11 files changed, 100 insertions, 34 deletions

diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp
index b7bc11416..85ee27333 100644
--- a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp
+++ b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp
@@ -59,6 +59,13 @@ std::string Image(EmitContext& ctx, IR::TextureInstInfo info,
     }
 }
 
+bool IsTextureMsaa(EmitContext& ctx, const IR::TextureInstInfo& info) {
+    if (info.type == TextureType::Buffer) {
+        return false;
+    }
+    return ctx.info.texture_descriptors.at(info.descriptor_index).is_multisample;
+}
+
 std::string_view TextureType(IR::TextureInstInfo info, bool is_ms = false) {
     if (info.is_depth) {
         switch (info.type) {
@@ -535,7 +542,8 @@ void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value&
                               ScalarS32 lod, [[maybe_unused]] const IR::Value& skip_mips) {
     const auto info{inst.Flags<IR::TextureInstInfo>()};
     const std::string texture{Texture(ctx, info, index)};
-    const std::string_view type{TextureType(info)};
+    const bool is_msaa{IsTextureMsaa(ctx, info)};
+    const std::string_view type{TextureType(info, is_msaa)};
     ctx.Add("TXQ {},{},{},{};", inst, lod, texture, type);
 }
 
diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp
index 4be2c25ec..f335c8af0 100644
--- a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp
+++ b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp
@@ -25,6 +25,13 @@ std::string Image(EmitContext& ctx, const IR::TextureInstInfo& info, const IR::V
     return fmt::format("img{}{}", def.binding, index_offset);
 }
 
+bool IsTextureMsaa(EmitContext& ctx, const IR::TextureInstInfo& info) {
+    if (info.type == TextureType::Buffer) {
+        return false;
+    }
+    return ctx.info.texture_descriptors.at(info.descriptor_index).is_multisample;
+}
+
 std::string CastToIntVec(std::string_view value, const IR::TextureInstInfo& info) {
     switch (info.type) {
     case TextureType::Color1D:
@@ -463,26 +470,33 @@ void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value&
                               std::string_view lod, const IR::Value& skip_mips_val) {
     const auto info{inst.Flags<IR::TextureInstInfo>()};
     const auto texture{Texture(ctx, info, index)};
+    const bool is_msaa{IsTextureMsaa(ctx, info)};
     const bool skip_mips{skip_mips_val.U1()};
-    const auto mips{
-        [&] { return skip_mips ? "0u" : fmt::format("uint(textureQueryLevels({}))", texture); }};
+    const auto mips{skip_mips ? "0u" : fmt::format("uint(textureQueryLevels({}))", texture)};
+    if (is_msaa && !skip_mips) {
+        throw NotImplementedException("EmitImageQueryDimensions MSAA QueryLevels");
+    }
+    if (info.type == TextureType::Buffer && !skip_mips) {
+        throw NotImplementedException("EmitImageQueryDimensions TextureType::Buffer QueryLevels");
+    }
+    const bool uses_lod{!is_msaa && info.type != TextureType::Buffer};
+    const auto lod_str{uses_lod ? fmt::format(",int({})", lod) : ""};
     switch (info.type) {
     case TextureType::Color1D:
-        return ctx.AddU32x4("{}=uvec4(uint(textureSize({},int({}))),0u,0u,{});", inst, texture, lod,
-                            mips());
+        return ctx.AddU32x4("{}=uvec4(uint(textureSize({}{})),0u,0u,{});", inst, texture, lod_str,
+                            mips);
     case TextureType::ColorArray1D:
     case TextureType::Color2D:
     case TextureType::ColorCube:
     case TextureType::Color2DRect:
-        return ctx.AddU32x4("{}=uvec4(uvec2(textureSize({},int({}))),0u,{});", inst, texture, lod,
-                            mips());
+        return ctx.AddU32x4("{}=uvec4(uvec2(textureSize({}{})),0u,{});", inst, texture, lod_str,
+                            mips);
     case TextureType::ColorArray2D:
     case TextureType::Color3D:
     case TextureType::ColorArrayCube:
-        return ctx.AddU32x4("{}=uvec4(uvec3(textureSize({},int({}))),{});", inst, texture, lod,
-                            mips());
+        return ctx.AddU32x4("{}=uvec4(uvec3(textureSize({}{})),{});", inst, texture, lod_str, mips);
     case TextureType::Buffer:
-        throw NotImplementedException("EmitImageQueryDimensions Texture buffers");
+        return ctx.AddU32x4("{}=uvec4(uint(textureSize({})),0u,0u,{});", inst, texture, mips);
     }
     throw LogicError("Unspecified image type {}", info.type.Value());
 }
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
index 3b969d915..02073c420 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
@@ -201,6 +201,13 @@ Id Image(EmitContext& ctx, const IR::Value& index, IR::TextureInstInfo info) {
     }
 }
 
+bool IsTextureMsaa(EmitContext& ctx, const IR::TextureInstInfo& info) {
+    if (info.type == TextureType::Buffer) {
+        return false;
+    }
+    return ctx.textures.at(info.descriptor_index).is_multisample;
+}
+
 Id Decorate(EmitContext& ctx, IR::Inst* inst, Id sample) {
     const auto info{inst->Flags<IR::TextureInstInfo>()};
     if (info.relaxed_precision != 0) {
@@ -452,24 +459,26 @@ Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& i
     const Id zero{ctx.u32_zero_value};
     const bool skip_mips{skip_mips_val.U1()};
     const auto mips{[&] { return skip_mips ? zero : ctx.OpImageQueryLevels(ctx.U32[1], image); }};
+    const bool is_msaa{IsTextureMsaa(ctx, info)};
+    const bool uses_lod{!is_msaa && info.type != TextureType::Buffer};
+    const auto query{[&](Id type) {
+        return uses_lod ? ctx.OpImageQuerySizeLod(type, image, lod)
+                        : ctx.OpImageQuerySize(type, image);
+    }};
     switch (info.type) {
     case TextureType::Color1D:
-        return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod),
-                                        zero, zero, mips());
+        return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[1]), zero, zero, mips());
     case TextureType::ColorArray1D:
     case TextureType::Color2D:
     case TextureType::ColorCube:
     case TextureType::Color2DRect:
-        return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[2], image, lod),
-                                        zero, mips());
+        return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[2]), zero, mips());
     case TextureType::ColorArray2D:
     case TextureType::Color3D:
     case TextureType::ColorArrayCube:
-        return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[3], image, lod),
-                                        mips());
+        return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[3]), mips());
     case TextureType::Buffer:
-        return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySize(ctx.U32[1], image), zero,
-                                        zero, mips());
+        return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[1]), zero, zero, mips());
     }
     throw LogicError("Unspecified image type {}", info.type.Value());
 }
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
index 3b97721e1..d48d4860e 100644
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
@@ -1288,6 +1288,7 @@ void EmitContext::DefineTextures(const Info& info, u32& binding, u32& scaling_in
             .pointer_type = pointer_type,
             .image_type = image_type,
             .count = desc.count,
+            .is_multisample = desc.is_multisample,
         });
         if (profile.supported_spirv >= 0x00010400) {
             interfaces.push_back(id);
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h
index dbc5c55b9..768a4fbb5 100644
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h
@@ -35,6 +35,7 @@ struct TextureDefinition {
     Id pointer_type;
     Id image_type;
     u32 count;
+    bool is_multisample;
 };
 
 struct TextureBufferDefinition {
diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp
index 26d066004..1a0cea9b7 100644
--- a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp
+++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp
@@ -30,7 +30,7 @@ bool ComputePipelineKey::operator==(const ComputePipelineKey& rhs) const noexcep
 ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cache_,
                                  BufferCache& buffer_cache_, ProgramManager& program_manager_,
                                  const Shader::Info& info_, std::string code,
-                                 std::vector<u32> code_v)
+                                 std::vector<u32> code_v, bool force_context_flush)
     : texture_cache{texture_cache_}, buffer_cache{buffer_cache_},
       program_manager{program_manager_}, info{info_} {
     switch (device.GetShaderBackend()) {
@@ -63,6 +63,15 @@ ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cac
     writes_global_memory = !use_storage_buffers &&
                            std::ranges::any_of(info.storage_buffers_descriptors,
                                                [](const auto& desc) { return desc.is_written; });
+    if (force_context_flush) {
+        std::scoped_lock lock{built_mutex};
+        built_fence.Create();
+        // Flush this context to ensure compilation commands and fence are in the GPU pipe.
+        glFlush();
+        built_condvar.notify_one();
+    } else {
+        is_built = true;
+    }
 }
 
 void ComputePipeline::Configure() {
@@ -142,6 +151,9 @@ void ComputePipeline::Configure() {
     }
     texture_cache.FillComputeImageViews(std::span(views.data(), views.size()));
 
+    if (!is_built) {
+        WaitForBuild();
+    }
     if (assembly_program.handle != 0) {
         program_manager.BindComputeAssemblyProgram(assembly_program.handle);
     } else {
@@ -223,4 +235,13 @@ void ComputePipeline::Configure() {
     }
 }
 
+void ComputePipeline::WaitForBuild() {
+    if (built_fence.handle == 0) {
+        std::unique_lock lock{built_mutex};
+        built_condvar.wait(lock, [this] { return built_fence.handle != 0; });
+    }
+    ASSERT(glClientWaitSync(built_fence.handle, 0, GL_TIMEOUT_IGNORED) != GL_WAIT_FAILED);
+    is_built = true;
+}
+
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.h b/src/video_core/renderer_opengl/gl_compute_pipeline.h
index 6534dec32..9bcc72b59 100644
--- a/src/video_core/renderer_opengl/gl_compute_pipeline.h
+++ b/src/video_core/renderer_opengl/gl_compute_pipeline.h
@@ -50,7 +50,8 @@ class ComputePipeline {
 public:
     explicit ComputePipeline(const Device& device, TextureCache& texture_cache_,
                              BufferCache& buffer_cache_, ProgramManager& program_manager_,
-                             const Shader::Info& info_, std::string code, std::vector<u32> code_v);
+                             const Shader::Info& info_, std::string code, std::vector<u32> code_v,
+                             bool force_context_flush = false);
 
     void Configure();
 
@@ -65,6 +66,8 @@ public:
     }
 
 private:
+    void WaitForBuild();
+
     TextureCache& texture_cache;
     BufferCache& buffer_cache;
     Tegra::MemoryManager* gpu_memory;
@@ -81,6 +84,11 @@ private:
 
     bool use_storage_buffers{};
     bool writes_global_memory{};
+
+    std::mutex built_mutex;
+    std::condition_variable built_condvar;
+    OGLSync built_fence{};
+    bool is_built{false};
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
index c115dabe1..29491e762 100644
--- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
+++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
@@ -176,7 +176,7 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c
                                    std::array<std::string, 5> sources,
                                    std::array<std::vector<u32>, 5> sources_spirv,
                                    const std::array<const Shader::Info*, 5>& infos,
-                                   const GraphicsPipelineKey& key_)
+                                   const GraphicsPipelineKey& key_, bool force_context_flush)
     : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_},
       state_tracker{state_tracker_}, key{key_} {
     if (shader_notify) {
@@ -231,7 +231,8 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c
     const bool in_parallel = thread_worker != nullptr;
     const auto backend = device.GetShaderBackend();
     auto func{[this, sources = std::move(sources), sources_spirv = std::move(sources_spirv),
-               shader_notify, backend, in_parallel](ShaderContext::Context*) mutable {
+               shader_notify, backend, in_parallel,
+               force_context_flush](ShaderContext::Context*) mutable {
         for (size_t stage = 0; stage < 5; ++stage) {
             switch (backend) {
             case Settings::ShaderBackend::GLSL:
@@ -251,7 +252,7 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c
                 break;
             }
         }
-        if (in_parallel) {
+        if (force_context_flush || in_parallel) {
             std::scoped_lock lock{built_mutex};
             built_fence.Create();
             // Flush this context to ensure compilation commands and fence are in the GPU pipe.
diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h
index 1c06b3655..7bab3be0a 100644
--- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h
+++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h
@@ -78,7 +78,7 @@ public:
                               std::array<std::string, 5> sources,
                               std::array<std::vector<u32>, 5> sources_spirv,
                               const std::array<const Shader::Info*, 5>& infos,
-                              const GraphicsPipelineKey& key_);
+                              const GraphicsPipelineKey& key_, bool force_context_flush = false);
 
     void Configure(bool is_indexed) {
         configure_func(this, is_indexed);
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 7dd854e0f..626ea7dcb 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -286,7 +286,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
         file.read(reinterpret_cast<char*>(&key), sizeof(key));
         queue_work([this, key, env = std::move(env), &state, &callback](Context* ctx) mutable {
             ctx->pools.ReleaseContents();
-            auto pipeline{CreateComputePipeline(ctx->pools, key, env)};
+            auto pipeline{CreateComputePipeline(ctx->pools, key, env, true)};
             std::scoped_lock lock{state.mutex};
             if (pipeline) {
                 compute_cache.emplace(key, std::move(pipeline));
@@ -307,7 +307,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
                 env_ptrs.push_back(&env);
             }
             ctx->pools.ReleaseContents();
-            auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false)};
+            auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false, true)};
             std::scoped_lock lock{state.mutex};
             if (pipeline) {
                 graphics_cache.emplace(key, std::move(pipeline));
@@ -439,7 +439,8 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline() {
 
 std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
     ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
-    std::span<Shader::Environment* const> envs, bool build_in_parallel) try {
+    std::span<Shader::Environment* const> envs, bool use_shader_workers,
+    bool force_context_flush) try {
     LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
     size_t env_index{};
     u32 total_storage_buffers{};
@@ -531,10 +532,10 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
         }
         previous_program = &program;
     }
-    auto* const thread_worker{build_in_parallel ? workers.get() : nullptr};
+    auto* const thread_worker{use_shader_workers ? workers.get() : nullptr};
     return std::make_unique<GraphicsPipeline>(device, texture_cache, buffer_cache, program_manager,
                                               state_tracker, thread_worker, &shader_notify, sources,
-                                              sources_spirv, infos, key);
+                                              sources_spirv, infos, key, force_context_flush);
 
 } catch (Shader::Exception& exception) {
     LOG_ERROR(Render_OpenGL, "{}", exception.what());
@@ -559,8 +560,8 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
 }
 
 std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
-    ShaderContext::ShaderPools& pools, const ComputePipelineKey& key,
-    Shader::Environment& env) try {
+    ShaderContext::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env,
+    bool force_context_flush) try {
     LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
@@ -589,7 +590,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
     }
 
     return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, program_manager,
-                                             program.info, code, code_spirv);
+                                             program.info, code, code_spirv, force_context_flush);
 } catch (Shader::Exception& exception) {
     LOG_ERROR(Render_OpenGL, "{}", exception.what());
     return nullptr;
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index f82420592..6b9732fca 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -50,14 +50,16 @@ private:
 
     std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
         ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
-        std::span<Shader::Environment* const> envs, bool build_in_parallel);
+        std::span<Shader::Environment* const> envs, bool use_shader_workers,
+        bool force_context_flush = false);
 
     std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineKey& key,
                                                            const VideoCommon::ShaderInfo* shader);
 
     std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderContext::ShaderPools& pools,
                                                            const ComputePipelineKey& key,
-                                                           Shader::Environment& env);
+                                                           Shader::Environment& env,
+                                                           bool force_context_flush = false);
 
     std::unique_ptr<ShaderWorker> CreateWorkers() const;