shader/texture: Deduce texture buffers from locker

Instead of specializing shaders to separate texture buffers from 1D
textures, use the locker to deduce them while they are being decoded.

6 files changed, 47 insertions, 105 deletions

diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 05f8e511b..b76de71ec 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -271,9 +271,9 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         const auto stage = static_cast<Maxwell::ShaderStage>(index == 0 ? 0 : index - 1);
         SetupDrawConstBuffers(stage, shader);
         SetupDrawGlobalMemory(stage, shader);
-        const auto texture_buffer_usage{SetupDrawTextures(stage, shader, base_bindings)};
+        SetupDrawTextures(stage, shader, base_bindings);
 
-        const ProgramVariant variant{base_bindings, primitive_mode, texture_buffer_usage};
+        const ProgramVariant variant{base_bindings, primitive_mode};
         const auto [program_handle, next_bindings] = shader->GetProgramHandle(variant);
 
         switch (program) {
@@ -303,7 +303,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         // When VertexA is enabled, we have dual vertex shaders
         if (program == Maxwell::ShaderProgram::VertexA) {
             // VertexB was combined with VertexA, so we skip the VertexB iteration
-            index++;
+            ++index;
         }
 
         base_bindings = next_bindings;
@@ -732,11 +732,10 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
     }
 
     auto kernel = shader_cache.GetComputeKernel(code_addr);
-    ProgramVariant variant;
-    variant.texture_buffer_usage = SetupComputeTextures(kernel);
+    SetupComputeTextures(kernel);
     SetupComputeImages(kernel);
 
-    const auto [program, next_bindings] = kernel->GetProgramHandle(variant);
+    const auto [program, next_bindings] = kernel->GetProgramHandle({});
     state.draw.shader_program = program;
     state.draw.program_pipeline = 0;
 
@@ -918,9 +917,8 @@ void RasterizerOpenGL::SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entr
     bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
 }
 
-TextureBufferUsage RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stage,
-                                                       const Shader& shader,
-                                                       BaseBindings base_bindings) {
+void RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stage, const Shader& shader,
+                                         BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& gpu = system.GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
@@ -929,8 +927,6 @@ TextureBufferUsage RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stag
     ASSERT_MSG(base_bindings.sampler + entries.size() <= std::size(state.textures),
                "Exceeded the number of active textures.");
 
-    TextureBufferUsage texture_buffer_usage{0};
-
     for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& entry = entries[bindpoint];
         const auto texture = [&] {
@@ -943,15 +939,11 @@ TextureBufferUsage RasterizerOpenGL::SetupDrawTextures(Maxwell::ShaderStage stag
             return maxwell3d.GetTextureInfo(tex_handle);
         }();
 
-        if (SetupTexture(base_bindings.sampler + bindpoint, texture, entry)) {
-            texture_buffer_usage.set(bindpoint);
-        }
+        SetupTexture(base_bindings.sampler + bindpoint, texture, entry);
     }
-
-    return texture_buffer_usage;
 }
 
-TextureBufferUsage RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
+void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& compute = system.GPU().KeplerCompute();
     const auto& entries = kernel->GetShaderEntries().samplers;
@@ -959,8 +951,6 @@ TextureBufferUsage RasterizerOpenGL::SetupComputeTextures(const Shader& kernel)
     ASSERT_MSG(entries.size() <= std::size(state.textures),
                "Exceeded the number of active textures.");
 
-    TextureBufferUsage texture_buffer_usage{0};
-
     for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& entry = entries[bindpoint];
         const auto texture = [&] {
@@ -972,34 +962,29 @@ TextureBufferUsage RasterizerOpenGL::SetupComputeTextures(const Shader& kernel)
             return compute.GetTextureInfo(tex_handle);
         }();
 
-        if (SetupTexture(bindpoint, texture, entry)) {
-            texture_buffer_usage.set(bindpoint);
-        }
+        SetupTexture(bindpoint, texture, entry);
     }
-
-    return texture_buffer_usage;
 }
 
-bool RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
+void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
                                     const GLShader::SamplerEntry& entry) {
-    state.samplers[binding] = sampler_cache.GetSampler(texture.tsc);
-
     const auto view = texture_cache.GetTextureSurface(texture.tic, entry);
     if (!view) {
         // Can occur when texture addr is null or its memory is unmapped/invalid
+        state.samplers[binding] = 0;
         state.textures[binding] = 0;
-        return false;
+        return;
     }
     state.textures[binding] = view->GetTexture();
 
     if (view->GetSurfaceParams().IsBuffer()) {
-        return true;
+        return;
     }
+    state.samplers[binding] = sampler_cache.GetSampler(texture.tsc);
 
     // Apply swizzle to textures that are not buffers.
     view->ApplySwizzle(texture.tic.x_source, texture.tic.y_source, texture.tic.z_source,
                        texture.tic.w_source);
-    return false;
 }
 
 void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index bd6fe5c3a..0e0819d59 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -107,16 +107,15 @@ private:
     /// Syncs all the state, shaders, render targets and textures setting before a draw call.
     void DrawPrelude();
 
-    /// Configures the current textures to use for the draw command. Returns shaders texture buffer
-    /// usage.
-    TextureBufferUsage SetupDrawTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                                         const Shader& shader, BaseBindings base_bindings);
+    /// Configures the current textures to use for the draw command.
+    void SetupDrawTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, const Shader& shader,
+                           BaseBindings base_bindings);
 
-    /// Configures the textures used in a compute shader. Returns texture buffer usage.
-    TextureBufferUsage SetupComputeTextures(const Shader& kernel);
+    /// Configures the textures used in a compute shader.
+    void SetupComputeTextures(const Shader& kernel);
 
-    /// Configures a texture. Returns true when the texture is a texture buffer.
-    bool SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
+    /// Configures a texture.
+    void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
                       const GLShader::SamplerEntry& entry);
 
     /// Configures images in a compute shader.
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 04a239a39..7ce06a978 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -270,7 +270,6 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ProgramTy
 
     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 = fmt::format(R"(// {}
 #version 430 core
@@ -317,17 +316,6 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ProgramTy
             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\n", i);
-    }
-    if (texture_buffer_usage.any()) {
-        source += '\n';
-    }
-
     if (program_type == ProgramType::Geometry) {
         const auto [glsl_topology, debug_name, max_vertices] =
             GetPrimitiveDescription(primitive_mode);
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 4f2b49170..51c80bf32 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -658,9 +658,11 @@ private:
             const std::string description{"layout (binding = SAMPLER_BINDING_" +
                                           std::to_string(sampler.GetIndex()) + ") uniform"};
             std::string sampler_type = [&]() {
+                if (sampler.IsBuffer()) {
+                    return "samplerBuffer";
+                }
                 switch (sampler.GetType()) {
                 case Tegra::Shader::TextureType::Texture1D:
-                    // Special cased, read below.
                     return "sampler1D";
                 case Tegra::Shader::TextureType::Texture2D:
                     return "sampler2D";
@@ -680,19 +682,7 @@ private:
                 sampler_type += "Shadow";
             }
 
-            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("{} samplerBuffer {};", description, 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);
-            }
+            code.AddLine("{} {} {};", description, sampler_type, name);
         }
         if (!samplers.empty()) {
             code.AddNewLine();
@@ -1749,27 +1739,14 @@ private:
                 expr += ", ";
         }
 
-        // Store a copy of the expression without the lod to be used with texture buffers
-        std::string expr_buffer = expr;
-
-        if (meta->lod) {
+        if (meta->lod && !meta->sampler.IsBuffer()) {
             expr += ", ";
             expr += Visit(meta->lod).AsInt();
         }
         expr += ')';
         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, Type::Float};
+        return {std::move(expr), Type::Float};
     }
 
     Expression ImageLoad(Operation operation) {
@@ -2214,10 +2191,6 @@ private:
         return GetDeclarationWithSuffix(static_cast<u32>(image.GetIndex()), "image");
     }
 
-    void EmitIfdefIsBuffer(const Sampler& sampler) {
-        code.AddLine("#ifdef SAMPLER_{}_IS_BUFFER", sampler.GetIndex());
-    }
-
     std::string GetDeclarationWithSuffix(u32 index, std::string_view name) const {
         return fmt::format("{}_{}_{}", name, index, suffix);
     }
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
index 184a565e6..3f4daf28d 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
@@ -28,34 +28,35 @@ using VideoCommon::Shader::KeyMap;
 
 namespace {
 
+using ShaderCacheVersionHash = std::array<u8, 64>;
+
+enum class TransferableEntryKind : u32 {
+    Raw,
+    Usage,
+};
+
 struct ConstBufferKey {
-    u32 cbuf;
-    u32 offset;
-    u32 value;
+    u32 cbuf{};
+    u32 offset{};
+    u32 value{};
 };
 
 struct BoundSamplerKey {
-    u32 offset;
-    Tegra::Engines::SamplerDescriptor sampler;
+    u32 offset{};
+    Tegra::Engines::SamplerDescriptor sampler{};
 };
 
 struct BindlessSamplerKey {
-    u32 cbuf;
-    u32 offset;
-    Tegra::Engines::SamplerDescriptor sampler;
-};
-
-using ShaderCacheVersionHash = std::array<u8, 64>;
-
-enum class TransferableEntryKind : u32 {
-    Raw,
-    Usage,
+    u32 cbuf{};
+    u32 offset{};
+    Tegra::Engines::SamplerDescriptor sampler{};
 };
 
-constexpr u32 NativeVersion = 5;
+constexpr u32 NativeVersion = 6;
 
 // Making sure sizes doesn't change by accident
 static_assert(sizeof(BaseBindings) == 16);
+static_assert(sizeof(ProgramVariant) == 20);
 
 ShaderCacheVersionHash GetShaderCacheVersionHash() {
     ShaderCacheVersionHash hash{};
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
index db23ada93..55311dc6d 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h
@@ -4,7 +4,6 @@
 
 #pragma once
 
-#include <bitset>
 #include <optional>
 #include <string>
 #include <tuple>
@@ -37,7 +36,6 @@ struct ShaderDiskCacheDump;
 
 using ProgramCode = std::vector<u64>;
 using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
-using TextureBufferUsage = std::bitset<64>;
 
 /// Allocated bindings used by an OpenGL shader program
 struct BaseBindings {
@@ -61,11 +59,10 @@ static_assert(std::is_trivially_copyable_v<BaseBindings>);
 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);
+        return std::tie(base_bindings, primitive_mode) ==
+               std::tie(rhs.base_bindings, rhs.primitive_mode);
     }
 
     bool operator!=(const ProgramVariant& rhs) const {
@@ -112,7 +109,6 @@ template <>
 struct hash<OpenGL::ProgramVariant> {
     std::size_t operator()(const OpenGL::ProgramVariant& variant) const noexcept {
         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);
     }
 };