3 files changed, 101 insertions, 91 deletions

diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp
index d54fb88c9..95ec1cdd9 100644
--- a/src/video_core/shader/decode/image.cpp
+++ b/src/video_core/shader/decode/image.cpp
@@ -41,11 +41,46 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
+    const auto GetCoordinates = [this, instr](Tegra::Shader::ImageType image_type) {
+        std::vector<Node> coords;
+        const std::size_t num_coords{GetImageTypeNumCoordinates(image_type)};
+        coords.reserve(num_coords);
+        for (std::size_t i = 0; i < num_coords; ++i) {
+            coords.push_back(GetRegister(instr.gpr8.Value() + i));
+        }
+        return coords;
+    };
+
     switch (opcode->get().GetId()) {
+    case OpCode::Id::SULD: {
+        UNIMPLEMENTED_IF(instr.suldst.mode != Tegra::Shader::SurfaceDataMode::P);
+        UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store !=
+                         Tegra::Shader::OutOfBoundsStore::Ignore);
+
+        const auto type{instr.suldst.image_type};
+        auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type)
+                                              : GetBindlessImage(instr.gpr39, type)};
+        image.MarkRead();
+
+        u32 indexer = 0;
+        for (u32 element = 0; element < 4; ++element) {
+            if (!instr.suldst.IsComponentEnabled(element)) {
+                continue;
+            }
+            MetaImage meta{image, {}, element};
+            Node value = Operation(OperationCode::ImageLoad, meta, GetCoordinates(type));
+            SetTemporary(bb, indexer++, std::move(value));
+        }
+        for (u32 i = 0; i < indexer; ++i) {
+            SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
+        }
+        break;
+    }
     case OpCode::Id::SUST: {
-        UNIMPLEMENTED_IF(instr.sust.mode != Tegra::Shader::SurfaceDataMode::P);
-        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
+        UNIMPLEMENTED_IF(instr.suldst.mode != Tegra::Shader::SurfaceDataMode::P);
+        UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store !=
+                         Tegra::Shader::OutOfBoundsStore::Ignore);
+        UNIMPLEMENTED_IF(instr.suldst.component_mask_selector != 0xf); // Ensure we have RGBA
 
         std::vector<Node> values;
         constexpr std::size_t hardcoded_size{4};
@@ -53,58 +88,51 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
             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};
-        auto& image{instr.sust.is_immediate ? GetImage(instr.image, type)
-                                            : GetBindlessImage(instr.gpr39, type)};
+        const auto type{instr.suldst.image_type};
+        auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type)
+                                              : GetBindlessImage(instr.gpr39, type)};
         image.MarkWrite();
 
-        MetaImage meta{image, values};
-        bb.push_back(Operation(OperationCode::ImageStore, meta, std::move(coords)));
+        MetaImage meta{image, std::move(values)};
+        bb.push_back(Operation(OperationCode::ImageStore, meta, GetCoordinates(type)));
         break;
     }
     case OpCode::Id::SUATOM: {
         UNIMPLEMENTED_IF(instr.suatom_d.is_ba != 0);
 
-        Node value = GetRegister(instr.gpr0);
-
-        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 OperationCode operation_code = [instr] {
-            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:
-                return OperationCode::AtomicImageOr;
-            case Tegra::Shader::ImageAtomicOperation::Xor:
-                return OperationCode::AtomicImageXor;
-            case Tegra::Shader::ImageAtomicOperation::Exch:
-                return OperationCode::AtomicImageExchange;
+            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::And:
+                    return OperationCode::AtomicImageAnd;
+                case Tegra::Shader::ImageAtomicOperation::Or:
+                    return OperationCode::AtomicImageOr;
+                case Tegra::Shader::ImageAtomicOperation::Xor:
+                    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;
+                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;
         }();
 
-        const auto& image{GetImage(instr.image, instr.suatom_d.image_type, instr.suatom_d.size)};
+        Node value = GetRegister(instr.gpr0);
+
+        const auto type = instr.suatom_d.image_type;
+        auto& image = GetImage(instr.image, type);
+        image.MarkAtomic();
+
         MetaImage meta{image, {std::move(value)}};
-        SetRegister(bb, instr.gpr0, Operation(operation_code, meta, std::move(coords)));
+        SetRegister(bb, instr.gpr0, Operation(operation_code, meta, GetCoordinates(type)));
         break;
     }
     default:
@@ -114,35 +142,32 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
-                          std::optional<Tegra::Shader::ImageAtomicSize> size) {
+Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
     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,
-                                  std::optional<Tegra::Shader::ImageAtomicSize> size) {
+Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) {
     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,
-                                     std::optional<Tegra::Shader::ImageAtomicSize> size) {
+Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type) {
     auto it = used_images.find(offset);
     if (it == used_images.end()) {
         return nullptr;
@@ -150,14 +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;
 }
 
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index abf2cb1ab..338bab17c 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -149,10 +149,10 @@ enum class OperationCode {
     TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
     TexelFetch,             /// (MetaTexture, int[N], int) -> float4
 
-    ImageStore,          /// (MetaImage, int[N] values) -> void
+    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
@@ -294,21 +294,18 @@ private:
 
 class Image final {
 public:
-    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}, size{size} {}
+    constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type)
+        : offset{offset}, index{index}, type{type}, is_bindless{false} {}
 
     constexpr explicit Image(u32 cbuf_index, u32 cbuf_offset, std::size_t index,
-                             Tegra::Shader::ImageType type,
-                             std::optional<Tegra::Shader::ImageAtomicSize> size)
+                             Tegra::Shader::ImageType type)
         : 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,
-                             std::optional<Tegra::Shader::ImageAtomicSize> size)
+                             bool is_bindless, bool is_written, bool is_read, bool is_atomic)
         : 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;
@@ -318,8 +315,10 @@ public:
         is_read = true;
     }
 
-    void SetSize(Tegra::Shader::ImageAtomicSize size_) {
-        size = size_;
+    void MarkAtomic() {
+        MarkWrite();
+        MarkRead();
+        is_atomic = true;
     }
 
     constexpr std::size_t GetOffset() const {
@@ -346,21 +345,17 @@ public:
         return is_read;
     }
 
-    constexpr std::pair<u32, u32> GetBindlessCBuf() const {
-        return {static_cast<u32>(offset >> 32), static_cast<u32>(offset)};
+    constexpr bool IsAtomic() const {
+        return is_atomic;
     }
 
-    constexpr bool IsSizeKnown() const {
-        return size.has_value();
-    }
-
-    constexpr Tegra::Shader::ImageAtomicSize GetSize() const {
-        return size.value();
+    constexpr std::pair<u32, u32> GetBindlessCBuf() const {
+        return {static_cast<u32>(offset >> 32), static_cast<u32>(offset)};
     }
 
     constexpr bool operator<(const Image& rhs) const {
-        return std::tie(offset, index, type, size, is_bindless) <
-               std::tie(rhs.offset, rhs.index, rhs.type, rhs.size, rhs.is_bindless);
+        return std::tie(offset, index, type, is_bindless) <
+               std::tie(rhs.offset, rhs.index, rhs.type, rhs.is_bindless);
     }
 
 private:
@@ -370,7 +365,7 @@ private:
     bool is_bindless{};
     bool is_written{};
     bool is_read{};
-    std::optional<Tegra::Shader::ImageAtomicSize> size{};
+    bool is_atomic{};
 };
 
 struct GlobalMemoryBase {
@@ -402,6 +397,7 @@ struct MetaTexture {
 struct MetaImage {
     const Image& image;
     std::vector<Node> values;
+    u32 element{};
 };
 
 /// Parameters that modify an operation but are not part of any particular operand
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 2f03d83ba..6f666ee30 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -284,16 +284,13 @@ private:
                                       bool is_shadow);
 
     /// Accesses an image.
-    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
-                    std::optional<Tegra::Shader::ImageAtomicSize> size = {});
+    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
 
     /// Access a bindless image sampler.
-    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type,
-                            std::optional<Tegra::Shader::ImageAtomicSize> size = {});
+    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type);
 
     /// Tries to access an existing image, updating it's state as needed
-    Image* TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
-                               std::optional<Tegra::Shader::ImageAtomicSize> size);
+    Image* TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type);
 
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);