[PATCH] D42380: [X86][SSE] LowerBUILD_VECTORAsVariablePermute - fix PSHUFB source/index operand ordering

Simon Pilgrim via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Jan 22 09:09:40 PST 2018


RKSimon created this revision.
RKSimon added reviewers: zvi, spatel, craig.topper.

As detailed in https://reviews.llvm.org/rL317463, PSHUFB (like most variable shuffle instructions) uses Op[0] for the source vector and Op[1] for the shuffle index vector, VPERMV works in reverse which is probably where the confusion comes from.


Repository:
  rL LLVM

https://reviews.llvm.org/D42380

Files:
  lib/Target/X86/X86ISelLowering.cpp
  test/CodeGen/X86/var-permute-128.ll


Index: test/CodeGen/X86/var-permute-128.ll
===================================================================
--- test/CodeGen/X86/var-permute-128.ll
+++ test/CodeGen/X86/var-permute-128.ll
@@ -207,13 +207,12 @@
 define <16 x i8> @var_shuffle_v16i8(<16 x i8> %v, <16 x i8> %indices) nounwind {
 ; SSSE3-LABEL: var_shuffle_v16i8:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    pshufb %xmm0, %xmm1
-; SSSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; AVX-LABEL: var_shuffle_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %index0 = extractelement <16 x i8> %indices, i32 0
   %index1 = extractelement <16 x i8> %indices, i32 1
Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -7903,8 +7903,9 @@
         DAG.getNode(ISD::INSERT_SUBVECTOR, SDLoc(SrcVec), VT, DAG.getUNDEF(VT),
                     SrcVec, DAG.getIntPtrConstant(0, SDLoc(SrcVec)));
   }
-  return DAG.getNode(VT == MVT::v16i8 ? X86ISD::PSHUFB : X86ISD::VPERMV,
-                     SDLoc(V), VT, IndicesVec, SrcVec);
+  if (VT == MVT::v16i8)
+    return DAG.getNode(X86ISD::PSHUFB, SDLoc(V), VT, SrcVec, IndicesVec);
+  return DAG.getNode(X86ISD::VPERMV, SDLoc(V), VT, IndicesVec, SrcVec);
 }
 
 SDValue


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D42380.130909.patch
Type: text/x-patch
Size: 1479 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180122/97bcc25b/attachment.bin>


More information about the llvm-commits mailing list