6 files changed, 51 insertions, 93 deletions

diff --git a/src/video_core/renderer_vulkan/vk_device.h b/src/video_core/renderer_vulkan/vk_device.h
index 30cd3e189..26a233db1 100644
--- a/src/video_core/renderer_vulkan/vk_device.h
+++ b/src/video_core/renderer_vulkan/vk_device.h
@@ -202,6 +202,7 @@ public:
         return reported_extensions;
     }
 
+    /// Returns true if the setting for async shader compilation is enabled.
     bool UseAsynchronousShaders() const {
         return use_asynchronous_shaders;
     }
@@ -255,7 +256,9 @@ private:
     bool ext_custom_border_color{};            ///< Support for VK_EXT_custom_border_color.
     bool ext_extended_dynamic_state{};         ///< Support for VK_EXT_extended_dynamic_state.
     bool nv_device_diagnostics_config{};       ///< Support for VK_NV_device_diagnostics_config.
-    bool use_asynchronous_shaders{};
+
+    // Asynchronous Graphics Pipeline setting
+    bool use_asynchronous_shaders{}; ///< Setting to use asynchronous shaders/graphics pipeline
 
     // Telemetry parameters
     std::string vendor_name;                      ///< Device's driver name.
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 7d51b9836..5dc4cd5af 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -78,14 +78,15 @@ VKGraphicsPipeline::VKGraphicsPipeline(const VKDevice& device, VKScheduler& sche
                                        const GraphicsPipelineCacheKey& key,
                                        vk::Span<VkDescriptorSetLayoutBinding> bindings,
                                        const SPIRVProgram& program)
-    : device{device}, scheduler{scheduler}, fixed_state{key.fixed_state}, hash{key.Hash()},
+    : device{device}, scheduler{scheduler}, hash{key.Hash()}, cache_key{key},
       descriptor_set_layout{CreateDescriptorSetLayout(bindings)},
       descriptor_allocator{descriptor_pool, *descriptor_set_layout},
       update_descriptor_queue{update_descriptor_queue}, layout{CreatePipelineLayout()},
       descriptor_template{CreateDescriptorUpdateTemplate(program)}, modules{CreateShaderModules(
                                                                         program)},
-      renderpass{renderpass_cache.GetRenderPass(key.renderpass_params)},
-      pipeline{CreatePipeline(key.renderpass_params, program)}, m_key{key} {}
+      renderpass{renderpass_cache.GetRenderPass(key.renderpass_params)}, pipeline{CreatePipeline(
+                                                                             key.renderpass_params,
+                                                                             program)} {}
 
 VKGraphicsPipeline::~VKGraphicsPipeline() = default;
 
@@ -180,7 +181,7 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
 
 vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
                                                 const SPIRVProgram& program) const {
-    const auto& state = fixed_state;
+    const auto& state = cache_key.fixed_state;
     const auto& viewport_swizzles = state.viewport_swizzles;
 
     FixedPipelineState::DynamicState dynamic;
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
index d50bd347c..9d462db0a 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
@@ -19,7 +19,27 @@ namespace Vulkan {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
-struct GraphicsPipelineCacheKey;
+struct GraphicsPipelineCacheKey {
+    RenderPassParams renderpass_params;
+    u32 padding;
+    std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
+    FixedPipelineState fixed_state;
+
+    std::size_t Hash() const noexcept;
+
+    bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept;
+
+    bool operator!=(const GraphicsPipelineCacheKey& rhs) const noexcept {
+        return !operator==(rhs);
+    }
+
+    std::size_t Size() const noexcept {
+        return sizeof(renderpass_params) + sizeof(padding) + sizeof(shaders) + fixed_state.Size();
+    }
+};
+static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>);
+static_assert(std::is_trivially_copyable_v<GraphicsPipelineCacheKey>);
+static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>);
 
 class VKDescriptorPool;
 class VKDevice;
@@ -54,8 +74,8 @@ public:
         return renderpass;
     }
 
-    const GraphicsPipelineCacheKey& GetCacheKey() const {
-        return m_key;
+    GraphicsPipelineCacheKey GetCacheKey() const {
+        return cache_key;
     }
 
 private:
@@ -74,8 +94,8 @@ private:
 
     const VKDevice& device;
     VKScheduler& scheduler;
-    const FixedPipelineState fixed_state;
     const u64 hash;
+    GraphicsPipelineCacheKey cache_key;
 
     vk::DescriptorSetLayout descriptor_set_layout;
     DescriptorAllocator descriptor_allocator;
@@ -86,8 +106,6 @@ private:
 
     VkRenderPass renderpass;
     vk::Pipeline pipeline;
-
-    const GraphicsPipelineCacheKey& m_key;
 };
 
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 1a8b2c62b..20ffbeb38 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -28,6 +28,7 @@
 #include "video_core/shader/compiler_settings.h"
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader_cache.h"
+#include "video_core/shader_notify.h"
 
 namespace Vulkan {
 
@@ -214,27 +215,31 @@ VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
     }
     last_graphics_key = key;
 
-    if (device.UseAsynchronousShaders()) {
+    if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(system.GPU())) {
         std::unique_lock lock{pipeline_cache};
         const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
         if (is_cache_miss) {
+            system.GPU().ShaderNotify().MarkSharderBuilding();
             LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
             const auto [program, bindings] = DecompileShaders(key.fixed_state);
-            async_shaders.QueueVulkanShader(this, bindings, program, key.renderpass_params,
-                                            key.padding, key.shaders, key.fixed_state);
+            async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool,
+                                            update_descriptor_queue, renderpass_cache, bindings,
+                                            program, key);
         }
-        last_graphics_pipeline = graphics_cache.at(key).get();
+        last_graphics_pipeline = pair->second.get();
         return last_graphics_pipeline;
     }
 
     const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
     auto& entry = pair->second;
     if (is_cache_miss) {
+        system.GPU().ShaderNotify().MarkSharderBuilding();
         LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
         const auto [program, bindings] = DecompileShaders(key.fixed_state);
         entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
                                                      update_descriptor_queue, renderpass_cache, key,
                                                      bindings, program);
+        system.GPU().ShaderNotify().MarkShaderComplete();
     }
     last_graphics_pipeline = entry.get();
     return last_graphics_pipeline;
@@ -294,14 +299,8 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
 
 void VKPipelineCache::EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline) {
     std::unique_lock lock{pipeline_cache};
-    const auto [pair, is_cache_miss] = graphics_cache.try_emplace(pipeline->GetCacheKey());
-    auto& entry = pair->second;
-    if (entry) {
-        LOG_INFO(Render_Vulkan, "Pipeline already here 0x{:016X}", pipeline->GetCacheKey().Hash());
-        duplicates.push_back(std::move(pipeline));
-    } else {
-        entry = std::move(pipeline);
-    }
+    graphics_cache.at(pipeline->GetCacheKey()) = std::move(pipeline);
+    system.GPU().ShaderNotify().MarkShaderComplete();
 }
 
 void VKPipelineCache::OnShaderRemoval(Shader* shader) {
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
index 777ef2038..c04829e77 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
@@ -44,28 +44,6 @@ class VKUpdateDescriptorQueue;
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
-struct GraphicsPipelineCacheKey {
-    RenderPassParams renderpass_params;
-    u32 padding;
-    std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
-    FixedPipelineState fixed_state;
-
-    std::size_t Hash() const noexcept;
-
-    bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept;
-
-    bool operator!=(const GraphicsPipelineCacheKey& rhs) const noexcept {
-        return !operator==(rhs);
-    }
-
-    std::size_t Size() const noexcept {
-        return sizeof(renderpass_params) + sizeof(padding) + sizeof(shaders) + fixed_state.Size();
-    }
-};
-static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>);
-static_assert(std::is_trivially_copyable_v<GraphicsPipelineCacheKey>);
-static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>);
-
 struct ComputePipelineCacheKey {
     GPUVAddr shader;
     u32 shared_memory_size;
@@ -158,41 +136,6 @@ public:
 
     VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
 
-    const VKDevice& GetDevice() const {
-        return device;
-    }
-
-    VKScheduler& GetScheduler() {
-        return scheduler;
-    }
-    const VKScheduler& GetScheduler() const {
-        return scheduler;
-    }
-
-    VKDescriptorPool& GetDescriptorPool() {
-        return descriptor_pool;
-    }
-
-    const VKDescriptorPool& GetDescriptorPool() const {
-        return descriptor_pool;
-    }
-
-    VKUpdateDescriptorQueue& GetUpdateDescriptorQueue() {
-        return update_descriptor_queue;
-    }
-
-    const VKUpdateDescriptorQueue& GetUpdateDescriptorQueue() const {
-        return update_descriptor_queue;
-    }
-
-    VKRenderPassCache& GetRenderpassCache() {
-        return renderpass_cache;
-    }
-
-    const VKRenderPassCache& GetRenderpassCache() const {
-        return renderpass_cache;
-    }
-
     void EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline);
 
 protected:
@@ -216,7 +159,6 @@ private:
 
     GraphicsPipelineCacheKey last_graphics_key;
     VKGraphicsPipeline* last_graphics_pipeline = nullptr;
-    std::vector<std::unique_ptr<VKGraphicsPipeline>> duplicates;
 
     std::mutex pipeline_cache;
     std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<VKGraphicsPipeline>>
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index fc1b51a96..720802ad5 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -14,6 +14,7 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/microprofile.h"
+#include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/settings.h"
 #include "video_core/engines/kepler_compute.h"
@@ -408,15 +409,10 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
 
         // Max worker threads we should allow
         constexpr u32 MAX_THREADS = 4;
-        // Amount of threads we should reserve for other parts of yuzu
-        constexpr u32 RESERVED_THREADS = 6;
-        // Get the amount of threads we can use(this can return zero)
-        const auto cpu_thread_count =
-            std::max(RESERVED_THREADS, std::thread::hardware_concurrency());
-        // Deduce how many "extra" threads we have to use.
-        const auto max_threads_unused = cpu_thread_count - RESERVED_THREADS;
+        // Deduce how many threads we can use
+        const auto threads_used = std::thread::hardware_concurrency() / 4;
         // Always allow at least 1 thread regardless of our settings
-        const auto max_worker_count = std::max(1u, max_threads_unused);
+        const auto max_worker_count = std::max(1U, threads_used);
         // Don't use more than MAX_THREADS
         const auto worker_count = std::min(max_worker_count, MAX_THREADS);
         async_shaders.AllocateWorkers(worker_count);
@@ -432,6 +428,8 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
 
     query_cache.UpdateCounters();
 
+    SCOPE_EXIT({ system.GPU().TickWork(); });
+
     const auto& gpu = system.GPU().Maxwell3D();
     GraphicsPipelineCacheKey key;
     key.fixed_state.Fill(gpu.regs, device.IsExtExtendedDynamicStateSupported());
@@ -458,10 +456,9 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
     key.renderpass_params = GetRenderPassParams(texceptions);
     key.padding = 0;
 
-    auto pipeline = pipeline_cache.GetGraphicsPipeline(key, async_shaders);
+    auto* pipeline = pipeline_cache.GetGraphicsPipeline(key, async_shaders);
     if (pipeline == nullptr || pipeline->GetHandle() == VK_NULL_HANDLE) {
         // Async graphics pipeline was not ready.
-        system.GPU().TickWork();
         return;
     }
 
@@ -488,8 +485,6 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
     });
 
     EndTransformFeedback();
-
-    system.GPU().TickWork();
 }
 
 void RasterizerVulkan::Clear() {