[PATCH] D18204: AVX512BW : Fix SRA v64i8 lowering.
Igor Breger via llvm-commits
llvm-commits at lists.llvm.org
Wed Mar 16 00:35:30 PDT 2016
igorb created this revision.
igorb added reviewers: AsafBadouh, delena.
igorb added a subscriber: llvm-commits.
igorb set the repository for this revision to rL LLVM.
AVX512BW: Fix SRA v64i8 lowering. Use PCMPGTM (cmp result in k register) for 512bit vector because PCMPGT supported only for 128/256bit.
Repository:
rL LLVM
http://reviews.llvm.org/D18204
Files:
lib/Target/X86/X86ISelLowering.cpp
test/CodeGen/X86/vector-shift-ashr-512.ll
Index: test/CodeGen/X86/vector-shift-ashr-512.ll
===================================================================
--- test/CodeGen/X86/vector-shift-ashr-512.ll
+++ test/CodeGen/X86/vector-shift-ashr-512.ll
@@ -376,3 +376,21 @@
%shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
ret <64 x i8> %shift
}
+
+define <64 x i8> @ashr_const7_v64i8(<64 x i8> %a) {
+; AVX512DQ-LABEL: ashr_const7_v64i8:
+; AVX512DQ: ## BB#0:
+; AVX512DQ-NEXT: vpxor %ymm2, %ymm2, %ymm2
+; AVX512DQ-NEXT: vpcmpgtb %ymm0, %ymm2, %ymm0
+; AVX512DQ-NEXT: vpcmpgtb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT: retq
+;
+; AVX512BW-LABEL: ashr_const7_v64i8:
+; AVX512BW: ## BB#0:
+; AVX512BW-NEXT: vpxord %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT: vpcmpgtb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT: vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT: retq
+ %res = ashr <64 x i8> %a, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+ ret <64 x i8> %res
+}
Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -19233,6 +19233,11 @@
// ashr(R, 7) === cmp_slt(R, 0)
if (Op.getOpcode() == ISD::SRA && ShiftAmt == 7) {
SDValue Zeros = getZeroVector(VT, Subtarget, DAG, dl);
+ if (VT.is512BitVector()) {
+ assert(VT == MVT::v64i8 && "Unexpected element type!");
+ SDValue CMP = DAG.getNode(X86ISD::PCMPGTM, dl, MVT::v64i1, Zeros, R);
+ return DAG.getNode(ISD::SIGN_EXTEND, dl, VT, CMP);
+ }
return DAG.getNode(X86ISD::PCMPGT, dl, VT, Zeros, R);
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D18204.50799.patch
Type: text/x-patch
Size: 2364 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160316/12cc4108/attachment.bin>
More information about the llvm-commits
mailing list