Merge pull request #3039 from ReinUsesLisp/cleanup-samplers

shader/node: Unpack bindless texture encoding

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -935,10 +935,9 @@ TextureBufferUsage RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stag
             if (!entry.IsBindless()) {
                 return maxwell3d.GetStageTexture(stage, entry.GetOffset());
             }
-            const auto cbuf = entry.GetBindlessCBuf();
-            Tegra::Texture::TextureHandle tex_handle;
-            Tegra::Engines::ShaderType shader_type = static_cast<Tegra::Engines::ShaderType>(stage);
-            tex_handle.raw = maxwell3d.AccessConstBuffer32(shader_type, cbuf.first, cbuf.second);
+            const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage);
+            const Tegra::Texture::TextureHandle tex_handle =
+                maxwell3d.AccessConstBuffer32(shader_type, entry.GetBuffer(), entry.GetOffset());
             return maxwell3d.GetTextureInfo(tex_handle);
         }();
 
@@ -966,10 +965,8 @@ TextureBufferUsage RasterizerOpenGL::SetupComputeTextures(const Shader& kernel)
             if (!entry.IsBindless()) {
                 return compute.GetTexture(entry.GetOffset());
             }
-            const auto cbuf = entry.GetBindlessCBuf();
-            Tegra::Texture::TextureHandle tex_handle;
-            tex_handle.raw = compute.AccessConstBuffer32(Tegra::Engines::ShaderType::Compute,
-                                                         cbuf.first, cbuf.second);
+            const Tegra::Texture::TextureHandle tex_handle = compute.AccessConstBuffer32(
+                Tegra::Engines::ShaderType::Compute, entry.GetBuffer(), entry.GetOffset());
             return compute.GetTextureInfo(tex_handle);
         }();
 
@@ -1012,10 +1009,8 @@ void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
             if (!entry.IsBindless()) {
                 return compute.GetTexture(entry.GetOffset()).tic;
             }
-            const auto cbuf = entry.GetBindlessCBuf();
-            Tegra::Texture::TextureHandle tex_handle;
-            tex_handle.raw = compute.AccessConstBuffer32(Tegra::Engines::ShaderType::Compute,
-                                                         cbuf.first, cbuf.second);
+            const Tegra::Texture::TextureHandle tex_handle = compute.AccessConstBuffer32(
+                Tegra::Engines::ShaderType::Compute, entry.GetBuffer(), entry.GetOffset());
             return compute.GetTextureInfo(tex_handle).tic;
         }();
         SetupImage(bindpoint, tic, entry);

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -735,7 +735,7 @@ private:
 
     void DeclareImages() {
         const auto& images{ir.GetImages()};
-        for (const auto& [offset, image] : images) {
+        for (const auto& image : images) {
             std::string qualifier = "coherent volatile";
             if (image.IsRead() && !image.IsWritten()) {
                 qualifier += " readonly";
@@ -2466,16 +2466,16 @@ ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir) {
         entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(),
                                            cbuf.first);
     }
+    for (const auto& [base, usage] : ir.GetGlobalMemory()) {
+        entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_read,
+                                                   usage.is_written);
+    }
     for (const auto& sampler : ir.GetSamplers()) {
         entries.samplers.emplace_back(sampler);
     }
-    for (const auto& [offset, image] : ir.GetImages()) {
+    for (const auto& image : ir.GetImages()) {
         entries.images.emplace_back(image);
     }
-    for (const auto& [base, usage] : ir.GetGlobalMemory()) {
-        entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_read,
-                                                   usage.is_written);
-    }
     entries.clip_distances = ir.GetClipDistances();
     entries.shader_length = ir.GetLength();
     return entries;

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -82,10 +82,9 @@ private:
 
 struct ShaderEntries {
     std::vector<ConstBufferEntry> const_buffers;
+    std::vector<GlobalMemoryEntry> global_memory_entries;
     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/shader/decode/image.cpp

@@ -143,39 +143,37 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
 }
 
 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 existing_image = TryUseExistingImage(offset, type)) {
-        return *existing_image;
-    }
-
-    const std::size_t next_index{used_images.size()};
-    return used_images.emplace(offset, Image{offset, next_index, type}).first->second;
-}
-
-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)) {
-        return *image;
-    }
+    const auto offset = static_cast<u32>(image.index.Value());
 
-    const std::size_t next_index{used_images.size()};
-    return used_images.emplace(cbuf_key, Image{cbuf_index, cbuf_offset, next_index, type})
-        .first->second;
+    const auto it =
+        std::find_if(std::begin(used_images), std::end(used_images),
+                     [offset](const Image& entry) { return entry.GetOffset() == offset; });
+    if (it != std::end(used_images)) {
+        ASSERT(!it->IsBindless() && it->GetType() == it->GetType());
+        return *it;
     }
 
-Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type) {
-    auto it = used_images.find(offset);
-    if (it == used_images.end()) {
-        return nullptr;
+    const auto next_index = static_cast<u32>(used_images.size());
+    return used_images.emplace_back(next_index, offset, type);
 }
-    auto& image = it->second;
-    ASSERT(image.GetType() == type);
 
-    return &image;
+Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) {
+    const Node image_register = GetRegister(reg);
+    const auto [base_image, buffer, offset] =
+        TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()));
+
+    const auto it =
+        std::find_if(std::begin(used_images), std::end(used_images),
+                     [buffer = buffer, offset = offset](const Image& entry) {
+                         return entry.GetBuffer() == buffer && entry.GetOffset() == offset;
+                     });
+    if (it != std::end(used_images)) {
+        ASSERT(it->IsBindless() && it->GetType() == it->GetType());
+        return *it;
+    }
+
+    const auto next_index = static_cast<u32>(used_images.size());
+    return used_images.emplace_back(next_index, offset, buffer, type);
 }
 
 } // namespace VideoCommon::Shader

src/video_core/shader/decode/texture.cpp

@@ -262,7 +262,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::TLDS: {
-        const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()};
+        const TextureType texture_type{instr.tlds.GetTextureType()};
         const bool is_array{instr.tlds.IsArrayTexture()};
 
         UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI),
@@ -293,77 +293,80 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
                                     std::optional<SamplerInfo> sampler_info) {
     const auto offset = static_cast<u32>(sampler.index.Value());
 
-    Tegra::Shader::TextureType type;
+    TextureType type;
     bool is_array;
     bool is_shadow;
     if (sampler_info) {
         type = sampler_info->type;
         is_array = sampler_info->is_array;
         is_shadow = sampler_info->is_shadow;
-    } else if (auto sampler = locker.ObtainBoundSampler(offset); sampler) {
+    } else if (const auto sampler = locker.ObtainBoundSampler(offset)) {
         type = sampler->texture_type.Value();
         is_array = sampler->is_array.Value() != 0;
         is_shadow = sampler->is_shadow.Value() != 0;
     } else {
-        type = Tegra::Shader::TextureType::Texture2D;
+        LOG_WARNING(HW_GPU, "Unknown sampler info");
+        type = TextureType::Texture2D;
         is_array = false;
         is_shadow = false;
     }
 
     // If this sampler has already been used, return the existing mapping.
-    const auto itr =
+    const auto it =
         std::find_if(used_samplers.begin(), used_samplers.end(),
-                     [&](const Sampler& entry) { return entry.GetOffset() == offset; });
-    if (itr != used_samplers.end()) {
-        ASSERT(itr->GetType() == type && itr->IsArray() == is_array &&
-               itr->IsShadow() == is_shadow);
-        return *itr;
+                     [offset](const Sampler& entry) { return entry.GetOffset() == offset; });
+    if (it != used_samplers.end()) {
+        ASSERT(!it->IsBindless() && it->GetType() == type && it->IsArray() == is_array &&
+               it->IsShadow() == is_shadow);
+        return *it;
     }
 
     // Otherwise create a new mapping for this sampler
-    const std::size_t next_index = used_samplers.size();
-    const Sampler entry{offset, next_index, type, is_array, is_shadow};
-    return *used_samplers.emplace(entry).first;
-} // namespace VideoCommon::Shader
+    const auto next_index = static_cast<u32>(used_samplers.size());
+    return used_samplers.emplace_back(Sampler(next_index, offset, type, is_array, is_shadow));
+}
 
 const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
                                             std::optional<SamplerInfo> sampler_info) {
     const Node sampler_register = GetRegister(reg);
-    const auto [base_sampler, cbuf_index, cbuf_offset] =
+    const auto [base_sampler, buffer, offset] =
         TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
     ASSERT(base_sampler != nullptr);
-    const auto cbuf_key = (static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset);
-    Tegra::Shader::TextureType type;
+
+    TextureType type;
     bool is_array;
     bool is_shadow;
     if (sampler_info) {
         type = sampler_info->type;
         is_array = sampler_info->is_array;
         is_shadow = sampler_info->is_shadow;
-    } else if (auto sampler = locker.ObtainBindlessSampler(cbuf_index, cbuf_offset); sampler) {
+    } else if (const auto sampler = locker.ObtainBindlessSampler(buffer, offset)) {
         type = sampler->texture_type.Value();
         is_array = sampler->is_array.Value() != 0;
         is_shadow = sampler->is_shadow.Value() != 0;
     } else {
-        type = Tegra::Shader::TextureType::Texture2D;
+        LOG_WARNING(HW_GPU, "Unknown sampler info");
+        type = TextureType::Texture2D;
         is_array = false;
         is_shadow = false;
     }
 
     // If this sampler has already been used, return the existing mapping.
-    const auto itr =
+    const auto it =
         std::find_if(used_samplers.begin(), used_samplers.end(),
-                     [&](const Sampler& entry) { return entry.GetOffset() == cbuf_key; });
-    if (itr != used_samplers.end()) {
-        ASSERT(itr->GetType() == type && itr->IsArray() == is_array &&
-               itr->IsShadow() == is_shadow);
-        return *itr;
+                     [buffer = buffer, offset = offset](const Sampler& entry) {
+                         return entry.GetBuffer() == buffer && entry.GetOffset() == offset;
+                     });
+    if (it != used_samplers.end()) {
+        ASSERT(it->IsBindless() && it->GetType() == type && it->IsArray() == is_array &&
+               it->IsShadow() == is_shadow);
+        return *it;
     }
 
     // Otherwise create a new mapping for this sampler
-    const std::size_t next_index = used_samplers.size();
-    const Sampler entry{cbuf_index, cbuf_offset, next_index, type, is_array, is_shadow};
-    return *used_samplers.emplace(entry).first;
+    const auto next_index = static_cast<u32>(used_samplers.size());
+    return used_samplers.emplace_back(
+        Sampler(next_index, offset, buffer, type, is_array, is_shadow));
 }
 
 void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) {

src/video_core/shader/node.h

@@ -230,62 +230,49 @@ using NodeBlock = std::vector<Node>;
 class Sampler {
 public:
     /// This constructor is for bound samplers
-    explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type,
+    constexpr explicit Sampler(u32 index, u32 offset, Tegra::Shader::TextureType type,
                                bool is_array, bool is_shadow)
-        : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow},
-          is_bindless{false} {}
+        : index{index}, offset{offset}, type{type}, is_array{is_array}, is_shadow{is_shadow} {}
 
     /// This constructor is for bindless samplers
-    explicit Sampler(u32 cbuf_index, u32 cbuf_offset, std::size_t index,
-                     Tegra::Shader::TextureType type, bool is_array, bool is_shadow)
-        : offset{(static_cast<u64>(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type},
-          is_array{is_array}, is_shadow{is_shadow}, is_bindless{true} {}
-
-    /// This constructor is for serialization/deserialization
-    explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type,
-                     bool is_array, bool is_shadow, bool is_bindless)
-        : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow},
-          is_bindless{is_bindless} {}
-
-    std::size_t GetOffset() const {
+    constexpr explicit Sampler(u32 index, u32 offset, u32 buffer, Tegra::Shader::TextureType type,
+                               bool is_array, bool is_shadow)
+        : index{index}, offset{offset}, buffer{buffer}, type{type}, is_array{is_array},
+          is_shadow{is_shadow}, is_bindless{true} {}
+
+    constexpr u32 GetIndex() const {
+        return index;
+    }
+
+    constexpr u32 GetOffset() const {
         return offset;
     }
 
-    std::size_t GetIndex() const {
-        return index;
+    constexpr u32 GetBuffer() const {
+        return buffer;
     }
 
-    Tegra::Shader::TextureType GetType() const {
+    constexpr Tegra::Shader::TextureType GetType() const {
         return type;
     }
 
-    bool IsArray() const {
+    constexpr bool IsArray() const {
         return is_array;
     }
 
-    bool IsShadow() const {
+    constexpr bool IsShadow() const {
         return is_shadow;
     }
 
-    bool IsBindless() const {
+    constexpr bool IsBindless() const {
         return is_bindless;
     }
 
-    std::pair<u32, u32> GetBindlessCBuf() const {
-        return {static_cast<u32>(offset >> 32), static_cast<u32>(offset)};
-    }
-
-    bool operator<(const Sampler& rhs) const {
-        return std::tie(index, offset, type, is_array, is_shadow, is_bindless) <
-               std::tie(rhs.index, rhs.offset, rhs.type, rhs.is_array, rhs.is_shadow,
-                        rhs.is_bindless);
-    }
-
 private:
-    /// Offset in TSC memory from which to read the sampler object, as specified by the sampling
-    /// instruction.
-    std::size_t offset{};
-    std::size_t index{}; ///< Value used to index into the generated GLSL sampler array.
+    u32 index{};  ///< Emulated index given for the this sampler.
+    u32 offset{}; ///< Offset in the const buffer from where the sampler is being read.
+    u32 buffer{}; ///< Buffer where the bindless sampler is being read (unused on bound samplers).
+
     Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc)
     bool is_array{};    ///< Whether the texture is being sampled as an array texture or not.
     bool is_shadow{};   ///< Whether the texture is being sampled as a depth texture or not.
@@ -294,18 +281,13 @@ private:
 
 class Image final {
 public:
-    constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type)
-        : offset{offset}, index{index}, type{type}, is_bindless{false} {}
+    /// This constructor is for bound images
+    constexpr explicit Image(u32 index, u32 offset, Tegra::Shader::ImageType type)
+        : index{index}, offset{offset}, type{type} {}
 
-    constexpr 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} {}
-
-    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_atomic)
-        : offset{offset}, index{index}, type{type}, is_bindless{is_bindless},
-          is_written{is_written}, is_read{is_read}, is_atomic{is_atomic} {}
+    /// This constructor is for bindless samplers
+    constexpr explicit Image(u32 index, u32 offset, u32 buffer, Tegra::Shader::ImageType type)
+        : index{index}, offset{offset}, buffer{buffer}, type{type}, is_bindless{true} {}
 
     void MarkWrite() {
         is_written = true;
@@ -321,12 +303,16 @@ public:
         is_atomic = true;
     }
 
-    constexpr std::size_t GetOffset() const {
+    constexpr u32 GetIndex() const {
+        return index;
+    }
+
+    constexpr u32 GetOffset() const {
         return offset;
     }
 
-    constexpr std::size_t GetIndex() const {
-        return index;
+    constexpr u32 GetBuffer() const {
+        return buffer;
     }
 
     constexpr Tegra::Shader::ImageType GetType() const {
@@ -349,18 +335,11 @@ public:
         return is_atomic;
     }
 
-    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, is_bindless) <
-               std::tie(rhs.offset, rhs.index, rhs.type, rhs.is_bindless);
-    }
-
 private:
-    u64 offset{};
-    std::size_t index{};
+    u32 index{};
+    u32 offset{};
+    u32 buffer{};
+
     Tegra::Shader::ImageType type{};
     bool is_bindless{};
     bool is_written{};

src/video_core/shader/shader_ir.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <array>
+#include <list>
 #include <map>
 #include <optional>
 #include <set>
@@ -95,11 +96,11 @@ public:
         return used_cbufs;
     }
 
-    const std::set<Sampler>& GetSamplers() const {
+    const std::list<Sampler>& GetSamplers() const {
         return used_samplers;
     }
 
-    const std::map<u64, Image>& GetImages() const {
+    const std::list<Image>& GetImages() const {
         return used_images;
     }
 
@@ -316,9 +317,6 @@ private:
     /// Access a bindless image sampler.
     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);
-
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);
 
@@ -402,8 +400,8 @@ private:
     std::set<Tegra::Shader::Attribute::Index> used_input_attributes;
     std::set<Tegra::Shader::Attribute::Index> used_output_attributes;
     std::map<u32, ConstBuffer> used_cbufs;
-    std::set<Sampler> used_samplers;
-    std::map<u64, Image> used_images;
+    std::list<Sampler> used_samplers;
+    std::list<Image> used_images;
     std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
     std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory;
     bool uses_layer{};

src/video_core/textures/texture.h

@@ -132,6 +132,8 @@ enum class SwizzleSource : u32 {
 };
 
 union TextureHandle {
+    TextureHandle(u32 raw) : raw{raw} {}
+
     u32 raw;
     BitField<0, 20, u32> tic_id;
     BitField<20, 12, u32> tsc_id;