From 61adaa941deca50f807c00d16fe499be6727115e Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Thu, 28 Mar 2019 14:09:08 +0100
Subject: [PATCH] [d3d11] Implement fast path for binding full constant buffers

Saves a few CPU cycles in the most common case where
we don't have to perform any sort of range check.
---
 src/d3d11/d3d11_context.cpp | 62 +++++++++++++++++++++++++------------
 src/d3d11/d3d11_context.h   |  7 +++++
 2 files changed, 50 insertions(+), 19 deletions(-)

diff --git a/src/d3d11/d3d11_context.cpp b/src/d3d11/d3d11_context.cpp
index 02d557281..07f4571b2 100644
--- a/src/d3d11/d3d11_context.cpp
+++ b/src/d3d11/d3d11_context.cpp
@@ -1657,8 +1657,7 @@ namespace dxvk {
       DxbcProgramType::VertexShader,
       m_state.vs.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -1670,7 +1669,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::VertexShader,
       m_state.vs.constantBuffers,
       StartSlot, NumBuffers,
@@ -1820,8 +1819,7 @@ namespace dxvk {
       DxbcProgramType::HullShader,
       m_state.hs.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -1833,7 +1831,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::HullShader,
       m_state.hs.constantBuffers,
       StartSlot, NumBuffers,
@@ -1969,8 +1967,7 @@ namespace dxvk {
       DxbcProgramType::DomainShader,
       m_state.ds.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -1982,7 +1979,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::DomainShader,
       m_state.ds.constantBuffers,
       StartSlot, NumBuffers,
@@ -2104,8 +2101,7 @@ namespace dxvk {
       DxbcProgramType::GeometryShader,
       m_state.gs.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -2117,7 +2113,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {  
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::GeometryShader,
       m_state.gs.constantBuffers,
       StartSlot, NumBuffers,
@@ -2253,8 +2249,7 @@ namespace dxvk {
       DxbcProgramType::PixelShader,
       m_state.ps.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -2266,7 +2261,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::PixelShader,
       m_state.ps.constantBuffers,
       StartSlot, NumBuffers,
@@ -2402,8 +2397,7 @@ namespace dxvk {
       DxbcProgramType::ComputeShader,
       m_state.cs.constantBuffers,
       StartSlot, NumBuffers,
-      ppConstantBuffers,
-      nullptr, nullptr);
+      ppConstantBuffers);
   }
 
 
@@ -2415,7 +2409,7 @@ namespace dxvk {
     const UINT*                             pNumConstants) {
     D3D10DeviceLock lock = LockContext();
     
-    SetConstantBuffers(
+    SetConstantBuffers1(
       DxbcProgramType::ComputeShader,
       m_state.cs.constantBuffers,
       StartSlot, NumBuffers,
@@ -3357,6 +3351,36 @@ namespace dxvk {
 
 
   void D3D11DeviceContext::SetConstantBuffers(
+          DxbcProgramType                   ShaderStage,
+          D3D11ConstantBufferBindings&      Bindings,
+          UINT                              StartSlot,
+          UINT                              NumBuffers,
+          ID3D11Buffer* const*              ppConstantBuffers) {
+    const uint32_t slotId = computeResourceSlotId(
+      ShaderStage, DxbcBindingType::ConstantBuffer,
+      StartSlot);
+    
+    for (uint32_t i = 0; i < NumBuffers; i++) {
+      auto newBuffer = static_cast<D3D11Buffer*>(ppConstantBuffers[i]);
+      
+      UINT constantBound = newBuffer
+        ? newBuffer->Desc()->ByteWidth / 16
+        : 0;
+      
+      if (Bindings[StartSlot + i].buffer         != newBuffer
+       || Bindings[StartSlot + i].constantBound  != constantBound) {
+        Bindings[StartSlot + i].buffer         = newBuffer;
+        Bindings[StartSlot + i].constantOffset = 0;
+        Bindings[StartSlot + i].constantCount  = constantBound;
+        Bindings[StartSlot + i].constantBound  = constantBound;
+        
+        BindConstantBuffer(slotId + i, &Bindings[StartSlot + i]);
+      }
+    }
+  }
+  
+  
+  void D3D11DeviceContext::SetConstantBuffers1(
           DxbcProgramType                   ShaderStage,
           D3D11ConstantBufferBindings&      Bindings,
           UINT                              StartSlot,
@@ -3378,7 +3402,7 @@ namespace dxvk {
       if (likely(newBuffer != nullptr)) {
         constantBound = newBuffer->Desc()->ByteWidth / 16;
 
-        if (pFirstConstant && pNumConstants) {
+        if (likely(pFirstConstant && pNumConstants)) {
           constantOffset  = pFirstConstant[i];
           constantCount   = pNumConstants [i];
 
diff --git a/src/d3d11/d3d11_context.h b/src/d3d11/d3d11_context.h
index fd49ae8e0..0a6e2a111 100644
--- a/src/d3d11/d3d11_context.h
+++ b/src/d3d11/d3d11_context.h
@@ -735,6 +735,13 @@ namespace dxvk {
             ID3D11Buffer*                     pBuffer);
     
     void SetConstantBuffers(
+            DxbcProgramType                   ShaderStage,
+            D3D11ConstantBufferBindings&      Bindings,
+            UINT                              StartSlot,
+            UINT                              NumBuffers,
+            ID3D11Buffer* const*              ppConstantBuffers);
+    
+    void SetConstantBuffers1(
             DxbcProgramType                   ShaderStage,
             D3D11ConstantBufferBindings&      Bindings,
             UINT                              StartSlot,