Quad:
+    case Tegra::Shader::UniformType::UnsignedQuad:
+        return 4;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented size={}!", static_cast<u32>(uniform_type));
+        return 1;
+    }
+}
+
+u32 GetStgMemorySize(Tegra::Shader::UniformType uniform_type) {
     switch (uniform_type) {
     case Tegra::Shader::UniformType::Single:
         return 1;
@@ -170,7 +186,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         const auto [real_address_base, base_address, descriptor] =
             TrackGlobalMemory(bb, instr, false);
 
-        const u32 count = GetUniformTypeElementsCount(type);
+        const u32 count = GetLdgMemorySize(type);
         if (!real_address_base || !base_address) {
             // Tracking failed, load zeroes.
             for (u32 i = 0; i < count; ++i) {
@@ -181,12 +197,22 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
-            const Node real_address =
-                Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
-            const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+            const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
+            Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+
+            if (type == Tegra::Shader::UniformType::UnsignedByte) {
+                // To handle unaligned loads get the byte used to dereferenced global memory
+                // and extract that byte from the loaded uint32.
+                Node byte = Operation(OperationCode::UBitwiseAnd, real_address, Immediate(3));
+                byte = Operation(OperationCode::ULogicalShiftLeft, std::move(byte), Immediate(3));
+
+                gmem = Operation(OperationCode::UBitfieldExtract, std::move(gmem), std::move(byte),
+                                 Immediate(8));
+            }
 
             SetTemporary(bb, i, gmem);
         }
+
         for (u32 i = 0; i < count; ++i) {
             SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
         }
@@ -276,7 +302,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             break;
         }
 
-        const u32 count = GetUniformTypeElementsCount(type);
+        const u32 count = GetStgMemorySize(type);
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
             const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index a4f1edd9a..38b3a4ba8 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -392,4 +392,42 @@ std::string SurfaceParams::TargetName() const {
     }
 }
 
+u32 SurfaceParams::GetBlockSize() const {
+    const u32 x = 64U << block_width;
+    const u32 y = 8U << block_height;
+    const u32 z = 1U << block_depth;
+    return x * y * z;
+}
+
+std::pair<u32, u32> SurfaceParams::GetBlockXY() const {
+    const u32 x_pixels = 64U / GetBytesPerPixel();
+    const u32 x = x_pixels << block_width;
+    const u32 y = 8U << block_height;
+    return {x, y};
+}
+
+std::tuple<u32, u32, u32> SurfaceParams::GetBlockOffsetXYZ(u32 offset) const {
+    const auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
+    const u32 block_size = GetBlockSize();
+    const u32 block_index = offset / block_size;
+    const u32 gob_offset = offset % block_size;
+    const u32 gob_index = gob_offset / static_cast<u32>(Tegra::Texture::GetGOBSize());
+    const u32 x_gob_pixels = 64U / GetBytesPerPixel();
+    const u32 x_block_pixels = x_gob_pixels << block_width;
+    const u32 y_block_pixels = 8U << block_height;
+    const u32 z_block_pixels = 1U << block_depth;
+    const u32 x_blocks = div_ceil(width, x_block_pixels);
+    const u32 y_blocks = div_ceil(height, y_block_pixels);
+    const u32 z_blocks = div_ceil(depth, z_block_pixels);
+    const u32 base_x = block_index % x_blocks;
+    const u32 base_y = (block_index / x_blocks) % y_blocks;
+    const u32 base_z = (block_index / (x_blocks * y_blocks)) % z_blocks;
+    u32 x = base_x * x_block_pixels;
+    u32 y = base_y * y_block_pixels;
+    u32 z = base_z * z_block_pixels;
+    z += gob_index >> block_height;
+    y += (gob_index * 8U) % y_block_pixels;
+    return {x, y, z};
+}
+
 } // namespace VideoCommon
diff --git a/src/video_core/texture_cache/surface_params.h b/src/video_core/texture_cache/surface_params.h
index 129817ad3..992b5c022 100644
--- a/src/video_core/texture_cache/surface_params.h
+++ b/src/video_core/texture_cache/surface_params.h
@@ -4,6 +4,8 @@
 
 #pragma once
 
+#include <utility>
+
 #include "common/alignment.h"
 #include "common/bit_util.h"
 #include "common/cityhash.h"
@@ -136,6 +138,15 @@ public:
 
     std::size_t GetConvertedMipmapSize(u32 level) const;
 
+    /// Get this texture Tegra Block size in guest memory layout
+    u32 GetBlockSize() const;
+
+    /// Get X, Y coordinates max sizes of a single block.
+    std::pair<u32, u32> GetBlockXY() const;
+
+    /// Get the offset in x, y, z coordinates from a memory offset
+    std::tuple<u32, u32, u32> GetBlockOffsetXYZ(u32 offset) const;
+
     /// Returns the size of a layer in bytes in guest memory.
     std::size_t GetGuestLayerSize() const {
         return GetLayerSize(false, false);
@@ -269,7 +280,8 @@ private:
 
     /// Returns the size of all mipmap levels and aligns as needed.
     std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const {
-        return GetLayerSize(as_host_size, uncompressed) * (layer_only ? 1U : depth);
+        return GetLayerSize(as_host_size, uncompressed) *
+               (layer_only ? 1U : (is_layered ? depth : 1U));
     }
 
     /// Returns the size of a layer
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 02d2e9136..f4c015635 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -616,6 +616,86 @@ private:
     }
 
     /**
+     * Takes care of managing 3D textures and its slices. Does HLE methods for reconstructing the 3D
+     * textures within the GPU if possible. Falls back to LLE when it isn't possible to use any of
+     * the HLE methods.
+     *
+     * @param overlaps          The overlapping surfaces registered in the cache.
+     * @param params            The parameters on the new surface.
+     * @param gpu_addr          The starting address of the new surface.
+     * @param cache_addr        The starting address of the new surface on physical memory.
+     * @param preserve_contents Indicates that the new surface should be loaded from memory or
+     *                          left blank.
+     */
+    std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps,
+                                                               const SurfaceParams& params,
+                                                               const GPUVAddr gpu_addr,
+                                                               const CacheAddr cache_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->GetCacheAddr() - cache_addr);
+                const auto [x, y, z] = params.GetBlockOffsetXYZ(offset);
+                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 {
+            for (const auto& surface : overlaps) {
+                if (!surface->MatchTarget(params.target)) {
+                    if (overlaps.size() == 1 && surface->GetCacheAddr() == cache_addr) {
+                        if (Settings::values.use_accurate_gpu_emulation) {
+                            return std::nullopt;
+                        }
+                        Unregister(surface);
+                        return InitializeSurface(gpu_addr, params, preserve_contents);
+                    }
+                    return std::nullopt;
+                }
+                if (surface->GetCacheAddr() != cache_addr) {
+                    continue;
+                }
+                if (surface->MatchesStructure(params) == MatchStructureResult::FullMatch) {
+                    return {{surface, surface->GetMainView()}};
+                }
+            }
+            return InitializeSurface(gpu_addr, params, preserve_contents);
+        }
+    }
+
+    /**
      * Gets the starting address and parameters of a candidate surface and tries
      * to find a matching surface within the cache. This is done in 3 big steps:
      *
@@ -687,6 +767,15 @@ private:
             }
         }
 
+        // Check if it's a 3D texture
+        if (params.block_depth > 0) {
+            auto surface =
+                Manage3DSurfaces(overlaps, params, gpu_addr, cache_addr, preserve_contents);
+            if (surface) {
+                return *surface;
+            }
+        }
+
         // Split cases between 1 overlap or many.
         if (overlaps.size() == 1) {
             TSurface current_surface = overlaps[0];
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index f1e3952bc..e5eac3f3b 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -12,6 +12,10 @@ namespace Tegra::Texture {
 
 // GOBSize constant. Calculated by 64 bytes in x multiplied by 8 y coords, represents
 // an small rect of (64/bytes_per_pixel)X8.
+inline std::size_t GetGOBSize() {
+    return 512;
+}
+
 inline std::size_t GetGOBSizeShift() {
     return 9;
 }