gl_shader_decompiler: Implement image binding settings

CMakeModules/GenerateSCMRev.cmake

@@ -70,6 +70,7 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/decode/half_set.cpp"
     "${VIDEO_CORE}/shader/decode/half_set_predicate.cpp"
     "${VIDEO_CORE}/shader/decode/hfma2.cpp"
+    "${VIDEO_CORE}/shader/decode/image.cpp"
     "${VIDEO_CORE}/shader/decode/integer_set.cpp"
     "${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
     "${VIDEO_CORE}/shader/decode/memory.cpp"

src/common/CMakeLists.txt

@@ -44,6 +44,7 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/decode/half_set.cpp"
       "${VIDEO_CORE}/shader/decode/half_set_predicate.cpp"
       "${VIDEO_CORE}/shader/decode/hfma2.cpp"
+      "${VIDEO_CORE}/shader/decode/image.cpp"
       "${VIDEO_CORE}/shader/decode/integer_set.cpp"
       "${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
       "${VIDEO_CORE}/shader/decode/memory.cpp"

src/video_core/CMakeLists.txt

@@ -89,6 +89,7 @@ add_library(video_core STATIC
     shader/decode/conversion.cpp
     shader/decode/memory.cpp
     shader/decode/texture.cpp
+    shader/decode/image.cpp
     shader/decode/float_set_predicate.cpp
     shader/decode/integer_set_predicate.cpp
     shader/decode/half_set_predicate.cpp

src/video_core/engines/maxwell_3d.cpp

@@ -432,14 +432,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
     Texture::TICEntry tic_entry;
     memory_manager.ReadBlockUnsafe(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry));
 
-    ASSERT_MSG(tic_entry.header_version == Texture::TICHeaderVersion::BlockLinear ||
-                   tic_entry.header_version == Texture::TICHeaderVersion::Pitch,
-               "TIC versions other than BlockLinear or Pitch are unimplemented");
-
-    const auto r_type = tic_entry.r_type.Value();
-    const auto g_type = tic_entry.g_type.Value();
-    const auto b_type = tic_entry.b_type.Value();
-    const auto a_type = tic_entry.a_type.Value();
+    const auto r_type{tic_entry.r_type.Value()};
+    const auto g_type{tic_entry.g_type.Value()};
+    const auto b_type{tic_entry.b_type.Value()};
+    const auto a_type{tic_entry.a_type.Value()};
 
     // TODO(Subv): Different data types for separate components are not supported
     ASSERT(r_type == g_type && r_type == b_type && r_type == a_type);

src/video_core/engines/shader_bytecode.h

@@ -122,6 +122,15 @@ union Sampler {
     u64 value{};
 };
 
+union Image {
+    Image() = default;
+
+    constexpr explicit Image(u64 value) : value{value} {}
+
+    BitField<36, 13, u64> index;
+    u64 value;
+};
+
 } // namespace Tegra::Shader
 
 namespace std {
@@ -340,6 +349,26 @@ enum class TextureMiscMode : u64 {
     PTP,
 };
 
+enum class SurfaceDataMode : u64 {
+    P = 0,
+    D_BA = 1,
+};
+
+enum class OutOfBoundsStore : u64 {
+    Ignore = 0,
+    Clamp = 1,
+    Trap = 2,
+};
+
+enum class ImageType : u64 {
+    Texture1D = 0,
+    TextureBuffer = 1,
+    Texture1DArray = 2,
+    Texture2D = 3,
+    Texture2DArray = 4,
+    Texture3D = 5,
+};
+
 enum class IsberdMode : u64 {
     None = 0,
     Patch = 1,
@@ -394,7 +423,7 @@ enum class LmemLoadCacheManagement : u64 {
     CV = 3,
 };
 
-enum class LmemStoreCacheManagement : u64 {
+enum class StoreCacheManagement : u64 {
     Default = 0,
     CG = 1,
     CS = 2,
@@ -796,7 +825,7 @@ union Instruction {
     } ld_l;
 
     union {
-        BitField<44, 2, LmemStoreCacheManagement> cache_management;
+        BitField<44, 2, StoreCacheManagement> cache_management;
     } st_l;
 
     union {
@@ -1207,6 +1236,20 @@ union Instruction {
         }
     } texs;
 
+    union {
+        BitField<28, 1, u64> is_array;
+        BitField<29, 2, TextureType> texture_type;
+        BitField<35, 1, u64> aoffi;
+        BitField<49, 1, u64> nodep_flag;
+        BitField<50, 1, u64> ms; // Multisample?
+        BitField<54, 1, u64> cl;
+        BitField<55, 1, u64> process_mode;
+
+        TextureProcessMode GetTextureProcessMode() const {
+            return process_mode == 0 ? TextureProcessMode::LZ : TextureProcessMode::LL;
+        }
+    } tld;
+
     union {
         BitField<49, 1, u64> nodep_flag;
         BitField<53, 4, u64> texture_info;
@@ -1256,6 +1299,35 @@ union Instruction {
         }
     } tlds;
 
+    union {
+        BitField<24, 2, StoreCacheManagement> cache_management;
+        BitField<33, 3, ImageType> image_type;
+        BitField<49, 2, OutOfBoundsStore> out_of_bounds_store;
+        BitField<51, 1, u64> is_immediate;
+        BitField<52, 1, SurfaceDataMode> mode;
+
+        BitField<20, 3, StoreType> store_data_layout;
+        BitField<20, 4, u64> component_mask_selector;
+
+        bool IsComponentEnabled(std::size_t component) const {
+            ASSERT(mode == SurfaceDataMode::P);
+            constexpr u8 R = 0b0001;
+            constexpr u8 G = 0b0010;
+            constexpr u8 B = 0b0100;
+            constexpr u8 A = 0b1000;
+            constexpr std::array<u8, 16> mask = {
+                0,       (R),         (G),         (R | G),        (B),     (R | B),
+                (G | B), (R | G | B), (A),         (R | A),        (G | A), (R | G | A),
+                (B | A), (R | B | A), (G | B | A), (R | G | B | A)};
+            return std::bitset<4>{mask.at(component_mask_selector)}.test(component);
+        }
+
+        StoreType GetStoreDataLayout() const {
+            ASSERT(mode == SurfaceDataMode::D_BA);
+            return store_data_layout;
+        }
+    } sust;
+
     union {
         BitField<20, 24, u64> target;
         BitField<5, 1, u64> constant_buffer;
@@ -1347,6 +1419,7 @@ union Instruction {
 
     Attribute attribute;
     Sampler sampler;
+    Image image;
 
     u64 value;
 };
@@ -1381,11 +1454,13 @@ public:
         TXQ,    // Texture Query
         TXQ_B,  // Texture Query Bindless
         TEXS,   // Texture Fetch with scalar/non-vec4 source/destinations
+        TLD,    // Texture Load
         TLDS,   // Texture Load with scalar/non-vec4 source/destinations
         TLD4,   // Texture Load 4
         TLD4S,  // Texture Load 4 with scalar / non - vec4 source / destinations
         TMML_B, // Texture Mip Map Level
         TMML,   // Texture Mip Map Level
+        SUST,   // Surface Store
         EXIT,
         IPA,
         OUT_R, // Emit vertex/primitive
@@ -1516,6 +1591,7 @@ public:
         Synch,
         Memory,
         Texture,
+        Image,
         FloatSet,
         FloatSetPredicate,
         IntegerSet,
@@ -1651,11 +1727,13 @@ private:
             INST("1101111101001---", Id::TXQ, Type::Texture, "TXQ"),
             INST("1101111101010---", Id::TXQ_B, Type::Texture, "TXQ_B"),
             INST("1101-00---------", Id::TEXS, Type::Texture, "TEXS"),
+            INST("11011100--11----", Id::TLD, Type::Texture, "TLD"),
             INST("1101101---------", Id::TLDS, Type::Texture, "TLDS"),
             INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"),
             INST("1101111100------", Id::TLD4S, Type::Texture, "TLD4S"),
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
+            INST("11101011001-----", Id::SUST, Type::Image, "SUST"),
             INST("111000110000----", Id::EXIT, Type::Trivial, "EXIT"),
             INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),

src/video_core/renderer_opengl/gl_device.cpp

@@ -24,6 +24,11 @@ Device::Device() {
     has_variable_aoffi = TestVariableAoffi();
 }
 
+Device::Device(std::nullptr_t) {
+    uniform_buffer_alignment = 0;
+    has_variable_aoffi = true;
+}
+
 bool Device::TestVariableAoffi() {
     const GLchar* AOFFI_TEST = R"(#version 430 core
 uniform sampler2D tex;

src/video_core/renderer_opengl/gl_device.h

@@ -10,7 +10,8 @@ namespace OpenGL {
 
 class Device {
 public:
-    Device();
+    explicit Device();
+    explicit Device(std::nullptr_t);
 
     std::size_t GetUniformBufferAlignment() const {
         return uniform_buffer_alignment;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -29,8 +29,10 @@
 namespace OpenGL {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using PixelFormat = VideoCore::Surface::PixelFormat;
-using SurfaceType = VideoCore::Surface::SurfaceType;
+
+using VideoCore::Surface::PixelFormat;
+using VideoCore::Surface::SurfaceTarget;
+using VideoCore::Surface::SurfaceType;
 
 MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Format Setup", MP_RGB(128, 128, 192));
 MICROPROFILE_DEFINE(OpenGL_VB, "OpenGL", "Vertex Buffer Setup", MP_RGB(128, 128, 192));
@@ -119,11 +121,6 @@ void RasterizerOpenGL::CheckExtensions() {
             Render_OpenGL,
             "Anisotropic filter is not supported! This can cause graphical issues in some games.");
     }
-    if (!GLAD_GL_ARB_buffer_storage) {
-        LOG_WARNING(
-            Render_OpenGL,
-            "Buffer storage control is not supported! This can cause performance degradation.");
-    }
 }
 
 GLuint RasterizerOpenGL::SetupVertexFormat() {
@@ -323,8 +320,14 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
                                  static_cast<GLsizeiptr>(sizeof(ubo)));
 
         Shader shader{shader_cache.GetStageProgram(program)};
-        const auto [program_handle, next_bindings] =
-            shader->GetProgramHandle(primitive_mode, base_bindings);
+
+        const auto stage_enum{static_cast<Maxwell::ShaderStage>(stage)};
+        SetupConstBuffers(stage_enum, shader, base_bindings);
+        SetupGlobalRegions(stage_enum, shader, base_bindings);
+        const auto texture_buffer_usage{SetupTextures(stage_enum, shader, base_bindings)};
+
+        const ProgramVariant variant{base_bindings, primitive_mode, texture_buffer_usage};
+        const auto [program_handle, next_bindings] = shader->GetProgramHandle(variant);
 
         switch (program) {
         case Maxwell::ShaderProgram::VertexA:
@@ -342,11 +345,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
                               shader_config.enable.Value(), shader_config.offset);
         }
 
-        const auto stage_enum = static_cast<Maxwell::ShaderStage>(stage);
-        SetupConstBuffers(stage_enum, shader, program_handle, base_bindings);
-        SetupGlobalRegions(stage_enum, shader, program_handle, base_bindings);
-        SetupTextures(stage_enum, shader, program_handle, base_bindings);
-
         // Workaround for Intel drivers.
         // When a clip distance is enabled but not set in the shader it crops parts of the screen
         // (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the
@@ -809,8 +807,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 }
 
 void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                                         const Shader& shader, GLuint program_handle,
-                                         BaseBindings base_bindings) {
+                                         const Shader& shader, BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
     const auto& gpu = system.GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
@@ -857,8 +854,7 @@ void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::Shader
 }
 
 void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                                          const Shader& shader, GLenum primitive_mode,
-                                          BaseBindings base_bindings) {
+                                          const Shader& shader, BaseBindings base_bindings) {
     const auto& entries = shader->GetShaderEntries().global_memory_entries;
     for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& entry{entries[bindpoint]};
@@ -871,8 +867,8 @@ void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::Shade
     }
 }
 
-void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
-                                     GLuint program_handle, BaseBindings base_bindings) {
+TextureBufferUsage RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
+                                                   BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& gpu = system.GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
@@ -881,6 +877,8 @@ void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& s
     ASSERT_MSG(base_bindings.sampler + entries.size() <= std::size(state.texture_units),
                "Exceeded the number of active textures.");
 
+    TextureBufferUsage texture_buffer_usage{0};
+
     for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& entry = entries[bindpoint];
         Tegra::Texture::FullTextureInfo texture;
@@ -894,18 +892,25 @@ void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& s
         }
         const u32 current_bindpoint = base_bindings.sampler + bindpoint;
 
-        state.texture_units[current_bindpoint].sampler = sampler_cache.GetSampler(texture.tsc);
+        auto& unit{state.texture_units[current_bindpoint]};
+        unit.sampler = sampler_cache.GetSampler(texture.tsc);
 
         if (Surface surface = res_cache.GetTextureSurface(texture, entry); surface) {
-            state.texture_units[current_bindpoint].texture =
-                surface->Texture(entry.IsArray()).handle;
+            if (surface->GetSurfaceParams().target == SurfaceTarget::TextureBuffer) {
+                // Record that this texture is a texture buffer.
+                texture_buffer_usage.set(bindpoint);
+            }
+
+            unit.texture = surface->Texture(entry.IsArray()).handle;
             surface->UpdateSwizzle(texture.tic.x_source, texture.tic.y_source, texture.tic.z_source,
                                    texture.tic.w_source);
         } else {
             // Can occur when texture addr is null or its memory is unmapped/invalid
-            state.texture_units[current_bindpoint].texture = 0;
+            unit.texture = 0;
         }
     }
+
+    return texture_buffer_usage;
 }
 
 void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -106,16 +106,16 @@ private:
 
     /// Configures the current constbuffers to use for the draw command.
     void SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, const Shader& shader,
-                           GLuint program_handle, BaseBindings base_bindings);
+                           BaseBindings base_bindings);
 
     /// Configures the current global memory entries to use for the draw command.
     void SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                            const Shader& shader, GLenum primitive_mode,
-                            BaseBindings base_bindings);
+                            const Shader& shader, BaseBindings base_bindings);
 
-    /// Configures the current textures to use for the draw command.
-    void SetupTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, const Shader& shader,
-                       GLuint program_handle, BaseBindings base_bindings);
+    /// Configures the current textures to use for the draw command. Returns shaders texture buffer
+    /// usage.
+    TextureBufferUsage SetupTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+                                     const Shader& shader, BaseBindings base_bindings);
 
     /// Syncs the viewport and depth range to match the guest state
     void SyncViewport(OpenGLState& current_state);

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -140,7 +140,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
 
     params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format));
     params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format));
-    if (!params.is_tiled) {
+    if (config.tic.IsLineal()) {
         params.pitch = config.tic.Pitch();
     }
     params.unaligned_height = config.tic.Height();
@@ -149,6 +149,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
 
     switch (params.target) {
     case SurfaceTarget::Texture1D:
+    case SurfaceTarget::TextureBuffer:
     case SurfaceTarget::Texture2D:
         params.depth = 1;
         break;
@@ -389,6 +390,8 @@ static GLenum SurfaceTargetToGL(SurfaceTarget target) {
     switch (target) {
     case SurfaceTarget::Texture1D:
         return GL_TEXTURE_1D;
+    case SurfaceTarget::TextureBuffer:
+        return GL_TEXTURE_BUFFER;
     case SurfaceTarget::Texture2D:
         return GL_TEXTURE_2D;
     case SurfaceTarget::Texture3D:
@@ -600,29 +603,35 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
 
     switch (params.target) {
     case SurfaceTarget::Texture1D:
-        glTextureStorage1D(texture.handle, params.max_mip_level, format_tuple.internal_format,
-                           width);
+        glTextureStorage1D(texture.handle, params.max_mip_level, gl_internal_format, width);
+        break;
+    case SurfaceTarget::TextureBuffer:
+        texture_buffer.Create();
+        glNamedBufferStorage(texture_buffer.handle,
+                             params.width * GetBytesPerPixel(params.pixel_format), nullptr,
+                             GL_DYNAMIC_STORAGE_BIT);
+        glTextureBuffer(texture.handle, gl_internal_format, texture_buffer.handle);
         break;
     case SurfaceTarget::Texture2D:
     case SurfaceTarget::TextureCubemap:
-        glTextureStorage2D(texture.handle, params.max_mip_level, format_tuple.internal_format,
-                           width, height);
+        glTextureStorage2D(texture.handle, params.max_mip_level, gl_internal_format, width, height);
         break;
     case SurfaceTarget::Texture3D:
     case SurfaceTarget::Texture2DArray:
     case SurfaceTarget::TextureCubeArray:
-        glTextureStorage3D(texture.handle, params.max_mip_level, format_tuple.internal_format,
-                           width, height, params.depth);
+        glTextureStorage3D(texture.handle, params.max_mip_level, gl_internal_format, width, height,
+                           params.depth);
         break;
     default:
         LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}",
                      static_cast<u32>(params.target));
         UNREACHABLE();
-        glTextureStorage2D(texture.handle, params.max_mip_level, format_tuple.internal_format,
-                           width, height);
+        glTextureStorage2D(texture.handle, params.max_mip_level, gl_internal_format, width, height);
     }
 
-    ApplyTextureDefaults(texture.handle, params.max_mip_level);
+    if (params.target != SurfaceTarget::TextureBuffer) {
+        ApplyTextureDefaults(texture.handle, params.max_mip_level);
+    }
 
     OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.gpu_addr, params.IdentityString());
 }
@@ -785,6 +794,13 @@ void CachedSurface::UploadGLMipmapTexture(RasterizerTemporaryMemory& res_cache_t
             glTextureSubImage1D(texture.handle, mip_map, x0, static_cast<GLsizei>(rect.GetWidth()),
                                 tuple.format, tuple.type, &gl_buffer[mip_map][buffer_offset]);
             break;
+        case SurfaceTarget::TextureBuffer:
+            ASSERT(mip_map == 0);
+            glNamedBufferSubData(texture_buffer.handle, x0,
+                                 static_cast<GLsizeiptr>(rect.GetWidth()) *
+                                     GetBytesPerPixel(params.pixel_format),
+                                 &gl_buffer[mip_map][buffer_offset]);
+            break;
         case SurfaceTarget::Texture2D:
             glTextureSubImage2D(texture.handle, mip_map, x0, y0,
                                 static_cast<GLsizei>(rect.GetWidth()),
@@ -860,6 +876,9 @@ void CachedSurface::UpdateSwizzle(Tegra::Texture::SwizzleSource swizzle_x,
                                   Tegra::Texture::SwizzleSource swizzle_y,
                                   Tegra::Texture::SwizzleSource swizzle_z,
                                   Tegra::Texture::SwizzleSource swizzle_w) {
+    if (params.target == SurfaceTarget::TextureBuffer) {
+        return;
+    }
     const GLenum new_x = MaxwellToGL::SwizzleSource(swizzle_x);
     const GLenum new_y = MaxwellToGL::SwizzleSource(swizzle_y);
     const GLenum new_z = MaxwellToGL::SwizzleSource(swizzle_z);

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -250,6 +250,8 @@ struct SurfaceParams {
         switch (target) {
         case SurfaceTarget::Texture1D:
             return "1D";
+        case SurfaceTarget::TextureBuffer:
+            return "Buffer";
         case SurfaceTarget::Texture2D:
             return "2D";
         case SurfaceTarget::Texture3D:
@@ -439,6 +441,7 @@ private:
 
     OGLTexture texture;
     OGLTexture discrepant_view;
+    OGLBuffer texture_buffer;
     SurfaceParams params{};
     GLenum gl_target{};
     GLenum gl_internal_format{};

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -164,8 +164,12 @@ GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgr
 }
 
 CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
-                               Maxwell::ShaderProgram program_type, BaseBindings base_bindings,
-                               GLenum primitive_mode, bool hint_retrievable = false) {
+                               Maxwell::ShaderProgram program_type, const ProgramVariant& variant,
+                               bool hint_retrievable = false) {
+    auto base_bindings{variant.base_bindings};
+    const auto primitive_mode{variant.primitive_mode};
+    const auto texture_buffer_usage{variant.texture_buffer_usage};
+
     std::string source = "#version 430 core\n";
     source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
 
@@ -181,6 +185,18 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
         source += fmt::format("#define SAMPLER_BINDING_{} {}\n", sampler.GetIndex(),
                               base_bindings.sampler++);
     }
+    for (const auto& image : entries.images) {
+        source +=
+            fmt::format("#define IMAGE_BINDING_{} {}\n", image.GetIndex(), base_bindings.image++);
+    }
+
+    // Transform 1D textures to texture samplers by declaring its preprocessor macros.
+    for (std::size_t i = 0; i < texture_buffer_usage.size(); ++i) {
+        if (!texture_buffer_usage.test(i)) {
+            continue;
+        }
+        source += fmt::format("#define SAMPLER_{}_IS_BUFFER", i);
+    }
 
     if (program_type == Maxwell::ShaderProgram::Geometry) {
         const auto [glsl_topology, debug_name, max_vertices] =
@@ -256,20 +272,18 @@ CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier,
     shader_length = entries.shader_length;
 }
 
-std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive_mode,
-                                                                BaseBindings base_bindings) {
+std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVariant& variant) {
     GLuint handle{};
     if (program_type == Maxwell::ShaderProgram::Geometry) {
-        handle = GetGeometryShader(primitive_mode, base_bindings);
+        handle = GetGeometryShader(variant);
     } else {
-        const auto [entry, is_cache_miss] = programs.try_emplace(base_bindings);
+        const auto [entry, is_cache_miss] = programs.try_emplace(variant);
         auto& program = entry->second;
         if (is_cache_miss) {
-            program = TryLoadProgram(primitive_mode, base_bindings);
+            program = TryLoadProgram(variant);
             if (!program) {
-                program =
-                    SpecializeShader(code, entries, program_type, base_bindings, primitive_mode);
-                disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
+                program = SpecializeShader(code, entries, program_type, variant);
+                disk_cache.SaveUsage(GetUsage(variant));
             }
 
             LabelGLObject(GL_PROGRAM, program->handle, cpu_addr);
@@ -278,6 +292,7 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive
         handle = program->handle;
     }
 
+    auto base_bindings{variant.base_bindings};
     base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + RESERVED_UBOS;
     base_bindings.gmem += static_cast<u32>(entries.global_memory_entries.size());
     base_bindings.sampler += static_cast<u32>(entries.samplers.size());
@@ -285,43 +300,42 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive
     return {handle, base_bindings};
 }
 
-GLuint CachedShader::GetGeometryShader(GLenum primitive_mode, BaseBindings base_bindings) {
-    const auto [entry, is_cache_miss] = geometry_programs.try_emplace(base_bindings);
+GLuint CachedShader::GetGeometryShader(const ProgramVariant& variant) {
+    const auto [entry, is_cache_miss] = geometry_programs.try_emplace(variant);
     auto& programs = entry->second;
 
-    switch (primitive_mode) {
+    switch (variant.primitive_mode) {
     case GL_POINTS:
-        return LazyGeometryProgram(programs.points, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.points, variant);
     case GL_LINES:
     case GL_LINE_STRIP:
-        return LazyGeometryProgram(programs.lines, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.lines, variant);
     case GL_LINES_ADJACENCY:
     case GL_LINE_STRIP_ADJACENCY:
-        return LazyGeometryProgram(programs.lines_adjacency, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.lines_adjacency, variant);
     case GL_TRIANGLES:
     case GL_TRIANGLE_STRIP:
     case GL_TRIANGLE_FAN:
-        return LazyGeometryProgram(programs.triangles, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.triangles, variant);
     case GL_TRIANGLES_ADJACENCY:
     case GL_TRIANGLE_STRIP_ADJACENCY:
-        return LazyGeometryProgram(programs.triangles_adjacency, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.triangles_adjacency, variant);
     default:
         UNREACHABLE_MSG("Unknown primitive mode.");
-        return LazyGeometryProgram(programs.points, base_bindings, primitive_mode);
+        return LazyGeometryProgram(programs.points, variant);
     }
 }
 
-GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBindings base_bindings,
-                                         GLenum primitive_mode) {
+GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program,
+                                         const ProgramVariant& variant) {
     if (target_program) {
         return target_program->handle;
     }
-    const auto [glsl_name, debug_name, vertices] = GetPrimitiveDescription(primitive_mode);
-    target_program = TryLoadProgram(primitive_mode, base_bindings);
+    const auto [glsl_name, debug_name, vertices] = GetPrimitiveDescription(variant.primitive_mode);
+    target_program = TryLoadProgram(variant);
     if (!target_program) {
-        target_program =
-            SpecializeShader(code, entries, program_type, base_bindings, primitive_mode);
-        disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
+        target_program = SpecializeShader(code, entries, program_type, variant);
+        disk_cache.SaveUsage(GetUsage(variant));
     }
 
     LabelGLObject(GL_PROGRAM, target_program->handle, cpu_addr, debug_name);
@@ -329,18 +343,19 @@ GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBind
     return target_program->handle;
 };
 
-CachedProgram CachedShader::TryLoadProgram(GLenum primitive_mode,
-                                           BaseBindings base_bindings) const {
-    const auto found = precompiled_programs.find(GetUsage(primitive_mode, base_bindings));
+CachedProgram CachedShader::TryLoadProgram(const ProgramVariant& variant) const {
+    const auto found = precompiled_programs.find(GetUsage(variant));
     if (found == precompiled_programs.end()) {
         return {};
     }
     return found->second;
 }
 
-ShaderDiskCacheUsage CachedShader::GetUsage(GLenum primitive_mode,
-                                            BaseBindings base_bindings) const {
-    return {unique_identifier, base_bindings, primitive_mode};
+ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant) const {
+    ShaderDiskCacheUsage usage;
+    usage.unique_identifier = unique_identifier;
+    usage.variant = variant;
+    return usage;
 }
 
 ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
@@ -394,7 +409,7 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
         }
         if (!shader) {
             shader = SpecializeShader(unspec.code, unspec.entries, unspec.program_type,
-                                      usage.bindings, usage.primitive, true);
+                                      usage.variant, true);
         }
         precompiled_programs.insert({usage, std::move(shader)});
 

src/video_core/renderer_opengl/gl_shader_cache.h

@@ -6,6 +6,7 @@
 
 #include <array>
 #include <atomic>
+#include <bitset>
 #include <memory>
 #include <set>
 #include <tuple>
@@ -22,7 +23,7 @@
 
 namespace Core {
 class System;
-} // namespace Core
+}
 
 namespace OpenGL {
 
@@ -63,8 +64,7 @@ public:
     }
 
     /// Gets the GL program handle for the shader
-    std::tuple<GLuint, BaseBindings> GetProgramHandle(GLenum primitive_mode,
-                                                      BaseBindings base_bindings);
+    std::tuple<GLuint, BaseBindings> GetProgramHandle(const ProgramVariant& variant);
 
 private:
     // Geometry programs. These are needed because GLSL needs an input topology but it's not
@@ -78,15 +78,14 @@ private:
         CachedProgram triangles_adjacency;
     };
 
-    GLuint GetGeometryShader(GLenum primitive_mode, BaseBindings base_bindings);
+    GLuint GetGeometryShader(const ProgramVariant& variant);
 
     /// Generates a geometry shader or returns one that already exists.
-    GLuint LazyGeometryProgram(CachedProgram& target_program, BaseBindings base_bindings,
-                               GLenum primitive_mode);
+    GLuint LazyGeometryProgram(CachedProgram& target_program, const ProgramVariant& variant);
 
-    CachedProgram TryLoadProgram(GLenum primitive_mode, BaseBindings base_bindings) const;
+    CachedProgram TryLoadProgram(const ProgramVariant& variant) const;
 
-    ShaderDiskCacheUsage GetUsage(GLenum primitive_mode, BaseBindings base_bindings) const;
+    ShaderDiskCacheUsage GetUsage(const ProgramVariant& variant) const;
 
     u8* host_ptr{};
     VAddr cpu_addr{};
@@ -100,8 +99,8 @@ private:
 
     std::string code;
 
-    std::unordered_map<BaseBindings, CachedProgram> programs;
-    std::unordered_map<BaseBindings, GeometryPrograms> geometry_programs;
+    std::unordered_map<ProgramVariant, CachedProgram> programs;
+    std::unordered_map<ProgramVariant, GeometryPrograms> geometry_programs;
 
     std::unordered_map<u32, GLuint> cbuf_resource_cache;
     std::unordered_map<u32, GLuint> gmem_resource_cache;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -152,6 +152,7 @@ public:
         DeclareConstantBuffers();
         DeclareGlobalMemory();
         DeclareSamplers();
+        DeclareImages();
 
         code.AddLine("void execute_" + suffix + "() {");
         ++code.scope;
@@ -204,6 +205,9 @@ public:
         for (const auto& sampler : ir.GetSamplers()) {
             entries.samplers.emplace_back(sampler);
         }
+        for (const auto& image : ir.GetImages()) {
+            entries.images.emplace_back(image);
+        }
         for (const auto& gmem_pair : ir.GetGlobalMemory()) {
             const auto& [base, usage] = gmem_pair;
             entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset,
@@ -402,9 +406,13 @@ private:
     void DeclareSamplers() {
         const auto& samplers = ir.GetSamplers();
         for (const auto& sampler : samplers) {
+            const std::string name{GetSampler(sampler)};
+            const std::string description{"layout (binding = SAMPLER_BINDING_" +
+                                          std::to_string(sampler.GetIndex()) + ") uniform "};
             std::string sampler_type = [&]() {
                 switch (sampler.GetType()) {
                 case Tegra::Shader::TextureType::Texture1D:
+                    // Special cased, read below.
                     return "sampler1D";
                 case Tegra::Shader::TextureType::Texture2D:
                     return "sampler2D";
@@ -422,13 +430,54 @@ private:
             if (sampler.IsShadow())
                 sampler_type += "Shadow";
 
-            code.AddLine("layout (binding = SAMPLER_BINDING_" + std::to_string(sampler.GetIndex()) +
-                         ") uniform " + sampler_type + ' ' + GetSampler(sampler) + ';');
+            if (sampler.GetType() == Tegra::Shader::TextureType::Texture1D) {
+                // 1D textures can be aliased to texture buffers, hide the declarations behind a
+                // preprocessor flag and use one or the other from the GPU state. This has to be
+                // done because shaders don't have enough information to determine the texture type.
+                EmitIfdefIsBuffer(sampler);
+                code.AddLine(description + "samplerBuffer " + name + ';');
+                code.AddLine("#else");
+                code.AddLine(description + sampler_type + ' ' + name + ';');
+                code.AddLine("#endif");
+            } else {
+                // The other texture types (2D, 3D and cubes) don't have this issue.
+                code.AddLine(description + sampler_type + ' ' + name + ';');
+            }
         }
         if (!samplers.empty())
             code.AddNewLine();
     }
 
+    void DeclareImages() {
+        const auto& images{ir.GetImages()};
+        for (const auto& image : images) {
+            const std::string image_type = [&]() {
+                switch (image.GetType()) {
+                case Tegra::Shader::ImageType::Texture1D:
+                    return "image1D";
+                case Tegra::Shader::ImageType::TextureBuffer:
+                    return "bufferImage";
+                case Tegra::Shader::ImageType::Texture1DArray:
+                    return "image1DArray";
+                case Tegra::Shader::ImageType::Texture2D:
+                    return "image2D";
+                case Tegra::Shader::ImageType::Texture2DArray:
+                    return "image2DArray";
+                case Tegra::Shader::ImageType::Texture3D:
+                    return "image3D";
+                default:
+                    UNREACHABLE();
+                    return "image1D";
+                }
+            }();
+            code.AddLine("layout (binding = IMAGE_BINDING_" + std::to_string(image.GetIndex()) +
+                         ") coherent volatile writeonly uniform " + image_type + ' ' +
+                         GetImage(image) + ';');
+        }
+        if (!images.empty())
+            code.AddNewLine();
+    }
+
     void VisitBlock(const NodeBlock& bb) {
         for (const Node node : bb) {
             if (const std::string expr = Visit(node); !expr.empty()) {
@@ -1313,13 +1362,61 @@ private:
             else if (next < count)
                 expr += ", ";
         }
+
+        // Store a copy of the expression without the lod to be used with texture buffers
+        std::string expr_buffer = expr;
+
         if (meta->lod) {
             expr += ", ";
             expr += CastOperand(Visit(meta->lod), Type::Int);
         }
         expr += ')';
+        expr += GetSwizzle(meta->element);
 
-        return expr + GetSwizzle(meta->element);
+        expr_buffer += ')';
+        expr_buffer += GetSwizzle(meta->element);
+
+        const std::string tmp{code.GenerateTemporary()};
+        EmitIfdefIsBuffer(meta->sampler);
+        code.AddLine("float " + tmp + " = " + expr_buffer + ';');
+        code.AddLine("#else");
+        code.AddLine("float " + tmp + " = " + expr + ';');
+        code.AddLine("#endif");
+
+        return tmp;
+    }
+
+    std::string ImageStore(Operation operation) {
+        constexpr std::array<const char*, 4> constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
+        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+
+        std::string expr = "imageStore(";
+        expr += GetImage(meta.image);
+        expr += ", ";
+
+        const std::size_t coords_count{operation.GetOperandsCount()};
+        expr += constructors.at(coords_count - 1);
+        for (std::size_t i = 0; i < coords_count; ++i) {
+            expr += VisitOperand(operation, i, Type::Int);
+            if (i + 1 < coords_count) {
+                expr += ", ";
+            }
+        }
+        expr += "), ";
+
+        const std::size_t values_count{meta.values.size()};
+        UNIMPLEMENTED_IF(values_count != 4);
+        expr += "vec4(";
+        for (std::size_t i = 0; i < values_count; ++i) {
+            expr += Visit(meta.values.at(i));
+            if (i + 1 < values_count) {
+                expr += ", ";
+            }
+        }
+        expr += "));";
+
+        code.AddLine(expr);
+        return {};
     }
 
     std::string Branch(Operation operation) {
@@ -1570,6 +1667,8 @@ private:
         &GLSLDecompiler::TextureQueryLod,
         &GLSLDecompiler::TexelFetch,
 
+        &GLSLDecompiler::ImageStore,
+
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::PushFlowStack,
         &GLSLDecompiler::PopFlowStack,
@@ -1636,6 +1735,14 @@ private:
         return GetDeclarationWithSuffix(static_cast<u32>(sampler.GetIndex()), "sampler");
     }
 
+    std::string GetImage(const Image& image) const {
+        return GetDeclarationWithSuffix(static_cast<u32>(image.GetIndex()), "image");
+    }
+
+    void EmitIfdefIsBuffer(const Sampler& sampler) {
+        code.AddLine(fmt::format("#ifdef SAMPLER_{}_IS_BUFFER", sampler.GetIndex()));
+    }
+
     std::string GetDeclarationWithSuffix(u32 index, const std::string& name) const {
         return name + '_' + std::to_string(index) + '_' + suffix;
     }

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -27,6 +27,7 @@ struct ShaderEntries;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 using ProgramResult = std::pair<std::string, ShaderEntries>;
 using SamplerEntry = VideoCommon::Shader::Sampler;
+using ImageEntry = VideoCommon::Shader::Image;
 
 class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
 public:
@@ -74,6 +75,7 @@ struct ShaderEntries {
     std::vector<ConstBufferEntry> const_buffers;
     std::vector<SamplerEntry> samplers;
     std::vector<SamplerEntry> bindless_samplers;
+    std::vector<ImageEntry> images;
     std::vector<GlobalMemoryEntry> global_memory_entries;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
     std::size_t shader_length{};

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -34,11 +34,11 @@ enum class PrecompiledEntryKind : u32 {
     Dump,
 };
 
-constexpr u32 NativeVersion = 1;
+constexpr u32 NativeVersion = 3;
 
 // Making sure sizes doesn't change by accident
-static_assert(sizeof(BaseBindings) == 12);
-static_assert(sizeof(ShaderDiskCacheUsage) == 24);
+static_assert(sizeof(BaseBindings) == 16);
+static_assert(sizeof(ShaderDiskCacheUsage) == 40);
 
 namespace {
 
@@ -286,7 +286,6 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
     if (!LoadObjectFromPrecompiled(code_size)) {
         return {};
     }
-
     std::vector<u8> code(code_size);
     if (!LoadArrayFromPrecompiled(code.data(), code.size())) {
         return {};
@@ -299,7 +298,6 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
     if (!LoadObjectFromPrecompiled(const_buffers_count)) {
         return {};
     }
-
     for (u32 i = 0; i < const_buffers_count; ++i) {
         u32 max_offset{};
         u32 index{};
@@ -315,7 +313,6 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
     if (!LoadObjectFromPrecompiled(samplers_count)) {
         return {};
     }
-
     for (u32 i = 0; i < samplers_count; ++i) {
         u64 offset{};
         u64 index{};
@@ -334,11 +331,28 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
                                             is_array != 0, is_shadow != 0, is_bindless != 0);
     }
 
+    u32 images_count{};
+    if (!LoadObjectFromPrecompiled(images_count)) {
+        return {};
+    }
+    for (u32 i = 0; i < images_count; ++i) {
+        u64 offset{};
+        u64 index{};
+        u32 type{};
+        u8 is_bindless{};
+        if (!LoadObjectFromPrecompiled(offset) || !LoadObjectFromPrecompiled(index) ||
+            !LoadObjectFromPrecompiled(type) || !LoadObjectFromPrecompiled(is_bindless)) {
+            return {};
+        }
+        entry.entries.images.emplace_back(
+            static_cast<std::size_t>(offset), static_cast<std::size_t>(index),
+            static_cast<Tegra::Shader::ImageType>(type), is_bindless != 0);
+    }
+
     u32 global_memory_count{};
     if (!LoadObjectFromPrecompiled(global_memory_count)) {
         return {};
     }
-
     for (u32 i = 0; i < global_memory_count; ++i) {
         u32 cbuf_index{};
         u32 cbuf_offset{};
@@ -363,7 +377,6 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
     if (!LoadObjectFromPrecompiled(shader_length)) {
         return {};
     }
-
     entry.entries.shader_length = static_cast<std::size_t>(shader_length);
 
     return entry;
@@ -403,6 +416,18 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
         }
     }
 
+    if (!SaveObjectToPrecompiled(static_cast<u32>(entries.images.size()))) {
+        return false;
+    }
+    for (const auto& image : entries.images) {
+        if (!SaveObjectToPrecompiled(static_cast<u64>(image.GetOffset())) ||
+            !SaveObjectToPrecompiled(static_cast<u64>(image.GetIndex())) ||
+            !SaveObjectToPrecompiled(static_cast<u32>(image.GetType())) ||
+            !SaveObjectToPrecompiled(static_cast<u8>(image.IsBindless() ? 1 : 0))) {
+            return false;
+        }
+    }
+
     if (!SaveObjectToPrecompiled(static_cast<u32>(entries.global_memory_entries.size()))) {
         return false;
     }

src/video_core/renderer_opengl/gl_shader_disk_cache.h

@@ -33,14 +33,18 @@ namespace OpenGL {
 using ProgramCode = std::vector<u64>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
-/// Allocated bindings used by an OpenGL shader program
+using TextureBufferUsage = std::bitset<64>;
+
+/// Allocated bindings used by an OpenGL shader program.
 struct BaseBindings {
     u32 cbuf{};
     u32 gmem{};
     u32 sampler{};
+    u32 image{};
 
     bool operator==(const BaseBindings& rhs) const {
-        return std::tie(cbuf, gmem, sampler) == std::tie(rhs.cbuf, rhs.gmem, rhs.sampler);
+        return std::tie(cbuf, gmem, sampler, image) ==
+               std::tie(rhs.cbuf, rhs.gmem, rhs.sampler, rhs.image);
     }
 
     bool operator!=(const BaseBindings& rhs) const {
@@ -48,15 +52,29 @@ struct BaseBindings {
     }
 };
 
-/// Describes how a shader is used
+/// Describes the different variants a single program can be compiled.
+struct ProgramVariant {
+    BaseBindings base_bindings;
+    GLenum primitive_mode{};
+    TextureBufferUsage texture_buffer_usage{};
+
+    bool operator==(const ProgramVariant& rhs) const {
+        return std::tie(base_bindings, primitive_mode, texture_buffer_usage) ==
+               std::tie(rhs.base_bindings, rhs.primitive_mode, rhs.texture_buffer_usage);
+    }
+
+    bool operator!=(const ProgramVariant& rhs) const {
+        return !operator==(rhs);
+    }
+};
+
+/// Describes how a shader is used.
 struct ShaderDiskCacheUsage {
     u64 unique_identifier{};
-    BaseBindings bindings;
-    GLenum primitive{};
+    ProgramVariant variant;
 
     bool operator==(const ShaderDiskCacheUsage& rhs) const {
-        return std::tie(unique_identifier, bindings, primitive) ==
-               std::tie(rhs.unique_identifier, rhs.bindings, rhs.primitive);
+        return std::tie(unique_identifier, variant) == std::tie(rhs.unique_identifier, rhs.variant);
     }
 
     bool operator!=(const ShaderDiskCacheUsage& rhs) const {
@@ -71,7 +89,19 @@ namespace std {
 template <>
 struct hash<OpenGL::BaseBindings> {
     std::size_t operator()(const OpenGL::BaseBindings& bindings) const {
-        return bindings.cbuf | bindings.gmem << 8 | bindings.sampler << 16;
+        return static_cast<std::size_t>(bindings.cbuf) ^
+               (static_cast<std::size_t>(bindings.gmem) << 8) ^
+               (static_cast<std::size_t>(bindings.sampler) << 16) ^
+               (static_cast<std::size_t>(bindings.image) << 24);
+    }
+};
+
+template <>
+struct hash<OpenGL::ProgramVariant> {
+    std::size_t operator()(const OpenGL::ProgramVariant& variant) const {
+        return std::hash<OpenGL::BaseBindings>()(variant.base_bindings) ^
+               std::hash<OpenGL::TextureBufferUsage>()(variant.texture_buffer_usage) ^
+               (static_cast<std::size_t>(variant.primitive_mode) << 6);
     }
 };
 
@@ -79,7 +109,7 @@ template <>
 struct hash<OpenGL::ShaderDiskCacheUsage> {
     std::size_t operator()(const OpenGL::ShaderDiskCacheUsage& usage) const {
         return static_cast<std::size_t>(usage.unique_identifier) ^
-               std::hash<OpenGL::BaseBindings>()(usage.bindings) ^ usage.primitive << 16;
+               std::hash<OpenGL::ProgramVariant>()(usage.variant);
     }
 };
 
@@ -264,16 +294,17 @@ private:
         return LoadArrayFromPrecompiled(&object, 1);
     }
 
-    // Copre system
     Core::System& system;
-    // Stored transferable shaders
-    std::map<u64, std::unordered_set<ShaderDiskCacheUsage>> transferable;
+
     // Stores whole precompiled cache which will be read from or saved to the precompiled chache
     // file
     FileSys::VectorVfsFile precompiled_cache_virtual_file;
     // Stores the current offset of the precompiled cache file for IO purposes
     std::size_t precompiled_cache_virtual_file_offset;
 
+    // Stored transferable shaders
+    std::unordered_map<u64, std::unordered_set<ShaderDiskCacheUsage>> transferable;
+
     // The cache has been loaded at boot
     bool tried_to_load{};
 };

src/video_core/renderer_opengl/gl_stream_buffer.cpp

@@ -15,7 +15,8 @@ MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
 
 namespace OpenGL {
 
-OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent)
+OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent,
+                                 bool use_persistent)
     : buffer_size(size) {
     gl_buffer.Create();
 
@@ -29,7 +30,7 @@ OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool p
         allocate_size *= 2;
     }
 
-    if (GLAD_GL_ARB_buffer_storage) {
+    if (use_persistent) {
         persistent = true;
         coherent = prefer_coherent;
         const GLbitfield flags =

src/video_core/renderer_opengl/gl_stream_buffer.h

@@ -13,7 +13,8 @@ namespace OpenGL {
 
 class OGLStreamBuffer : private NonCopyable {
 public:
-    explicit OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent = false);
+    explicit OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent = false,
+                             bool use_persistent = true);
     ~OGLStreamBuffer();
 
     GLuint GetHandle() const;

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -472,7 +472,6 @@ static void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum
     }
 }
 
-/// Initialize the renderer
 bool RendererOpenGL::Init() {
     Core::Frontend::ScopeAcquireWindowContext acquire_context{render_window};
 

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -930,6 +930,11 @@ private:
         return {};
     }
 
+    Id ImageStore(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id Branch(Operation operation) {
         const auto target = std::get_if<ImmediateNode>(operation[0]);
         UNIMPLEMENTED_IF(!target);
@@ -1282,6 +1287,8 @@ private:
         &SPIRVDecompiler::TextureQueryLod,
         &SPIRVDecompiler::TexelFetch,
 
+        &SPIRVDecompiler::ImageStore,
+
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::PushFlowStack,
         &SPIRVDecompiler::PopFlowStack,

src/video_core/shader/decode.cpp

@@ -168,6 +168,7 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
         {OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
         {OpCode::Type::Memory, &ShaderIR::DecodeMemory},
         {OpCode::Type::Texture, &ShaderIR::DecodeTexture},
+        {OpCode::Type::Image, &ShaderIR::DecodeImage},
         {OpCode::Type::FloatSetPredicate, &ShaderIR::DecodeFloatSetPredicate},
         {OpCode::Type::IntegerSetPredicate, &ShaderIR::DecodeIntegerSetPredicate},
         {OpCode::Type::HalfSetPredicate, &ShaderIR::DecodeHalfSetPredicate},

src/video_core/shader/decode/image.cpp

@@ -0,0 +1,115 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/engines/shader_bytecode.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+
+using Tegra::Shader::Instruction;
+using Tegra::Shader::OpCode;
+
+namespace {
+std::size_t GetImageTypeNumCoordinates(Tegra::Shader::ImageType image_type) {
+    switch (image_type) {
+    case Tegra::Shader::ImageType::Texture1D:
+    case Tegra::Shader::ImageType::TextureBuffer:
+        return 1;
+    case Tegra::Shader::ImageType::Texture1DArray:
+    case Tegra::Shader::ImageType::Texture2D:
+        return 2;
+    case Tegra::Shader::ImageType::Texture2DArray:
+    case Tegra::Shader::ImageType::Texture3D:
+        return 3;
+    }
+    UNREACHABLE();
+    return 1;
+}
+} // Anonymous namespace
+
+u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
+    const Instruction instr = {program_code[pc]};
+    const auto opcode = OpCode::Decode(instr);
+
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::SUST: {
+        UNIMPLEMENTED_IF(instr.sust.mode != Tegra::Shader::SurfaceDataMode::P);
+        UNIMPLEMENTED_IF(instr.sust.image_type == Tegra::Shader::ImageType::TextureBuffer);
+        UNIMPLEMENTED_IF(instr.sust.out_of_bounds_store != Tegra::Shader::OutOfBoundsStore::Ignore);
+        UNIMPLEMENTED_IF(instr.sust.component_mask_selector != 0xf); // Ensure we have an RGBA store
+
+        std::vector<Node> values;
+        constexpr std::size_t hardcoded_size{4};
+        for (std::size_t i = 0; i < hardcoded_size; ++i) {
+            values.push_back(GetRegister(instr.gpr0.Value() + i));
+        }
+
+        std::vector<Node> coords;
+        const std::size_t num_coords{GetImageTypeNumCoordinates(instr.sust.image_type)};
+        for (std::size_t i = 0; i < num_coords; ++i) {
+            coords.push_back(GetRegister(instr.gpr8.Value() + i));
+        }
+
+        const auto type{instr.sust.image_type};
+        const auto& image{instr.sust.is_immediate ? GetImage(instr.image, type)
+                                                  : GetBindlessImage(instr.gpr39, type)};
+        MetaImage meta{image, values};
+        const Node store{Operation(OperationCode::ImageStore, meta, std::move(coords))};
+        bb.push_back(store);
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unhandled conversion instruction: {}", opcode->get().GetName());
+    }
+
+    return pc;
+}
+
+const Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
+    const auto offset{static_cast<std::size_t>(image.index.Value())};
+
+    // If this image has already been used, return the existing mapping.
+    const auto itr{std::find_if(used_images.begin(), used_images.end(),
+                                [=](const Image& entry) { return entry.GetOffset() == offset; })};
+    if (itr != used_images.end()) {
+        ASSERT(itr->GetType() == type);
+        return *itr;
+    }
+
+    // Otherwise create a new mapping for this image.
+    const std::size_t next_index{used_images.size()};
+    const Image entry{offset, next_index, type};
+    return *used_images.emplace(entry).first;
+}
+
+const Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg,
+                                        Tegra::Shader::ImageType type) {
+    const Node image_register{GetRegister(reg)};
+    const Node base_image{
+        TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()))};
+    const auto cbuf{std::get_if<CbufNode>(base_image)};
+    const auto cbuf_offset_imm{std::get_if<ImmediateNode>(cbuf->GetOffset())};
+    const auto cbuf_offset{cbuf_offset_imm->GetValue()};
+    const auto cbuf_index{cbuf->GetIndex()};
+    const auto cbuf_key{(static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset)};
+
+    // If this image has already been used, return the existing mapping.
+    const auto itr{std::find_if(used_images.begin(), used_images.end(),
+                                [=](const Image& entry) { return entry.GetOffset() == cbuf_key; })};
+    if (itr != used_images.end()) {
+        ASSERT(itr->GetType() == type);
+        return *itr;
+    }
+
+    // Otherwise create a new mapping for this image.
+    const std::size_t next_index{used_images.size()};
+    const Image entry{cbuf_index, cbuf_offset, next_index, type};
+    return *used_images.emplace(entry).first;
+}
+
+} // namespace VideoCommon::Shader

src/video_core/shader/decode/texture.cpp

@@ -244,6 +244,18 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
         }
         break;
     }
+    case OpCode::Id::TLD: {
+        UNIMPLEMENTED_IF_MSG(instr.tld.aoffi, "AOFFI is not implemented");
+        UNIMPLEMENTED_IF_MSG(instr.tld.ms, "MS is not implemented");
+        UNIMPLEMENTED_IF_MSG(instr.tld.cl, "CL is not implemented");
+
+        if (instr.tld.nodep_flag) {
+            LOG_WARNING(HW_GPU, "TLD.NODEP implementation is incomplete");
+        }
+
+        WriteTexInstructionFloat(bb, instr, GetTldCode(instr));
+        break;
+    }
     case OpCode::Id::TLDS: {
         const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()};
         const bool is_array{instr.tlds.IsArrayTexture()};
@@ -574,6 +586,38 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
     return values;
 }
 
+Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
+    const auto texture_type{instr.tld.texture_type};
+    const bool is_array{instr.tld.is_array};
+    const bool lod_enabled{instr.tld.GetTextureProcessMode() == TextureProcessMode::LL};
+    const std::size_t coord_count{GetCoordCount(texture_type)};
+
+    u64 gpr8_cursor{instr.gpr8.Value()};
+    const Node array_register{is_array ? GetRegister(gpr8_cursor++) : nullptr};
+
+    std::vector<Node> coords;
+    for (std::size_t i = 0; i < coord_count; ++i) {
+        coords.push_back(GetRegister(gpr8_cursor++));
+    }
+
+    u64 gpr20_cursor{instr.gpr20.Value()};
+    // const Node bindless_register{is_bindless ? GetRegister(gpr20_cursor++) : nullptr};
+    const Node lod{lod_enabled ? GetRegister(gpr20_cursor++) : Immediate(0u)};
+    // const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
+    // const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
+
+    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
+
+    Node4 values;
+    for (u32 element = 0; element < values.size(); ++element) {
+        auto coords_copy = coords;
+        MetaTexture meta{sampler, array_register, {}, {}, {}, lod, {}, element};
+        values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
+    }
+
+    return values;
+}
+
 Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
     const std::size_t type_coord_count = GetCoordCount(texture_type);
     const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;

src/video_core/shader/shader_ir.h

@@ -172,6 +172,8 @@ enum class OperationCode {
     TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
     TexelFetch,             /// (MetaTexture, int[N], int) -> float4
 
+    ImageStore, /// (MetaImage, float[N] coords) -> void
+
     Branch,        /// (uint branch_target) -> void
     PushFlowStack, /// (uint branch_target) -> void
     PopFlowStack,  /// () -> void
@@ -267,6 +269,48 @@ private:
     bool is_bindless{}; ///< Whether this sampler belongs to a bindless texture or not.
 };
 
+class Image {
+public:
+    explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type)
+        : offset{offset}, index{index}, type{type}, is_bindless{false} {}
+
+    explicit Image(u32 cbuf_index, u32 cbuf_offset, std::size_t index,
+                   Tegra::Shader::ImageType type)
+        : offset{(static_cast<u64>(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type},
+          is_bindless{true} {}
+
+    explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type,
+                   bool is_bindless)
+        : offset{offset}, index{index}, type{type}, is_bindless{is_bindless} {}
+
+    std::size_t GetOffset() const {
+        return offset;
+    }
+
+    std::size_t GetIndex() const {
+        return index;
+    }
+
+    Tegra::Shader::ImageType GetType() const {
+        return type;
+    }
+
+    bool IsBindless() const {
+        return is_bindless;
+    }
+
+    bool operator<(const Image& rhs) const {
+        return std::tie(offset, index, type, is_bindless) <
+               std::tie(rhs.offset, rhs.index, rhs.type, rhs.is_bindless);
+    }
+
+private:
+    std::size_t offset{};
+    std::size_t index{};
+    Tegra::Shader::ImageType type{};
+    bool is_bindless{};
+};
+
 class ConstBuffer {
 public:
     explicit ConstBuffer(u32 max_offset, bool is_indirect)
@@ -328,10 +372,15 @@ struct MetaTexture {
     u32 element{};
 };
 
+struct MetaImage {
+    const Image& image;
+    std::vector<Node> values;
+};
+
 inline constexpr MetaArithmetic PRECISE = {true};
 inline constexpr MetaArithmetic NO_PRECISE = {false};
 
-using Meta = std::variant<MetaArithmetic, MetaTexture, Tegra::Shader::HalfType>;
+using Meta = std::variant<MetaArithmetic, MetaTexture, MetaImage, Tegra::Shader::HalfType>;
 
 /// Holds any kind of operation that can be done in the IR
 class OperationNode final {
@@ -602,6 +651,10 @@ public:
         return used_samplers;
     }
 
+    const std::set<Image>& GetImages() const {
+        return used_images;
+    }
+
     const std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances>& GetClipDistances()
         const {
         return used_clip_distances;
@@ -648,6 +701,7 @@ private:
     u32 DecodeConversion(NodeBlock& bb, u32 pc);
     u32 DecodeMemory(NodeBlock& bb, u32 pc);
     u32 DecodeTexture(NodeBlock& bb, u32 pc);
+    u32 DecodeImage(NodeBlock& bb, u32 pc);
     u32 DecodeFloatSetPredicate(NodeBlock& bb, u32 pc);
     u32 DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc);
     u32 DecodeHalfSetPredicate(NodeBlock& bb, u32 pc);
@@ -767,6 +821,12 @@ private:
                                       Tegra::Shader::TextureType type, bool is_array,
                                       bool is_shadow);
 
+    /// Accesses an image.
+    const Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
+
+    /// Access a bindless image sampler.
+    const Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type);
+
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);
 
@@ -790,6 +850,8 @@ private:
     Node4 GetTld4Code(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
                       bool depth_compare, bool is_array, bool is_aoffi);
 
+    Node4 GetTldCode(Tegra::Shader::Instruction instr);
+
     Node4 GetTldsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
                       bool is_array);
 
@@ -877,6 +939,7 @@ private:
     std::set<Tegra::Shader::Attribute::Index> used_output_attributes;
     std::map<u32, ConstBuffer> used_cbufs;
     std::set<Sampler> used_samplers;
+    std::set<Image> used_images;
     std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
     std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory;
 

src/video_core/surface.cpp

@@ -12,6 +12,8 @@ SurfaceTarget SurfaceTargetFromTextureType(Tegra::Texture::TextureType texture_t
     switch (texture_type) {
     case Tegra::Texture::TextureType::Texture1D:
         return SurfaceTarget::Texture1D;
+    case Tegra::Texture::TextureType::Texture1DBuffer:
+        return SurfaceTarget::TextureBuffer;
     case Tegra::Texture::TextureType::Texture2D:
     case Tegra::Texture::TextureType::Texture2DNoMipmap:
         return SurfaceTarget::Texture2D;
@@ -35,6 +37,7 @@ SurfaceTarget SurfaceTargetFromTextureType(Tegra::Texture::TextureType texture_t
 bool SurfaceTargetIsLayered(SurfaceTarget target) {
     switch (target) {
     case SurfaceTarget::Texture1D:
+    case SurfaceTarget::TextureBuffer:
     case SurfaceTarget::Texture2D:
     case SurfaceTarget::Texture3D:
         return false;
@@ -53,6 +56,7 @@ bool SurfaceTargetIsLayered(SurfaceTarget target) {
 bool SurfaceTargetIsArray(SurfaceTarget target) {
     switch (target) {
     case SurfaceTarget::Texture1D:
+    case SurfaceTarget::TextureBuffer:
     case SurfaceTarget::Texture2D:
     case SurfaceTarget::Texture3D:
     case SurfaceTarget::TextureCubemap:

src/video_core/surface.h

@@ -114,6 +114,7 @@ enum class SurfaceType {
 
 enum class SurfaceTarget {
     Texture1D,
+    TextureBuffer,
     Texture2D,
     Texture3D,
     Texture1DArray,

src/video_core/textures/texture.h

@@ -172,12 +172,16 @@ struct TICEntry {
         BitField<26, 1, u32> use_header_opt_control;
         BitField<27, 1, u32> depth_texture;
         BitField<28, 4, u32> max_mip_level;
+
+        BitField<0, 16, u32> buffer_high_width_minus_one;
     };
     union {
         BitField<0, 16, u32> width_minus_1;
         BitField<22, 1, u32> srgb_conversion;
         BitField<23, 4, TextureType> texture_type;
         BitField<29, 3, u32> border_size;
+
+        BitField<0, 16, u32> buffer_low_width_minus_one;
     };
     union {
         BitField<0, 16, u32> height_minus_1;
@@ -206,7 +210,10 @@ struct TICEntry {
     }
 
     u32 Width() const {
-        return width_minus_1 + 1;
+        if (header_version != TICHeaderVersion::OneDBuffer) {
+            return width_minus_1 + 1;
+        }
+        return (buffer_high_width_minus_one << 16) | buffer_low_width_minus_one;
     }
 
     u32 Height() const {
@@ -240,6 +247,15 @@ struct TICEntry {
                header_version == TICHeaderVersion::BlockLinearColorKey;
     }
 
+    bool IsLineal() const {
+        return header_version == TICHeaderVersion::Pitch ||
+               header_version == TICHeaderVersion::PitchColorKey;
+    }
+
+    bool IsBuffer() const {
+        return header_version == TICHeaderVersion::OneDBuffer;
+    }
+
     bool IsSrgbConversionEnabled() const {
         return srgb_conversion != 0;
     }

src/yuzu/main.cpp

@@ -742,6 +742,8 @@ void GMainWindow::OnDisplayTitleBars(bool show) {
 QStringList GMainWindow::GetUnsupportedGLExtensions() {
     QStringList unsupported_ext;
 
+    if (!GLAD_GL_ARB_buffer_storage)
+        unsupported_ext.append("ARB_buffer_storage");
     if (!GLAD_GL_ARB_direct_state_access)
         unsupported_ext.append("ARB_direct_state_access");
     if (!GLAD_GL_ARB_vertex_type_10f_11f_11f_rev)

src/yuzu_cmd/emu_window/emu_window_sdl2.cpp

@@ -142,6 +142,8 @@ void EmuWindow_SDL2::Fullscreen() {
 bool EmuWindow_SDL2::SupportsRequiredGLExtensions() {
     std::vector<std::string> unsupported_ext;
 
+    if (!GLAD_GL_ARB_buffer_storage)
+        unsupported_ext.push_back("ARB_buffer_storage");
     if (!GLAD_GL_ARB_direct_state_access)
         unsupported_ext.push_back("ARB_direct_state_access");
     if (!GLAD_GL_ARB_vertex_type_10f_11f_11f_rev)