OGL: Re-implement async shader compiling

Source/Core/VideoCommon/DriverDetails.cpp

@@ -102,8 +102,6 @@ static BugInfo m_known_bugs[] = {
      true},
     {API_OPENGL, OS_LINUX, VENDOR_MESA, DRIVER_I965, Family::UNKNOWN,
      BUG_SHARED_CONTEXT_SHADER_COMPILATION, -1.0, -1.0, true},
-    {API_OPENGL, OS_LINUX, VENDOR_MESA, DRIVER_NOUVEAU, Family::UNKNOWN,
-     BUG_SHARED_CONTEXT_SHADER_COMPILATION, -1.0, -1.0, true},
     {API_VULKAN, OS_ALL, VENDOR_NVIDIA, DRIVER_NVIDIA, Family::UNKNOWN, BUG_BROKEN_MSAA_CLEAR, -1.0,
      -1.0, true},
     {API_VULKAN, OS_ALL, VENDOR_IMGTEC, DRIVER_IMGTEC, Family::UNKNOWN,

Source/Core/VideoCommon/DriverDetails.h

@@ -252,8 +252,10 @@ enum Bug
   // the negated value to a temporary variable then using that in the bitwise op.
   BUG_BROKEN_BITWISE_OP_NEGATION,
 
-  // Bug: Shaders are recompiled on the main thread after being previously compiled on
-  // a worker thread on Mesa i965.
+  // BUG: The GPU shader code appears to be context-specific on Mesa/i965.
+  // This means that if we compiled the ubershaders asynchronously, they will be recompiled
+  // on the main thread the first time they are used, causing stutter. For now, disable
+  // asynchronous compilation on Mesa i965.
   // Started version: -1
   // Ended Version: -1
   BUG_SHARED_CONTEXT_SHADER_COMPILATION,

Source/Core/VideoCommon/RenderBase.cpp

@@ -1027,3 +1027,8 @@ bool Renderer::UseVertexDepthRange() const
   // in the vertex shader.
   return fabs(xfmem.viewport.zRange) > 16777215.0f || fabs(xfmem.viewport.farZ) > 16777215.0f;
 }
+
+std::unique_ptr<VideoCommon::AsyncShaderCompiler> Renderer::CreateAsyncShaderCompiler()
+{
+  return std::make_unique<VideoCommon::AsyncShaderCompiler>();
+}

Source/Core/VideoCommon/RenderBase.h

@@ -28,6 +28,7 @@
 #include "Common/Flag.h"
 #include "Common/MathUtil.h"
 #include "VideoCommon/AVIDump.h"
+#include "VideoCommon/AsyncShaderCompiler.h"
 #include "VideoCommon/BPMemory.h"
 #include "VideoCommon/FPSCounter.h"
 #include "VideoCommon/RenderState.h"
@@ -189,6 +190,8 @@ public:
   void ResizeSurface(int new_width, int new_height);
   bool UseVertexDepthRange() const;
 
+  virtual std::unique_ptr<VideoCommon::AsyncShaderCompiler> CreateAsyncShaderCompiler();
+
   virtual void Shutdown();
 
   // Drawing utility shaders.

Source/Core/VideoCommon/ShaderCache.cpp

@@ -27,7 +27,7 @@ bool ShaderCache::Initialize()
   m_efb_multisamples = g_ActiveConfig.iMultisamples;
 
   // Create the async compiler, and start the worker threads.
-  m_async_shader_compiler = std::make_unique<VideoCommon::AsyncShaderCompiler>();
+  m_async_shader_compiler = g_renderer->CreateAsyncShaderCompiler();
   m_async_shader_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderPrecompilerThreads());
 
   // Load shader and UID caches.

Source/Core/VideoCommon/VideoConfig.cpp

@@ -187,8 +187,7 @@ static u32 GetNumAutoShaderCompilerThreads()
 
 u32 VideoConfig::GetShaderCompilerThreads() const
 {
-  // videocommon shader cache is currently broken on OGL, needs multiple contexts.
-  if (backend_info.api_type == APIType::OpenGL)
+  if (!backend_info.bSupportsBackgroundCompiling)
     return 0;
 
   if (iShaderCompilerThreads >= 0)
@@ -199,8 +198,7 @@ u32 VideoConfig::GetShaderCompilerThreads() const
 
 u32 VideoConfig::GetShaderPrecompilerThreads() const
 {
-  // videocommon shader cache is currently broken on OGL, needs multiple contexts.
-  if (backend_info.api_type == APIType::OpenGL)
+  if (!backend_info.bSupportsBackgroundCompiling)
     return 0;
 
   if (iShaderPrecompilerThreads >= 0)

Source/Core/VideoCommon/VideoConfig.h

@@ -227,6 +227,7 @@ struct VideoConfig final
     bool bSupportsDynamicSamplerIndexing;  // Needed by UberShaders, so must stay in VideoCommon
     bool bSupportsBPTCTextures;
     bool bSupportsFramebufferFetch;  // Used as an alternative to dual-source blend on GLES
+    bool bSupportsBackgroundCompiling;
   } backend_info;
 
   // Utility