From e5a06d3f4a103a54cd4eb51970fedee405d1d698 Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Mon, 28 Jan 2019 21:14:17 +0100
Subject: [PATCH] [dxbc] Only load requested components from constant buffers

Results in better performance compared to loading the entire vector
on RADV. Suggested by Samuel Pitoiset.
---
 src/dxbc/dxbc_compiler.cpp | 57 ++++++++++++++++++++++++++++++++++++--
 src/dxbc/dxbc_compiler.h   |  4 +++
 2 files changed, 59 insertions(+), 2 deletions(-)

diff --git a/src/dxbc/dxbc_compiler.cpp b/src/dxbc/dxbc_compiler.cpp
index 6417641c6..bd3f4a3c8 100644
--- a/src/dxbc/dxbc_compiler.cpp
+++ b/src/dxbc/dxbc_compiler.cpp
@@ -5187,14 +5187,65 @@ namespace dxvk {
       m_module.opStore(ptr.id, tmp.id);
     }
   }
-  
-  
+
+
   DxbcRegisterValue DxbcCompiler::emitRegisterLoadRaw(
     const DxbcRegister&           reg) {
     return emitValueLoad(emitGetOperandPtr(reg));
   }
   
   
+  DxbcRegisterValue DxbcCompiler::emitConstantBufferLoad(
+    const DxbcRegister&           reg,
+          DxbcRegMask             writeMask) {
+    DxbcRegisterPointer ptr = emitGetOperandPtr(reg);
+
+    std::array<uint32_t, 4> ccomps = { 0, 0, 0, 0 };
+    std::array<uint32_t, 4> scomps = { 0, 0, 0, 0 };
+    uint32_t                scount = 0;
+
+    for (uint32_t i = 0; i < 4; i++) {
+      uint32_t sindex = reg.swizzle[i];
+
+      if (!writeMask[i] || ccomps[sindex])
+        continue;
+      
+      uint32_t componentId = m_module.constu32(sindex);
+      uint32_t componentPtr = m_module.opAccessChain(
+        m_module.defPointerType(
+          getScalarTypeId(DxbcScalarType::Float32),
+          spv::StorageClassUniformConstant),
+        ptr.id, 1, &componentId);
+      
+      ccomps[sindex] = m_module.opLoad(
+        getScalarTypeId(DxbcScalarType::Float32),
+        componentPtr);
+    }
+
+    for (uint32_t i = 0; i < 4; i++) {
+      uint32_t sindex = reg.swizzle[i];
+      
+      if (writeMask[i])
+        scomps[scount++] = ccomps[sindex];
+    }
+    
+    DxbcRegisterValue result;
+    result.type.ctype  = DxbcScalarType::Float32;
+    result.type.ccount = scount;
+    result.id = scomps[0];
+    
+    if (scount > 1) {
+      result.id = m_module.opCompositeConstruct(
+        getVectorTypeId(result.type),
+        scount, scomps.data());
+    }
+
+    result = emitRegisterBitcast(result, reg.dataType);
+    result = emitSrcOperandModifiers(result, reg.modifiers);
+    return result;
+  }
+  
+  
   DxbcRegisterValue DxbcCompiler::emitRegisterLoad(
     const DxbcRegister&           reg,
           DxbcRegMask             writeMask) {
@@ -5236,6 +5287,8 @@ namespace dxvk {
       
       // Cast constants to the requested type
       return emitRegisterBitcast(result, reg.dataType);
+    } else if (reg.type == DxbcOperandType::ConstantBuffer) {
+      return emitConstantBufferLoad(reg, writeMask);
     } else {
       // Load operand from the operand pointer
       DxbcRegisterValue result = emitRegisterLoadRaw(reg);
diff --git a/src/dxbc/dxbc_compiler.h b/src/dxbc/dxbc_compiler.h
index 28ac3d03c..3b717169d 100644
--- a/src/dxbc/dxbc_compiler.h
+++ b/src/dxbc/dxbc_compiler.h
@@ -972,6 +972,10 @@ namespace dxvk {
     DxbcRegisterValue emitRegisterLoadRaw(
       const DxbcRegister&           reg);
     
+    DxbcRegisterValue emitConstantBufferLoad(
+      const DxbcRegister&           reg,
+            DxbcRegMask             writeMask);
+    
     DxbcRegisterValue emitRegisterLoad(
       const DxbcRegister&           reg,
             DxbcRegMask             writeMask);