replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[X86] Auto-upgrade AVX1 vbroadcast intrinsics 

They are replaced with the same IR that is generated for the
vector-initializers in avxintrin.h.

The test verifies that we get back the original instruction.  I haven't seen
this approach to be used in other auto-upgrade tests (i.e. llc + FileCheck)
but I think it's the most direct way to test this case.  I believe this should
work because llc upgrades calls during parsing.  (Other tests mostly check
that assembling and disassembling yields the upgraded IR.)

llvm-svn: 209863
This commit is contained in:
Adam Nemet 2014-05-29 23:35:33 +00:00
parent 1316a0e0e0
commit 39066800e9
2 changed files with 57 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/IR/AutoUpgrade.cpp
View File

@ -114,6 +114,9 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
Name == "x86.avx.movnt.pd.256" ||
Name == "x86.avx.movnt.ps.256" ||
Name == "x86.sse42.crc32.64.8" ||
Name == "x86.avx.vbroadcast.ss" ||
Name == "x86.avx.vbroadcast.ss.256" ||
Name == "x86.avx.vbroadcast.sd.256" ||
(Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
NewFn = nullptr;
return true;
@ -335,6 +338,19 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
Rep = Builder.CreateCall2(CRC32, Trunc0, CI->getArgOperand(1));
Rep = Builder.CreateZExt(Rep, CI->getType(), "");
} else if (Name.startswith("llvm.x86.avx.vbroadcast")) {
// Replace broadcasts with a series of insertelements.
Type *VecTy = CI->getType();
Type *EltTy = VecTy->getVectorElementType();
unsigned EltNum = VecTy->getVectorNumElements();
Value *Cast = Builder.CreateBitCast(CI->getArgOperand(0),
EltTy->getPointerTo());
Value *Load = Builder.CreateLoad(Cast);
Type *I32Ty = Type::getInt32Ty(C);
Rep = UndefValue::get(VecTy);
for (unsigned I = 0; I < EltNum; ++I)
Rep = Builder.CreateInsertElement(Rep, Load,
ConstantInt::get(I32Ty, I));
} else {
bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
if (Name == "llvm.x86.avx.vpermil.pd.256")

41
llvm/test/Bitcode/x86-upgrade-avx-vbroadcast.ll Normal file
View File

@ -0,0 +1,41 @@
; RUN: llc -mattr=+avx < %s | FileCheck %s
; Check that we properly upgrade the AVX vbroadcast intrinsics to IR. The
; expectation is that we should still get the original instruction back that
; maps to the intrinsic.
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.9.0"
; CHECK-LABEL: test_mm_broadcast_ss:
define <4 x float> @test_mm_broadcast_ss(float* readonly %__a){
entry:
%0 = bitcast float* %__a to i8*
; CHECK: vbroadcastss (%{{.*}}), %xmm
%1 = tail call <4 x float> @llvm.x86.avx.vbroadcast.ss(i8* %0)
ret <4 x float> %1
}
; CHECK-LABEL: test_mm256_broadcast_sd:
define <4 x double> @test_mm256_broadcast_sd(double* readonly %__a) {
entry:
%0 = bitcast double* %__a to i8*
; CHECK: vbroadcastsd (%{{.*}}), %ymm
%1 = tail call <4 x double> @llvm.x86.avx.vbroadcast.sd.256(i8* %0)
ret <4 x double> %1
}
; CHECK-LABEL: test_mm256_broadcast_ss:
define <8 x float> @test_mm256_broadcast_ss(float* readonly %__a) {
entry:
%0 = bitcast float* %__a to i8*
; CHECK: vbroadcastss (%{{.*}}), %ymm
%1 = tail call <8 x float> @llvm.x86.avx.vbroadcast.ss.256(i8* %0)
ret <8 x float> %1
}
declare <8 x float> @llvm.x86.avx.vbroadcast.ss.256(i8*)
declare <4 x double> @llvm.x86.avx.vbroadcast.sd.256(i8*)
declare <4 x float> @llvm.x86.avx.vbroadcast.ss(i8*)