From: Tanya Lattner <tonic@nondot.org>
Date: Thu, 16 Aug 2012 00:10:13 +0000 (+0000)
Subject: Convert loads and stores of vec3 to vec4 to achieve better code generation. Add test... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=c58dcdc8facb646d88675bb6fbcb5c787166c4be;p=clang

Convert loads and stores of vec3 to vec4 to achieve better code generation. Add test case.


git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@162002 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/CGExpr.cpp b/lib/CodeGen/CGExpr.cpp
index ecee7b4931..1fe4c18bad 100644
--- a/lib/CodeGen/CGExpr.cpp
+++ b/lib/CodeGen/CGExpr.cpp
@@ -938,6 +938,50 @@ llvm::MDNode *CodeGenFunction::getRangeForLoadFromType(QualType Ty) {
 llvm::Value *CodeGenFunction::EmitLoadOfScalar(llvm::Value *Addr, bool Volatile,
                                               unsigned Alignment, QualType Ty,
                                               llvm::MDNode *TBAAInfo) {
+  
+  // For better performance, handle vector loads differently.
+  if (Ty->isVectorType()) {
+    llvm::Value *V;
+    const llvm::Type *EltTy =
+    cast<llvm::PointerType>(Addr->getType())->getElementType();
+    
+    const llvm::VectorType *VTy = cast<llvm::VectorType>(EltTy);
+      
+    // Handle vectors of size 3, like size 4 for better performance.
+    if (VTy->getNumElements() == 3) {
+        
+      // Bitcast to vec4 type.
+      llvm::VectorType *vec4Ty = llvm::VectorType::get(VTy->getElementType(),
+                                                         4);
+      llvm::PointerType *ptVec4Ty =
+      llvm::PointerType::get(vec4Ty,
+                             (cast<llvm::PointerType>(
+                                      Addr->getType()))->getAddressSpace());
+      llvm::Value *Cast = Builder.CreateBitCast(Addr, ptVec4Ty,
+                                                "castToVec4");
+      // Now load value.
+      llvm::Value *LoadVal = Builder.CreateLoad(Cast, Volatile, "loadVec4");
+        
+      // Shuffle vector to get vec3.
+      llvm::SmallVector<llvm::Constant*, 3> Mask;
+      Mask.push_back(llvm::ConstantInt::get(
+                                    llvm::Type::getInt32Ty(getLLVMContext()),
+                                            0));
+      Mask.push_back(llvm::ConstantInt::get(
+                                    llvm::Type::getInt32Ty(getLLVMContext()),
+                                            1));
+      Mask.push_back(llvm::ConstantInt::get(
+                                     llvm::Type::getInt32Ty(getLLVMContext()),
+                                            2));
+        
+      llvm::Value *MaskV = llvm::ConstantVector::get(Mask);
+      V = Builder.CreateShuffleVector(LoadVal,
+                                      llvm::UndefValue::get(vec4Ty),
+                                      MaskV, "extractVec");
+      return EmitFromMemory(V, Ty);
+    }
+  }
+  
   llvm::LoadInst *Load = Builder.CreateLoad(Addr);
   if (Volatile)
     Load->setVolatile(true);
@@ -984,6 +1028,42 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, llvm::Value *Addr,
                                         QualType Ty,
                                         llvm::MDNode *TBAAInfo,
                                         bool isInit) {
+  
+  // Handle vectors differently to get better performance.
+  if (Ty->isVectorType()) {
+    llvm::Type *SrcTy = Value->getType();
+    llvm::VectorType *VecTy = cast<llvm::VectorType>(SrcTy);
+    // Handle vec3 special.
+    if (VecTy->getNumElements() == 3) {
+      llvm::LLVMContext &VMContext = getLLVMContext();
+      
+      // Our source is a vec3, do a shuffle vector to make it a vec4.
+      llvm::SmallVector<llvm::Constant*, 4> Mask;
+      Mask.push_back(llvm::ConstantInt::get(
+                                            llvm::Type::getInt32Ty(VMContext),
+                                            0));
+      Mask.push_back(llvm::ConstantInt::get(
+                                            llvm::Type::getInt32Ty(VMContext),
+                                            1));
+      Mask.push_back(llvm::ConstantInt::get(
+                                            llvm::Type::getInt32Ty(VMContext),
+                                            2));
+      Mask.push_back(llvm::UndefValue::get(llvm::Type::getInt32Ty(VMContext)));
+      
+      llvm::Value *MaskV = llvm::ConstantVector::get(Mask);
+      Value = Builder.CreateShuffleVector(Value,
+                                          llvm::UndefValue::get(VecTy),
+                                          MaskV, "extractVec");
+      SrcTy = llvm::VectorType::get(VecTy->getElementType(), 4);
+    }
+    llvm::PointerType *DstPtr = cast<llvm::PointerType>(Addr->getType());
+    if (DstPtr->getElementType() != SrcTy) {
+      llvm::Type *MemTy =
+      llvm::PointerType::get(SrcTy, DstPtr->getAddressSpace());
+      Addr = Builder.CreateBitCast(Addr, MemTy, "storetmp");
+    }
+  }
+  
   Value = EmitToMemory(Value, Ty);
   
   llvm::StoreInst *Store = Builder.CreateStore(Value, Addr, Volatile);
diff --git a/test/CodeGen/alignment.c b/test/CodeGen/alignment.c
index 8882c91d03..5051a98db9 100644
--- a/test/CodeGen/alignment.c
+++ b/test/CodeGen/alignment.c
@@ -43,7 +43,8 @@ void test3(packedfloat3 *p) {
   *p = (packedfloat3) { 3.2f, 2.3f, 0.1f };
 }
 // CHECK: @test3(
-// CHECK: store <3 x float> {{.*}}, align 4
+// CHECK: bitcast <3 x float>* %.compoundliteral to <4 x float>*
+// CHECK: store <4 x float> {{.*}}, align 4
 // CHECK: ret void
 
 
diff --git a/test/CodeGenOpenCL/vectorLoadStore.cl b/test/CodeGenOpenCL/vectorLoadStore.cl
new file mode 100644
index 0000000000..ba0405d098
--- /dev/null
+++ b/test/CodeGenOpenCL/vectorLoadStore.cl
@@ -0,0 +1,9 @@
+// RUN: %clang_cc1 %s -emit-llvm -O0 -o - | FileCheck %s
+
+typedef char char3 __attribute((ext_vector_type(3)));;
+
+// Check for optimized vec3 load/store which treats vec3 as vec4.
+void foo(char3 *P, char3 *Q) {
+  *P = *Q;
+  // CHECK: %extractVec = shufflevector <4 x i8> %loadVec4, <4 x i8> undef, <3 x i32> <i32 0, i32 1, i32 2>
+}