maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

CodeGen: Support for Cast Operations in vector code generation 

llvm-svn: 139097
This commit is contained in:
Tobias Grosser 2011-09-04 11:45:52 +00:00
parent 8b00a51164
commit c92151516f
2 changed files with 53 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
polly/lib/CodeGeneration.cpp
View File

@ -409,6 +409,21 @@ public:
vectorMap[load] = newLoad;
}
void copyUnaryInst(const UnaryInstruction *Inst, ValueMapT &BBMap,
ValueMapT &VectorMap, int VectorDimension,
int VectorWidth) {
Value *NewOperand = getOperand(Inst->getOperand(0), BBMap, &VectorMap);
NewOperand = makeVectorOperand(NewOperand, VectorWidth);
if (const CastInst *Cast = dyn_cast<CastInst>(Inst)) {
VectorType *DestType = VectorType::get(Inst->getType(), VectorWidth);
VectorMap[Inst] = Builder.CreateCast(Cast->getOpcode(), NewOperand,
DestType);
} else
llvm_unreachable("Can not generate vector code for instruction");
return;
}
void copyBinInst(const BinaryOperator *Inst, ValueMapT &BBMap,
ValueMapT &vectorMap, int vectorDimension, int vectorWidth) {
Value *opZero = Inst->getOperand(0);
@ -529,7 +544,11 @@ public:
}
if (isVectorBlock() && hasVectorOperands(Inst, vectorMap)) {
if (const BinaryOperator *binaryInst = dyn_cast<BinaryOperator>(Inst))
if (const UnaryInstruction *UnaryInst = dyn_cast<UnaryInstruction>(Inst))
copyUnaryInst(UnaryInst, BBMap, vectorMap, vectorDimension,
vectorWidth);
else if
(const BinaryOperator *binaryInst = dyn_cast<BinaryOperator>(Inst))
copyBinInst(binaryInst, BBMap, vectorMap, vectorDimension, vectorWidth);
else if (const StoreInst *store = dyn_cast<StoreInst>(Inst))
copyVectorStore(store, BBMap, vectorMap, scalarMaps, vectorDimension,

33
polly/test/CodeGen/simple_vec_cast.ll Normal file
View File

@ -0,0 +1,33 @@
; RUN: opt %loadPolly -basicaa -polly-codegen -enable-polly-vector -dce -S %s | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-unknown-linux-gnu"
@A = common global [1024 x float] zeroinitializer, align 16
@B = common global [1024 x double] zeroinitializer, align 16
define void @simple_vec_const() nounwind {
bb:
br label %bb1
bb1: ; preds = %bb3, %bb
%indvar = phi i64 [ %indvar.next, %bb3 ], [ 0, %bb ]
%scevgep = getelementptr [1024 x double]* @B, i64 0, i64 %indvar
%exitcond = icmp ne i64 %indvar, 4
br i1 %exitcond, label %bb2, label %bb4
bb2: ; preds = %bb1
%tmp = load float* getelementptr inbounds ([1024 x float]* @A, i64 0, i64 0), align 16
%tmp2 = fpext float %tmp to double
store double %tmp2, double* %scevgep, align 4
br label %bb3
bb3: ; preds = %bb2
%indvar.next = add i64 %indvar, 1
br label %bb1
bb4: ; preds = %bb1
ret void
}
; CHECK: fpext <4 x float> %tmp_p_splat to <4 x double>