From 48aad8dc05f027c21aa0e8a68d827006d9f7a196 Mon Sep 17 00:00:00 2001
From: ReinUsesLisp <reinuseslisp@airmail.cc>
Date: Sat, 5 Jun 2021 22:10:02 -0300
Subject: vk_pipeline_cache: Add asynchronous shaders

---
 .../renderer_vulkan/vk_graphics_pipeline.h         |  6 ++++-
 .../renderer_vulkan/vk_pipeline_cache.cpp          | 26 ++++++++++++++++++++--
 src/video_core/renderer_vulkan/vk_pipeline_cache.h |  4 ++++
 3 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
index 8c81c28a8..3f8895927 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h
@@ -87,7 +87,7 @@ public:
         configure_func(this, is_indexed);
     }
 
-    GraphicsPipeline* Next(const GraphicsPipelineCacheKey& current_key) noexcept {
+    [[nodiscard]] GraphicsPipeline* Next(const GraphicsPipelineCacheKey& current_key) noexcept {
         if (key == current_key) {
             return this;
         }
@@ -96,6 +96,10 @@ public:
                                            : nullptr;
     }
 
+    [[nodiscard]] bool IsBuilt() const noexcept {
+        return is_built.load(std::memory_order::relaxed);
+    }
+
     template <typename Spec>
     static auto MakeConfigureSpecFunc() {
         return [](GraphicsPipeline* pipeline, bool is_indexed) {
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index a7f3619a5..741ed1a98 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -240,6 +240,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
       device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_},
       update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_},
       buffer_cache{buffer_cache_}, texture_cache{texture_cache_},
+      use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
       workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"),
       serialization_thread(1, "yuzu:PipelineSerialization") {
     const auto& float_control{device.FloatControlProperties()};
@@ -303,7 +304,7 @@ GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() {
         GraphicsPipeline* const next{current_pipeline->Next(graphics_key)};
         if (next) {
             current_pipeline = next;
-            return current_pipeline;
+            return BuiltPipeline(current_pipeline);
         }
     }
     const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
@@ -318,7 +319,7 @@ GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() {
         current_pipeline->AddTransition(pipeline.get());
     }
     current_pipeline = pipeline.get();
-    return current_pipeline;
+    return BuiltPipeline(current_pipeline);
 }
 
 ComputePipeline* PipelineCache::CurrentComputePipeline() {
@@ -415,6 +416,27 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     workers.WaitForRequests();
 }
 
+GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const noexcept {
+    if (pipeline->IsBuilt()) {
+        return pipeline;
+    }
+    if (!use_asynchronous_shaders) {
+        return pipeline;
+    }
+    // If something is using depth, we can assume that games are not rendering anything which
+    // will be used one time.
+    if (maxwell3d.regs.zeta_enable) {
+        return nullptr;
+    }
+    // If games are using a small index count, we can assume these are full screen quads.
+    // Usually these shaders are only used once for building textures so we can assume they
+    // can't be built async
+    if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) {
+        return pipeline;
+    }
+    return nullptr;
+}
+
 std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     ShaderPools& pools, const GraphicsPipelineCacheKey& key,
     std::span<Shader::Environment* const> envs, bool build_in_parallel) try {
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
index 4116cc73f..869c63baf 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
@@ -115,6 +115,8 @@ public:
                            const VideoCore::DiskResourceLoadCallback& callback);
 
 private:
+    [[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept;
+
     std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
 
     std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
@@ -140,6 +142,8 @@ private:
     GraphicsPipelineCacheKey graphics_key{};
     GraphicsPipeline* current_pipeline{};
 
+    bool use_asynchronous_shaders{};
+
     std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<ComputePipeline>> compute_cache;
     std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<GraphicsPipeline>> graphics_cache;
 
-- 
cgit v1.2.3