From 590307b94c5865c0a75a18e9a593d251ce38407a Mon Sep 17 00:00:00 2001
From: Stenzek <stenzek@gmail.com>
Date: Tue, 27 Feb 2018 19:38:00 +1000
Subject: [PATCH] ShaderCache: Use memcmp for comparing pipeline UIDs

As these are stored in a map, operator< will become a hot function when
doing lookups, which happen every frame. std::tie generated a rather
large function here with quite a few branches.
---
 Source/Core/VideoCommon/ShaderCache.h       | 38 ++++++++++++---------
 Source/Core/VideoCommon/VertexManagerBase.h |  4 +--
 2 files changed, 24 insertions(+), 18 deletions(-)

diff --git a/Source/Core/VideoCommon/ShaderCache.h b/Source/Core/VideoCommon/ShaderCache.h
index 922cd4d5fa..6516edd8d8 100644
--- a/Source/Core/VideoCommon/ShaderCache.h
+++ b/Source/Core/VideoCommon/ShaderCache.h
@@ -6,6 +6,7 @@
 
 #include <array>
 #include <cstddef>
+#include <cstring>
 #include <map>
 #include <memory>
 #include <optional>
@@ -42,19 +43,23 @@ struct GXPipelineConfig
   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::tie(vertex_format, vs_uid, gs_uid, ps_uid, rasterization_state.hex, depth_state.hex,
-                    blending_state.hex) < std::tie(rhs.vertex_format, rhs.vs_uid, rhs.gs_uid,
-                                                   rhs.ps_uid, rhs.rasterization_state.hex,
-                                                   rhs.depth_state.hex, rhs.blending_state.hex);
+    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
   }
   bool operator==(const GXPipelineConfig& rhs) const
   {
-    return std::tie(vertex_format, vs_uid, gs_uid, ps_uid, rasterization_state.hex, depth_state.hex,
-                    blending_state.hex) == std::tie(rhs.vertex_format, rhs.vs_uid, rhs.gs_uid,
-                                                    rhs.ps_uid, rhs.rasterization_state.hex,
-                                                    rhs.depth_state.hex, rhs.blending_state.hex);
+    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
   }
   bool operator!=(const GXPipelineConfig& rhs) const { return !operator==(rhs); }
 };
@@ -68,19 +73,20 @@ struct GXUberPipelineConfig
   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::tie(vertex_format, vs_uid, gs_uid, ps_uid, rasterization_state.hex, depth_state.hex,
-                    blending_state.hex) < std::tie(rhs.vertex_format, rhs.vs_uid, rhs.gs_uid,
-                                                   rhs.ps_uid, rhs.rasterization_state.hex,
-                                                   rhs.depth_state.hex, rhs.blending_state.hex);
+    return std::memcmp(this, &rhs, sizeof(*this)) < 0;
   }
   bool operator==(const GXUberPipelineConfig& rhs) const
   {
-    return std::tie(vertex_format, vs_uid, gs_uid, ps_uid, rasterization_state.hex, depth_state.hex,
-                    blending_state.hex) == std::tie(rhs.vertex_format, rhs.vs_uid, rhs.gs_uid,
-                                                    rhs.ps_uid, rhs.rasterization_state.hex,
-                                                    rhs.depth_state.hex, rhs.blending_state.hex);
+    return std::memcmp(this, &rhs, sizeof(*this)) == 0;
   }
   bool operator!=(const GXUberPipelineConfig& rhs) const { return !operator==(rhs); }
 };
diff --git a/Source/Core/VideoCommon/VertexManagerBase.h b/Source/Core/VideoCommon/VertexManagerBase.h
index 20a63a3e2a..eac95f614f 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::GXPipelineConfig m_current_pipeline_config;
+  VideoCommon::GXUberPipelineConfig 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;