[llvm] r312894 - Fixed a bug in splitting Scatter operation in the Type Legalizer.
Elena Demikhovsky via llvm-commits
llvm-commits at lists.llvm.org
Sun Sep 10 23:18:15 PDT 2017
Author: delena
Date: Sun Sep 10 23:18:15 2017
New Revision: 312894
URL: http://llvm.org/viewvc/llvm-project?rev=312894&view=rev
Log:
Fixed a bug in splitting Scatter operation in the Type Legalizer.
After the split of the Scatter operation, the order of the new instructions is well defined - Lo goes before Hi. Otherwise the semantic of Scatter (from LSB to MSB) is broken.
I'm chaining 2 nodes to prevent reordering.
Differential Revision https://reviews.llvm.org/D37670
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
llvm/trunk/test/CodeGen/X86/scatter-schedule.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp?rev=312894&r1=312893&r2=312894&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp Sun Sep 10 23:18:15 2017
@@ -1970,13 +1970,12 @@ SDValue DAGTypeLegalizer::SplitVecOp_MSC
MachineMemOperand::MOStore, HiMemVT.getStoreSize(),
Alignment, N->getAAInfo(), N->getRanges());
- SDValue OpsHi[] = {Ch, DataHi, MaskHi, Ptr, IndexHi};
- Hi = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataHi.getValueType(),
- DL, OpsHi, MMO);
-
- // Build a factor node to remember that this store is independent of the
- // other one.
- return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
+ // The order of the Scatter operation after split is well defined. The "Hi"
+ // part comes after the "Lo". So these two operations should be chained one
+ // after another.
+ SDValue OpsHi[] = {Lo, DataHi, MaskHi, Ptr, IndexHi};
+ return DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataHi.getValueType(),
+ DL, OpsHi, MMO);
}
SDValue DAGTypeLegalizer::SplitVecOp_STORE(StoreSDNode *N, unsigned OpNo) {
Modified: llvm/trunk/test/CodeGen/X86/scatter-schedule.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/scatter-schedule.ll?rev=312894&r1=312893&r2=312894&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/scatter-schedule.ll (original)
+++ llvm/trunk/test/CodeGen/X86/scatter-schedule.ll Sun Sep 10 23:18:15 2017
@@ -5,16 +5,15 @@ target triple = "x86_64-unknown-linux-gn
; This test checks the order of scatter operations after split.
; The right order is "from LSB to MSB", otherwise the semantic is broken.
-; The submitted version of the test demonstrates the bug.
define void @test(i64 %x272, <16 x i32*> %x335, <16 x i32> %x270) {
; CHECK-LABEL: test:
; CHECK: # BB#0:
-; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
; CHECK-NEXT: kxnorw %k0, %k0, %k1
; CHECK-NEXT: kxnorw %k0, %k0, %k2
-; CHECK-NEXT: vpscatterqd %ymm3, (,%zmm1) {%k2}
-; CHECK-NEXT: vpscatterqd %ymm2, (,%zmm0) {%k1}
+; CHECK-NEXT: vpscatterqd %ymm2, (,%zmm0) {%k2}
+; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm0
+; CHECK-NEXT: vpscatterqd %ymm0, (,%zmm1) {%k1}
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
call void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32> %x270, <16 x i32*> %x335, i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
More information about the llvm-commits
mailing list