VideoBackends: Support a different number of threads for precompiling

At runtime, we only really want a single shader compiler thread.
However, for initial boots, we can use a higher number to speed things
up.

Source/Core/Core/Config/GraphicsSettings.cpp

@@ -85,6 +85,8 @@ const ConfigInfo<bool> GFX_PRECOMPILE_UBER_SHADERS{
     {System::GFX, "Settings", "PrecompileUberShaders"}, true};
 const ConfigInfo<int> GFX_SHADER_COMPILER_THREADS{
     {System::GFX, "Settings", "ShaderCompilerThreads"}, 1};
+const ConfigInfo<int> GFX_SHADER_PRECOMPILER_THREADS{
+    {System::GFX, "Settings", "ShaderPrecompilerThreads"}, 1};
 const ConfigInfo<bool> GFX_FORCE_VERTEX_UBER_SHADERS{
     {System::GFX, "Settings", "ForceVertexUberShaders"}, false};
 const ConfigInfo<bool> GFX_FORCE_PIXEL_UBER_SHADERS{

Source/Core/Core/Config/GraphicsSettings.h

@@ -63,6 +63,7 @@ extern const ConfigInfo<bool> GFX_BACKGROUND_SHADER_COMPILING;
 extern const ConfigInfo<bool> GFX_DISABLE_SPECIALIZED_SHADERS;
 extern const ConfigInfo<bool> GFX_PRECOMPILE_UBER_SHADERS;
 extern const ConfigInfo<int> GFX_SHADER_COMPILER_THREADS;
+extern const ConfigInfo<int> GFX_SHADER_PRECOMPILER_THREADS;
 extern const ConfigInfo<bool> GFX_FORCE_VERTEX_UBER_SHADERS;
 extern const ConfigInfo<bool> GFX_FORCE_PIXEL_UBER_SHADERS;
 

Source/Core/Core/ConfigLoaders/IsSettingSaveable.cpp

@@ -46,6 +46,7 @@ bool IsSettingSaveable(const Config::ConfigLocation& config_location)
       Config::GFX_BACKGROUND_SHADER_COMPILING.location,
       Config::GFX_DISABLE_SPECIALIZED_SHADERS.location,
       Config::GFX_PRECOMPILE_UBER_SHADERS.location, Config::GFX_SHADER_COMPILER_THREADS.location,
+      Config::GFX_SHADER_PRECOMPILER_THREADS.location,
       Config::GFX_FORCE_VERTEX_UBER_SHADERS.location, Config::GFX_FORCE_PIXEL_UBER_SHADERS.location,
 
       Config::GFX_SW_ZCOMPLOC.location, Config::GFX_SW_ZFREEZE.location,

Source/Core/VideoBackends/D3D/PixelShaderCache.cpp

@@ -513,7 +513,7 @@ void PixelShaderCache::Init()
     LoadShaderCache();
 
   if (g_ActiveConfig.CanPrecompileUberShaders())
-    PrecompileUberShaders();
+    QueueUberShaderCompiles();
 }
 
 void PixelShaderCache::LoadShaderCache()
@@ -538,7 +538,7 @@ void PixelShaderCache::Reload()
     LoadShaderCache();
 
   if (g_ActiveConfig.CanPrecompileUberShaders())
-    PrecompileUberShaders();
+    QueueUberShaderCompiles();
 }
 
 // ONLY to be used during shutdown.
@@ -757,7 +757,7 @@ bool PixelShaderCache::InsertShader(const UberShader::PixelShaderUid& uid,
   return (shader != nullptr);
 }
 
-void PixelShaderCache::PrecompileUberShaders()
+void PixelShaderCache::QueueUberShaderCompiles()
 {
   UberShader::EnumeratePixelShaderUids([&](const UberShader::PixelShaderUid& uid) {
     if (UberPixelShaders.find(uid) != UberPixelShaders.end())

Source/Core/VideoBackends/D3D/PixelShaderCache.h

@@ -28,7 +28,7 @@ public:
   static bool InsertByteCode(const UberShader::PixelShaderUid& uid, const u8* data, size_t len);
   static bool InsertShader(const PixelShaderUid& uid, ID3D11PixelShader* shader);
   static bool InsertShader(const UberShader::PixelShaderUid& uid, ID3D11PixelShader* shader);
-  static void PrecompileUberShaders();
+  static void QueueUberShaderCompiles();
 
   static ID3D11Buffer* GetConstantBuffer();
 

Source/Core/VideoBackends/D3D/VertexShaderCache.cpp

@@ -174,11 +174,12 @@ void VertexShaderCache::Init()
     LoadShaderCache();
 
   g_async_compiler = std::make_unique<VideoCommon::AsyncShaderCompiler>();
-  if (g_ActiveConfig.GetShaderCompilerThreads() > 0)
-    g_async_compiler->StartWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
+  g_async_compiler->ResizeWorkerThreads(g_ActiveConfig.CanPrecompileUberShaders() ?
+                                            g_ActiveConfig.GetShaderPrecompilerThreads() :
+                                            g_ActiveConfig.GetShaderCompilerThreads());
 
   if (g_ActiveConfig.CanPrecompileUberShaders())
-    PrecompileUberShaders();
+    QueueUberShaderCompiles();
 }
 
 void VertexShaderCache::LoadShaderCache()
@@ -206,7 +207,7 @@ void VertexShaderCache::Reload()
     LoadShaderCache();
 
   if (g_ActiveConfig.CanPrecompileUberShaders())
-    PrecompileUberShaders();
+    QueueUberShaderCompiles();
 }
 
 void VertexShaderCache::Clear()
@@ -426,7 +427,7 @@ void VertexShaderCache::RetreiveAsyncShaders()
   g_async_compiler->RetrieveWorkItems();
 }
 
-void VertexShaderCache::PrecompileUberShaders()
+void VertexShaderCache::QueueUberShaderCompiles()
 {
   UberShader::EnumerateVertexShaderUids([&](const UberShader::VertexShaderUid& uid) {
     if (ubervshaders.find(uid) != ubervshaders.end())
@@ -435,13 +436,19 @@ void VertexShaderCache::PrecompileUberShaders()
     g_async_compiler->QueueWorkItem(
         g_async_compiler->CreateWorkItem<UberVertexShaderCompilerWorkItem>(uid));
   });
+}
 
+void VertexShaderCache::WaitForBackgroundCompilesToComplete()
+{
   g_async_compiler->WaitUntilCompletion([](size_t completed, size_t total) {
     Host_UpdateProgressDialog(GetStringT("Compiling shaders...").c_str(),
                               static_cast<int>(completed), static_cast<int>(total));
   });
   g_async_compiler->RetrieveWorkItems();
   Host_UpdateProgressDialog("", -1, -1);
+
+  // Switch from precompile -> runtime compiler threads.
+  g_async_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
 }
 
 VertexShaderCache::VertexShaderCompilerWorkItem::VertexShaderCompilerWorkItem(

Source/Core/VideoBackends/D3D/VertexShaderCache.h

@@ -27,7 +27,8 @@ public:
   static bool SetShader(D3DVertexFormat* vertex_format);
   static bool SetUberShader(D3DVertexFormat* vertex_format);
   static void RetreiveAsyncShaders();
-  static void PrecompileUberShaders();
+  static void QueueUberShaderCompiles();
+  static void WaitForBackgroundCompilesToComplete();
 
   static ID3D11Buffer*& GetConstantBuffer();
 

Source/Core/VideoBackends/D3D/main.cpp

@@ -161,6 +161,7 @@ void VideoBackend::Video_Prepare()
   VertexShaderCache::Init();
   PixelShaderCache::Init();
   GeometryShaderCache::Init();
+  VertexShaderCache::WaitForBackgroundCompilesToComplete();
   D3D::InitUtils();
   BBox::Init();
 }

Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp

@@ -607,17 +607,8 @@ void ProgramShaderCache::Init()
   // on the main thread the first time they are used, causing stutter. Nouveau has been
   // reported to crash if draw calls are invoked on the shared context threads. For now,
   // disable asynchronous compilation on Mesa.
-  if (!DriverDetails::HasBug(DriverDetails::BUG_SHARED_CONTEXT_SHADER_COMPILATION) &&
-      g_ActiveConfig.GetShaderCompilerThreads() > 0)
-  {
+  if (!DriverDetails::HasBug(DriverDetails::BUG_SHARED_CONTEXT_SHADER_COMPILATION))
     s_async_compiler = std::make_unique<SharedContextAsyncShaderCompiler>();
-    s_async_compiler->StartWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
-    if (!s_async_compiler->HasWorkerThreads())
-    {
-      // No point using the async compiler without workers.
-      s_async_compiler.reset();
-    }
-  }
 
   // Read our shader cache, only if supported and enabled
   if (g_ogl_config.bSupportsGLSLCache && g_ActiveConfig.bShaderCache)
@@ -630,7 +621,19 @@ void ProgramShaderCache::Init()
   last_uber_entry = nullptr;
 
   if (g_ActiveConfig.CanPrecompileUberShaders())
+  {
+    if (s_async_compiler)
+      s_async_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderPrecompilerThreads());
     PrecompileUberShaders();
+  }
+
+  if (s_async_compiler)
+  {
+    // No point using the async compiler without workers.
+    s_async_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
+    if (!s_async_compiler->HasWorkerThreads())
+      s_async_compiler.reset();
+  }
 }
 
 void ProgramShaderCache::RetrieveAsyncShaders()

Source/Core/VideoBackends/Vulkan/ShaderCache.cpp

@@ -63,9 +63,9 @@ bool ShaderCache::Initialize()
     return false;
 
   m_async_shader_compiler = std::make_unique<VideoCommon::AsyncShaderCompiler>();
-  if (g_ActiveConfig.GetShaderCompilerThreads() > 0)
-    m_async_shader_compiler->StartWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
-
+  m_async_shader_compiler->ResizeWorkerThreads(g_ActiveConfig.CanPrecompileUberShaders() ?
+                                                   g_ActiveConfig.GetShaderPrecompilerThreads() :
+                                                   g_ActiveConfig.GetShaderCompilerThreads());
   return true;
 }
 
@@ -1236,6 +1236,9 @@ void ShaderCache::PrecompileUberShaders()
   });
 
   WaitForBackgroundCompilesToComplete();
+
+  // Switch to the runtime/background thread config.
+  m_async_shader_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
 }
 
 void ShaderCache::WaitForBackgroundCompilesToComplete()

Source/Core/VideoCommon/AsyncShaderCompiler.cpp

@@ -106,8 +106,11 @@ void AsyncShaderCompiler::WaitUntilCompletion(
   }
 }
 
-void AsyncShaderCompiler::StartWorkerThreads(u32 num_worker_threads)
+bool AsyncShaderCompiler::StartWorkerThreads(u32 num_worker_threads)
 {
+  if (num_worker_threads == 0)
+    return true;
+
   for (u32 i = 0; i < num_worker_threads; i++)
   {
     void* thread_param = nullptr;
@@ -131,6 +134,17 @@ void AsyncShaderCompiler::StartWorkerThreads(u32 num_worker_threads)
 
     m_worker_threads.push_back(std::move(thr));
   }
+
+  return HasWorkerThreads();
+}
+
+bool AsyncShaderCompiler::ResizeWorkerThreads(u32 num_worker_threads)
+{
+  if (m_worker_threads.size() == num_worker_threads)
+    return true;
+
+  StopWorkerThreads();
+  return StartWorkerThreads(num_worker_threads);
 }
 
 bool AsyncShaderCompiler::HasWorkerThreads() const
@@ -140,6 +154,9 @@ bool AsyncShaderCompiler::HasWorkerThreads() const
 
 void AsyncShaderCompiler::StopWorkerThreads()
 {
+  if (!HasWorkerThreads())
+    return;
+
   // Signal worker threads to stop, and wake all of them.
   {
     std::lock_guard<std::mutex> guard(m_pending_work_lock);
@@ -151,6 +168,7 @@ void AsyncShaderCompiler::StopWorkerThreads()
   for (std::thread& thr : m_worker_threads)
     thr.join();
   m_worker_threads.clear();
+  m_exit_flag.Clear();
 }
 
 bool AsyncShaderCompiler::WorkerThreadInitMainThread(void** param)

Source/Core/VideoCommon/AsyncShaderCompiler.h

@@ -52,7 +52,8 @@ public:
   void WaitUntilCompletion(const std::function<void(size_t, size_t)>& progress_callback);
 
   // Needed because of calling virtual methods in shutdown procedure.
-  void StartWorkerThreads(u32 num_worker_threads);
+  bool StartWorkerThreads(u32 num_worker_threads);
+  bool ResizeWorkerThreads(u32 num_worker_threads);
   bool HasWorkerThreads() const;
   void StopWorkerThreads();
 

Source/Core/VideoCommon/VideoConfig.cpp

@@ -98,6 +98,7 @@ void VideoConfig::Refresh()
   bDisableSpecializedShaders = Config::Get(Config::GFX_DISABLE_SPECIALIZED_SHADERS);
   bPrecompileUberShaders = Config::Get(Config::GFX_PRECOMPILE_UBER_SHADERS);
   iShaderCompilerThreads = Config::Get(Config::GFX_SHADER_COMPILER_THREADS);
+  iShaderPrecompilerThreads = Config::Get(Config::GFX_SHADER_PRECOMPILER_THREADS);
   bForceVertexUberShaders = Config::Get(Config::GFX_FORCE_VERTEX_UBER_SHADERS);
   bForcePixelUberShaders = Config::Get(Config::GFX_FORCE_PIXEL_UBER_SHADERS);
 
@@ -196,13 +197,26 @@ bool VideoConfig::IsVSync()
   return bVSync && !Core::GetIsThrottlerTempDisabled();
 }
 
+static u32 GetNumAutoShaderCompilerThreads()
+{
+  // Automatic number. We use clamp(cpus - 3, 1, 4).
+  return static_cast<u32>(std::min(std::max(cpu_info.num_cores - 3, 1), 4));
+}
+
 u32 VideoConfig::GetShaderCompilerThreads() const
 {
   if (iShaderCompilerThreads >= 0)
     return static_cast<u32>(iShaderCompilerThreads);
+  else
+    return GetNumAutoShaderCompilerThreads();
+}
 
-  // Automatic number. We use clamp(cpus - 3, 1, 4).
-  return static_cast<u32>(std::min(std::max(cpu_info.num_cores - 3, 1), 4));
+u32 VideoConfig::GetShaderPrecompilerThreads() const
+{
+  if (iShaderPrecompilerThreads >= 0)
+    return static_cast<u32>(iShaderPrecompilerThreads);
+  else
+    return GetNumAutoShaderCompilerThreads();
 }
 
 bool VideoConfig::CanUseUberShaders() const

Source/Core/VideoCommon/VideoConfig.h

@@ -192,6 +192,7 @@ struct VideoConfig final
   // 0 disables background compilation.
   // -1 uses an automatic number based on the CPU threads.
   int iShaderCompilerThreads;
+  int iShaderPrecompilerThreads;
 
   // Temporary toggling of ubershaders, for debugging
   bool bForceVertexUberShaders;
@@ -256,6 +257,7 @@ struct VideoConfig final
   }
   bool UseVertexRounding() const { return bVertexRounding && iEFBScale != SCALE_1X; }
   u32 GetShaderCompilerThreads() const;
+  u32 GetShaderPrecompilerThreads() const;
   bool CanUseUberShaders() const;
   bool CanPrecompileUberShaders() const;
   bool CanBackgroundCompileShaders() const;