33 files changed, 1020 insertions, 391 deletions

diff --git a/CMakeModules/GenerateSCMRev.cmake b/CMakeModules/GenerateSCMRev.cmake
index 8c13a94fb..83e4e9df2 100644
--- a/CMakeModules/GenerateSCMRev.cmake
+++ b/CMakeModules/GenerateSCMRev.cmake
@@ -102,6 +102,8 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/shader_ir.cpp"
     "${VIDEO_CORE}/shader/shader_ir.h"
     "${VIDEO_CORE}/shader/track.cpp"
+    "${VIDEO_CORE}/shader/transform_feedback.cpp"
+    "${VIDEO_CORE}/shader/transform_feedback.h"
 )
 set(COMBINED "")
 foreach (F IN LISTS HASH_FILES)
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 1f621fb1f..fbebed715 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -83,6 +83,8 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/shader_ir.cpp"
       "${VIDEO_CORE}/shader/shader_ir.h"
       "${VIDEO_CORE}/shader/track.cpp"
+      "${VIDEO_CORE}/shader/transform_feedback.cpp"
+      "${VIDEO_CORE}/shader/transform_feedback.h"
       # and also check that the scm_rev files haven't changed
       "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in"
       "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.h"
diff --git a/src/common/page_table.cpp b/src/common/page_table.cpp
index 69b7abc54..566b57b62 100644
--- a/src/common/page_table.cpp
+++ b/src/common/page_table.cpp
@@ -16,7 +16,6 @@ void PageTable::Resize(std::size_t address_space_width_in_bits) {
 
     pointers.resize(num_page_table_entries);
     attributes.resize(num_page_table_entries);
-    backing_addr.resize(num_page_table_entries);
 
     // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
     // vector size is subsequently decreased (via resize), the vector might not automatically
@@ -25,6 +24,17 @@ void PageTable::Resize(std::size_t address_space_width_in_bits) {
 
     pointers.shrink_to_fit();
     attributes.shrink_to_fit();
+}
+
+BackingPageTable::BackingPageTable(std::size_t page_size_in_bits) : PageTable{page_size_in_bits} {}
+
+BackingPageTable::~BackingPageTable() = default;
+
+void BackingPageTable::Resize(std::size_t address_space_width_in_bits) {
+    PageTable::Resize(address_space_width_in_bits);
+    const std::size_t num_page_table_entries = 1ULL
+                                               << (address_space_width_in_bits - page_size_in_bits);
+    backing_addr.resize(num_page_table_entries);
     backing_addr.shrink_to_fit();
 }
 
diff --git a/src/common/page_table.h b/src/common/page_table.h
index 8b8ff0bb8..dbc272ab7 100644
--- a/src/common/page_table.h
+++ b/src/common/page_table.h
@@ -76,9 +76,20 @@ struct PageTable {
      */
     std::vector<PageType> attributes;
 
-    std::vector<u64> backing_addr;
-
     const std::size_t page_size_in_bits{};
 };
 
+/**
+ * A more advanced Page Table with the ability to save a backing address when using it
+ * depends on another MMU.
+ */
+struct BackingPageTable : PageTable {
+    explicit BackingPageTable(std::size_t page_size_in_bits);
+    ~BackingPageTable();
+
+    void Resize(std::size_t address_space_width_in_bits);
+
+    std::vector<u64> backing_addr;
+};
+
 } // namespace Common
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 0101e5f0e..91df062d7 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -129,6 +129,8 @@ add_library(video_core STATIC
     shader/shader_ir.cpp
     shader/shader_ir.h
     shader/track.cpp
+    shader/transform_feedback.cpp
+    shader/transform_feedback.h
     surface.cpp
     surface.h
     texture_cache/format_lookup_table.cpp
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 8752a1cfb..8a9e9992e 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -628,19 +628,26 @@ public:
             float depth_range_far;
         };
 
-        struct alignas(32) TransformFeedbackBinding {
+        struct TransformFeedbackBinding {
             u32 buffer_enable;
             u32 address_high;
             u32 address_low;
             s32 buffer_size;
             s32 buffer_offset;
+            INSERT_UNION_PADDING_WORDS(3);
+
+            GPUVAddr Address() const {
+                return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                             address_low);
+            }
         };
         static_assert(sizeof(TransformFeedbackBinding) == 32);
 
-        struct alignas(16) TransformFeedbackLayout {
+        struct TransformFeedbackLayout {
             u32 stream;
             u32 varying_count;
             u32 stride;
+            INSERT_UNION_PADDING_WORDS(1);
         };
         static_assert(sizeof(TransformFeedbackLayout) == 16);
 
@@ -652,6 +659,10 @@ public:
             return shader_config[index].enable != 0;
         }
 
+        bool IsShaderConfigEnabled(Regs::ShaderProgram type) const {
+            return IsShaderConfigEnabled(static_cast<std::size_t>(type));
+        }
+
         union {
             struct {
                 INSERT_UNION_PADDING_WORDS(0x45);
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index c9bc83cd7..eba42deb4 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -911,14 +911,9 @@ union Instruction {
     } fadd32i;
 
     union {
-        BitField<20, 8, u64> shift_position;
-        BitField<28, 8, u64> shift_length;
-        BitField<48, 1, u64> negate_b;
-        BitField<49, 1, u64> negate_a;
-
-        u64 GetLeftShiftValue() const {
-            return 32 - (shift_position + shift_length);
-        }
+        BitField<40, 1, u64> brev;
+        BitField<47, 1, u64> rd_cc;
+        BitField<48, 1, u64> is_signed;
     } bfe;
 
     union {
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index ba8c9d665..64acb17df 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -39,6 +39,7 @@ enum class RenderTargetFormat : u32 {
     RGBA32_FLOAT = 0xC0,
     RGBA32_UINT = 0xC2,
     RGBA16_UNORM = 0xC6,
+    RGBA16_SNORM = 0xC7,
     RGBA16_UINT = 0xC9,
     RGBA16_FLOAT = 0xCA,
     RG32_FLOAT = 0xCB,
diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h
index aea010087..073bdb491 100644
--- a/src/video_core/memory_manager.h
+++ b/src/video_core/memory_manager.h
@@ -174,7 +174,7 @@ private:
     /// End of address space, based on address space in bits.
     static constexpr GPUVAddr address_space_end{1ULL << address_space_width};
 
-    Common::PageTable page_table{page_bits};
+    Common::BackingPageTable page_table{page_bits};
     VMAMap vma_map;
     VideoCore::RasterizerInterface& rasterizer;
 
diff --git a/src/video_core/morton.cpp b/src/video_core/morton.cpp
index f2c83266e..6d522c318 100644
--- a/src/video_core/morton.cpp
+++ b/src/video_core/morton.cpp
@@ -51,6 +51,7 @@ static constexpr ConversionArray morton_to_linear_fns = {
     MortonCopy<true, PixelFormat::R8UI>,
     MortonCopy<true, PixelFormat::RGBA16F>,
     MortonCopy<true, PixelFormat::RGBA16U>,
+    MortonCopy<true, PixelFormat::RGBA16S>,
     MortonCopy<true, PixelFormat::RGBA16UI>,
     MortonCopy<true, PixelFormat::R11FG11FB10F>,
     MortonCopy<true, PixelFormat::RGBA32UI>,
@@ -131,6 +132,7 @@ static constexpr ConversionArray linear_to_morton_fns = {
     MortonCopy<false, PixelFormat::R8U>,
     MortonCopy<false, PixelFormat::R8UI>,
     MortonCopy<false, PixelFormat::RGBA16F>,
+    MortonCopy<false, PixelFormat::RGBA16S>,
     MortonCopy<false, PixelFormat::RGBA16U>,
     MortonCopy<false, PixelFormat::RGBA16UI>,
     MortonCopy<false, PixelFormat::R11FG11FB10F>,
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 8a2db8e36..1af4268a4 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -496,7 +496,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     SyncCullMode();
     SyncPrimitiveRestart();
     SyncScissorTest();
-    SyncTransformFeedback();
     SyncPointState();
     SyncPolygonOffset();
     SyncAlphaTest();
@@ -569,7 +568,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
         glTextureBarrier();
     }
 
-    ++num_queued_commands;
+    BeginTransformFeedback(primitive_mode);
 
     const GLuint base_instance = static_cast<GLuint>(gpu.regs.vb_base_instance);
     const GLsizei num_instances =
@@ -608,6 +607,10 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
                                               num_instances, base_instance);
         }
     }
+
+    EndTransformFeedback();
+
+    ++num_queued_commands;
 }
 
 void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
@@ -1290,11 +1293,6 @@ void RasterizerOpenGL::SyncScissorTest() {
     }
 }
 
-void RasterizerOpenGL::SyncTransformFeedback() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
-    UNIMPLEMENTED_IF_MSG(regs.tfb_enabled != 0, "Transform feedbacks are not implemented");
-}
-
 void RasterizerOpenGL::SyncPointState() {
     auto& gpu = system.GPU().Maxwell3D();
     auto& flags = gpu.dirty.flags;
@@ -1370,4 +1368,62 @@ void RasterizerOpenGL::SyncFramebufferSRGB() {
     oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb);
 }
 
+void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    if (regs.tfb_enabled == 0) {
+        return;
+    }
+
+    UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
+                     regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
+                     regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
+
+    for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) {
+        const auto& binding = regs.tfb_bindings[index];
+        if (!binding.buffer_enable) {
+            if (enabled_transform_feedback_buffers[index]) {
+                glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), 0, 0,
+                                  0);
+            }
+            enabled_transform_feedback_buffers[index] = false;
+            continue;
+        }
+        enabled_transform_feedback_buffers[index] = true;
+
+        auto& tfb_buffer = transform_feedback_buffers[index];
+        tfb_buffer.Create();
+
+        const GLuint handle = tfb_buffer.handle;
+        const std::size_t size = binding.buffer_size;
+        glNamedBufferData(handle, static_cast<GLsizeiptr>(size), nullptr, GL_STREAM_COPY);
+        glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), handle, 0,
+                          static_cast<GLsizeiptr>(size));
+    }
+
+    glBeginTransformFeedback(GL_POINTS);
+}
+
+void RasterizerOpenGL::EndTransformFeedback() {
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    if (regs.tfb_enabled == 0) {
+        return;
+    }
+
+    glEndTransformFeedback();
+
+    for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) {
+        const auto& binding = regs.tfb_bindings[index];
+        if (!binding.buffer_enable) {
+            continue;
+        }
+        UNIMPLEMENTED_IF(binding.buffer_offset != 0);
+
+        const GLuint handle = transform_feedback_buffers[index].handle;
+        const GPUVAddr gpu_addr = binding.Address();
+        const std::size_t size = binding.buffer_size;
+        const auto [dest_buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
+        glCopyNamedBufferSubData(handle, *dest_buffer, 0, offset, static_cast<GLsizeiptr>(size));
+    }
+}
+
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index e6424f5d2..2d3be2437 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -168,9 +168,6 @@ private:
     /// Syncs the scissor test state to match the guest state
     void SyncScissorTest();
 
-    /// Syncs the transform feedback state to match the guest state
-    void SyncTransformFeedback();
-
     /// Syncs the point state to match the guest state
     void SyncPointState();
 
@@ -192,6 +189,12 @@ private:
     /// Syncs the framebuffer sRGB state to match the guest state
     void SyncFramebufferSRGB();
 
+    /// Begin a transform feedback
+    void BeginTransformFeedback(GLenum primitive_mode);
+
+    /// End a transform feedback
+    void EndTransformFeedback();
+
     /// Check for extension that are not strictly required but are needed for correct emulation
     void CheckExtensions();
 
@@ -229,6 +232,11 @@ private:
     BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER};
     BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER};
 
+    std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers>
+        transform_feedback_buffers;
+    std::bitset<Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers>
+        enabled_transform_feedback_buffers;
+
     /// Number of commands queued to the OpenGL driver. Reseted on flush.
     std::size_t num_queued_commands = 0;
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 19d6f3dcb..849839fe3 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -23,6 +23,7 @@
 #include "video_core/shader/ast.h"
 #include "video_core/shader/node.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader/transform_feedback.h"
 
 namespace OpenGL {
 
@@ -36,6 +37,7 @@ using Tegra::Shader::IpaInterpMode;
 using Tegra::Shader::IpaMode;
 using Tegra::Shader::IpaSampleMode;
 using Tegra::Shader::Register;
+using VideoCommon::Shader::BuildTransformFeedback;
 using VideoCommon::Shader::Registry;
 
 using namespace std::string_literals;
@@ -49,6 +51,11 @@ class ExprDecompiler;
 
 enum class Type { Void, Bool, Bool2, Float, Int, Uint, HalfFloat };
 
+constexpr std::array FLOAT_TYPES{"float", "vec2", "vec3", "vec4"};
+
+constexpr std::string_view INPUT_ATTRIBUTE_NAME = "in_attr";
+constexpr std::string_view OUTPUT_ATTRIBUTE_NAME = "out_attr";
+
 struct TextureOffset {};
 struct TextureDerivates {};
 using TextureArgument = std::pair<Type, Node>;
@@ -390,12 +397,22 @@ std::string FlowStackTopName(MetaStackClass stack) {
     return stage == ShaderType::Vertex;
 }
 
+struct GenericVaryingDescription {
+    std::string name;
+    u8 first_element = 0;
+    bool is_scalar = false;
+};
+
 class GLSLDecompiler final {
 public:
     explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry,
                             ShaderType stage, std::string_view identifier, std::string_view suffix)
         : device{device}, ir{ir}, registry{registry}, stage{stage},
-          identifier{identifier}, suffix{suffix}, header{ir.GetHeader()} {}
+          identifier{identifier}, suffix{suffix}, header{ir.GetHeader()} {
+        if (stage != ShaderType::Compute) {
+            transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo());
+        }
+    }
 
     void Decompile() {
         DeclareHeader();
@@ -403,17 +420,17 @@ public:
         DeclareGeometry();
         DeclareFragment();
         DeclareCompute();
-        DeclareRegisters();
-        DeclareCustomVariables();
-        DeclarePredicates();
-        DeclareLocalMemory();
-        DeclareInternalFlags();
         DeclareInputAttributes();
         DeclareOutputAttributes();
-        DeclareConstantBuffers();
-        DeclareGlobalMemory();
-        DeclareSamplers();
         DeclareImages();
+        DeclareSamplers();
+        DeclareGlobalMemory();
+        DeclareConstantBuffers();
+        DeclareLocalMemory();
+        DeclareRegisters();
+        DeclarePredicates();
+        DeclareInternalFlags();
+        DeclareCustomVariables();
         DeclarePhysicalAttributeReader();
 
         code.AddLine("void main() {{");
@@ -485,7 +502,7 @@ private:
         if (!identifier.empty()) {
             code.AddLine("// {}", identifier);
         }
-        code.AddLine("#version 430 core");
+        code.AddLine("#version 440 core");
         code.AddLine("#extension GL_ARB_separate_shader_objects : enable");
         if (device.HasShaderBallot()) {
             code.AddLine("#extension GL_ARB_shader_ballot : require");
@@ -570,7 +587,13 @@ private:
         code.AddLine("out gl_PerVertex {{");
         ++code.scope;
 
-        code.AddLine("vec4 gl_Position;");
+        auto pos_xfb = GetTransformFeedbackDecoration(Attribute::Index::Position);
+        if (!pos_xfb.empty()) {
+            pos_xfb = fmt::format("layout ({}) ", pos_xfb);
+        }
+        const char* pos_type =
+            FLOAT_TYPES.at(GetNumComponents(Attribute::Index::Position).value_or(4) - 1);
+        code.AddLine("{}{} gl_Position;", pos_xfb, pos_type);
 
         for (const auto attribute : ir.GetOutputAttributes()) {
             if (attribute == Attribute::Index::ClipDistances0123 ||
@@ -703,7 +726,7 @@ private:
     void DeclareInputAttribute(Attribute::Index index, bool skip_unused) {
         const u32 location{GetGenericAttributeIndex(index)};
 
-        std::string name{GetInputAttribute(index)};
+        std::string name{GetGenericInputAttribute(index)};
         if (stage == ShaderType::Geometry) {
             name = "gs_" + name + "[]";
         }
@@ -740,9 +763,59 @@ private:
         }
     }
 
+    std::optional<std::size_t> GetNumComponents(Attribute::Index index, u8 element = 0) const {
+        const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element);
+        const auto it = transform_feedback.find(location);
+        if (it == transform_feedback.end()) {
+            return {};
+        }
+        return it->second.components;
+    }
+
+    std::string GetTransformFeedbackDecoration(Attribute::Index index, u8 element = 0) const {
+        const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element);
+        const auto it = transform_feedback.find(location);
+        if (it == transform_feedback.end()) {
+            return {};
+        }
+
+        const VaryingTFB& tfb = it->second;
+        return fmt::format("xfb_buffer = {}, xfb_offset = {}, xfb_stride = {}", tfb.buffer,
+                           tfb.offset, tfb.stride);
+    }
+
     void DeclareOutputAttribute(Attribute::Index index) {
-        const u32 location{GetGenericAttributeIndex(index)};
-        code.AddLine("layout (location = {}) out vec4 {};", location, GetOutputAttribute(index));
+        static constexpr std::string_view swizzle = "xyzw";
+        u8 element = 0;
+        while (element < 4) {
+            auto xfb = GetTransformFeedbackDecoration(index, element);
+            if (!xfb.empty()) {
+                xfb = fmt::format(", {}", xfb);
+            }
+            const std::size_t remainder = 4 - element;
+            const std::size_t num_components = GetNumComponents(index, element).value_or(remainder);
+            const char* const type = FLOAT_TYPES.at(num_components - 1);
+
+            const u32 location = GetGenericAttributeIndex(index);
+
+            GenericVaryingDescription description;
+            description.first_element = static_cast<u8>(element);
+            description.is_scalar = num_components == 1;
+            description.name = AppendSuffix(location, OUTPUT_ATTRIBUTE_NAME);
+            if (element != 0 || num_components != 4) {
+                const std::string_view name_swizzle = swizzle.substr(element, num_components);
+                description.name = fmt::format("{}_{}", description.name, name_swizzle);
+            }
+            for (std::size_t i = 0; i < num_components; ++i) {
+                const u8 offset = static_cast<u8>(location * 4 + element + i);
+                varying_description.insert({offset, description});
+            }
+
+            code.AddLine("layout (location = {}, component = {}{}) out {} {};", location, element,
+                         xfb, type, description.name);
+
+            element = static_cast<u8>(static_cast<std::size_t>(element) + num_components);
+        }
     }
 
     void DeclareConstantBuffers() {
@@ -1095,7 +1168,7 @@ private:
             return {"0", Type::Int};
         default:
             if (IsGenericAttribute(attribute)) {
-                return {GeometryPass(GetInputAttribute(attribute)) + GetSwizzle(element),
+                return {GeometryPass(GetGenericInputAttribute(attribute)) + GetSwizzle(element),
                         Type::Float};
             }
             break;
@@ -1164,8 +1237,7 @@ private:
             return {{fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4), Type::Float}};
         default:
             if (IsGenericAttribute(attribute)) {
-                return {
-                    {GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()), Type::Float}};
+                return {{GetGenericOutputAttribute(attribute, abuf->GetElement()), Type::Float}};
             }
             UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
             return {};
@@ -1937,16 +2009,19 @@ private:
         expr += GetSampler(meta->sampler);
         expr += ", ";
 
-        expr += constructors.at(operation.GetOperandsCount() - 1);
+        expr += constructors.at(operation.GetOperandsCount() + (meta->array ? 1 : 0) - 1);
         expr += '(';
         for (std::size_t i = 0; i < count; ++i) {
-            expr += VisitOperand(operation, i).AsInt();
-            const std::size_t next = i + 1;
-            if (next == count)
-                expr += ')';
-            else if (next < count)
+            if (i > 0) {
                 expr += ", ";
+            }
+            expr += VisitOperand(operation, i).AsInt();
+        }
+        if (meta->array) {
+            expr += ", ";
+            expr += Visit(meta->array).AsInt();
         }
+        expr += ')';
 
         if (meta->lod && !meta->sampler.IsBuffer()) {
             expr += ", ";
@@ -2376,27 +2451,34 @@ private:
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
     std::string GetRegister(u32 index) const {
-        return GetDeclarationWithSuffix(index, "gpr");
+        return AppendSuffix(index, "gpr");
     }
 
     std::string GetCustomVariable(u32 index) const {
-        return GetDeclarationWithSuffix(index, "custom_var");
+        return AppendSuffix(index, "custom_var");
     }
 
     std::string GetPredicate(Tegra::Shader::Pred pred) const {
-        return GetDeclarationWithSuffix(static_cast<u32>(pred), "pred");
+        return AppendSuffix(static_cast<u32>(pred), "pred");
     }
 
-    std::string GetInputAttribute(Attribute::Index attribute) const {
-        return GetDeclarationWithSuffix(GetGenericAttributeIndex(attribute), "input_attr");
+    std::string GetGenericInputAttribute(Attribute::Index attribute) const {
+        return AppendSuffix(GetGenericAttributeIndex(attribute), INPUT_ATTRIBUTE_NAME);
     }
 
-    std::string GetOutputAttribute(Attribute::Index attribute) const {
-        return GetDeclarationWithSuffix(GetGenericAttributeIndex(attribute), "output_attr");
+    std::unordered_map<u8, GenericVaryingDescription> varying_description;
+
+    std::string GetGenericOutputAttribute(Attribute::Index attribute, std::size_t element) const {
+        const u8 offset = static_cast<u8>(GetGenericAttributeIndex(attribute) * 4 + element);
+        const auto& description = varying_description.at(offset);
+        if (description.is_scalar) {
+            return description.name;
+        }
+        return fmt::format("{}[{}]", description.name, element - description.first_element);
     }
 
     std::string GetConstBuffer(u32 index) const {
-        return GetDeclarationWithSuffix(index, "cbuf");
+        return AppendSuffix(index, "cbuf");
     }
 
     std::string GetGlobalMemory(const GlobalMemoryBase& descriptor) const {
@@ -2409,7 +2491,7 @@ private:
     }
 
     std::string GetConstBufferBlock(u32 index) const {
-        return GetDeclarationWithSuffix(index, "cbuf_block");
+        return AppendSuffix(index, "cbuf_block");
     }
 
     std::string GetLocalMemory() const {
@@ -2434,14 +2516,14 @@ private:
     }
 
     std::string GetSampler(const Sampler& sampler) const {
-        return GetDeclarationWithSuffix(static_cast<u32>(sampler.GetIndex()), "sampler");
+        return AppendSuffix(static_cast<u32>(sampler.GetIndex()), "sampler");
     }
 
     std::string GetImage(const Image& image) const {
-        return GetDeclarationWithSuffix(static_cast<u32>(image.GetIndex()), "image");
+        return AppendSuffix(static_cast<u32>(image.GetIndex()), "image");
     }
 
-    std::string GetDeclarationWithSuffix(u32 index, std::string_view name) const {
+    std::string AppendSuffix(u32 index, std::string_view name) const {
         if (suffix.empty()) {
             return fmt::format("{}{}", name, index);
         } else {
@@ -2477,6 +2559,7 @@ private:
     const std::string_view identifier;
     const std::string_view suffix;
     const Header header;
+    std::unordered_map<u8, VaryingTFB> transform_feedback;
 
     ShaderWriter code;
 
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 2d3838a7a..f424e3000 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -53,6 +53,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
     {GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE, false},                             // R8UI
     {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT, false},                                    // RGBA16F
     {GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT, false},                                 // RGBA16U
+    {GL_RGBA16_SNORM, GL_RGBA, GL_SHORT, false},                                    // RGBA16S
     {GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT, false},                       // RGBA16UI
     {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, false},            // R11FG11FB10F
     {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT, false},                         // RGBA32UI
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 12333e8c9..fca5e3ec0 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -5,8 +5,11 @@
 #include <algorithm>
 #include <cstddef>
 #include <cstdlib>
+#include <cstring>
 #include <memory>
+
 #include <glad/glad.h>
+
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/microprofile.h"
@@ -25,6 +28,8 @@
 
 namespace OpenGL {
 
+namespace {
+
 // If the size of this is too small, it ends up creating a soft cap on FPS as the renderer will have
 // to wait on available presentation frames.
 constexpr std::size_t SWAP_CHAIN_SIZE = 3;
@@ -41,124 +46,6 @@ struct Frame {
     bool is_srgb{};                   /// Framebuffer is sRGB or RGB
 };
 
-/**
- * For smooth Vsync rendering, we want to always present the latest frame that the core generates,
- * but also make sure that rendering happens at the pace that the frontend dictates. This is a
- * helper class that the renderer uses to sync frames between the render thread and the presentation
- * thread
- */
-class FrameMailbox {
-public:
-    std::mutex swap_chain_lock;
-    std::condition_variable present_cv;
-    std::array<Frame, SWAP_CHAIN_SIZE> swap_chain{};
-    std::queue<Frame*> free_queue;
-    std::deque<Frame*> present_queue;
-    Frame* previous_frame{};
-
-    FrameMailbox() {
-        for (auto& frame : swap_chain) {
-            free_queue.push(&frame);
-        }
-    }
-
-    ~FrameMailbox() {
-        // lock the mutex and clear out the present and free_queues and notify any people who are
-        // blocked to prevent deadlock on shutdown
-        std::scoped_lock lock{swap_chain_lock};
-        std::queue<Frame*>().swap(free_queue);
-        present_queue.clear();
-        present_cv.notify_all();
-    }
-
-    void ReloadPresentFrame(Frame* frame, u32 height, u32 width) {
-        frame->present.Release();
-        frame->present.Create();
-        GLint previous_draw_fbo{};
-        glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &previous_draw_fbo);
-        glBindFramebuffer(GL_FRAMEBUFFER, frame->present.handle);
-        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER,
-                                  frame->color.handle);
-        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
-            LOG_CRITICAL(Render_OpenGL, "Failed to recreate present FBO!");
-        }
-        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, previous_draw_fbo);
-        frame->color_reloaded = false;
-    }
-
-    void ReloadRenderFrame(Frame* frame, u32 width, u32 height) {
-        // Recreate the color texture attachment
-        frame->color.Release();
-        frame->color.Create();
-        const GLenum internal_format = frame->is_srgb ? GL_SRGB8 : GL_RGB8;
-        glNamedRenderbufferStorage(frame->color.handle, internal_format, width, height);
-
-        // Recreate the FBO for the render target
-        frame->render.Release();
-        frame->render.Create();
-        glBindFramebuffer(GL_FRAMEBUFFER, frame->render.handle);
-        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER,
-                                  frame->color.handle);
-        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
-            LOG_CRITICAL(Render_OpenGL, "Failed to recreate render FBO!");
-        }
-
-        frame->width = width;
-        frame->height = height;
-        frame->color_reloaded = true;
-    }
-
-    Frame* GetRenderFrame() {
-        std::unique_lock lock{swap_chain_lock};
-
-        // If theres no free frames, we will reuse the oldest render frame
-        if (free_queue.empty()) {
-            auto frame = present_queue.back();
-            present_queue.pop_back();
-            return frame;
-        }
-
-        Frame* frame = free_queue.front();
-        free_queue.pop();
-        return frame;
-    }
-
-    void ReleaseRenderFrame(Frame* frame) {
-        std::unique_lock lock{swap_chain_lock};
-        present_queue.push_front(frame);
-        present_cv.notify_one();
-    }
-
-    Frame* TryGetPresentFrame(int timeout_ms) {
-        std::unique_lock lock{swap_chain_lock};
-        // wait for new entries in the present_queue
-        present_cv.wait_for(lock, std::chrono::milliseconds(timeout_ms),
-                            [&] { return !present_queue.empty(); });
-        if (present_queue.empty()) {
-            // timed out waiting for a frame to draw so return the previous frame
-            return previous_frame;
-        }
-
-        // free the previous frame and add it back to the free queue
-        if (previous_frame) {
-            free_queue.push(previous_frame);
-        }
-
-        // the newest entries are pushed to the front of the queue
-        Frame* frame = present_queue.front();
-        present_queue.pop_front();
-        // remove all old entries from the present queue and move them back to the free_queue
-        for (auto f : present_queue) {
-            free_queue.push(f);
-        }
-        present_queue.clear();
-        previous_frame = frame;
-        return frame;
-    }
-};
-
-namespace {
-
 constexpr char VERTEX_SHADER[] = R"(
 #version 430 core
 
@@ -211,6 +98,24 @@ struct ScreenRectVertex {
     std::array<GLfloat, 2> tex_coord;
 };
 
+/// Returns true if any debug tool is attached
+bool HasDebugTool() {
+    const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
+    if (nsight) {
+        return true;
+    }
+
+    GLint num_extensions;
+    glGetIntegerv(GL_NUM_EXTENSIONS, &num_extensions);
+    for (GLuint index = 0; index < static_cast<GLuint>(num_extensions); ++index) {
+        const auto name = reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, index));
+        if (!std::strcmp(name, "GL_EXT_debug_tool")) {
+            return true;
+        }
+    }
+    return false;
+}
+
 /**
  * Defines a 1:1 pixel ortographic projection matrix with (0,0) on the top-left
  * corner and (width, height) on the lower-bottom.
@@ -294,6 +199,153 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit
 
 } // Anonymous namespace
 
+/**
+ * For smooth Vsync rendering, we want to always present the latest frame that the core generates,
+ * but also make sure that rendering happens at the pace that the frontend dictates. This is a
+ * helper class that the renderer uses to sync frames between the render thread and the presentation
+ * thread
+ */
+class FrameMailbox {
+public:
+    std::mutex swap_chain_lock;
+    std::condition_variable present_cv;
+    std::array<Frame, SWAP_CHAIN_SIZE> swap_chain{};
+    std::queue<Frame*> free_queue;
+    std::deque<Frame*> present_queue;
+    Frame* previous_frame{};
+
+    FrameMailbox() : has_debug_tool{HasDebugTool()} {
+        for (auto& frame : swap_chain) {
+            free_queue.push(&frame);
+        }
+    }
+
+    ~FrameMailbox() {
+        // lock the mutex and clear out the present and free_queues and notify any people who are
+        // blocked to prevent deadlock on shutdown
+        std::scoped_lock lock{swap_chain_lock};
+        std::queue<Frame*>().swap(free_queue);
+        present_queue.clear();
+        present_cv.notify_all();
+    }
+
+    void ReloadPresentFrame(Frame* frame, u32 height, u32 width) {
+        frame->present.Release();
+        frame->present.Create();
+        GLint previous_draw_fbo{};
+        glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &previous_draw_fbo);
+        glBindFramebuffer(GL_FRAMEBUFFER, frame->present.handle);
+        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER,
+                                  frame->color.handle);
+        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
+            LOG_CRITICAL(Render_OpenGL, "Failed to recreate present FBO!");
+        }
+        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, previous_draw_fbo);
+        frame->color_reloaded = false;
+    }
+
+    void ReloadRenderFrame(Frame* frame, u32 width, u32 height) {
+        // Recreate the color texture attachment
+        frame->color.Release();
+        frame->color.Create();
+        const GLenum internal_format = frame->is_srgb ? GL_SRGB8 : GL_RGB8;
+        glNamedRenderbufferStorage(frame->color.handle, internal_format, width, height);
+
+        // Recreate the FBO for the render target
+        frame->render.Release();
+        frame->render.Create();
+        glBindFramebuffer(GL_FRAMEBUFFER, frame->render.handle);
+        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER,
+                                  frame->color.handle);
+        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
+            LOG_CRITICAL(Render_OpenGL, "Failed to recreate render FBO!");
+        }
+
+        frame->width = width;
+        frame->height = height;
+        frame->color_reloaded = true;
+    }
+
+    Frame* GetRenderFrame() {
+        std::unique_lock lock{swap_chain_lock};
+
+        // If theres no free frames, we will reuse the oldest render frame
+        if (free_queue.empty()) {
+            auto frame = present_queue.back();
+            present_queue.pop_back();
+            return frame;
+        }
+
+        Frame* frame = free_queue.front();
+        free_queue.pop();
+        return frame;
+    }
+
+    void ReleaseRenderFrame(Frame* frame) {
+        std::unique_lock lock{swap_chain_lock};
+        present_queue.push_front(frame);
+        present_cv.notify_one();
+
+        DebugNotifyNextFrame();
+    }
+
+    Frame* TryGetPresentFrame(int timeout_ms) {
+        DebugWaitForNextFrame();
+
+        std::unique_lock lock{swap_chain_lock};
+        // wait for new entries in the present_queue
+        present_cv.wait_for(lock, std::chrono::milliseconds(timeout_ms),
+                            [&] { return !present_queue.empty(); });
+        if (present_queue.empty()) {
+            // timed out waiting for a frame to draw so return the previous frame
+            return previous_frame;
+        }
+
+        // free the previous frame and add it back to the free queue
+        if (previous_frame) {
+            free_queue.push(previous_frame);
+        }
+
+        // the newest entries are pushed to the front of the queue
+        Frame* frame = present_queue.front();
+        present_queue.pop_front();
+        // remove all old entries from the present queue and move them back to the free_queue
+        for (auto f : present_queue) {
+            free_queue.push(f);
+        }
+        present_queue.clear();
+        previous_frame = frame;
+        return frame;
+    }
+
+private:
+    std::mutex debug_synch_mutex;
+    std::condition_variable debug_synch_condition;
+    std::atomic_int frame_for_debug{};
+    const bool has_debug_tool; // When true, using a GPU debugger, so keep frames in lock-step
+
+    /// Signal that a new frame is available (called from GPU thread)
+    void DebugNotifyNextFrame() {
+        if (!has_debug_tool) {
+            return;
+        }
+        frame_for_debug++;
+        std::lock_guard lock{debug_synch_mutex};
+        debug_synch_condition.notify_one();
+    }
+
+    /// Wait for a new frame to be available (called from presentation thread)
+    void DebugWaitForNextFrame() {
+        if (!has_debug_tool) {
+            return;
+        }
+        const int last_frame = frame_for_debug;
+        std::unique_lock lock{debug_synch_mutex};
+        debug_synch_condition.wait(lock,
+                                   [this, last_frame] { return frame_for_debug > last_frame; });
+    }
+};
+
 RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system)
     : VideoCore::RendererBase{emu_window}, emu_window{emu_window}, system{system},
       frame_mailbox{std::make_unique<FrameMailbox>()} {}
diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
index df3ac707c..0e2e5e6c7 100644
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
@@ -125,6 +125,7 @@ struct FormatTuple {
     {vk::Format::eR8Uint, Attachable | Storage},                 // R8UI
     {vk::Format::eR16G16B16A16Sfloat, Attachable | Storage},     // RGBA16F
     {vk::Format::eR16G16B16A16Unorm, Attachable | Storage},      // RGBA16U
+    {vk::Format::eR16G16B16A16Snorm, Attachable | Storage},      // RGBA16S
     {vk::Format::eR16G16B16A16Uint, Attachable | Storage},       // RGBA16UI
     {vk::Format::eB10G11R11UfloatPack32, Attachable | Storage},  // R11FG11FB10F
     {vk::Format::eR32G32B32A32Uint, Attachable | Storage},       // RGBA32UI
@@ -331,6 +332,8 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
             return vk::Format::eR16G16B16Unorm;
         case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
             return vk::Format::eR16G16B16A16Unorm;
+        case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
+            return vk::Format::eA2B10G10R10UnormPack32;
         default:
             break;
         }
@@ -364,6 +367,10 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
             return vk::Format::eR8G8B8A8Uint;
         case Maxwell::VertexAttribute::Size::Size_32:
             return vk::Format::eR32Uint;
+        case Maxwell::VertexAttribute::Size::Size_32_32:
+            return vk::Format::eR32G32Uint;
+        case Maxwell::VertexAttribute::Size::Size_32_32_32:
+            return vk::Format::eR32G32B32Uint;
         case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
             return vk::Format::eR32G32B32A32Uint;
         default:
diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp
index 886bde3b9..28d2fbc4f 100644
--- a/src/video_core/renderer_vulkan/vk_device.cpp
+++ b/src/video_core/renderer_vulkan/vk_device.cpp
@@ -107,8 +107,7 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
     features.occlusionQueryPrecise = true;
     features.fragmentStoresAndAtomics = true;
     features.shaderImageGatherExtended = true;
-    features.shaderStorageImageReadWithoutFormat =
-        is_shader_storage_img_read_without_format_supported;
+    features.shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported;
     features.shaderStorageImageWriteWithoutFormat = true;
     features.textureCompressionASTC_LDR = is_optimal_astc_supported;
 
@@ -148,6 +147,15 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
         LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes");
     }
 
+    vk::PhysicalDeviceTransformFeedbackFeaturesEXT transform_feedback;
+    if (ext_transform_feedback) {
+        transform_feedback.transformFeedback = true;
+        transform_feedback.geometryStreams = true;
+        SetNext(next, transform_feedback);
+    } else {
+        LOG_INFO(Render_Vulkan, "Device doesn't support transform feedbacks");
+    }
+
     if (!ext_depth_range_unrestricted) {
         LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
     }
@@ -385,7 +393,7 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
         }
     };
 
-    extensions.reserve(14);
+    extensions.reserve(15);
     extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
     extensions.push_back(VK_KHR_16BIT_STORAGE_EXTENSION_NAME);
     extensions.push_back(VK_KHR_8BIT_STORAGE_EXTENSION_NAME);
@@ -397,18 +405,22 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
 
     [[maybe_unused]] const bool nsight =
         std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
-    bool khr_shader_float16_int8{};
-    bool ext_subgroup_size_control{};
+    bool has_khr_shader_float16_int8{};
+    bool has_ext_subgroup_size_control{};
+    bool has_ext_transform_feedback{};
     for (const auto& extension : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) {
         Test(extension, khr_uniform_buffer_standard_layout,
              VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true);
-        Test(extension, khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false);
+        Test(extension, has_khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME,
+             false);
         Test(extension, ext_depth_range_unrestricted,
              VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true);
         Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true);
         Test(extension, ext_shader_viewport_index_layer,
              VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true);
-        Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME,
+        Test(extension, has_ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME,
+             false);
+        Test(extension, has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME,
              false);
         if (Settings::values.renderer_debug) {
             Test(extension, nv_device_diagnostic_checkpoints,
@@ -416,13 +428,13 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
         }
     }
 
-    if (khr_shader_float16_int8) {
+    if (has_khr_shader_float16_int8) {
         is_float16_supported =
             GetFeatures<vk::PhysicalDeviceFloat16Int8FeaturesKHR>(physical, dldi).shaderFloat16;
         extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME);
     }
 
-    if (ext_subgroup_size_control) {
+    if (has_ext_subgroup_size_control) {
         const auto features =
             GetFeatures<vk::PhysicalDeviceSubgroupSizeControlFeaturesEXT>(physical, dldi);
         const auto properties =
@@ -439,6 +451,20 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
         is_warp_potentially_bigger = true;
     }
 
+    if (has_ext_transform_feedback) {
+        const auto features =
+            GetFeatures<vk::PhysicalDeviceTransformFeedbackFeaturesEXT>(physical, dldi);
+        const auto properties =
+            GetProperties<vk::PhysicalDeviceTransformFeedbackPropertiesEXT>(physical, dldi);
+
+        if (features.transformFeedback && features.geometryStreams &&
+            properties.maxTransformFeedbackStreams >= 4 && properties.maxTransformFeedbackBuffers &&
+            properties.transformFeedbackQueries && properties.transformFeedbackDraw) {
+            extensions.push_back(VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME);
+            ext_transform_feedback = true;
+        }
+    }
+
     return extensions;
 }
 
@@ -467,8 +493,7 @@ void VKDevice::SetupFamilies(const vk::DispatchLoaderDynamic& dldi, vk::SurfaceK
 
 void VKDevice::SetupFeatures(const vk::DispatchLoaderDynamic& dldi) {
     const auto supported_features{physical.getFeatures(dldi)};
-    is_shader_storage_img_read_without_format_supported =
-        supported_features.shaderStorageImageReadWithoutFormat;
+    is_formatless_image_load_supported = supported_features.shaderStorageImageReadWithoutFormat;
     is_optimal_astc_supported = IsOptimalAstcSupported(supported_features, dldi);
 }
 
@@ -510,6 +535,7 @@ std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperti
                                         vk::Format::eR32G32Sfloat,
                                         vk::Format::eR32G32Uint,
                                         vk::Format::eR16G16B16A16Uint,
+                                        vk::Format::eR16G16B16A16Snorm,
                                         vk::Format::eR16G16B16A16Unorm,
                                         vk::Format::eR16G16Unorm,
                                         vk::Format::eR16G16Snorm,
diff --git a/src/video_core/renderer_vulkan/vk_device.h b/src/video_core/renderer_vulkan/vk_device.h
index 2c27ad730..6e656517f 100644
--- a/src/video_core/renderer_vulkan/vk_device.h
+++ b/src/video_core/renderer_vulkan/vk_device.h
@@ -122,11 +122,6 @@ public:
         return properties.limits.maxPushConstantsSize;
     }
 
-    /// Returns true if Shader storage Image Read Without Format supported.
-    bool IsShaderStorageImageReadWithoutFormatSupported() const {
-        return is_shader_storage_img_read_without_format_supported;
-    }
-
     /// Returns true if ASTC is natively supported.
     bool IsOptimalAstcSupported() const {
         return is_optimal_astc_supported;
@@ -147,6 +142,11 @@ public:
         return (guest_warp_stages & stage) != vk::ShaderStageFlags{};
     }
 
+    /// Returns true if formatless image load is supported.
+    bool IsFormatlessImageLoadSupported() const {
+        return is_formatless_image_load_supported;
+    }
+
     /// Returns true if the device supports VK_EXT_scalar_block_layout.
     bool IsKhrUniformBufferStandardLayoutSupported() const {
         return khr_uniform_buffer_standard_layout;
@@ -167,6 +167,11 @@ public:
         return ext_shader_viewport_index_layer;
     }
 
+    /// Returns true if the device supports VK_EXT_transform_feedback.
+    bool IsExtTransformFeedbackSupported() const {
+        return ext_transform_feedback;
+    }
+
     /// Returns true if the device supports VK_NV_device_diagnostic_checkpoints.
     bool IsNvDeviceDiagnosticCheckpoints() const {
         return nv_device_diagnostic_checkpoints;
@@ -214,26 +219,26 @@ private:
     static std::unordered_map<vk::Format, vk::FormatProperties> GetFormatProperties(
         const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical);
 
-    const vk::PhysicalDevice physical;         ///< Physical device.
-    vk::DispatchLoaderDynamic dld;             ///< Device function pointers.
-    vk::PhysicalDeviceProperties properties;   ///< Device properties.
-    UniqueDevice logical;                      ///< Logical device.
-    vk::Queue graphics_queue;                  ///< Main graphics queue.
-    vk::Queue present_queue;                   ///< Main present queue.
-    u32 graphics_family{};                     ///< Main graphics queue family index.
-    u32 present_family{};                      ///< Main present queue family index.
-    vk::DriverIdKHR driver_id{};               ///< Driver ID.
-    vk::ShaderStageFlags guest_warp_stages{};  ///< Stages where the guest warp size can be forced.
-    bool is_optimal_astc_supported{};          ///< Support for native ASTC.
-    bool is_float16_supported{};               ///< Support for float16 arithmetics.
-    bool is_warp_potentially_bigger{};         ///< Host warp size can be bigger than guest.
+    const vk::PhysicalDevice physical;        ///< Physical device.
+    vk::DispatchLoaderDynamic dld;            ///< Device function pointers.
+    vk::PhysicalDeviceProperties properties;  ///< Device properties.
+    UniqueDevice logical;                     ///< Logical device.
+    vk::Queue graphics_queue;                 ///< Main graphics queue.
+    vk::Queue present_queue;                  ///< Main present queue.
+    u32 graphics_family{};                    ///< Main graphics queue family index.
+    u32 present_family{};                     ///< Main present queue family index.
+    vk::DriverIdKHR driver_id{};              ///< Driver ID.
+    vk::ShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced.ed
+    bool is_optimal_astc_supported{};         ///< Support for native ASTC.
+    bool is_float16_supported{};              ///< Support for float16 arithmetics.
+    bool is_warp_potentially_bigger{};        ///< Host warp size can be bigger than guest.
+    bool is_formatless_image_load_supported{}; ///< Support for shader image read without format.
     bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs.
     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 ext_transform_feedback{};             ///< Support for VK_EXT_transform_feedback.
     bool nv_device_diagnostic_checkpoints{};   ///< Support for VK_NV_device_diagnostic_checkpoints.
-    bool is_shader_storage_img_read_without_format_supported{}; ///< Support for shader storage
-                                                                ///< image read without format
 
     // Telemetry parameters
     std::string vendor_name;                      ///< Device's driver name.
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index ebf85f311..91e7b7791 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -179,10 +179,11 @@ Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine(
 VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
                                  const VKDevice& device, VKScheduler& scheduler,
                                  VKDescriptorPool& descriptor_pool,
-                                 VKUpdateDescriptorQueue& update_descriptor_queue)
+                                 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(device) {}
+      renderpass_cache{renderpass_cache} {}
 
 VKPipelineCache::~VKPipelineCache() = default;
 
@@ -273,9 +274,9 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
     specialization.workgroup_size = key.workgroup_size;
     specialization.shared_memory_size = key.shared_memory_size;
 
-    const SPIRVShader spirv_shader{
-        Decompile(device, shader->GetIR(), ShaderType::Compute, specialization),
-        shader->GetEntries()};
+    const SPIRVShader spirv_shader{Decompile(device, shader->GetIR(), ShaderType::Compute,
+                                             shader->GetRegistry(), specialization),
+                                   shader->GetEntries()};
     entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool,
                                                 update_descriptor_queue, spirv_shader);
     return *entry;
@@ -324,8 +325,7 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
     const auto& gpu = system.GPU().Maxwell3D();
 
     Specialization specialization;
-    specialization.primitive_topology = fixed_state.input_assembly.topology;
-    if (specialization.primitive_topology == Maxwell::PrimitiveTopology::Points) {
+    if (fixed_state.input_assembly.topology == Maxwell::PrimitiveTopology::Points) {
         ASSERT(fixed_state.input_assembly.point_size != 0.0f);
         specialization.point_size = fixed_state.input_assembly.point_size;
     }
@@ -333,9 +333,6 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
         specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].type;
     }
     specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;
-    specialization.tessellation.primitive = fixed_state.tessellation.primitive;
-    specialization.tessellation.spacing = fixed_state.tessellation.spacing;
-    specialization.tessellation.clockwise = fixed_state.tessellation.clockwise;
 
     SPIRVProgram program;
     std::vector<vk::DescriptorSetLayoutBinding> bindings;
@@ -356,8 +353,9 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
         const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
         const auto program_type = GetShaderType(program_enum);
         const auto& entries = shader->GetEntries();
-        program[stage] = {Decompile(device, shader->GetIR(), program_type, specialization),
-                          entries};
+        program[stage] = {
+            Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
+            entries};
 
         if (program_enum == Maxwell::ShaderProgram::VertexA) {
             // VertexB was combined with VertexA, so we skip the VertexB iteration
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
index e292526bb..c4c112290 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
@@ -132,6 +132,10 @@ public:
         return shader_ir;
     }
 
+    const VideoCommon::Shader::Registry& GetRegistry() const {
+        return registry;
+    }
+
     const VideoCommon::Shader::ShaderIR& GetIR() const {
         return shader_ir;
     }
@@ -157,7 +161,8 @@ public:
     explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
                              const VKDevice& device, VKScheduler& scheduler,
                              VKDescriptorPool& descriptor_pool,
-                             VKUpdateDescriptorQueue& update_descriptor_queue);
+                             VKUpdateDescriptorQueue& update_descriptor_queue,
+                             VKRenderPassCache& renderpass_cache);
     ~VKPipelineCache();
 
     std::array<Shader, Maxwell::MaxShaderProgram> GetShaders();
@@ -180,8 +185,7 @@ private:
     VKScheduler& scheduler;
     VKDescriptorPool& descriptor_pool;
     VKUpdateDescriptorQueue& update_descriptor_queue;
-
-    VKRenderPassCache renderpass_cache;
+    VKRenderPassCache& renderpass_cache;
 
     std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
 
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 2bcb17b56..755aad643 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -287,12 +287,13 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
       screen_info{screen_info}, device{device}, resource_manager{resource_manager},
       memory_manager{memory_manager}, state_tracker{state_tracker}, scheduler{scheduler},
       staging_pool(device, memory_manager, scheduler), descriptor_pool(device),
-      update_descriptor_queue(device, scheduler),
+      update_descriptor_queue(device, scheduler), renderpass_cache(device),
       quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
       uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
       texture_cache(system, *this, device, resource_manager, memory_manager, scheduler,
                     staging_pool),
-      pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue),
+      pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue,
+                     renderpass_cache),
       buffer_cache(*this, system, device, memory_manager, scheduler, staging_pool),
       sampler_cache(device), query_cache(system, *this, device, scheduler) {
     scheduler.SetQueryCache(query_cache);
@@ -347,6 +348,8 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
             [&pipeline](auto cmdbuf, auto& dld) { cmdbuf.setCheckpointNV(&pipeline, dld); });
     }
 
+    BeginTransformFeedback();
+
     const auto pipeline_layout = pipeline.GetLayout();
     const auto descriptor_set = pipeline.CommitDescriptorSet();
     scheduler.Record([pipeline_layout, descriptor_set, draw_params](auto cmdbuf, auto& dld) {
@@ -356,18 +359,23 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
         }
         draw_params.Draw(cmdbuf, dld);
     });
+
+    EndTransformFeedback();
 }
 
 void RasterizerVulkan::Clear() {
     MICROPROFILE_SCOPE(Vulkan_Clearing);
 
-    query_cache.UpdateCounters();
-
     const auto& gpu = system.GPU().Maxwell3D();
     if (!system.GPU().Maxwell3D().ShouldExecute()) {
         return;
     }
 
+    sampled_views.clear();
+    image_views.clear();
+
+    query_cache.UpdateCounters();
+
     const auto& regs = gpu.regs;
     const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B ||
                            regs.clear_buffers.A;
@@ -376,52 +384,54 @@ void RasterizerVulkan::Clear() {
     if (!use_color && !use_depth && !use_stencil) {
         return;
     }
-    // Clearing images requires to be out of a renderpass
-    scheduler.RequestOutsideRenderPassOperationContext();
 
-    // TODO(Rodrigo): Implement clears rendering a quad or using beginning a renderpass.
+    [[maybe_unused]] const auto texceptions = UpdateAttachments();
+    DEBUG_ASSERT(texceptions.none());
+    SetupImageTransitions(0, color_attachments, zeta_attachment);
 
-    if (use_color) {
-        View color_view;
-        {
-            MICROPROFILE_SCOPE(Vulkan_RenderTargets);
-            color_view = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT.Value(), false);
-        }
+    const vk::RenderPass renderpass = renderpass_cache.GetRenderPass(GetRenderPassParams(0));
+    const auto [framebuffer, render_area] = ConfigureFramebuffers(renderpass);
+    scheduler.RequestRenderpass({renderpass, framebuffer, {{0, 0}, render_area}, 0, nullptr});
+
+    const auto& scissor = regs.scissor_test[0];
+    const vk::Offset2D scissor_offset(scissor.min_x, scissor.min_y);
+    vk::Extent2D scissor_extent{scissor.max_x - scissor.min_x, scissor.max_y - scissor.min_y};
+    scissor_extent.width = std::min(scissor_extent.width, render_area.width);
+    scissor_extent.height = std::min(scissor_extent.height, render_area.height);
 
-        color_view->Transition(vk::ImageLayout::eTransferDstOptimal,
-                               vk::PipelineStageFlagBits::eTransfer,
-                               vk::AccessFlagBits::eTransferWrite);
+    const u32 layer = regs.clear_buffers.layer;
+    const vk::ClearRect clear_rect({scissor_offset, scissor_extent}, layer, 1);
 
+    if (use_color) {
         const std::array clear_color = {regs.clear_color[0], regs.clear_color[1],
                                         regs.clear_color[2], regs.clear_color[3]};
-        const vk::ClearColorValue clear(clear_color);
-        scheduler.Record([image = color_view->GetImage(),
-                          subresource = color_view->GetImageSubresourceRange(),
-                          clear](auto cmdbuf, auto& dld) {
-            cmdbuf.clearColorImage(image, vk::ImageLayout::eTransferDstOptimal, clear, subresource,
-                                   dld);
+        const vk::ClearValue clear_value{clear_color};
+        const u32 color_attachment = regs.clear_buffers.RT;
+        scheduler.Record([color_attachment, clear_value, clear_rect](auto cmdbuf, auto& dld) {
+            const vk::ClearAttachment attachment(vk::ImageAspectFlagBits::eColor, color_attachment,
+                                                 clear_value);
+            cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld);
         });
     }
-    if (use_depth || use_stencil) {
-        View zeta_surface;
-        {
-            MICROPROFILE_SCOPE(Vulkan_RenderTargets);
-            zeta_surface = texture_cache.GetDepthBufferSurface(false);
-        }
 
-        zeta_surface->Transition(vk::ImageLayout::eTransferDstOptimal,
-                                 vk::PipelineStageFlagBits::eTransfer,
-                                 vk::AccessFlagBits::eTransferWrite);
-
-        const vk::ClearDepthStencilValue clear(regs.clear_depth,
-                                               static_cast<u32>(regs.clear_stencil));
-        scheduler.Record([image = zeta_surface->GetImage(),
-                          subresource = zeta_surface->GetImageSubresourceRange(),
-                          clear](auto cmdbuf, auto& dld) {
-            cmdbuf.clearDepthStencilImage(image, vk::ImageLayout::eTransferDstOptimal, clear,
-                                          subresource, dld);
-        });
+    if (!use_depth && !use_stencil) {
+        return;
+    }
+    vk::ImageAspectFlags aspect_flags;
+    if (use_depth) {
+        aspect_flags |= vk::ImageAspectFlagBits::eDepth;
     }
+    if (use_stencil) {
+        aspect_flags |= vk::ImageAspectFlagBits::eStencil;
+    }
+
+    scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
+                      clear_rect, aspect_flags](auto cmdbuf, auto& dld) {
+        const vk::ClearDepthStencilValue clear_zeta(clear_depth, clear_stencil);
+        const vk::ClearValue clear_value{clear_zeta};
+        const vk::ClearAttachment attachment(aspect_flags, 0, clear_value);
+        cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld);
+    });
 }
 
 void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
@@ -738,6 +748,44 @@ void RasterizerVulkan::UpdateDynamicStates() {
     UpdateStencilFaces(regs);
 }
 
+void RasterizerVulkan::BeginTransformFeedback() {
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    if (regs.tfb_enabled == 0) {
+        return;
+    }
+
+    UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
+                     regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
+                     regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
+
+    UNIMPLEMENTED_IF(regs.tfb_bindings[1].buffer_enable);
+    UNIMPLEMENTED_IF(regs.tfb_bindings[2].buffer_enable);
+    UNIMPLEMENTED_IF(regs.tfb_bindings[3].buffer_enable);
+
+    const auto& binding = regs.tfb_bindings[0];
+    UNIMPLEMENTED_IF(binding.buffer_enable == 0);
+    UNIMPLEMENTED_IF(binding.buffer_offset != 0);
+
+    const GPUVAddr gpu_addr = binding.Address();
+    const std::size_t size = binding.buffer_size;
+    const auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
+
+    scheduler.Record([buffer = *buffer, offset = offset, size](auto cmdbuf, auto& dld) {
+        cmdbuf.bindTransformFeedbackBuffersEXT(0, {buffer}, {offset}, {size}, dld);
+        cmdbuf.beginTransformFeedbackEXT(0, {}, {}, dld);
+    });
+}
+
+void RasterizerVulkan::EndTransformFeedback() {
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    if (regs.tfb_enabled == 0) {
+        return;
+    }
+
+    scheduler.Record(
+        [](auto cmdbuf, auto& dld) { cmdbuf.endTransformFeedbackEXT(0, {}, {}, dld); });
+}
+
 void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex_input,
                                          BufferBindings& buffer_bindings) {
     const auto& regs = system.GPU().Maxwell3D().regs;
@@ -1109,7 +1157,7 @@ std::size_t RasterizerVulkan::CalculateVertexArraysSize() const {
         // This implementation assumes that all attributes are used in the shader.
         const GPUVAddr start{regs.vertex_array[index].StartAddress()};
         const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
-        DEBUG_ASSERT(end > start);
+        DEBUG_ASSERT(end >= start);
 
         size += (end - start + 1) * regs.vertex_array[index].enable;
     }
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 96ea05f0a..3185868e9 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -169,6 +169,10 @@ private:
 
     void UpdateDynamicStates();
 
+    void BeginTransformFeedback();
+
+    void EndTransformFeedback();
+
     bool WalkAttachmentOverlaps(const CachedSurfaceView& attachment);
 
     void SetupVertexArrays(FixedPipelineState::VertexInput& vertex_input,
@@ -249,6 +253,7 @@ private:
     VKStagingBufferPool staging_pool;
     VKDescriptorPool descriptor_pool;
     VKUpdateDescriptorQueue update_descriptor_queue;
+    VKRenderPassCache renderpass_cache;
     QuadArrayPass quad_array_pass;
     Uint8Pass uint8_pass;
 
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index cfcca5af0..51ecb5567 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -5,7 +5,9 @@
 #include <functional>
 #include <limits>
 #include <map>
+#include <optional>
 #include <type_traits>
+#include <unordered_map>
 #include <utility>
 
 #include <fmt/format.h>
@@ -24,6 +26,7 @@
 #include "video_core/renderer_vulkan/vk_shader_decompiler.h"
 #include "video_core/shader/node.h"
 #include "video_core/shader/shader_ir.h"
+#include "video_core/shader/transform_feedback.h"
 
 namespace Vulkan {
 
@@ -93,6 +96,12 @@ struct VertexIndices {
     std::optional<u32> clip_distances;
 };
 
+struct GenericVaryingDescription {
+    Id id = nullptr;
+    u32 first_element = 0;
+    bool is_scalar = false;
+};
+
 spv::Dim GetSamplerDim(const Sampler& sampler) {
     ASSERT(!sampler.IsBuffer());
     switch (sampler.GetType()) {
@@ -266,9 +275,13 @@ bool IsPrecise(Operation operand) {
 class SPIRVDecompiler final : public Sirit::Module {
 public:
     explicit SPIRVDecompiler(const VKDevice& device, const ShaderIR& ir, ShaderType stage,
-                             const Specialization& specialization)
+                             const Registry& registry, const Specialization& specialization)
         : Module(0x00010300), device{device}, ir{ir}, stage{stage}, header{ir.GetHeader()},
-          specialization{specialization} {
+          registry{registry}, specialization{specialization} {
+        if (stage != ShaderType::Compute) {
+            transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo());
+        }
+
         AddCapability(spv::Capability::Shader);
         AddCapability(spv::Capability::UniformAndStorageBuffer16BitAccess);
         AddCapability(spv::Capability::ImageQuery);
@@ -286,6 +299,15 @@ public:
         AddExtension("SPV_KHR_variable_pointers");
         AddExtension("SPV_KHR_shader_draw_parameters");
 
+        if (!transform_feedback.empty()) {
+            if (device.IsExtTransformFeedbackSupported()) {
+                AddCapability(spv::Capability::TransformFeedback);
+            } else {
+                LOG_ERROR(Render_Vulkan, "Shader requires transform feedbacks but these are not "
+                                         "supported on this device");
+            }
+        }
+
         if (ir.UsesLayer() || ir.UsesViewportIndex()) {
             if (ir.UsesViewportIndex()) {
                 AddCapability(spv::Capability::MultiViewport);
@@ -296,7 +318,7 @@ public:
             }
         }
 
-        if (device.IsShaderStorageImageReadWithoutFormatSupported()) {
+        if (device.IsFormatlessImageLoadSupported()) {
             AddCapability(spv::Capability::StorageImageReadWithoutFormat);
         }
 
@@ -318,25 +340,29 @@ public:
             AddExecutionMode(main, spv::ExecutionMode::OutputVertices,
                              header.common2.threads_per_input_primitive);
             break;
-        case ShaderType::TesselationEval:
+        case ShaderType::TesselationEval: {
+            const auto& info = registry.GetGraphicsInfo();
             AddCapability(spv::Capability::Tessellation);
             AddEntryPoint(spv::ExecutionModel::TessellationEvaluation, main, "main", interfaces);
-            AddExecutionMode(main, GetExecutionMode(specialization.tessellation.primitive));
-            AddExecutionMode(main, GetExecutionMode(specialization.tessellation.spacing));
-            AddExecutionMode(main, specialization.tessellation.clockwise
+            AddExecutionMode(main, GetExecutionMode(info.tessellation_primitive));
+            AddExecutionMode(main, GetExecutionMode(info.tessellation_spacing));
+            AddExecutionMode(main, info.tessellation_clockwise
                                        ? spv::ExecutionMode::VertexOrderCw
                                        : spv::ExecutionMode::VertexOrderCcw);
             break;
-        case ShaderType::Geometry:
+        }
+        case ShaderType::Geometry: {
+            const auto& info = registry.GetGraphicsInfo();
             AddCapability(spv::Capability::Geometry);
             AddEntryPoint(spv::ExecutionModel::Geometry, main, "main", interfaces);
-            AddExecutionMode(main, GetExecutionMode(specialization.primitive_topology));
+            AddExecutionMode(main, GetExecutionMode(info.primitive_topology));
             AddExecutionMode(main, GetExecutionMode(header.common3.output_topology));
             AddExecutionMode(main, spv::ExecutionMode::OutputVertices,
                              header.common4.max_output_vertices);
             // TODO(Rodrigo): Where can we get this info from?
             AddExecutionMode(main, spv::ExecutionMode::Invocations, 1U);
             break;
+        }
         case ShaderType::Fragment:
             AddEntryPoint(spv::ExecutionModel::Fragment, main, "main", interfaces);
             AddExecutionMode(main, spv::ExecutionMode::OriginUpperLeft);
@@ -545,7 +571,8 @@ private:
         if (stage != ShaderType::Geometry) {
             return;
         }
-        const u32 num_input = GetNumPrimitiveTopologyVertices(specialization.primitive_topology);
+        const auto& info = registry.GetGraphicsInfo();
+        const u32 num_input = GetNumPrimitiveTopologyVertices(info.primitive_topology);
         DeclareInputVertexArray(num_input);
         DeclareOutputVertex();
     }
@@ -742,12 +769,34 @@ private:
     }
 
     void DeclareOutputAttributes() {
+        if (stage == ShaderType::Compute || stage == ShaderType::Fragment) {
+            return;
+        }
+
+        UNIMPLEMENTED_IF(registry.GetGraphicsInfo().tfb_enabled && stage != ShaderType::Vertex);
         for (const auto index : ir.GetOutputAttributes()) {
             if (!IsGenericAttribute(index)) {
                 continue;
             }
-            const u32 location = GetGenericAttributeLocation(index);
-            Id type = t_float4;
+            DeclareOutputAttribute(index);
+        }
+    }
+
+    void DeclareOutputAttribute(Attribute::Index index) {
+        static constexpr std::string_view swizzle = "xyzw";
+
+        const u32 location = GetGenericAttributeLocation(index);
+        u8 element = 0;
+        while (element < 4) {
+            const std::size_t remainder = 4 - element;
+
+            std::size_t num_components = remainder;
+            const std::optional tfb = GetTransformFeedbackInfo(index, element);
+            if (tfb) {
+                num_components = tfb->components;
+            }
+
+            Id type = GetTypeVectorDefinitionLut(Type::Float).at(num_components - 1);
             Id varying_default = v_varying_default;
             if (IsOutputAttributeArray()) {
                 const u32 num = GetNumOutputVertices();
@@ -760,13 +809,45 @@ private:
             }
             type = TypePointer(spv::StorageClass::Output, type);
 
+            std::string name = fmt::format("out_attr{}", location);
+            if (num_components < 4 || element > 0) {
+                name = fmt::format("{}_{}", name, swizzle.substr(element, num_components));
+            }
+
             const Id id = OpVariable(type, spv::StorageClass::Output, varying_default);
-            Name(AddGlobalVariable(id), fmt::format("out_attr{}", location));
-            output_attributes.emplace(index, id);
+            Name(AddGlobalVariable(id), name);
+
+            GenericVaryingDescription description;
+            description.id = id;
+            description.first_element = element;
+            description.is_scalar = num_components == 1;
+            for (u32 i = 0; i < num_components; ++i) {
+                const u8 offset = static_cast<u8>(static_cast<u32>(index) * 4 + element + i);
+                output_attributes.emplace(offset, description);
+            }
             interfaces.push_back(id);
 
             Decorate(id, spv::Decoration::Location, location);
+            if (element > 0) {
+                Decorate(id, spv::Decoration::Component, static_cast<u32>(element));
+            }
+            if (tfb && device.IsExtTransformFeedbackSupported()) {
+                Decorate(id, spv::Decoration::XfbBuffer, static_cast<u32>(tfb->buffer));
+                Decorate(id, spv::Decoration::XfbStride, static_cast<u32>(tfb->stride));
+                Decorate(id, spv::Decoration::Offset, static_cast<u32>(tfb->offset));
+            }
+
+            element = static_cast<u8>(static_cast<std::size_t>(element) + num_components);
+        }
+    }
+
+    std::optional<VaryingTFB> GetTransformFeedbackInfo(Attribute::Index index, u8 element = 0) {
+        const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element);
+        const auto it = transform_feedback.find(location);
+        if (it == transform_feedback.end()) {
+            return {};
         }
+        return it->second;
     }
 
     u32 DeclareConstantBuffers(u32 binding) {
@@ -898,7 +979,7 @@ private:
     u32 GetNumInputVertices() const {
         switch (stage) {
         case ShaderType::Geometry:
-            return GetNumPrimitiveTopologyVertices(specialization.primitive_topology);
+            return GetNumPrimitiveTopologyVertices(registry.GetGraphicsInfo().primitive_topology);
         case ShaderType::TesselationControl:
         case ShaderType::TesselationEval:
             return NumInputPatches;
@@ -1346,8 +1427,14 @@ private:
                 }
                 default:
                     if (IsGenericAttribute(attribute)) {
-                        const Id composite = output_attributes.at(attribute);
-                        return {ArrayPass(t_out_float, composite, {element}), Type::Float};
+                        const u8 offset = static_cast<u8>(static_cast<u8>(attribute) * 4 + element);
+                        const GenericVaryingDescription description = output_attributes.at(offset);
+                        const Id composite = description.id;
+                        std::vector<u32> indices;
+                        if (!description.is_scalar) {
+                            indices.push_back(element - description.first_element);
+                        }
+                        return {ArrayPass(t_out_float, composite, indices), Type::Float};
                     }
                     UNIMPLEMENTED_MSG("Unhandled output attribute: {}",
                                       static_cast<u32>(attribute));
@@ -1793,7 +1880,7 @@ private:
     }
 
     Expression ImageLoad(Operation operation) {
-        if (!device.IsShaderStorageImageReadWithoutFormatSupported()) {
+        if (!device.IsFormatlessImageLoadSupported()) {
             return {v_float_zero, Type::Float};
         }
 
@@ -2258,11 +2345,11 @@ private:
     std::array<Id, 4> GetTypeVectorDefinitionLut(Type type) const {
         switch (type) {
         case Type::Float:
-            return {nullptr, t_float2, t_float3, t_float4};
+            return {t_float, t_float2, t_float3, t_float4};
         case Type::Int:
-            return {nullptr, t_int2, t_int3, t_int4};
+            return {t_int, t_int2, t_int3, t_int4};
         case Type::Uint:
-            return {nullptr, t_uint2, t_uint3, t_uint4};
+            return {t_uint, t_uint2, t_uint3, t_uint4};
         default:
             UNIMPLEMENTED();
             return {};
@@ -2495,7 +2582,9 @@ private:
     const ShaderIR& ir;
     const ShaderType stage;
     const Tegra::Shader::Header header;
+    const Registry& registry;
     const Specialization& specialization;
+    std::unordered_map<u8, VaryingTFB> transform_feedback;
 
     const Id t_void = Name(TypeVoid(), "void");
 
@@ -2584,7 +2673,7 @@ private:
     Id shared_memory{};
     std::array<Id, INTERNAL_FLAGS_COUNT> internal_flags{};
     std::map<Attribute::Index, Id> input_attributes;
-    std::map<Attribute::Index, Id> output_attributes;
+    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;
@@ -2870,8 +2959,9 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
 }
 
 std::vector<u32> Decompile(const VKDevice& device, const VideoCommon::Shader::ShaderIR& ir,
-                           ShaderType stage, const Specialization& specialization) {
-    return SPIRVDecompiler(device, ir, stage, specialization).Assemble();
+                           ShaderType stage, const VideoCommon::Shader::Registry& registry,
+                           const Specialization& specialization) {
+    return SPIRVDecompiler(device, ir, stage, registry, specialization).Assemble();
 }
 
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index f5dc14d9e..ffea4709e 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -15,6 +15,7 @@
 #include "common/common_types.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/shader_type.h"
+#include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"
 
 namespace Vulkan {
@@ -91,17 +92,9 @@ struct Specialization final {
     u32 shared_memory_size{};
 
     // Graphics specific
-    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 {
-        Maxwell::TessellationPrimitive primitive{};
-        Maxwell::TessellationSpacing spacing{};
-        bool clockwise{};
-    } tessellation;
 };
 // Old gcc versions don't consider this trivially copyable.
 // static_assert(std::is_trivially_copyable_v<Specialization>);
@@ -114,6 +107,8 @@ struct SPIRVShader {
 ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir);
 
 std::vector<u32> Decompile(const VKDevice& device, const VideoCommon::Shader::ShaderIR& ir,
-                           Tegra::Engines::ShaderType stage, const Specialization& specialization);
+                           Tegra::Engines::ShaderType stage,
+                           const VideoCommon::Shader::Registry& registry,
+                           const Specialization& specialization);
 
 } // namespace Vulkan
diff --git a/src/video_core/shader/decode/bfe.cpp b/src/video_core/shader/decode/bfe.cpp
index e02bcd097..8e3b46e8e 100644
--- a/src/video_core/shader/decode/bfe.cpp
+++ b/src/video_core/shader/decode/bfe.cpp
@@ -17,33 +17,60 @@ u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    UNIMPLEMENTED_IF(instr.bfe.negate_b);
-
     Node op_a = GetRegister(instr.gpr8);
-    op_a = GetOperandAbsNegInteger(op_a, false, instr.bfe.negate_a, false);
-
-    switch (opcode->get().GetId()) {
-    case OpCode::Id::BFE_IMM: {
-        UNIMPLEMENTED_IF_MSG(instr.generates_cc,
-                             "Condition codes generation in BFE is not implemented");
+    Node op_b = [&] {
+        switch (opcode->get().GetId()) {
+        case OpCode::Id::BFE_R:
+            return GetRegister(instr.gpr20);
+        case OpCode::Id::BFE_C:
+            return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
+        case OpCode::Id::BFE_IMM:
+            return Immediate(instr.alu.GetSignedImm20_20());
+        default:
+            UNREACHABLE();
+            return Immediate(0);
+        }
+    }();
 
-        const Node inner_shift_imm = Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue()));
-        const Node outer_shift_imm =
-            Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue() + instr.bfe.shift_position));
+    UNIMPLEMENTED_IF_MSG(instr.bfe.rd_cc, "Condition codes in BFE is not implemented");
 
-        const Node inner_shift =
-            Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, inner_shift_imm);
-        const Node outer_shift =
-            Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, inner_shift, outer_shift_imm);
+    const bool is_signed = instr.bfe.is_signed;
 
-        SetInternalFlagsFromInteger(bb, outer_shift, instr.generates_cc);
-        SetRegister(bb, instr.gpr0, outer_shift);
-        break;
-    }
-    default:
-        UNIMPLEMENTED_MSG("Unhandled BFE instruction: {}", opcode->get().GetName());
+    // using reverse parallel method in
+    // https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel
+    // note for later if possible to implement faster method.
+    if (instr.bfe.brev) {
+        const auto swap = [&](u32 s, u32 mask) {
+            Node v1 =
+                SignedOperation(OperationCode::ILogicalShiftRight, is_signed, op_a, Immediate(s));
+            if (mask != 0) {
+                v1 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v1),
+                                     Immediate(mask));
+            }
+            Node v2 = op_a;
+            if (mask != 0) {
+                v2 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v2),
+                                     Immediate(mask));
+            }
+            v2 = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, std::move(v2),
+                                 Immediate(s));
+            return SignedOperation(OperationCode::IBitwiseOr, is_signed, std::move(v1),
+                                   std::move(v2));
+        };
+        op_a = swap(1, 0x55555555U);
+        op_a = swap(2, 0x33333333U);
+        op_a = swap(4, 0x0F0F0F0FU);
+        op_a = swap(8, 0x00FF00FFU);
+        op_a = swap(16, 0);
     }
 
+    const auto offset = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b,
+                                        Immediate(0), Immediate(8));
+    const auto bits = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b,
+                                      Immediate(8), Immediate(8));
+    auto result = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_a, offset, bits);
+    SetRegister(bb, instr.gpr0, std::move(result));
+
     return pc;
 }
 
diff --git a/src/video_core/shader/node_helper.cpp b/src/video_core/shader/node_helper.cpp
index b3dcd291c..76c56abb5 100644
--- a/src/video_core/shader/node_helper.cpp
+++ b/src/video_core/shader/node_helper.cpp
@@ -68,6 +68,8 @@ OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed)
         return OperationCode::UBitwiseXor;
     case OperationCode::IBitwiseNot:
         return OperationCode::UBitwiseNot;
+    case OperationCode::IBitfieldExtract:
+        return OperationCode::UBitfieldExtract;
     case OperationCode::IBitfieldInsert:
         return OperationCode::UBitfieldInsert;
     case OperationCode::IBitCount:
diff --git a/src/video_core/shader/transform_feedback.cpp b/src/video_core/shader/transform_feedback.cpp
new file mode 100644
index 000000000..22a933761
--- /dev/null
+++ b/src/video_core/shader/transform_feedback.cpp
@@ -0,0 +1,115 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <array>
+#include <unordered_map>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/shader/registry.h"
+#include "video_core/shader/transform_feedback.h"
+
+namespace VideoCommon::Shader {
+
+namespace {
+
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+// TODO(Rodrigo): Change this to constexpr std::unordered_set in C++20
+
+/// Attribute offsets that describe a vector
+constexpr std::array VECTORS = {
+    28,  // gl_Position
+    32,  // Generic 0
+    36,  // Generic 1
+    40,  // Generic 2
+    44,  // Generic 3
+    48,  // Generic 4
+    52,  // Generic 5
+    56,  // Generic 6
+    60,  // Generic 7
+    64,  // Generic 8
+    68,  // Generic 9
+    72,  // Generic 10
+    76,  // Generic 11
+    80,  // Generic 12
+    84,  // Generic 13
+    88,  // Generic 14
+    92,  // Generic 15
+    96,  // Generic 16
+    100, // Generic 17
+    104, // Generic 18
+    108, // Generic 19
+    112, // Generic 20
+    116, // Generic 21
+    120, // Generic 22
+    124, // Generic 23
+    128, // Generic 24
+    132, // Generic 25
+    136, // Generic 26
+    140, // Generic 27
+    144, // Generic 28
+    148, // Generic 29
+    152, // Generic 30
+    156, // Generic 31
+    160, // gl_FrontColor
+    164, // gl_FrontSecondaryColor
+    160, // gl_BackColor
+    164, // gl_BackSecondaryColor
+    192, // gl_TexCoord[0]
+    196, // gl_TexCoord[1]
+    200, // gl_TexCoord[2]
+    204, // gl_TexCoord[3]
+    208, // gl_TexCoord[4]
+    212, // gl_TexCoord[5]
+    216, // gl_TexCoord[6]
+    220, // gl_TexCoord[7]
+};
+} // namespace
+
+std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info) {
+
+    std::unordered_map<u8, VaryingTFB> tfb;
+
+    for (std::size_t buffer = 0; buffer < Maxwell::NumTransformFeedbackBuffers; ++buffer) {
+        const auto& locations = info.tfb_varying_locs[buffer];
+        const auto& layout = info.tfb_layouts[buffer];
+        const std::size_t varying_count = layout.varying_count;
+
+        std::size_t highest = 0;
+
+        for (std::size_t offset = 0; offset < varying_count; ++offset) {
+            const std::size_t base_offset = offset;
+            const u8 location = locations[offset];
+
+            VaryingTFB varying;
+            varying.buffer = layout.stream;
+            varying.stride = layout.stride;
+            varying.offset = offset * sizeof(u32);
+            varying.components = 1;
+
+            if (std::find(VECTORS.begin(), VECTORS.end(), location / 4 * 4) != VECTORS.end()) {
+                UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB");
+
+                const u8 base_index = location / 4;
+                while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) {
+                    ++offset;
+                    ++varying.components;
+                }
+            }
+
+            [[maybe_unused]] const bool inserted = tfb.emplace(location, varying).second;
+            UNIMPLEMENTED_IF_MSG(!inserted, "Varying already stored");
+
+            highest = std::max(highest, (base_offset + varying.components) * sizeof(u32));
+        }
+
+        UNIMPLEMENTED_IF(highest != layout.stride);
+    }
+    return tfb;
+}
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/transform_feedback.h b/src/video_core/shader/transform_feedback.h
new file mode 100644
index 000000000..77d05f64c
--- /dev/null
+++ b/src/video_core/shader/transform_feedback.h
@@ -0,0 +1,23 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <unordered_map>
+
+#include "common/common_types.h"
+#include "video_core/shader/registry.h"
+
+namespace VideoCommon::Shader {
+
+struct VaryingTFB {
+    std::size_t buffer;
+    std::size_t stride;
+    std::size_t offset;
+    std::size_t components;
+};
+
+std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info);
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp
index 9707c353d..cc7181229 100644
--- a/src/video_core/surface.cpp
+++ b/src/video_core/surface.cpp
@@ -111,6 +111,8 @@ PixelFormat PixelFormatFromRenderTargetFormat(Tegra::RenderTargetFormat format)
         return PixelFormat::RGBA16F;
     case Tegra::RenderTargetFormat::RGBA16_UNORM:
         return PixelFormat::RGBA16U;
+    case Tegra::RenderTargetFormat::RGBA16_SNORM:
+        return PixelFormat::RGBA16S;
     case Tegra::RenderTargetFormat::RGBA16_UINT:
         return PixelFormat::RGBA16UI;
     case Tegra::RenderTargetFormat::RGBA32_FLOAT:
diff --git a/src/video_core/surface.h b/src/video_core/surface.h
index d88109e5a..ae8817465 100644
--- a/src/video_core/surface.h
+++ b/src/video_core/surface.h
@@ -25,82 +25,83 @@ enum class PixelFormat {
     R8UI = 7,
     RGBA16F = 8,
     RGBA16U = 9,
-    RGBA16UI = 10,
-    R11FG11FB10F = 11,
-    RGBA32UI = 12,
-    DXT1 = 13,
-    DXT23 = 14,
-    DXT45 = 15,
-    DXN1 = 16, // This is also known as BC4
-    DXN2UNORM = 17,
-    DXN2SNORM = 18,
-    BC7U = 19,
-    BC6H_UF16 = 20,
-    BC6H_SF16 = 21,
-    ASTC_2D_4X4 = 22,
-    BGRA8 = 23,
-    RGBA32F = 24,
-    RG32F = 25,
-    R32F = 26,
-    R16F = 27,
-    R16U = 28,
-    R16S = 29,
-    R16UI = 30,
-    R16I = 31,
-    RG16 = 32,
-    RG16F = 33,
-    RG16UI = 34,
-    RG16I = 35,
-    RG16S = 36,
-    RGB32F = 37,
-    RGBA8_SRGB = 38,
-    RG8U = 39,
-    RG8S = 40,
-    RG32UI = 41,
-    RGBX16F = 42,
-    R32UI = 43,
-    R32I = 44,
-    ASTC_2D_8X8 = 45,
-    ASTC_2D_8X5 = 46,
-    ASTC_2D_5X4 = 47,
-    BGRA8_SRGB = 48,
-    DXT1_SRGB = 49,
-    DXT23_SRGB = 50,
-    DXT45_SRGB = 51,
-    BC7U_SRGB = 52,
-    R4G4B4A4U = 53,
-    ASTC_2D_4X4_SRGB = 54,
-    ASTC_2D_8X8_SRGB = 55,
-    ASTC_2D_8X5_SRGB = 56,
-    ASTC_2D_5X4_SRGB = 57,
-    ASTC_2D_5X5 = 58,
-    ASTC_2D_5X5_SRGB = 59,
-    ASTC_2D_10X8 = 60,
-    ASTC_2D_10X8_SRGB = 61,
-    ASTC_2D_6X6 = 62,
-    ASTC_2D_6X6_SRGB = 63,
-    ASTC_2D_10X10 = 64,
-    ASTC_2D_10X10_SRGB = 65,
-    ASTC_2D_12X12 = 66,
-    ASTC_2D_12X12_SRGB = 67,
-    ASTC_2D_8X6 = 68,
-    ASTC_2D_8X6_SRGB = 69,
-    ASTC_2D_6X5 = 70,
-    ASTC_2D_6X5_SRGB = 71,
-    E5B9G9R9F = 72,
+    RGBA16S = 10,
+    RGBA16UI = 11,
+    R11FG11FB10F = 12,
+    RGBA32UI = 13,
+    DXT1 = 14,
+    DXT23 = 15,
+    DXT45 = 16,
+    DXN1 = 17, // This is also known as BC4
+    DXN2UNORM = 18,
+    DXN2SNORM = 19,
+    BC7U = 20,
+    BC6H_UF16 = 21,
+    BC6H_SF16 = 22,
+    ASTC_2D_4X4 = 23,
+    BGRA8 = 24,
+    RGBA32F = 25,
+    RG32F = 26,
+    R32F = 27,
+    R16F = 28,
+    R16U = 29,
+    R16S = 30,
+    R16UI = 31,
+    R16I = 32,
+    RG16 = 33,
+    RG16F = 34,
+    RG16UI = 35,
+    RG16I = 36,
+    RG16S = 37,
+    RGB32F = 38,
+    RGBA8_SRGB = 39,
+    RG8U = 40,
+    RG8S = 41,
+    RG32UI = 42,
+    RGBX16F = 43,
+    R32UI = 44,
+    R32I = 45,
+    ASTC_2D_8X8 = 46,
+    ASTC_2D_8X5 = 47,
+    ASTC_2D_5X4 = 48,
+    BGRA8_SRGB = 49,
+    DXT1_SRGB = 50,
+    DXT23_SRGB = 51,
+    DXT45_SRGB = 52,
+    BC7U_SRGB = 53,
+    R4G4B4A4U = 54,
+    ASTC_2D_4X4_SRGB = 55,
+    ASTC_2D_8X8_SRGB = 56,
+    ASTC_2D_8X5_SRGB = 57,
+    ASTC_2D_5X4_SRGB = 58,
+    ASTC_2D_5X5 = 59,
+    ASTC_2D_5X5_SRGB = 60,
+    ASTC_2D_10X8 = 61,
+    ASTC_2D_10X8_SRGB = 62,
+    ASTC_2D_6X6 = 63,
+    ASTC_2D_6X6_SRGB = 64,
+    ASTC_2D_10X10 = 65,
+    ASTC_2D_10X10_SRGB = 66,
+    ASTC_2D_12X12 = 67,
+    ASTC_2D_12X12_SRGB = 68,
+    ASTC_2D_8X6 = 69,
+    ASTC_2D_8X6_SRGB = 70,
+    ASTC_2D_6X5 = 71,
+    ASTC_2D_6X5_SRGB = 72,
+    E5B9G9R9F = 73,
 
     MaxColorFormat,
 
     // Depth formats
-    Z32F = 73,
-    Z16 = 74,
+    Z32F = 74,
+    Z16 = 75,
 
     MaxDepthFormat,
 
     // DepthStencil formats
-    Z24S8 = 75,
-    S8Z24 = 76,
-    Z32FS8 = 77,
+    Z24S8 = 76,
+    S8Z24 = 77,
+    Z32FS8 = 78,
 
     MaxDepthStencilFormat,
 
@@ -138,6 +139,7 @@ constexpr std::array<u32, MaxPixelFormat> compression_factor_shift_table = {{
     0, // R8UI
     0, // RGBA16F
     0, // RGBA16U
+    0, // RGBA16S
     0, // RGBA16UI
     0, // R11FG11FB10F
     0, // RGBA32UI
@@ -235,6 +237,7 @@ constexpr std::array<u32, MaxPixelFormat> block_width_table = {{
     1,  // R8UI
     1,  // RGBA16F
     1,  // RGBA16U
+    1,  // RGBA16S
     1,  // RGBA16UI
     1,  // R11FG11FB10F
     1,  // RGBA32UI
@@ -324,6 +327,7 @@ constexpr std::array<u32, MaxPixelFormat> block_height_table = {{
     1,  // R8UI
     1,  // RGBA16F
     1,  // RGBA16U
+    1,  // RGBA16S
     1,  // RGBA16UI
     1,  // R11FG11FB10F
     1,  // RGBA32UI
@@ -413,6 +417,7 @@ constexpr std::array<u32, MaxPixelFormat> bpp_table = {{
     8,   // R8UI
     64,  // RGBA16F
     64,  // RGBA16U
+    64,  // RGBA16S
     64,  // RGBA16UI
     32,  // R11FG11FB10F
     128, // RGBA32UI
@@ -517,6 +522,7 @@ constexpr std::array<SurfaceCompression, MaxPixelFormat> compression_type_table
     SurfaceCompression::None,       // R8UI
     SurfaceCompression::None,       // RGBA16F
     SurfaceCompression::None,       // RGBA16U
+    SurfaceCompression::None,       // RGBA16S
     SurfaceCompression::None,       // RGBA16UI
     SurfaceCompression::None,       // R11FG11FB10F
     SurfaceCompression::None,       // RGBA32UI
diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp
index cc3ad8417..e151c26c4 100644
--- a/src/video_core/texture_cache/format_lookup_table.cpp
+++ b/src/video_core/texture_cache/format_lookup_table.cpp
@@ -41,7 +41,7 @@ struct Table {
     ComponentType alpha_component;
     bool is_srgb;
 };
-constexpr std::array<Table, 75> DefinitionTable = {{
+constexpr std::array<Table, 76> DefinitionTable = {{
     {TextureFormat::A8R8G8B8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ABGR8U},
     {TextureFormat::A8R8G8B8, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::ABGR8S},
     {TextureFormat::A8R8G8B8, C, UINT, UINT, UINT, UINT, PixelFormat::ABGR8UI},
@@ -61,6 +61,7 @@ constexpr std::array<Table, 75> DefinitionTable = {{
     {TextureFormat::G8R8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::RG8U},
     {TextureFormat::G8R8, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::RG8S},
 
+    {TextureFormat::R16_G16_B16_A16, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::RGBA16S},
     {TextureFormat::R16_G16_B16_A16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::RGBA16U},
     {TextureFormat::R16_G16_B16_A16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RGBA16F},
     {TextureFormat::R16_G16_B16_A16, C, UINT, UINT, UINT, UINT, PixelFormat::RGBA16UI},
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index f00839313..9931c5ef7 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -113,8 +113,10 @@ SurfaceParams SurfaceParams::CreateForTexture(const FormatLookupTable& lookup_ta
         params.height = tic.Height();
         params.depth = tic.Depth();
         params.pitch = params.is_tiled ? 0 : tic.Pitch();
-        if (params.target == SurfaceTarget::TextureCubemap ||
-            params.target == SurfaceTarget::TextureCubeArray) {
+        if (params.target == SurfaceTarget::Texture2D && params.depth > 1) {
+            params.depth = 1;
+        } else if (params.target == SurfaceTarget::TextureCubemap ||
+                   params.target == SurfaceTarget::TextureCubeArray) {
             params.depth *= 6;
         }
         params.num_levels = tic.max_mip_level + 1;
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 51373b687..6cdbe63d0 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -104,6 +104,11 @@ public:
         if (!cache_addr) {
             return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
         }
+
+        if (!IsTypeCompatible(tic.texture_type, entry)) {
+            return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
+        }
+
         const auto params{SurfaceParams::CreateForTexture(format_lookup_table, tic, entry)};
         const auto [surface, view] = GetSurface(gpu_addr, cache_addr, params, true, false);
         if (guard_samplers) {
@@ -914,13 +919,15 @@ private:
         params.width = 1;
         params.height = 1;
         params.depth = 1;
+        if (target == SurfaceTarget::TextureCubemap || target == SurfaceTarget::TextureCubeArray) {
+            params.depth = 6;
+        }
         params.pitch = 4;
         params.num_levels = 1;
         params.emulated_levels = 1;
-        params.pixel_format = VideoCore::Surface::PixelFormat::RGBA16F;
+        params.pixel_format = VideoCore::Surface::PixelFormat::R8U;
         params.type = VideoCore::Surface::SurfaceType::ColorTexture;
         auto surface = CreateSurface(0ULL, params);
-        invalid_memory.clear();
         invalid_memory.resize(surface->GetHostSizeInBytes(), 0U);
         surface->UploadTexture(invalid_memory);
         surface->MarkAsModified(false, Tick());
@@ -1082,6 +1089,36 @@ private:
         return siblings_table[static_cast<std::size_t>(format)];
     }
 
+    /// Returns true the shader sampler entry is compatible with the TIC texture type.
+    static bool IsTypeCompatible(Tegra::Texture::TextureType tic_type,
+                                 const VideoCommon::Shader::Sampler& entry) {
+        const auto shader_type = entry.GetType();
+        switch (tic_type) {
+        case Tegra::Texture::TextureType::Texture1D:
+        case Tegra::Texture::TextureType::Texture1DArray:
+            return shader_type == Tegra::Shader::TextureType::Texture1D;
+        case Tegra::Texture::TextureType::Texture1DBuffer:
+            // TODO(Rodrigo): Assume as valid for now
+            return true;
+        case Tegra::Texture::TextureType::Texture2D:
+        case Tegra::Texture::TextureType::Texture2DNoMipmap:
+            return shader_type == Tegra::Shader::TextureType::Texture2D;
+        case Tegra::Texture::TextureType::Texture2DArray:
+            return shader_type == Tegra::Shader::TextureType::Texture2D ||
+                   shader_type == Tegra::Shader::TextureType::TextureCube;
+        case Tegra::Texture::TextureType::Texture3D:
+            return shader_type == Tegra::Shader::TextureType::Texture3D;
+        case Tegra::Texture::TextureType::TextureCubeArray:
+        case Tegra::Texture::TextureType::TextureCubemap:
+            if (shader_type == Tegra::Shader::TextureType::TextureCube) {
+                return true;
+            }
+            return shader_type == Tegra::Shader::TextureType::Texture2D && entry.IsArray();
+        }
+        UNREACHABLE();
+        return true;
+    }
+
     struct FramebufferTargetInfo {
         TSurface target;
         TView view;