This is an archive of the discontinued LLVM Phabricator instance.

Differential D21521

[X86][SSE] Add cost model for BSWAP of vectors
ClosedPublic

Authored by RKSimon on Jun 20 2016, 10:47 AM.

Details

Reviewers
spatel
delena
andreadb
mkuper
hfinkel
Commits
rG356e823b51c3: [X86][SSE] Add cost model for BSWAP of vectors
rL273217: [X86][SSE] Add cost model for BSWAP of vectors
Summary

The BSWAP of vector types is quite efficiently implemented using vector shuffles on SSE/AVX targets, we should reflect the typical cost of this to encourage vectorization.

Also, we're not making much use of the intrinsic costings on any target - for instance why do we not use this for CTPOP instead of the rather limited getPopcntSupport() approach? CTLZ/CTTZ would probably benefit as well.

Diff Detail

Repository
rL LLVM

Event Timeline

RKSimon updated this revision to Diff 61267.Jun 20 2016, 10:47 AM
RKSimon retitled this revision from to [X86][SSE] Add cost model for BSWAP of vectors.
RKSimon updated this object.
RKSimon added reviewers: hfinkel, delena, mkuper, andreadb, spatel.
RKSimon set the repository for this revision to rL LLVM.
RKSimon added a subscriber: llvm-commits.
mkuper accepted this revision.Jun 20 2016, 11:02 AM
mkuper edited edge metadata.

I assume the costs are based on the lowering we have in test/CodeGen/X86/bswap-vector.ll ?
In any case, this LGTM.

(Out of curiosity, do you see this get hit in practice?)

This revision is now accepted and ready to land.Jun 20 2016, 11:02 AM
RKSimon added a comment.Jun 20 2016, 3:57 PM
In D21521#462314, @mkuper wrote:

I assume the costs are based on the lowering we have in test/CodeGen/X86/bswap-vector.ll ?
In any case, this LGTM.

(Out of curiosity, do you see this get hit in practice?)

Thanks Michael - yes the costs are based off the codegen from bswap-vector.ll

I have a couple of examples where this patch (along with enabling the loadcombine pass) can help vectorize the loading/conversion of big endian float data.

Closed by commit rL273217: [X86][SSE] Add cost model for BSWAP of vectors (authored by RKSimon). · Explain WhyJun 20 2016, 4:15 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Target/
X86/
X86TargetTransformInfo.cpp
X86TargetTransformInfo.cpp (revision 273167)
27 lines
test/
Analysis/
CostModel/
X86/
bswap.ll
bswap.ll (revision 273167)
68 lines
scalarize.ll
scalarize.ll (revision 273167)
8 lines
Transforms/
SLPVectorizer/
X86/
bswap.ll
bswap.ll (revision 273167)
202 lines

Diff 61267

lib/Target/X86/X86TargetTransformInfo.cpp

Show First 20 Lines • Show All 945 Lines • ▼ Show 20 Lines static const CostTblEntry XOPCostTbl[] = {
{ ISD::BITREVERSE, MVT::i32, 3 }, { ISD::BITREVERSE, MVT::i32, 3 },
{ ISD::BITREVERSE, MVT::i16, 3 }, { ISD::BITREVERSE, MVT::i16, 3 },
{ ISD::BITREVERSE, MVT::i8, 3 } { ISD::BITREVERSE, MVT::i8, 3 }
}; };
static const CostTblEntry AVX2CostTbl[] = { static const CostTblEntry AVX2CostTbl[] = {
{ ISD::BITREVERSE, MVT::v4i64, 5 }, { ISD::BITREVERSE, MVT::v4i64, 5 },
{ ISD::BITREVERSE, MVT::v8i32, 5 }, { ISD::BITREVERSE, MVT::v8i32, 5 },
{ ISD::BITREVERSE, MVT::v16i16, 5 }, { ISD::BITREVERSE, MVT::v16i16, 5 },
{ ISD::BITREVERSE, MVT::v32i8, 5 } { ISD::BITREVERSE, MVT::v32i8, 5 },
{ ISD::BSWAP, MVT::v4i64, 1 },
{ ISD::BSWAP, MVT::v8i32, 1 },
{ ISD::BSWAP, MVT::v16i16, 1 }
}; };
static const CostTblEntry AVX1CostTbl[] = { static const CostTblEntry AVX1CostTbl[] = {
{ ISD::BITREVERSE, MVT::v4i64, 10 }, { ISD::BITREVERSE, MVT::v4i64, 10 },
{ ISD::BITREVERSE, MVT::v8i32, 10 }, { ISD::BITREVERSE, MVT::v8i32, 10 },
{ ISD::BITREVERSE, MVT::v16i16, 10 }, { ISD::BITREVERSE, MVT::v16i16, 10 },
{ ISD::BITREVERSE, MVT::v32i8, 10 } { ISD::BITREVERSE, MVT::v32i8, 10 },
{ ISD::BSWAP, MVT::v4i64, 4 },
{ ISD::BSWAP, MVT::v8i32, 4 },
{ ISD::BSWAP, MVT::v16i16, 4 }
}; };
static const CostTblEntry SSSE3CostTbl[] = { static const CostTblEntry SSSE3CostTbl[] = {
{ ISD::BITREVERSE, MVT::v2i64, 5 }, { ISD::BITREVERSE, MVT::v2i64, 5 },
{ ISD::BITREVERSE, MVT::v4i32, 5 }, { ISD::BITREVERSE, MVT::v4i32, 5 },
{ ISD::BITREVERSE, MVT::v8i16, 5 }, { ISD::BITREVERSE, MVT::v8i16, 5 },
{ ISD::BITREVERSE, MVT::v16i8, 5 } { ISD::BITREVERSE, MVT::v16i8, 5 },
{ ISD::BSWAP, MVT::v2i64, 1 },
{ ISD::BSWAP, MVT::v4i32, 1 },
{ ISD::BSWAP, MVT::v8i16, 1 }
};
static const CostTblEntry SSE2CostTbl[] = {
{ ISD::BSWAP, MVT::v2i64, 7 },
{ ISD::BSWAP, MVT::v4i32, 7 },
{ ISD::BSWAP, MVT::v8i16, 7 }
}; };
unsigned ISD = ISD::DELETED_NODE; unsigned ISD = ISD::DELETED_NODE;
switch (IID) { switch (IID) {
default: default:
break; break;
case Intrinsic::bitreverse: case Intrinsic::bitreverse:
ISD = ISD::BITREVERSE; ISD = ISD::BITREVERSE;
break; break;
case Intrinsic::bswap:
ISD = ISD::BSWAP;
break;
} }
// Legalize the type. // Legalize the type.
std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, RetTy); std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, RetTy);
MVT MTy = LT.second; MVT MTy = LT.second;
// Attempt to lookup cost. // Attempt to lookup cost.
if (ST->hasXOP()) if (ST->hasXOP())
if (const auto *Entry = CostTableLookup(XOPCostTbl, ISD, MTy)) if (const auto *Entry = CostTableLookup(XOPCostTbl, ISD, MTy))
return LT.first * Entry->Cost; return LT.first * Entry->Cost;
if (ST->hasAVX2()) if (ST->hasAVX2())
if (const auto *Entry = CostTableLookup(AVX2CostTbl, ISD, MTy)) if (const auto *Entry = CostTableLookup(AVX2CostTbl, ISD, MTy))
return LT.first * Entry->Cost; return LT.first * Entry->Cost;
if (ST->hasAVX()) if (ST->hasAVX())
if (const auto *Entry = CostTableLookup(AVX1CostTbl, ISD, MTy)) if (const auto *Entry = CostTableLookup(AVX1CostTbl, ISD, MTy))
return LT.first * Entry->Cost; return LT.first * Entry->Cost;
if (ST->hasSSSE3()) if (ST->hasSSSE3())
if (const auto *Entry = CostTableLookup(SSSE3CostTbl, ISD, MTy)) if (const auto *Entry = CostTableLookup(SSSE3CostTbl, ISD, MTy))
return LT.first * Entry->Cost; return LT.first * Entry->Cost;
if (ST->hasSSE2())
if (const auto *Entry = CostTableLookup(SSE2CostTbl, ISD, MTy))
return LT.first * Entry->Cost;
return BaseT::getIntrinsicInstrCost(IID, RetTy, Tys, FMF); return BaseT::getIntrinsicInstrCost(IID, RetTy, Tys, FMF);
} }
int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy, int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
ArrayRef<Value *> Args, FastMathFlags FMF) { ArrayRef<Value *> Args, FastMathFlags FMF) {
return BaseT::getIntrinsicInstrCost(IID, RetTy, Args, FMF); return BaseT::getIntrinsicInstrCost(IID, RetTy, Args, FMF);
} }
▲ Show 20 Lines • Show All 583 Lines • Show Last 20 Lines

test/Analysis/CostModel/X86/bswap.ll

; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE2 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE42 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX2 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=XOP -check-prefix=XOPAVX ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=XOP -check-prefix=XOPAVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=XOP -check-prefix=XOPAVX2 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=XOP -check-prefix=XOPAVX2
; Verify the cost of vector bswap instructions. ; Verify the cost of vector bswap instructions.
declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>) declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>) declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>) declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>) declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>) declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>) declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
define <2 x i64> @var_bswap_v2i64(<2 x i64> %a) { define <2 x i64> @var_bswap_v2i64(<2 x i64> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v2i64': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v2i64':
; SSE2: Found an estimated cost of 6 for instruction: %bswap ; SSE2: Found an estimated cost of 7 for instruction: %bswap
; SSE42: Found an estimated cost of 6 for instruction: %bswap ; SSE42: Found an estimated cost of 1 for instruction: %bswap
; AVX: Found an estimated cost of 6 for instruction: %bswap ; AVX: Found an estimated cost of 1 for instruction: %bswap
; AVX2: Found an estimated cost of 6 for instruction: %bswap ; XOP: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 6 for instruction: %bswap
%bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a) %bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a)
ret <2 x i64> %bswap ret <2 x i64> %bswap
} }
define <4 x i64> @var_bswap_v4i64(<4 x i64> %a) { define <4 x i64> @var_bswap_v4i64(<4 x i64> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v4i64': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v4i64':
; SSE2: Found an estimated cost of 12 for instruction: %bswap ; SSE2: Found an estimated cost of 14 for instruction: %bswap
; SSE42: Found an estimated cost of 12 for instruction: %bswap ; SSE42: Found an estimated cost of 2 for instruction: %bswap
; AVX: Found an estimated cost of 12 for instruction: %bswap ; AVX1: Found an estimated cost of 4 for instruction: %bswap
; AVX2: Found an estimated cost of 12 for instruction: %bswap ; AVX2: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 12 for instruction: %bswap ; XOPAVX1: Found an estimated cost of 4 for instruction: %bswap
; XOPAVX2: Found an estimated cost of 1 for instruction: %bswap
%bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a) %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
ret <4 x i64> %bswap ret <4 x i64> %bswap
} }
define <4 x i32> @var_bswap_v4i32(<4 x i32> %a) { define <4 x i32> @var_bswap_v4i32(<4 x i32> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v4i32': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v4i32':
; SSE2: Found an estimated cost of 12 for instruction: %bswap ; SSE2: Found an estimated cost of 7 for instruction: %bswap
; SSE42: Found an estimated cost of 12 for instruction: %bswap ; SSE42: Found an estimated cost of 1 for instruction: %bswap
; AVX: Found an estimated cost of 12 for instruction: %bswap ; AVX: Found an estimated cost of 1 for instruction: %bswap
; AVX2: Found an estimated cost of 12 for instruction: %bswap ; XOP: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 12 for instruction: %bswap
%bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a) %bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a)
ret <4 x i32> %bswap ret <4 x i32> %bswap
} }
define <8 x i32> @var_bswap_v8i32(<8 x i32> %a) { define <8 x i32> @var_bswap_v8i32(<8 x i32> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v8i32': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v8i32':
; SSE2: Found an estimated cost of 24 for instruction: %bswap ; SSE2: Found an estimated cost of 14 for instruction: %bswap
; SSE42: Found an estimated cost of 24 for instruction: %bswap ; SSE42: Found an estimated cost of 2 for instruction: %bswap
; AVX: Found an estimated cost of 24 for instruction: %bswap ; AVX1: Found an estimated cost of 4 for instruction: %bswap
; AVX2: Found an estimated cost of 24 for instruction: %bswap ; AVX2: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 24 for instruction: %bswap ; XOPAVX1: Found an estimated cost of 4 for instruction: %bswap
; XOPAVX2: Found an estimated cost of 1 for instruction: %bswap
%bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a) %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
ret <8 x i32> %bswap ret <8 x i32> %bswap
} }
define <8 x i16> @var_bswap_v8i16(<8 x i16> %a) { define <8 x i16> @var_bswap_v8i16(<8 x i16> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v8i16': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v8i16':
; SSE2: Found an estimated cost of 24 for instruction: %bswap ; SSE2: Found an estimated cost of 7 for instruction: %bswap
; SSE42: Found an estimated cost of 24 for instruction: %bswap ; SSE42: Found an estimated cost of 1 for instruction: %bswap
; AVX: Found an estimated cost of 24 for instruction: %bswap ; AVX: Found an estimated cost of 1 for instruction: %bswap
; AVX2: Found an estimated cost of 24 for instruction: %bswap ; XOP: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 24 for instruction: %bswap
%bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a) %bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a)
ret <8 x i16> %bswap ret <8 x i16> %bswap
} }
define <16 x i16> @var_bswap_v16i16(<16 x i16> %a) { define <16 x i16> @var_bswap_v16i16(<16 x i16> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_bswap_v16i16': ; CHECK: 'Cost Model Analysis' for function 'var_bswap_v16i16':
; SSE2: Found an estimated cost of 48 for instruction: %bswap ; SSE2: Found an estimated cost of 14 for instruction: %bswap
; SSE42: Found an estimated cost of 48 for instruction: %bswap ; SSE42: Found an estimated cost of 2 for instruction: %bswap
; AVX: Found an estimated cost of 48 for instruction: %bswap ; AVX1: Found an estimated cost of 4 for instruction: %bswap
; AVX2: Found an estimated cost of 48 for instruction: %bswap ; AVX2: Found an estimated cost of 1 for instruction: %bswap
; XOP: Found an estimated cost of 48 for instruction: %bswap ; XOPAVX1: Found an estimated cost of 4 for instruction: %bswap
; XOPAVX2: Found an estimated cost of 1 for instruction: %bswap
%bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a) %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
ret <16 x i16> %bswap ret <16 x i16> %bswap
} }
No newline at end of file

test/Analysis/CostModel/X86/scalarize.ll

Show All 15 Lines
declare %i4 @llvm.cttz.v4i32(%i4) declare %i4 @llvm.cttz.v4i32(%i4)
declare %i8 @llvm.cttz.v2i64(%i8) declare %i8 @llvm.cttz.v2i64(%i8)
; CHECK32-LABEL: test_scalarized_intrinsics ; CHECK32-LABEL: test_scalarized_intrinsics
; CHECK64-LABEL: test_scalarized_intrinsics ; CHECK64-LABEL: test_scalarized_intrinsics
define void @test_scalarized_intrinsics() { define void @test_scalarized_intrinsics() {
%r1 = add %i8 undef, undef %r1 = add %i8 undef, undef
; CHECK32: cost of 12 {{.*}}bswap.v4i32 ; CHECK32: cost of 1 {{.*}}bswap.v4i32
; CHECK64: cost of 12 {{.*}}bswap.v4i32 ; CHECK64: cost of 1 {{.*}}bswap.v4i32
%r2 = call %i4 @llvm.bswap.v4i32(%i4 undef) %r2 = call %i4 @llvm.bswap.v4i32(%i4 undef)
; CHECK32: cost of 10 {{.*}}bswap.v2i64 ; CHECK32: cost of 1 {{.*}}bswap.v2i64
; CHECK64: cost of 6 {{.*}}bswap.v2i64 ; CHECK64: cost of 1 {{.*}}bswap.v2i64
%r3 = call %i8 @llvm.bswap.v2i64(%i8 undef) %r3 = call %i8 @llvm.bswap.v2i64(%i8 undef)
; CHECK32: cost of 12 {{.*}}cttz.v4i32 ; CHECK32: cost of 12 {{.*}}cttz.v4i32
; CHECK64: cost of 12 {{.*}}cttz.v4i32 ; CHECK64: cost of 12 {{.*}}cttz.v4i32
%r4 = call %i4 @llvm.cttz.v4i32(%i4 undef) %r4 = call %i4 @llvm.cttz.v4i32(%i4 undef)
; CHECK32: cost of 10 {{.*}}cttz.v2i64 ; CHECK32: cost of 10 {{.*}}cttz.v2i64
; CHECK64: cost of 6 {{.*}}cttz.v2i64 ; CHECK64: cost of 6 {{.*}}cttz.v2i64
%r5 = call %i8 @llvm.cttz.v2i64(%i8 undef) %r5 = call %i8 @llvm.cttz.v2i64(%i8 undef)
; CHECK32: ret ; CHECK32: ret
; CHECK64: ret ; CHECK64: ret
ret void ret void
} }

test/Transforms/SLPVectorizer/X86/bswap.ll

Show All 11 Lines
@src16 = common global [16 x i16] zeroinitializer, align 32 @src16 = common global [16 x i16] zeroinitializer, align 32
@dst16 = common global [16 x i16] zeroinitializer, align 32 @dst16 = common global [16 x i16] zeroinitializer, align 32
declare i64 @llvm.bswap.i64(i64) declare i64 @llvm.bswap.i64(i64)
declare i32 @llvm.bswap.i32(i32) declare i32 @llvm.bswap.i32(i32)
declare i16 @llvm.bswap.i16(i16) declare i16 @llvm.bswap.i16(i16)
define void @bswap_2i64() #0 { define void @bswap_2i64() #0 {
; CHECK-LABEL: @bswap_2i64( ; SSE-LABEL: @bswap_2i64(
; CHECK-NEXT: [[LD0:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 0), align 8 ; SSE-NEXT: [[LD0:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 0), align 8
; CHECK-NEXT: [[LD1:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 1), align 8 ; SSE-NEXT: [[LD1:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 1), align 8
; CHECK-NEXT: [[BSWAP0:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD0]]) ; SSE-NEXT: [[BSWAP0:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD0]])
; CHECK-NEXT: [[BSWAP1:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD1]]) ; SSE-NEXT: [[BSWAP1:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD1]])
; CHECK-NEXT: store i64 [[BSWAP0]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 0), align 8 ; SSE-NEXT: store i64 [[BSWAP0]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 0), align 8
; CHECK-NEXT: store i64 [[BSWAP1]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 1), align 8 ; SSE-NEXT: store i64 [[BSWAP1]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 1), align 8
; CHECK-NEXT: ret void ; SSE-NEXT: ret void
;
; AVX-LABEL: @bswap_2i64(
; AVX-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* bitcast ([4 x i64]* @src64 to <2 x i64>*), align 8
; AVX-NEXT: [[TMP2:%.*]] = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> [[TMP1]])
; AVX-NEXT: store <2 x i64> [[TMP2]], <2 x i64>* bitcast ([4 x i64]* @dst64 to <2 x i64>*), align 8
; AVX-NEXT: ret void
; ;
%ld0 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 0), align 8 %ld0 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 0), align 8
%ld1 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 1), align 8 %ld1 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i32 0, i64 1), align 8
%bswap0 = call i64 @llvm.bswap.i64(i64 %ld0) %bswap0 = call i64 @llvm.bswap.i64(i64 %ld0)
%bswap1 = call i64 @llvm.bswap.i64(i64 %ld1) %bswap1 = call i64 @llvm.bswap.i64(i64 %ld1)
store i64 %bswap0, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 0), align 8 store i64 %bswap0, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 0), align 8
store i64 %bswap1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 1), align 8 store i64 %bswap1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i32 0, i64 1), align 8
ret void ret void
} }
define void @bswap_4i64() #0 { define void @bswap_4i64() #0 {
; CHECK-LABEL: @bswap_4i64( ; SSE-LABEL: @bswap_4i64(
; CHECK-NEXT: [[LD0:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 0), align 4 ; SSE-NEXT: [[LD0:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 0), align 4
; CHECK-NEXT: [[LD1:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 1), align 4 ; SSE-NEXT: [[LD1:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 1), align 4
; CHECK-NEXT: [[LD2:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 2), align 4 ; SSE-NEXT: [[LD2:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 2), align 4
; CHECK-NEXT: [[LD3:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 3), align 4 ; SSE-NEXT: [[LD3:%.*]] = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 3), align 4
; CHECK-NEXT: [[BSWAP0:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD0]]) ; SSE-NEXT: [[BSWAP0:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD0]])
; CHECK-NEXT: [[BSWAP1:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD1]]) ; SSE-NEXT: [[BSWAP1:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD1]])
; CHECK-NEXT: [[BSWAP2:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD2]]) ; SSE-NEXT: [[BSWAP2:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD2]])
; CHECK-NEXT: [[BSWAP3:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD3]]) ; SSE-NEXT: [[BSWAP3:%.*]] = call i64 @llvm.bswap.i64(i64 [[LD3]])
; CHECK-NEXT: store i64 [[BSWAP0]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 0), align 4 ; SSE-NEXT: store i64 [[BSWAP0]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 0), align 4
; CHECK-NEXT: store i64 [[BSWAP1]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 1), align 4 ; SSE-NEXT: store i64 [[BSWAP1]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 1), align 4
; CHECK-NEXT: store i64 [[BSWAP2]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 2), align 4 ; SSE-NEXT: store i64 [[BSWAP2]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 2), align 4
; CHECK-NEXT: store i64 [[BSWAP3]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 3), align 4 ; SSE-NEXT: store i64 [[BSWAP3]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 3), align 4
; CHECK-NEXT: ret void ; SSE-NEXT: ret void
;
; AVX-LABEL: @bswap_4i64(
; AVX-NEXT: [[TMP1:%.*]] = load <4 x i64>, <4 x i64>* bitcast ([4 x i64]* @src64 to <4 x i64>*), align 4
; AVX-NEXT: [[TMP2:%.*]] = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> [[TMP1]])
; AVX-NEXT: store <4 x i64> [[TMP2]], <4 x i64>* bitcast ([4 x i64]* @dst64 to <4 x i64>*), align 4
; AVX-NEXT: ret void
; ;
%ld0 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 0), align 4 %ld0 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 0), align 4
%ld1 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 1), align 4 %ld1 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 1), align 4
%ld2 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 2), align 4 %ld2 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 2), align 4
%ld3 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 3), align 4 %ld3 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @src64, i64 0, i64 3), align 4
%bswap0 = call i64 @llvm.bswap.i64(i64 %ld0) %bswap0 = call i64 @llvm.bswap.i64(i64 %ld0)
%bswap1 = call i64 @llvm.bswap.i64(i64 %ld1) %bswap1 = call i64 @llvm.bswap.i64(i64 %ld1)
%bswap2 = call i64 @llvm.bswap.i64(i64 %ld2) %bswap2 = call i64 @llvm.bswap.i64(i64 %ld2)
%bswap3 = call i64 @llvm.bswap.i64(i64 %ld3) %bswap3 = call i64 @llvm.bswap.i64(i64 %ld3)
store i64 %bswap0, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 0), align 4 store i64 %bswap0, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 0), align 4
store i64 %bswap1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 1), align 4 store i64 %bswap1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 1), align 4
store i64 %bswap2, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 2), align 4 store i64 %bswap2, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 2), align 4
store i64 %bswap3, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 3), align 4 store i64 %bswap3, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @dst64, i64 0, i64 3), align 4
ret void ret void
} }
define void @bswap_4i32() #0 { define void @bswap_4i32() #0 {
; CHECK-LABEL: @bswap_4i32( ; CHECK-LABEL: @bswap_4i32(
; CHECK-NEXT: [[LD0:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 4 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([8 x i32]* @src32 to <4 x i32>*), align 4
; CHECK-NEXT: [[LD1:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 4 ; CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[LD2:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 4 ; CHECK-NEXT: store <4 x i32> [[TMP2]], <4 x i32>* bitcast ([8 x i32]* @dst32 to <4 x i32>*), align 4
; CHECK-NEXT: [[LD3:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 4
; CHECK-NEXT: [[BSWAP0:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD0]])
; CHECK-NEXT: [[BSWAP1:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD1]])
; CHECK-NEXT: [[BSWAP2:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD2]])
; CHECK-NEXT: [[BSWAP3:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD3]])
; CHECK-NEXT: store i32 [[BSWAP0]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 0), align 4
; CHECK-NEXT: store i32 [[BSWAP1]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 1), align 4
; CHECK-NEXT: store i32 [[BSWAP2]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 2), align 4
; CHECK-NEXT: store i32 [[BSWAP3]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 3), align 4
; CHECK-NEXT: ret void ; CHECK-NEXT: ret void
; ;
%ld0 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 4 %ld0 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 4
%ld1 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 4 %ld1 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 4
%ld2 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 4 %ld2 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 4
%ld3 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 4 %ld3 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 4
%bswap0 = call i32 @llvm.bswap.i32(i32 %ld0) %bswap0 = call i32 @llvm.bswap.i32(i32 %ld0)
%bswap1 = call i32 @llvm.bswap.i32(i32 %ld1) %bswap1 = call i32 @llvm.bswap.i32(i32 %ld1)
%bswap2 = call i32 @llvm.bswap.i32(i32 %ld2) %bswap2 = call i32 @llvm.bswap.i32(i32 %ld2)
%bswap3 = call i32 @llvm.bswap.i32(i32 %ld3) %bswap3 = call i32 @llvm.bswap.i32(i32 %ld3)
store i32 %bswap0, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 0), align 4 store i32 %bswap0, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 0), align 4
store i32 %bswap1, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 1), align 4 store i32 %bswap1, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 1), align 4
store i32 %bswap2, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 2), align 4 store i32 %bswap2, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 2), align 4
store i32 %bswap3, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 3), align 4 store i32 %bswap3, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 3), align 4
ret void ret void
} }
define void @bswap_8i32() #0 { define void @bswap_8i32() #0 {
; CHECK-LABEL: @bswap_8i32( ; SSE-LABEL: @bswap_8i32(
; CHECK-NEXT: [[LD0:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 2 ; SSE-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([8 x i32]* @src32 to <4 x i32>*), align 2
; CHECK-NEXT: [[LD1:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 2 ; SSE-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 4) to <4 x i32>*), align 2
; CHECK-NEXT: [[LD2:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 2 ; SSE-NEXT: [[TMP3:%.*]] = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: [[LD3:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 2 ; SSE-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> [[TMP2]])
; CHECK-NEXT: [[LD4:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 4), align 2 ; SSE-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* bitcast ([8 x i32]* @dst32 to <4 x i32>*), align 2
; CHECK-NEXT: [[LD5:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 5), align 2 ; SSE-NEXT: store <4 x i32> [[TMP4]], <4 x i32>* bitcast (i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 4) to <4 x i32>*), align 2
; CHECK-NEXT: [[LD6:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 6), align 2 ; SSE-NEXT: ret void
; CHECK-NEXT: [[LD7:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 7), align 2 ;
; CHECK-NEXT: [[BSWAP0:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD0]]) ; AVX-LABEL: @bswap_8i32(
; CHECK-NEXT: [[BSWAP1:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD1]]) ; AVX-NEXT: [[TMP1:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([8 x i32]* @src32 to <8 x i32>*), align 2
; CHECK-NEXT: [[BSWAP2:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD2]]) ; AVX-NEXT: [[TMP2:%.*]] = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> [[TMP1]])
; CHECK-NEXT: [[BSWAP3:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD3]]) ; AVX-NEXT: store <8 x i32> [[TMP2]], <8 x i32>* bitcast ([8 x i32]* @dst32 to <8 x i32>*), align 2
; CHECK-NEXT: [[BSWAP4:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD4]]) ; AVX-NEXT: ret void
; CHECK-NEXT: [[BSWAP5:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD5]])
; CHECK-NEXT: [[BSWAP6:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD6]])
; CHECK-NEXT: [[BSWAP7:%.*]] = call i32 @llvm.bswap.i32(i32 [[LD7]])
; CHECK-NEXT: store i32 [[BSWAP0]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 0), align 2
; CHECK-NEXT: store i32 [[BSWAP1]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 1), align 2
; CHECK-NEXT: store i32 [[BSWAP2]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 2), align 2
; CHECK-NEXT: store i32 [[BSWAP3]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 3), align 2
; CHECK-NEXT: store i32 [[BSWAP4]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 4), align 2
; CHECK-NEXT: store i32 [[BSWAP5]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 5), align 2
; CHECK-NEXT: store i32 [[BSWAP6]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 6), align 2
; CHECK-NEXT: store i32 [[BSWAP7]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 7), align 2
; CHECK-NEXT: ret void
; ;
%ld0 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 2 %ld0 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 2
%ld1 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 2 %ld1 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 2
%ld2 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 2 %ld2 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 2
%ld3 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 2 %ld3 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 2
%ld4 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 4), align 2 %ld4 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 4), align 2
%ld5 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 5), align 2 %ld5 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 5), align 2
%ld6 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 6), align 2 %ld6 = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 6), align 2
Show All 14 Lines;
store i32 %bswap5, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 5), align 2 store i32 %bswap5, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 5), align 2
store i32 %bswap6, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 6), align 2 store i32 %bswap6, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 6), align 2
store i32 %bswap7, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 7), align 2 store i32 %bswap7, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 7), align 2
ret void ret void
} }
define void @bswap_8i16() #0 { define void @bswap_8i16() #0 {
; CHECK-LABEL: @bswap_8i16( ; CHECK-LABEL: @bswap_8i16(
; CHECK-NEXT: [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
; CHECK-NEXT: [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2 ; CHECK-NEXT: [[TMP2:%.*]] = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> [[TMP1]])
; CHECK-NEXT: [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2 ; CHECK-NEXT: store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
; CHECK-NEXT: [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
; CHECK-NEXT: [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
; CHECK-NEXT: [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
; CHECK-NEXT: [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
; CHECK-NEXT: [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2
; CHECK-NEXT: [[BSWAP0:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD0]])
; CHECK-NEXT: [[BSWAP1:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD1]])
; CHECK-NEXT: [[BSWAP2:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD2]])
; CHECK-NEXT: [[BSWAP3:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD3]])
; CHECK-NEXT: [[BSWAP4:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD4]])
; CHECK-NEXT: [[BSWAP5:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD5]])
; CHECK-NEXT: [[BSWAP6:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD6]])
; CHECK-NEXT: [[BSWAP7:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD7]])
; CHECK-NEXT: store i16 [[BSWAP0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
; CHECK-NEXT: store i16 [[BSWAP1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
; CHECK-NEXT: store i16 [[BSWAP2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
; CHECK-NEXT: store i16 [[BSWAP3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
; CHECK-NEXT: store i16 [[BSWAP4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
; CHECK-NEXT: store i16 [[BSWAP5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
; CHECK-NEXT: store i16 [[BSWAP6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
; CHECK-NEXT: store i16 [[BSWAP7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
; CHECK-NEXT: ret void ; CHECK-NEXT: ret void
; ;
%ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2 %ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
%ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2 %ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
%ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2 %ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
%ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2 %ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
%ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2 %ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
%ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2 %ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
Show All 14 Lines;
store i16 %bswap4, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2 store i16 %bswap4, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
store i16 %bswap5, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2 store i16 %bswap5, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
store i16 %bswap6, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2 store i16 %bswap6, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
store i16 %bswap7, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2 store i16 %bswap7, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
ret void ret void
} }
define void @bswap_16i16() #0 { define void @bswap_16i16() #0 {
; CHECK-LABEL: @bswap_16i16( ; SSE-LABEL: @bswap_16i16(
; CHECK-NEXT: [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2 ; SSE-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
; CHECK-NEXT: [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2 ; SSE-NEXT: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8) to <8 x i16>*), align 2
; CHECK-NEXT: [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2 ; SSE-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> [[TMP1]])
; CHECK-NEXT: [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2 ; SSE-NEXT: [[TMP4:%.*]] = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> [[TMP2]])
; CHECK-NEXT: [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2 ; SSE-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
; CHECK-NEXT: [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2 ; SSE-NEXT: store <8 x i16> [[TMP4]], <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8) to <8 x i16>*), align 2
; CHECK-NEXT: [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2 ; SSE-NEXT: ret void
; CHECK-NEXT: [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2 ;
; CHECK-NEXT: [[LD8:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8), align 2 ; AVX-LABEL: @bswap_16i16(
; CHECK-NEXT: [[LD9:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 9), align 2 ; AVX-NEXT: [[TMP1:%.*]] = load <16 x i16>, <16 x i16>* bitcast ([16 x i16]* @src16 to <16 x i16>*), align 2
; CHECK-NEXT: [[LD10:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 10), align 2 ; AVX-NEXT: [[TMP2:%.*]] = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> [[TMP1]])
; CHECK-NEXT: [[LD11:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 11), align 2 ; AVX-NEXT: store <16 x i16> [[TMP2]], <16 x i16>* bitcast ([16 x i16]* @dst16 to <16 x i16>*), align 2
; CHECK-NEXT: [[LD12:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 12), align 2 ; AVX-NEXT: ret void
; CHECK-NEXT: [[LD13:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 13), align 2
; CHECK-NEXT: [[LD14:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 14), align 2
; CHECK-NEXT: [[LD15:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 15), align 2
; CHECK-NEXT: [[BSWAP0:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD0]])
; CHECK-NEXT: [[BSWAP1:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD1]])
; CHECK-NEXT: [[BSWAP2:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD2]])
; CHECK-NEXT: [[BSWAP3:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD3]])
; CHECK-NEXT: [[BSWAP4:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD4]])
; CHECK-NEXT: [[BSWAP5:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD5]])
; CHECK-NEXT: [[BSWAP6:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD6]])
; CHECK-NEXT: [[BSWAP7:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD7]])
; CHECK-NEXT: [[BSWAP8:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD8]])
; CHECK-NEXT: [[BSWAP9:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD9]])
; CHECK-NEXT: [[BSWAP10:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD10]])
; CHECK-NEXT: [[BSWAP11:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD11]])
; CHECK-NEXT: [[BSWAP12:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD12]])
; CHECK-NEXT: [[BSWAP13:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD13]])
; CHECK-NEXT: [[BSWAP14:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD14]])
; CHECK-NEXT: [[BSWAP15:%.*]] = call i16 @llvm.bswap.i16(i16 [[LD15]])
; CHECK-NEXT: store i16 [[BSWAP0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
; CHECK-NEXT: store i16 [[BSWAP1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
; CHECK-NEXT: store i16 [[BSWAP2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
; CHECK-NEXT: store i16 [[BSWAP3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
; CHECK-NEXT: store i16 [[BSWAP4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
; CHECK-NEXT: store i16 [[BSWAP5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
; CHECK-NEXT: store i16 [[BSWAP6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
; CHECK-NEXT: store i16 [[BSWAP7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
; CHECK-NEXT: store i16 [[BSWAP8]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8), align 2
; CHECK-NEXT: store i16 [[BSWAP9]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 9), align 2
; CHECK-NEXT: store i16 [[BSWAP10]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 10), align 2
; CHECK-NEXT: store i16 [[BSWAP11]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 11), align 2
; CHECK-NEXT: store i16 [[BSWAP12]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 12), align 2
; CHECK-NEXT: store i16 [[BSWAP13]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 13), align 2
; CHECK-NEXT: store i16 [[BSWAP14]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 14), align 2
; CHECK-NEXT: store i16 [[BSWAP15]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 15), align 2
; CHECK-NEXT: ret void
; ;
%ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2 %ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
%ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2 %ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
%ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2 %ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
%ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2 %ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
%ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2 %ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
%ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2 %ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
%ld6 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2 %ld6 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
▲ Show 20 Lines • Show All 45 Lines • Show Last 20 Lines