From: Craig Topper <craig.topper@intel.com>
Date: Sat, 17 Aug 2019 22:46:15 +0000 (+0000)
Subject: [X86] Add a one use check to the combineStore code that handles v16i16->v16i8 truncat... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=b90d51eebbe53789773168228e50694076daf980;p=llvm

[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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369200 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 207782bc80b..37d84892791 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/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/test/CodeGen/X86/avx512-trunc.ll b/test/CodeGen/X86/avx512-trunc.ll
index 572b6065b2f..45384a4df4f 100644
--- a/test/CodeGen/X86/avx512-trunc.ll
+++ b/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: