gl_shader_decompiler: Use uint for images and fix SUATOM

In the process remove implementation of SUATOM.MIN and SUATOM.MAX as
these require a distinction between U32 and S32. These have to be
implemented with imageCompSwap loop.

src/video_core/engines/shader_bytecode.h

@@ -544,7 +544,7 @@ enum class VoteOperation : u64 {
     Eq = 2,  // allThreadsEqualNV
 };
 
-enum class ImageAtomicSize : u64 {
+enum class ImageAtomicOperationType : u64 {
     U32 = 0,
     S32 = 1,
     U64 = 2,
@@ -1431,7 +1431,7 @@ union Instruction {
 
     union {
         BitField<28, 1, u64> is_ba;
-        BitField<51, 3, ImageAtomicSize> size;
+        BitField<51, 3, ImageAtomicOperationType> operation_type;
         BitField<33, 3, ImageType> image_type;
         BitField<29, 4, ImageAtomicOperation> operation;
         BitField<49, 2, OutOfBoundsStore> out_of_bounds_store;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -716,36 +716,20 @@ private:
             const char* image_type = [&] {
                 switch (image.GetType()) {
                 case Tegra::Shader::ImageType::Texture1D:
-                    return "image1D";
+                    return "1D";
                 case Tegra::Shader::ImageType::TextureBuffer:
-                    return "imageBuffer";
+                    return "Buffer";
                 case Tegra::Shader::ImageType::Texture1DArray:
-                    return "image1DArray";
+                    return "1DArray";
                 case Tegra::Shader::ImageType::Texture2D:
-                    return "image2D";
+                    return "2D";
                 case Tegra::Shader::ImageType::Texture2DArray:
-                    return "image2DArray";
+                    return "2DArray";
                 case Tegra::Shader::ImageType::Texture3D:
-                    return "image3D";
+                    return "3D";
                 default:
                     UNREACHABLE();
-                    return "image1D";
+                    return "1D";
-                }
-            }();
-
-            const auto [type_prefix, format] = [&]() -> std::pair<const char*, const char*> {
-                if (!image.IsSizeKnown()) {
-                    return {"", ""};
-                }
-                switch (image.GetSize()) {
-                case Tegra::Shader::ImageAtomicSize::U32:
-                    return {"u", "r32ui, "};
-                case Tegra::Shader::ImageAtomicSize::S32:
-                    return {"i", "r32i, "};
-                default:
-                    UNIMPLEMENTED_MSG("Unimplemented atomic size={}",
-                                      static_cast<u32>(image.GetSize()));
-                    return {"", ""};
                 }
             }();
 
@@ -756,8 +740,12 @@ private:
                 qualifier += " writeonly";
             }
 
-            code.AddLine("layout (binding = IMAGE_BINDING_{}) {} uniform "
+            std::string format;
-                         "{} {};",
+            if (image.IsAtomic()) {
+                format = "r32ui, ";
+            }
+
+            code.AddLine("layout ({}binding = IMAGE_BINDING_{}) {} uniform uimage{} {};", format,
                          image.GetIndex(), qualifier, image_type, GetImage(image));
         }
         if (!images.empty()) {
@@ -1225,28 +1213,13 @@ private:
     }
 
     std::string BuildImageValues(Operation operation) {
+        constexpr std::array constructors{"uint", "uvec2", "uvec3", "uvec4"};
         const auto meta{std::get<MetaImage>(operation.GetMeta())};
-        const auto [constructors, type] = [&]() -> std::pair<std::array<const char*, 4>, Type> {
-            constexpr std::array float_constructors{"float", "vec2", "vec3", "vec4"};
-            if (!meta.image.IsSizeKnown()) {
-                return {float_constructors, Type::Float};
-            }
-            switch (meta.image.GetSize()) {
-            case Tegra::Shader::ImageAtomicSize::U32:
-                return {{"uint", "uvec2", "uvec3", "uvec4"}, Type::Uint};
-            case Tegra::Shader::ImageAtomicSize::S32:
-                return {{"int", "ivec2", "ivec3", "ivec4"}, Type::Uint};
-            default:
-                UNIMPLEMENTED_MSG("Unimplemented image size={}",
-                                  static_cast<u32>(meta.image.GetSize()));
-                return {float_constructors, Type::Float};
-            }
-        }();
 
         const std::size_t values_count{meta.values.size()};
         std::string expr = fmt::format("{}(", constructors.at(values_count - 1));
         for (std::size_t i = 0; i < values_count; ++i) {
-            expr += Visit(meta.values.at(i)).As(type);
+            expr += Visit(meta.values.at(i)).AsUint();
             if (i + 1 < values_count) {
                 expr += ", ";
             }
@@ -1255,29 +1228,6 @@ private:
         return expr;
     }
 
-    Expression AtomicImage(Operation operation, const char* opname) {
-        constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
-        const auto meta{std::get<MetaImage>(operation.GetMeta())};
-        ASSERT(meta.values.size() == 1);
-        ASSERT(meta.image.IsSizeKnown());
-
-        const auto type = [&]() {
-            switch (const auto size = meta.image.GetSize()) {
-            case Tegra::Shader::ImageAtomicSize::U32:
-                return Type::Uint;
-            case Tegra::Shader::ImageAtomicSize::S32:
-                return Type::Int;
-            default:
-                UNIMPLEMENTED_MSG("Unimplemented image size={}", static_cast<u32>(size));
-                return Type::Uint;
-            }
-        }();
-
-        return {fmt::format("{}({}, {}, {})", opname, GetImage(meta.image),
-                            BuildIntegerCoordinates(operation), Visit(meta.values[0]).As(type)),
-                type};
-    }
-
     Expression Assign(Operation operation) {
         const Node& dest = operation[0];
         const Node& src = operation[1];
@@ -1810,7 +1760,7 @@ private:
         const auto meta{std::get<MetaImage>(operation.GetMeta())};
         return {fmt::format("imageLoad({}, {}){}", GetImage(meta.image),
                             BuildIntegerCoordinates(operation), GetSwizzle(meta.element)),
-                Type::Float};
+                Type::Uint};
     }
 
     Expression ImageStore(Operation operation) {
@@ -1820,31 +1770,14 @@ private:
         return {};
     }
 
-    Expression AtomicImageAdd(Operation operation) {
+    template <const std::string_view& opname>
-        return AtomicImage(operation, "imageAtomicAdd");
+    Expression AtomicImage(Operation operation) {
-    }
+        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        ASSERT(meta.values.size() == 1);
 
-    Expression AtomicImageMin(Operation operation) {
+        return {fmt::format("imageAtomic{}({}, {}, {})", opname, GetImage(meta.image),
-        return AtomicImage(operation, "imageAtomicMin");
+                            BuildIntegerCoordinates(operation), Visit(meta.values[0]).AsUint()),
-    }
+                Type::Uint};
-
-    Expression AtomicImageMax(Operation operation) {
-        return AtomicImage(operation, "imageAtomicMax");
-    }
-    Expression AtomicImageAnd(Operation operation) {
-        return AtomicImage(operation, "imageAtomicAnd");
-    }
-
-    Expression AtomicImageOr(Operation operation) {
-        return AtomicImage(operation, "imageAtomicOr");
-    }
-
-    Expression AtomicImageXor(Operation operation) {
-        return AtomicImage(operation, "imageAtomicXor");
-    }
-
-    Expression AtomicImageExchange(Operation operation) {
-        return AtomicImage(operation, "imageAtomicExchange");
     }
 
     Expression Branch(Operation operation) {
@@ -2039,6 +1972,12 @@ private:
         Func() = delete;
         ~Func() = delete;
 
+        static constexpr std::string_view Add = "Add";
+        static constexpr std::string_view And = "And";
+        static constexpr std::string_view Or = "Or";
+        static constexpr std::string_view Xor = "Xor";
+        static constexpr std::string_view Exchange = "Exchange";
+
         static constexpr std::string_view ShuffleIndexed = "shuffleNV";
         static constexpr std::string_view ShuffleUp = "shuffleUpNV";
         static constexpr std::string_view ShuffleDown = "shuffleDownNV";
@@ -2178,13 +2117,12 @@ private:
 
         &GLSLDecompiler::ImageLoad,
         &GLSLDecompiler::ImageStore,
-        &GLSLDecompiler::AtomicImageAdd,
+
-        &GLSLDecompiler::AtomicImageMin,
+        &GLSLDecompiler::AtomicImage<Func::Add>,
-        &GLSLDecompiler::AtomicImageMax,
+        &GLSLDecompiler::AtomicImage<Func::And>,
-        &GLSLDecompiler::AtomicImageAnd,
+        &GLSLDecompiler::AtomicImage<Func::Or>,
-        &GLSLDecompiler::AtomicImageOr,
+        &GLSLDecompiler::AtomicImage<Func::Xor>,
-        &GLSLDecompiler::AtomicImageXor,
+        &GLSLDecompiler::AtomicImage<Func::Exchange>,
-        &GLSLDecompiler::AtomicImageExchange,
 
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::BranchIndirect,

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -343,20 +343,17 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
         u8 is_bindless{};
         u8 is_written{};
         u8 is_read{};
-        u8 is_size_known{};
+        u8 is_atomic{};
-        u32 size{};
         if (!LoadObjectFromPrecompiled(offset) || !LoadObjectFromPrecompiled(index) ||
             !LoadObjectFromPrecompiled(type) || !LoadObjectFromPrecompiled(is_bindless) ||
             !LoadObjectFromPrecompiled(is_written) || !LoadObjectFromPrecompiled(is_read) ||
-            !LoadObjectFromPrecompiled(is_size_known) || !LoadObjectFromPrecompiled(size)) {
+            !LoadObjectFromPrecompiled(is_atomic)) {
             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, is_written != 0,
-            is_read != 0,
+            is_read != 0, is_atomic != 0);
-            is_size_known ? std::make_optional(static_cast<Tegra::Shader::ImageAtomicSize>(size))
-                          : std::nullopt);
     }
 
     u32 global_memory_count{};
@@ -429,14 +426,13 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
         return false;
     }
     for (const auto& image : entries.images) {
-        const u32 size = image.IsSizeKnown() ? static_cast<u32>(image.GetSize()) : 0U;
         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)) ||
             !SaveObjectToPrecompiled(static_cast<u8>(image.IsWritten() ? 1 : 0)) ||
             !SaveObjectToPrecompiled(static_cast<u8>(image.IsRead() ? 1 : 0)) ||
-            !SaveObjectToPrecompiled(image.IsSizeKnown()) || !SaveObjectToPrecompiled(size)) {
+            !SaveObjectToPrecompiled(static_cast<u8>(image.IsAtomic() ? 1 : 0))) {
             return false;
         }
     }

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -955,16 +955,6 @@ private:
         return {};
     }
 
-    Id AtomicImageMin(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id AtomicImageMax(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
     Id AtomicImageAnd(Operation operation) {
         UNIMPLEMENTED();
         return {};
@@ -1449,8 +1439,6 @@ private:
         &SPIRVDecompiler::ImageLoad,
         &SPIRVDecompiler::ImageStore,
         &SPIRVDecompiler::AtomicImageAdd,
-        &SPIRVDecompiler::AtomicImageMin,
-        &SPIRVDecompiler::AtomicImageMax,
         &SPIRVDecompiler::AtomicImageAnd,
         &SPIRVDecompiler::AtomicImageOr,
         &SPIRVDecompiler::AtomicImageXor,

src/video_core/shader/decode/image.cpp

@@ -101,13 +101,12 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
         UNIMPLEMENTED_IF(instr.suatom_d.is_ba != 0);
 
         const OperationCode operation_code = [instr] {
+            switch (instr.suatom_d.operation_type) {
+            case Tegra::Shader::ImageAtomicOperationType::S32:
+            case Tegra::Shader::ImageAtomicOperationType::U32:
                 switch (instr.suatom_d.operation) {
                 case Tegra::Shader::ImageAtomicOperation::Add:
                     return OperationCode::AtomicImageAdd;
-            case Tegra::Shader::ImageAtomicOperation::Min:
-                return OperationCode::AtomicImageMin;
-            case Tegra::Shader::ImageAtomicOperation::Max:
-                return OperationCode::AtomicImageMax;
                 case Tegra::Shader::ImageAtomicOperation::And:
                     return OperationCode::AtomicImageAnd;
                 case Tegra::Shader::ImageAtomicOperation::Or:
@@ -116,17 +115,21 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
                     return OperationCode::AtomicImageXor;
                 case Tegra::Shader::ImageAtomicOperation::Exch:
                     return OperationCode::AtomicImageExchange;
-            default:
-                UNIMPLEMENTED_MSG("Unimplemented operation={}",
-                                  static_cast<u32>(instr.suatom_d.operation.Value()));
-                return OperationCode::AtomicImageAdd;
                 }
+            default:
+                break;
+            }
+            UNIMPLEMENTED_MSG("Unimplemented operation={} type={}",
+                              static_cast<u64>(instr.suatom_d.operation.Value()),
+                              static_cast<u64>(instr.suatom_d.operation_type.Value()));
+            return OperationCode::AtomicImageAdd;
         }();
 
         Node value = GetRegister(instr.gpr0);
 
         const auto type = instr.suatom_d.image_type;
-        const auto& image{GetImage(instr.image, type, instr.suatom_d.size)};
+        auto& image = GetImage(instr.image, type);
+        image.MarkAtomic();
 
         MetaImage meta{image, {std::move(value)}};
         SetRegister(bb, instr.gpr0, Operation(operation_code, meta, GetCoordinates(type)));
@@ -139,35 +142,32 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
+Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
-                          std::optional<Tegra::Shader::ImageAtomicSize> size) {
     const auto offset{static_cast<std::size_t>(image.index.Value())};
-    if (const auto image = TryUseExistingImage(offset, type, size)) {
+    if (const auto image = TryUseExistingImage(offset, type)) {
         return *image;
     }
 
     const std::size_t next_index{used_images.size()};
-    return used_images.emplace(offset, Image{offset, next_index, type, size}).first->second;
+    return used_images.emplace(offset, Image{offset, next_index, type}).first->second;
 }
 
-Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type,
+Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) {
-                                  std::optional<Tegra::Shader::ImageAtomicSize> size) {
     const Node image_register{GetRegister(reg)};
     const auto [base_image, cbuf_index, cbuf_offset]{
         TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()))};
     const auto cbuf_key{(static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset)};
 
-    if (const auto image = TryUseExistingImage(cbuf_key, type, size)) {
+    if (const auto image = TryUseExistingImage(cbuf_key, type)) {
         return *image;
     }
 
     const std::size_t next_index{used_images.size()};
-    return used_images.emplace(cbuf_key, Image{cbuf_index, cbuf_offset, next_index, type, size})
+    return used_images.emplace(cbuf_key, Image{cbuf_index, cbuf_offset, next_index, type})
         .first->second;
 }
 
-Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
+Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type) {
-                                     std::optional<Tegra::Shader::ImageAtomicSize> size) {
     auto it = used_images.find(offset);
     if (it == used_images.end()) {
         return nullptr;
@@ -175,14 +175,6 @@ Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
     auto& image = it->second;
     ASSERT(image.GetType() == type);
 
-    if (size) {
-        // We know the size, if it's known it has to be the same as before, otherwise we can set it.
-        if (image.IsSizeKnown()) {
-            ASSERT(image.GetSize() == size);
-        } else {
-            image.SetSize(*size);
-        }
-    }
     return &image;
 }
 

src/video_core/shader/node.h

@@ -151,9 +151,8 @@ enum class OperationCode {
 
     ImageLoad,  /// (MetaImage, int[N] coords) -> void
     ImageStore, /// (MetaImage, int[N] coords) -> void
+
     AtomicImageAdd,      /// (MetaImage, int[N] coords) -> void
-    AtomicImageMin,      /// (MetaImage, int[N] coords) -> void
-    AtomicImageMax,      /// (MetaImage, int[N] coords) -> void
     AtomicImageAnd,      /// (MetaImage, int[N] coords) -> void
     AtomicImageOr,       /// (MetaImage, int[N] coords) -> void
     AtomicImageXor,      /// (MetaImage, int[N] coords) -> void
@@ -295,21 +294,18 @@ private:
 
 class Image final {
 public:
-    constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type,
+    constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type)
-                             std::optional<Tegra::Shader::ImageAtomicSize> size)
+        : offset{offset}, index{index}, type{type}, is_bindless{false} {}
-        : offset{offset}, index{index}, type{type}, is_bindless{false}, size{size} {}
 
     constexpr explicit Image(u32 cbuf_index, u32 cbuf_offset, std::size_t index,
-                             Tegra::Shader::ImageType type,
+                             Tegra::Shader::ImageType type)
-                             std::optional<Tegra::Shader::ImageAtomicSize> size)
         : offset{(static_cast<u64>(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type},
-          is_bindless{true}, size{size} {}
+          is_bindless{true} {}
 
     constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type,
-                             bool is_bindless, bool is_written, bool is_read,
+                             bool is_bindless, bool is_written, bool is_read, bool is_atomic)
-                             std::optional<Tegra::Shader::ImageAtomicSize> size)
         : offset{offset}, index{index}, type{type}, is_bindless{is_bindless},
-          is_written{is_written}, is_read{is_read}, size{size} {}
+          is_written{is_written}, is_read{is_read}, is_atomic{is_atomic} {}
 
     void MarkWrite() {
         is_written = true;
@@ -319,8 +315,10 @@ public:
         is_read = true;
     }
 
-    void SetSize(Tegra::Shader::ImageAtomicSize size_) {
+    void MarkAtomic() {
-        size = size_;
+        MarkWrite();
+        MarkRead();
+        is_atomic = true;
     }
 
     constexpr std::size_t GetOffset() const {
@@ -347,21 +345,17 @@ public:
         return is_read;
     }
 
+    constexpr bool IsAtomic() const {
+        return is_atomic;
+    }
+
     constexpr std::pair<u32, u32> GetBindlessCBuf() const {
         return {static_cast<u32>(offset >> 32), static_cast<u32>(offset)};
     }
 
-    constexpr bool IsSizeKnown() const {
-        return size.has_value();
-    }
-
-    constexpr Tegra::Shader::ImageAtomicSize GetSize() const {
-        return size.value();
-    }
-
     constexpr bool operator<(const Image& rhs) const {
-        return std::tie(offset, index, type, size, is_bindless) <
+        return std::tie(offset, index, type, is_bindless) <
-               std::tie(rhs.offset, rhs.index, rhs.type, rhs.size, rhs.is_bindless);
+               std::tie(rhs.offset, rhs.index, rhs.type, rhs.is_bindless);
     }
 
 private:
@@ -371,7 +365,7 @@ private:
     bool is_bindless{};
     bool is_written{};
     bool is_read{};
-    std::optional<Tegra::Shader::ImageAtomicSize> size{};
+    bool is_atomic{};
 };
 
 struct GlobalMemoryBase {

src/video_core/shader/shader_ir.h

@@ -276,16 +276,13 @@ private:
                                       bool is_shadow);
 
     /// Accesses an image.
-    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
+    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
-                    std::optional<Tegra::Shader::ImageAtomicSize> size = {});
 
     /// Access a bindless image sampler.
-    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type,
+    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type);
-                            std::optional<Tegra::Shader::ImageAtomicSize> size = {});
 
     /// Tries to access an existing image, updating it's state as needed
-    Image* TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
+    Image* TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type);
-                               std::optional<Tegra::Shader::ImageAtomicSize> size);
 
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);