diff options
| author | Craig Topper <craig.topper@intel.com> | 2019-08-17 22:46:15 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2019-08-17 22:46:15 +0000 |
| commit | 31f829f0cd4dbc00066ff0c5929f07a46e204102 (patch) | |
| tree | a07e4f4a153ba98107ee68b4be3ae97e73f542fc | |
| parent | a8dad5c79b64523e4c39d2ee32739f8894d40129 (diff) | |
| download | bcm5719-llvm-31f829f0cd4dbc00066ff0c5929f07a46e204102.tar.gz bcm5719-llvm-31f829f0cd4dbc00066ff0c5929f07a46e204102.zip | |
[X86] Add a one use check to the combineStore code that handles v16i16->v16i8 truncate+store by extending to v16i32 and then emitting a v16i32->v16i8 truncstore.
This prevent us from emitting a separate truncate and a truncating
store instruction.
llvm-svn: 369200
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 2 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512-trunc.ll | 20 |
2 files changed, 21 insertions, 1 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 207782bc80b..37d84892791 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -39581,7 +39581,7 @@ static SDValue combineStore(SDNode *N, SelectionDAG &DAG, St->getValue().getOpcode() == ISD::TRUNCATE && St->getValue().getOperand(0).getValueType() == MVT::v16i16 && TLI.isTruncStoreLegal(MVT::v16i32, MVT::v16i8) && - !DCI.isBeforeLegalizeOps()) { + St->getValue().hasOneUse() && !DCI.isBeforeLegalizeOps()) { SDValue Ext = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::v16i32, St->getValue()); return DAG.getTruncStore(St->getChain(), dl, Ext, St->getBasePtr(), MVT::v16i8, St->getMemOperand()); diff --git a/llvm/test/CodeGen/X86/avx512-trunc.ll b/llvm/test/CodeGen/X86/avx512-trunc.ll index 572b6065b2f..45384a4df4f 100644 --- a/llvm/test/CodeGen/X86/avx512-trunc.ll +++ b/llvm/test/CodeGen/X86/avx512-trunc.ll @@ -514,6 +514,26 @@ define void @trunc_wb_256_mem(<16 x i16> %i, <16 x i8>* %res) #0 { ret void } +define <16 x i8> @trunc_wb_256_mem_and_ret(<16 x i16> %i, <16 x i8>* %res) #0 { +; KNL-LABEL: trunc_wb_256_mem_and_ret: +; KNL: ## %bb.0: +; KNL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero +; KNL-NEXT: vpmovdb %zmm0, %xmm0 +; KNL-NEXT: vmovdqa %xmm0, (%rdi) +; KNL-NEXT: vzeroupper +; KNL-NEXT: retq +; +; SKX-LABEL: trunc_wb_256_mem_and_ret: +; SKX: ## %bb.0: +; SKX-NEXT: vpmovwb %ymm0, %xmm0 +; SKX-NEXT: vmovdqa %xmm0, (%rdi) +; SKX-NEXT: vzeroupper +; SKX-NEXT: retq + %x = trunc <16 x i16> %i to <16 x i8> + store <16 x i8> %x, <16 x i8>* %res + ret <16 x i8> %x +} + define <8 x i8> @trunc_wb_128(<8 x i16> %i) #0 { ; ALL-LABEL: trunc_wb_128: ; ALL: ## %bb.0: |
