ShaderCache: Implement background shader compilation

This enables shaders to be compiled while the game is starting, instead
of blocking startup. If a shader is needed before it is compiled,
emulation will block.

15 files changed, 202 insertions, 145 deletions

diff --git a/Source/Core/Core/Config/GraphicsSettings.cpp b/Source/Core/Core/Config/GraphicsSettings.cpp
index b06ccdaf41..10a70c6020 100644
--- a/Source/Core/Core/Config/GraphicsSettings.cpp
+++ b/Source/Core/Core/Config/GraphicsSettings.cpp
@@ -76,6 +76,8 @@ const ConfigInfo<bool> GFX_BACKEND_MULTITHREADING{
 const ConfigInfo<int> GFX_COMMAND_BUFFER_EXECUTE_INTERVAL{
     {System::GFX, "Settings", "CommandBufferExecuteInterval"}, 100};
 const ConfigInfo<bool> GFX_SHADER_CACHE{{System::GFX, "Settings", "ShaderCache"}, true};
+const ConfigInfo<bool> GFX_WAIT_FOR_SHADERS_BEFORE_STARTING{
+    {System::GFX, "Settings", "WaitForShadersBeforeStarting"}, false};
 const ConfigInfo<int> GFX_UBERSHADER_MODE{{System::GFX, "Settings", "UberShaderMode"},
                                           static_cast<int>(UberShaderMode::Disabled)};
 const ConfigInfo<int> GFX_SHADER_COMPILER_THREADS{
diff --git a/Source/Core/Core/Config/GraphicsSettings.h b/Source/Core/Core/Config/GraphicsSettings.h
index 5f07d1906a..5c1684777b 100644
--- a/Source/Core/Core/Config/GraphicsSettings.h
+++ b/Source/Core/Core/Config/GraphicsSettings.h
@@ -59,6 +59,7 @@ extern const ConfigInfo<bool> GFX_ENABLE_VALIDATION_LAYER;
 extern const ConfigInfo<bool> GFX_BACKEND_MULTITHREADING;
 extern const ConfigInfo<int> GFX_COMMAND_BUFFER_EXECUTE_INTERVAL;
 extern const ConfigInfo<bool> GFX_SHADER_CACHE;
+extern const ConfigInfo<bool> GFX_WAIT_FOR_SHADERS_BEFORE_STARTING;
 extern const ConfigInfo<int> GFX_UBERSHADER_MODE;
 extern const ConfigInfo<int> GFX_SHADER_COMPILER_THREADS;
 extern const ConfigInfo<int> GFX_SHADER_PRECOMPILER_THREADS;
diff --git a/Source/Core/Core/ConfigLoaders/IsSettingSaveable.cpp b/Source/Core/Core/ConfigLoaders/IsSettingSaveable.cpp
index abc30b24d3..ef71c085ab 100644
--- a/Source/Core/Core/ConfigLoaders/IsSettingSaveable.cpp
+++ b/Source/Core/Core/ConfigLoaders/IsSettingSaveable.cpp
@@ -46,8 +46,8 @@ bool IsSettingSaveable(const Config::ConfigLocation& config_location)
       Config::GFX_DISABLE_FOG.location, Config::GFX_BORDERLESS_FULLSCREEN.location,
       Config::GFX_ENABLE_VALIDATION_LAYER.location, Config::GFX_BACKEND_MULTITHREADING.location,
       Config::GFX_COMMAND_BUFFER_EXECUTE_INTERVAL.location, Config::GFX_SHADER_CACHE.location,
-      Config::GFX_UBERSHADER_MODE.location, Config::GFX_SHADER_COMPILER_THREADS.location,
-      Config::GFX_SHADER_PRECOMPILER_THREADS.location,
+      Config::GFX_WAIT_FOR_SHADERS_BEFORE_STARTING.location, Config::GFX_UBERSHADER_MODE.location,
+      Config::GFX_SHADER_COMPILER_THREADS.location, Config::GFX_SHADER_PRECOMPILER_THREADS.location,
 
       Config::GFX_SW_ZCOMPLOC.location, Config::GFX_SW_ZFREEZE.location,
       Config::GFX_SW_DUMP_OBJECTS.location, Config::GFX_SW_DUMP_TEV_STAGES.location,
diff --git a/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.cpp b/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.cpp
index e76fc6a35c..21281488b8 100644
--- a/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.cpp
+++ b/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.cpp
@@ -87,6 +87,8 @@ void GeneralWidget::CreateWidgets()
   m_keep_window_top = new QCheckBox(tr("Keep Window on Top"));
   m_hide_cursor = new QCheckBox(tr("Hide Mouse Cursor"));
   m_render_main_window = new QCheckBox(tr("Render to Main Window"));
+  m_wait_for_shaders = new GraphicsBool(tr("Immediately Compile Shaders"),
+                                        Config::GFX_WAIT_FOR_SHADERS_BEFORE_STARTING);
 
   m_options_box->setLayout(m_options_layout);
 
@@ -101,6 +103,7 @@ void GeneralWidget::CreateWidgets()
 
   m_options_layout->addWidget(m_hide_cursor, 3, 0);
   m_options_layout->addWidget(m_render_main_window, 3, 1);
+  m_options_layout->addWidget(m_wait_for_shaders, 4, 0);
 
   main_layout->addWidget(m_video_box);
   main_layout->addWidget(m_options_box);
@@ -265,6 +268,12 @@ void GeneralWidget::AddDescriptions()
   static const char* TR_SHOW_NETPLAY_MESSAGES_DESCRIPTION =
       QT_TR_NOOP("When playing on NetPlay, show chat messages, buffer changes and "
                  "desync alerts.\n\nIf unsure, leave this unchecked.");
+  static const char* TR_WAIT_FOR_SHADERS_DESCRIPTION = QT_TR_NOOP(
+      "Waits for all shaders to finish compiling before starting a game. Enabling this "
+      "option may reduce stuttering or hitching for a short time after the game is "
+      "started, at the cost of a longer delay before the game starts.\n\nFor systems "
+      "with two or fewer cores, it is recommended to enable this option, as a large "
+      "shader queue may reduce frame rates. Otherwise, if unsure, leave this unchecked.");
 
   AddDescription(m_backend_combo, TR_BACKEND_DESCRIPTION);
 #ifdef _WIN32
@@ -282,6 +291,7 @@ void GeneralWidget::AddDescriptions()
   AddDescription(m_show_messages, TR_SHOW_FPS_DESCRIPTION);
   AddDescription(m_keep_window_top, TR_KEEP_WINDOW_ON_TOP_DESCRIPTION);
   AddDescription(m_show_messages, TR_SHOW_NETPLAY_MESSAGES_DESCRIPTION);
+  AddDescription(m_wait_for_shaders, TR_WAIT_FOR_SHADERS_DESCRIPTION);
 }
 void GeneralWidget::OnBackendChanged(const QString& backend_name)
 {
diff --git a/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.h b/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.h
index e611950dec..8b2d22f331 100644
--- a/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.h
+++ b/Source/Core/DolphinQt2/Config/Graphics/GeneralWidget.h
@@ -52,6 +52,7 @@ private:
   QCheckBox* m_keep_window_top;
   QCheckBox* m_hide_cursor;
   QCheckBox* m_render_main_window;
+  QCheckBox* m_wait_for_shaders;
 
   X11Utils::XRRConfiguration* m_xrr_config;
 };
diff --git a/Source/Core/DolphinWX/VideoConfigDiag.cpp b/Source/Core/DolphinWX/VideoConfigDiag.cpp
index 874508459e..f1a57f8b8c 100644
--- a/Source/Core/DolphinWX/VideoConfigDiag.cpp
+++ b/Source/Core/DolphinWX/VideoConfigDiag.cpp
@@ -317,6 +317,12 @@ static wxString ubershader_desc =
                 "stuttering. Balances performance and smoothness.\n\n"
                 "Exclusive: Ubershaders will always be used. Only recommended for high-end "
                 "systems.");
+static wxString wait_for_shaders_desc =
+    wxTRANSLATE("Waits for all shaders to finish compiling before starting a game. Enabling this "
+                "option may reduce stuttering or hitching for a short time after the game is "
+                "started, at the cost of a longer delay before the game starts.\n\nFor systems "
+                "with two or fewer cores, it is recommended to enable this option, as a large "
+                "shader queue may reduce frame rates. Otherwise, if unsure, leave this unchecked.");
 
 VideoConfigDiag::VideoConfigDiag(wxWindow* parent, const std::string& title)
     : wxDialog(parent, wxID_ANY, wxString::Format(_("Dolphin %s Graphics Configuration"),
@@ -442,6 +448,10 @@ VideoConfigDiag::VideoConfigDiag(wxWindow* parent, const std::string& title)
                                         wxGetTranslation(backend_multithreading_desc),
                                         Config::GFX_BACKEND_MULTITHREADING));
         }
+
+        szr_other->Add(CreateCheckBox(page_general, _("Immediately Compile Shaders"),
+                                      wxGetTranslation(wait_for_shaders_desc),
+                                      Config::GFX_WAIT_FOR_SHADERS_BEFORE_STARTING));
       }
 
       wxStaticBoxSizer* const group_basic =
diff --git a/Source/Core/VideoCommon/AsyncShaderCompiler.cpp b/Source/Core/VideoCommon/AsyncShaderCompiler.cpp
index 59ef9762f1..e605828ed5 100644
--- a/Source/Core/VideoCommon/AsyncShaderCompiler.cpp
+++ b/Source/Core/VideoCommon/AsyncShaderCompiler.cpp
@@ -18,7 +18,6 @@ AsyncShaderCompiler::~AsyncShaderCompiler()
   // Pending work can be left at shutdown.
   // The work item classes are expected to clean up after themselves.
   _assert_(!HasWorkerThreads());
-  _assert_(m_completed_work.empty());
 }
 
 void AsyncShaderCompiler::QueueWorkItem(WorkItemPtr item)
diff --git a/Source/Core/VideoCommon/GXPipelineTypes.h b/Source/Core/VideoCommon/GXPipelineTypes.h
new file mode 100644
index 0000000000..165e9c67f8
--- /dev/null
+++ b/Source/Core/VideoCommon/GXPipelineTypes.h
@@ -0,0 +1,75 @@
+// Copyright 2018 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "VideoCommon/GeometryShaderGen.h"
+#include "VideoCommon/PixelShaderGen.h"
+#include "VideoCommon/RenderState.h"
+#include "VideoCommon/UberShaderPixel.h"
+#include "VideoCommon/UberShaderVertex.h"
+#include "VideoCommon/VertexShaderGen.h"
+
+class NativeVertexFormat;
+
+namespace VideoCommon
+{
+struct GXPipelineUid
+{
+  const NativeVertexFormat* vertex_format;
+  VertexShaderUid vs_uid;
+  GeometryShaderUid gs_uid;
+  PixelShaderUid ps_uid;
+  RasterizationState rasterization_state;
+  DepthState depth_state;
+  BlendingState blending_state;
+
+  // We use memcmp() for comparing pipelines as std::tie generates a large number of instructions,
+  // and this map lookup can happen every draw call. However, as using memcmp() will also compare
+  // any padding bytes, we have to ensure these are zeroed out.
+  GXPipelineUid() { std::memset(this, 0, sizeof(*this)); }
+  GXPipelineUid(const GXPipelineUid& rhs) { std::memcpy(this, &rhs, sizeof(*this)); }
+  GXPipelineUid& operator=(const GXPipelineUid& rhs)
+  {
+    std::memcpy(this, &rhs, sizeof(*this));
+    return *this;
+  }
+  bool operator<(const GXPipelineUid& rhs) const
+  {
+    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
+  }
+  bool operator==(const GXPipelineUid& rhs) const
+  {
+    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
+  }
+  bool operator!=(const GXPipelineUid& rhs) const { return !operator==(rhs); }
+};
+struct GXUberPipelineUid
+{
+  const NativeVertexFormat* vertex_format;
+  UberShader::VertexShaderUid vs_uid;
+  GeometryShaderUid gs_uid;
+  UberShader::PixelShaderUid ps_uid;
+  RasterizationState rasterization_state;
+  DepthState depth_state;
+  BlendingState blending_state;
+
+  GXUberPipelineUid() { std::memset(this, 0, sizeof(*this)); }
+  GXUberPipelineUid(const GXUberPipelineUid& rhs) { std::memcpy(this, &rhs, sizeof(*this)); }
+  GXUberPipelineUid& operator=(const GXUberPipelineUid& rhs)
+  {
+    std::memcpy(this, &rhs, sizeof(*this));
+    return *this;
+  }
+  bool operator<(const GXUberPipelineUid& rhs) const
+  {
+    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
+  }
+  bool operator==(const GXUberPipelineUid& rhs) const
+  {
+    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
+  }
+  bool operator!=(const GXUberPipelineUid& rhs) const { return !operator==(rhs); }
+};
+}  // namespace VideoCommon
diff --git a/Source/Core/VideoCommon/ShaderCache.cpp b/Source/Core/VideoCommon/ShaderCache.cpp
index 43a31dfd6b..512adbd12a 100644
--- a/Source/Core/VideoCommon/ShaderCache.cpp
+++ b/Source/Core/VideoCommon/ShaderCache.cpp
@@ -43,6 +43,8 @@ bool ShaderCache::Initialize()
 
   // Compile all known UIDs.
   CompileMissingPipelines();
+  if (g_ActiveConfig.bWaitForShadersBeforeStarting)
+    WaitForAsyncCompiler();
 
   // Switch to the runtime shader compiler thread configuration.
   m_async_shader_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
@@ -61,9 +63,7 @@ void ShaderCache::SetHostConfig(const ShaderHostConfig& host_config, u32 efb_mul
 
 void ShaderCache::Reload()
 {
-  m_async_shader_compiler->WaitUntilCompletion();
-  m_async_shader_compiler->RetrieveWorkItems();
-
+  WaitForAsyncCompiler();
   InvalidateCachedPipelines();
   ClearShaderCaches();
 
@@ -77,6 +77,8 @@ void ShaderCache::Reload()
   // UIDs are still be in the map. Therefore, when these are rebuilt, the shaders will also
   // be recompiled.
   CompileMissingPipelines();
+  if (g_ActiveConfig.bWaitForShadersBeforeStarting)
+    WaitForAsyncCompiler();
   m_async_shader_compiler->ResizeWorkerThreads(g_ActiveConfig.GetShaderCompilerThreads());
 }
 
@@ -87,14 +89,14 @@ void ShaderCache::RetrieveAsyncShaders()
 
 void ShaderCache::Shutdown()
 {
+  // This may leave shaders uncommitted to the cache, but it's better than blocking shutdown
+  // until everything has finished compiling.
   m_async_shader_compiler->StopWorkerThreads();
-  m_async_shader_compiler->RetrieveWorkItems();
-
   ClearShaderCaches();
   ClearPipelineCaches();
 }
 
-const AbstractPipeline* ShaderCache::GetPipelineForUid(const GXPipelineConfig& uid)
+const AbstractPipeline* ShaderCache::GetPipelineForUid(const GXPipelineUid& uid)
 {
   auto it = m_gx_pipeline_cache.find(uid);
   if (it != m_gx_pipeline_cache.end() && !it->second.second)
@@ -109,8 +111,7 @@ const AbstractPipeline* ShaderCache::GetPipelineForUid(const GXPipelineConfig& u
   return InsertGXPipeline(uid, std::move(pipeline));
 }
 
-std::optional<const AbstractPipeline*>
-ShaderCache::GetPipelineForUidAsync(const GXPipelineConfig& uid)
+std::optional<const AbstractPipeline*> ShaderCache::GetPipelineForUidAsync(const GXPipelineUid& uid)
 {
   auto it = m_gx_pipeline_cache.find(uid);
   if (it != m_gx_pipeline_cache.end())
@@ -159,7 +160,7 @@ ShaderCache::GetPipelineForUidAsync(const GXPipelineConfig& uid)
   return {};
 }
 
-const AbstractPipeline* ShaderCache::GetUberPipelineForUid(const GXUberPipelineConfig& uid)
+const AbstractPipeline* ShaderCache::GetUberPipelineForUid(const GXUberPipelineUid& uid)
 {
   auto it = m_gx_uber_pipeline_cache.find(uid);
   if (it != m_gx_uber_pipeline_cache.end() && !it->second.second)
@@ -172,12 +173,16 @@ const AbstractPipeline* ShaderCache::GetUberPipelineForUid(const GXUberPipelineC
   return InsertGXUberPipeline(uid, std::move(pipeline));
 }
 
-void ShaderCache::WaitForAsyncCompiler(const std::string& msg)
+void ShaderCache::WaitForAsyncCompiler()
 {
-  m_async_shader_compiler->WaitUntilCompletion([&msg](size_t completed, size_t total) {
-    Host_UpdateProgressDialog(msg.c_str(), static_cast<int>(completed), static_cast<int>(total));
-  });
-  m_async_shader_compiler->RetrieveWorkItems();
+  while (m_async_shader_compiler->HasPendingWork())
+  {
+    m_async_shader_compiler->WaitUntilCompletion([](size_t completed, size_t total) {
+      Host_UpdateProgressDialog(GetStringT("Compiling shaders...").c_str(),
+                                static_cast<int>(completed), static_cast<int>(total));
+    });
+    m_async_shader_compiler->RetrieveWorkItems();
+  }
   Host_UpdateProgressDialog("", -1, -1);
 }
 
@@ -274,7 +279,7 @@ void ShaderCache::LoadPipelineUIDCache()
     CacheReader(ShaderCache* shader_cache_) : shader_cache(shader_cache_) {}
     void Read(const GXPipelineDiskCacheUid& key, const u8* data, u32 data_size)
     {
-      GXPipelineConfig config = {};
+      GXPipelineUid config = {};
       config.vertex_format = VertexLoaderManager::GetOrCreateMatchingFormat(key.vertex_decl);
       config.vs_uid = key.vs_uid;
       config.gs_uid = key.gs_uid;
@@ -315,8 +320,6 @@ void ShaderCache::CompileMissingPipelines()
     if (!it.second.second)
       QueueUberPipelineCompile(it.first);
   }
-
-  WaitForAsyncCompiler(GetStringT("Compiling shaders..."));
 }
 
 void ShaderCache::InvalidateCachedPipelines()
@@ -508,8 +511,7 @@ AbstractPipelineConfig ShaderCache::GetGXPipelineConfig(
   return config;
 }
 
-std::optional<AbstractPipelineConfig>
-ShaderCache::GetGXPipelineConfig(const GXPipelineConfig& config)
+std::optional<AbstractPipelineConfig> ShaderCache::GetGXPipelineConfig(const GXPipelineUid& config)
 {
   const AbstractShader* vs;
   auto vs_iter = m_vs_cache.shader_map.find(config.vs_uid);
@@ -545,7 +547,7 @@ ShaderCache::GetGXPipelineConfig(const GXPipelineConfig& config)
 }
 
 std::optional<AbstractPipelineConfig>
-ShaderCache::GetGXUberPipelineConfig(const GXUberPipelineConfig& config)
+ShaderCache::GetGXUberPipelineConfig(const GXUberPipelineUid& config)
 {
   const AbstractShader* vs;
   auto vs_iter = m_uber_vs_cache.shader_map.find(config.vs_uid);
@@ -580,7 +582,7 @@ ShaderCache::GetGXUberPipelineConfig(const GXUberPipelineConfig& config)
                              config.depth_state, config.blending_state);
 }
 
-const AbstractPipeline* ShaderCache::InsertGXPipeline(const GXPipelineConfig& config,
+const AbstractPipeline* ShaderCache::InsertGXPipeline(const GXPipelineUid& config,
                                                       std::unique_ptr<AbstractPipeline> pipeline)
 {
   auto& entry = m_gx_pipeline_cache[config];
@@ -592,7 +594,7 @@ const AbstractPipeline* ShaderCache::InsertGXPipeline(const GXPipelineConfig& co
 }
 
 const AbstractPipeline*
-ShaderCache::InsertGXUberPipeline(const GXUberPipelineConfig& config,
+ShaderCache::InsertGXUberPipeline(const GXUberPipelineUid& config,
                                   std::unique_ptr<AbstractPipeline> pipeline)
 {
   auto& entry = m_gx_uber_pipeline_cache[config];
@@ -603,7 +605,7 @@ ShaderCache::InsertGXUberPipeline(const GXUberPipelineConfig& config,
   return entry.first.get();
 }
 
-void ShaderCache::AppendGXPipelineUID(const GXPipelineConfig& config)
+void ShaderCache::AppendGXPipelineUID(const GXPipelineUid& config)
 {
   // Convert to disk format.
   GXPipelineDiskCacheUid disk_uid = {};
@@ -633,7 +635,7 @@ void ShaderCache::QueueVertexShaderCompile(const VertexShaderUid& uid)
       return true;
     }
 
-    virtual void Retrieve() override { shader_cache->InsertVertexShader(uid, std::move(shader)); }
+    void Retrieve() override { shader_cache->InsertVertexShader(uid, std::move(shader)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractShader> shader;
@@ -661,11 +663,7 @@ void ShaderCache::QueueVertexUberShaderCompile(const UberShader::VertexShaderUid
       return true;
     }
 
-    virtual void Retrieve() override
-    {
-      shader_cache->InsertVertexUberShader(uid, std::move(shader));
-    }
-
+    void Retrieve() override { shader_cache->InsertVertexUberShader(uid, std::move(shader)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractShader> shader;
@@ -693,7 +691,7 @@ void ShaderCache::QueuePixelShaderCompile(const PixelShaderUid& uid)
       return true;
     }
 
-    virtual void Retrieve() override { shader_cache->InsertPixelShader(uid, std::move(shader)); }
+    void Retrieve() override { shader_cache->InsertPixelShader(uid, std::move(shader)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractShader> shader;
@@ -721,11 +719,7 @@ void ShaderCache::QueuePixelUberShaderCompile(const UberShader::PixelShaderUid&
       return true;
     }
 
-    virtual void Retrieve() override
-    {
-      shader_cache->InsertPixelUberShader(uid, std::move(shader));
-    }
-
+    void Retrieve() override { shader_cache->InsertPixelUberShader(uid, std::move(shader)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractShader> shader;
@@ -737,12 +731,12 @@ void ShaderCache::QueuePixelUberShaderCompile(const UberShader::PixelShaderUid&
   m_async_shader_compiler->QueueWorkItem(std::move(wi));
 }
 
-void ShaderCache::QueuePipelineCompile(const GXPipelineConfig& uid)
+void ShaderCache::QueuePipelineCompile(const GXPipelineUid& uid)
 {
   class PipelineWorkItem final : public AsyncShaderCompiler::WorkItem
   {
   public:
-    PipelineWorkItem(ShaderCache* shader_cache_, const GXPipelineConfig& uid_,
+    PipelineWorkItem(ShaderCache* shader_cache_, const GXPipelineUid& uid_,
                      const AbstractPipelineConfig& config_)
         : shader_cache(shader_cache_), uid(uid_), config(config_)
     {
@@ -754,11 +748,11 @@ void ShaderCache::QueuePipelineCompile(const GXPipelineConfig& uid)
       return true;
     }
 
-    virtual void Retrieve() override { shader_cache->InsertGXPipeline(uid, std::move(pipeline)); }
+    void Retrieve() override { shader_cache->InsertGXPipeline(uid, std::move(pipeline)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractPipeline> pipeline;
-    GXPipelineConfig uid;
+    GXPipelineUid uid;
     AbstractPipelineConfig config;
   };
 
@@ -775,13 +769,15 @@ void ShaderCache::QueuePipelineCompile(const GXPipelineConfig& uid)
   m_gx_pipeline_cache[uid].second = true;
 }
 
-void ShaderCache::QueueUberPipelineCompile(const GXUberPipelineConfig& uid)
+void ShaderCache::QueueUberPipelineCompile(const GXUberPipelineUid& uid)
 {
-  class UberPipelineWorkItem final : public AsyncShaderCompiler::WorkItem
+  // Since the shaders may not be compiled at pipelines request time, we do this in two passes.
+  // This is necessary because we can't access the caches in the worker thread.
+  class UberPipelineCompilePass final : public AsyncShaderCompiler::WorkItem
   {
   public:
-    UberPipelineWorkItem(ShaderCache* shader_cache_, const GXUberPipelineConfig& uid_,
-                         const AbstractPipelineConfig& config_)
+    UberPipelineCompilePass(ShaderCache* shader_cache_, const GXUberPipelineUid& uid_,
+                            const AbstractPipelineConfig& config_)
         : shader_cache(shader_cache_), uid(uid_), config(config_)
     {
     }
@@ -792,27 +788,43 @@ void ShaderCache::QueueUberPipelineCompile(const GXUberPipelineConfig& uid)
       return true;
     }
 
-    virtual void Retrieve() override
-    {
-      shader_cache->InsertGXUberPipeline(uid, std::move(pipeline));
-    }
-
+    void Retrieve() override { shader_cache->InsertGXUberPipeline(uid, std::move(pipeline)); }
   private:
     ShaderCache* shader_cache;
     std::unique_ptr<AbstractPipeline> pipeline;
-    GXUberPipelineConfig uid;
+    GXUberPipelineUid uid;
     AbstractPipelineConfig config;
   };
-
-  auto config = GetGXUberPipelineConfig(uid);
-  if (!config)
+  class UberPipelinePreparePass final : public AsyncShaderCompiler::WorkItem
   {
-    // One or more stages failed to compile.
-    InsertGXUberPipeline(uid, nullptr);
-    return;
-  }
+  public:
+    UberPipelinePreparePass(ShaderCache* shader_cache_, const GXUberPipelineUid& uid_)
+        : shader_cache(shader_cache_), uid(uid_)
+    {
+    }
+
+    bool Compile() override { return true; }
+    void Retrieve() override
+    {
+      auto config = shader_cache->GetGXUberPipelineConfig(uid);
+      if (!config)
+      {
+        // One or more stages failed to compile.
+        shader_cache->InsertGXUberPipeline(uid, nullptr);
+        return;
+      }
+
+      auto wi = shader_cache->m_async_shader_compiler->CreateWorkItem<UberPipelineCompilePass>(
+          shader_cache, uid, *config);
+      shader_cache->m_async_shader_compiler->QueueWorkItem(std::move(wi));
+    }
 
-  auto wi = m_async_shader_compiler->CreateWorkItem<UberPipelineWorkItem>(this, uid, *config);
+  private:
+    ShaderCache* shader_cache;
+    GXUberPipelineUid uid;
+  };
+
+  auto wi = m_async_shader_compiler->CreateWorkItem<UberPipelinePreparePass>(this, uid);
   m_async_shader_compiler->QueueWorkItem(std::move(wi));
   m_gx_uber_pipeline_cache[uid].second = true;
 }
@@ -841,9 +853,6 @@ void ShaderCache::PrecompileUberShaders()
       QueuePixelUberShaderCompile(puid);
   });
 
-  // Wait for shaders to finish compiling.
-  WaitForAsyncCompiler(GetStringT("Compiling uber shaders..."));
-
   // Create a dummy vertex format with no attributes.
   // All attributes will be enabled in GetUberVertexFormat.
   PortableVertexDeclaration dummy_vertex_decl = {};
@@ -856,7 +865,7 @@ void ShaderCache::PrecompileUberShaders()
   auto QueueDummyPipeline = [&](const UberShader::VertexShaderUid& vs_uid,
                                 const GeometryShaderUid& gs_uid,
                                 const UberShader::PixelShaderUid& ps_uid) {
-    GXUberPipelineConfig config;
+    GXUberPipelineUid config;
     config.vertex_format = dummy_vertex_format;
     config.vs_uid = vs_uid;
     config.gs_uid = gs_uid;
diff --git a/Source/Core/VideoCommon/ShaderCache.h b/Source/Core/VideoCommon/ShaderCache.h
index 6516edd8d8..8681df19be 100644
--- a/Source/Core/VideoCommon/ShaderCache.h
+++ b/Source/Core/VideoCommon/ShaderCache.h
@@ -19,10 +19,10 @@
 
 #include "VideoCommon/AbstractPipeline.h"
 #include "VideoCommon/AbstractShader.h"
-#include "VideoCommon/NativeVertexFormat.h"
-
 #include "VideoCommon/AsyncShaderCompiler.h"
+#include "VideoCommon/GXPipelineTypes.h"
 #include "VideoCommon/GeometryShaderGen.h"
+#include "VideoCommon/NativeVertexFormat.h"
 #include "VideoCommon/PixelShaderGen.h"
 #include "VideoCommon/RenderState.h"
 #include "VideoCommon/UberShaderPixel.h"
@@ -33,64 +33,6 @@ class NativeVertexFormat;
 
 namespace VideoCommon
 {
-struct GXPipelineConfig
-{
-  const NativeVertexFormat* vertex_format;
-  VertexShaderUid vs_uid;
-  GeometryShaderUid gs_uid;
-  PixelShaderUid ps_uid;
-  RasterizationState rasterization_state;
-  DepthState depth_state;
-  BlendingState blending_state;
-
-  // We use memcmp() for comparing pipelines as std::tie generates a large number of instructions,
-  // and this map lookup can happen every draw call. However, as using memcmp() will also compare
-  // any padding bytes, we have to ensure these are zeroed out.
-  GXPipelineConfig() { std::memset(this, 0, sizeof(*this)); }
-  GXPipelineConfig(const GXPipelineConfig& rhs) { std::memcpy(this, &rhs, sizeof(*this)); }
-  GXPipelineConfig& operator=(const GXPipelineConfig& rhs)
-  {
-    std::memcpy(this, &rhs, sizeof(*this));
-    return *this;
-  }
-  bool operator<(const GXPipelineConfig& rhs) const
-  {
-    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
-  }
-  bool operator==(const GXPipelineConfig& rhs) const
-  {
-    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
-  }
-  bool operator!=(const GXPipelineConfig& rhs) const { return !operator==(rhs); }
-};
-struct GXUberPipelineConfig
-{
-  const NativeVertexFormat* vertex_format;
-  UberShader::VertexShaderUid vs_uid;
-  GeometryShaderUid gs_uid;
-  UberShader::PixelShaderUid ps_uid;
-  RasterizationState rasterization_state;
-  DepthState depth_state;
-  BlendingState blending_state;
-
-  GXUberPipelineConfig() { std::memset(this, 0, sizeof(*this)); }
-  GXUberPipelineConfig(const GXUberPipelineConfig& rhs) { std::memcpy(this, &rhs, sizeof(*this)); }
-  GXUberPipelineConfig& operator=(const GXUberPipelineConfig& rhs)
-  {
-    std::memcpy(this, &rhs, sizeof(*this));
-    return *this;
-  }
-  bool operator<(const GXUberPipelineConfig& rhs) const
-  {
-    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
-  }
-  bool operator==(const GXUberPipelineConfig& rhs) const
-  {
-    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
-  }
-  bool operator!=(const GXUberPipelineConfig& rhs) const { return !operator==(rhs); }
-};
-
 class ShaderCache final
 {
 public:
@@ -114,15 +56,15 @@ public:
   std::string GetUtilityShaderHeader() const;
 
   // Accesses ShaderGen shader caches
-  const AbstractPipeline* GetPipelineForUid(const GXPipelineConfig& uid);
-  const AbstractPipeline* GetUberPipelineForUid(const GXUberPipelineConfig& uid);
+  const AbstractPipeline* GetPipelineForUid(const GXPipelineUid& uid);
+  const AbstractPipeline* GetUberPipelineForUid(const GXUberPipelineUid& uid);
 
   // Accesses ShaderGen shader caches asynchronously.
   // The optional will be empty if this pipeline is now background compiling.
-  std::optional<const AbstractPipeline*> GetPipelineForUidAsync(const GXPipelineConfig& uid);
+  std::optional<const AbstractPipeline*> GetPipelineForUidAsync(const GXPipelineUid& uid);
 
 private:
-  void WaitForAsyncCompiler(const std::string& msg);
+  void WaitForAsyncCompiler();
   void LoadShaderCaches();
   void ClearShaderCaches();
   void LoadPipelineUIDCache();
@@ -155,21 +97,21 @@ private:
                       const AbstractShader* geometry_shader, const AbstractShader* pixel_shader,
                       const RasterizationState& rasterization_state, const DepthState& depth_state,
                       const BlendingState& blending_state);
-  std::optional<AbstractPipelineConfig> GetGXPipelineConfig(const GXPipelineConfig& uid);
-  std::optional<AbstractPipelineConfig> GetGXUberPipelineConfig(const GXUberPipelineConfig& uid);
-  const AbstractPipeline* InsertGXPipeline(const GXPipelineConfig& config,
+  std::optional<AbstractPipelineConfig> GetGXPipelineConfig(const GXPipelineUid& uid);
+  std::optional<AbstractPipelineConfig> GetGXUberPipelineConfig(const GXUberPipelineUid& uid);
+  const AbstractPipeline* InsertGXPipeline(const GXPipelineUid& config,
                                            std::unique_ptr<AbstractPipeline> pipeline);
-  const AbstractPipeline* InsertGXUberPipeline(const GXUberPipelineConfig& config,
+  const AbstractPipeline* InsertGXUberPipeline(const GXUberPipelineUid& config,
                                                std::unique_ptr<AbstractPipeline> pipeline);
-  void AppendGXPipelineUID(const GXPipelineConfig& config);
+  void AppendGXPipelineUID(const GXPipelineUid& config);
 
   // ASync Compiler Methods
   void QueueVertexShaderCompile(const VertexShaderUid& uid);
   void QueueVertexUberShaderCompile(const UberShader::VertexShaderUid& uid);
   void QueuePixelShaderCompile(const PixelShaderUid& uid);
   void QueuePixelUberShaderCompile(const UberShader::PixelShaderUid& uid);
-  void QueuePipelineCompile(const GXPipelineConfig& uid);
-  void QueueUberPipelineCompile(const GXUberPipelineConfig& uid);
+  void QueuePipelineCompile(const GXPipelineUid& uid);
+  void QueueUberPipelineCompile(const GXUberPipelineUid& uid);
 
   // Configuration bits.
   APIType m_api_type = APIType::Nothing;
@@ -196,10 +138,8 @@ private:
   ShaderModuleCache<UberShader::PixelShaderUid> m_uber_ps_cache;
 
   // GX Pipeline Caches - .first - pipeline, .second - pending
-  // TODO: Use unordered_map for speed.
-  std::map<GXPipelineConfig, std::pair<std::unique_ptr<AbstractPipeline>, bool>>
-      m_gx_pipeline_cache;
-  std::map<GXUberPipelineConfig, std::pair<std::unique_ptr<AbstractPipeline>, bool>>
+  std::map<GXPipelineUid, std::pair<std::unique_ptr<AbstractPipeline>, bool>> m_gx_pipeline_cache;
+  std::map<GXUberPipelineUid, std::pair<std::unique_ptr<AbstractPipeline>, bool>>
       m_gx_uber_pipeline_cache;
 
   // Disk cache of pipeline UIDs
diff --git a/Source/Core/VideoCommon/VertexManagerBase.h b/Source/Core/VideoCommon/VertexManagerBase.h
index eac95f614f..4d07b08411 100644
--- a/Source/Core/VideoCommon/VertexManagerBase.h
+++ b/Source/Core/VideoCommon/VertexManagerBase.h
@@ -83,8 +83,8 @@ protected:
   Slope m_zslope = {};
   void CalculateZSlope(NativeVertexFormat* format);
 
-  VideoCommon::GXPipelineConfig m_current_pipeline_config;
-  VideoCommon::GXUberPipelineConfig m_current_uber_pipeline_config;
+  VideoCommon::GXPipelineUid m_current_pipeline_config;
+  VideoCommon::GXUberPipelineUid m_current_uber_pipeline_config;
   const AbstractPipeline* m_current_pipeline_object = nullptr;
   PrimitiveType m_current_primitive_type = PrimitiveType::Points;
   bool m_pipeline_config_changed = true;
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj b/Source/Core/VideoCommon/VideoCommon.vcxproj
index 4cc6611ee8..b84bfb6d66 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj
@@ -120,6 +120,7 @@
     <ClInclude Include="Fifo.h" />
     <ClInclude Include="FPSCounter.h" />
     <ClInclude Include="FramebufferManagerBase.h" />
+    <ClInclude Include="GXPipelineTypes.h" />
     <ClInclude Include="ShaderCache.h" />
     <ClInclude Include="UberShaderCommon.h" />
     <ClInclude Include="UberShaderPixel.h" />
@@ -190,4 +191,4 @@
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
   </ImportGroup>
-</Project>
+</Project>
\ No newline at end of file
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
index ca0fe0029b..f356cc9ef4 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
@@ -198,7 +198,7 @@
       <Filter>Base</Filter>
     </ClCompile>
     <ClCompile Include="ShaderCache.cpp">
-      <Filter>Shader Managers</Filter>
+      <Filter>Shader Generators</Filter>
     </ClCompile>
   </ItemGroup>
   <ItemGroup>
@@ -381,11 +381,14 @@
     <ClInclude Include="AbstractFramebuffer.h">
       <Filter>Base</Filter>
     </ClInclude>
+    <ClInclude Include="GXPipelineTypes.h">
+      <Filter>Shader Generators</Filter>
+    </ClInclude>
     <ClInclude Include="ShaderCache.h">
-      <Filter>Shader Managers</Filter>
+      <Filter>Shader Generators</Filter>
     </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Text Include="CMakeLists.txt" />
   </ItemGroup>
-</Project>
+</Project>
\ No newline at end of file
diff --git a/Source/Core/VideoCommon/VideoConfig.cpp b/Source/Core/VideoCommon/VideoConfig.cpp
index c14a00757a..271223ba3d 100644
--- a/Source/Core/VideoCommon/VideoConfig.cpp
+++ b/Source/Core/VideoCommon/VideoConfig.cpp
@@ -102,6 +102,7 @@ void VideoConfig::Refresh()
   bBackendMultithreading = Config::Get(Config::GFX_BACKEND_MULTITHREADING);
   iCommandBufferExecuteInterval = Config::Get(Config::GFX_COMMAND_BUFFER_EXECUTE_INTERVAL);
   bShaderCache = Config::Get(Config::GFX_SHADER_CACHE);
+  bWaitForShadersBeforeStarting = Config::Get(Config::GFX_WAIT_FOR_SHADERS_BEFORE_STARTING);
   iUberShaderMode = static_cast<UberShaderMode>(Config::Get(Config::GFX_UBERSHADER_MODE));
   iShaderCompilerThreads = Config::Get(Config::GFX_SHADER_COMPILER_THREADS);
   iShaderPrecompilerThreads = Config::Get(Config::GFX_SHADER_PRECOMPILER_THREADS);
@@ -196,6 +197,10 @@ u32 VideoConfig::GetShaderCompilerThreads() const
 
 u32 VideoConfig::GetShaderPrecompilerThreads() const
 {
+  // When using background compilation, always keep the same thread count.
+  if (bWaitForShadersBeforeStarting)
+    return GetShaderCompilerThreads();
+
   if (!backend_info.bSupportsBackgroundCompiling)
     return 0;
 
diff --git a/Source/Core/VideoCommon/VideoConfig.h b/Source/Core/VideoCommon/VideoConfig.h
index 2b69919528..c8bded371b 100644
--- a/Source/Core/VideoCommon/VideoConfig.h
+++ b/Source/Core/VideoCommon/VideoConfig.h
@@ -169,6 +169,7 @@ struct VideoConfig final
   int iCommandBufferExecuteInterval;
 
   // Shader compilation settings.
+  bool bWaitForShadersBeforeStarting;
   UberShaderMode iUberShaderMode;
 
   // Number of shader compiler threads.