[llvm-branch-commits] [llvm] da72df4 - [SVE] Fix bug in SVEIntrinsicOpts::optimizePTest
Hans Wennborg via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Aug 18 01:07:17 PDT 2020
Author: David Sherwood
Date: 2020-08-18T10:02:50+02:00
New Revision: da72df44005c909e40af867843168c4c456baa3d
URL: https://github.com/llvm/llvm-project/commit/da72df44005c909e40af867843168c4c456baa3d
DIFF: https://github.com/llvm/llvm-project/commit/da72df44005c909e40af867843168c4c456baa3d.diff
LOG: [SVE] Fix bug in SVEIntrinsicOpts::optimizePTest
The code wasn't taking into account that the two operands
passed to ptest could be identical and was trying to erase
them twice.
Differential Revision: https://reviews.llvm.org/D85892
(cherry picked from commit 6c7957c9901714b7ad0a8d2743a8c431b57fd0c9)
Added:
Modified:
llvm/lib/Target/AArch64/SVEIntrinsicOpts.cpp
llvm/test/CodeGen/AArch64/sve-intrinsic-opts-ptest.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/SVEIntrinsicOpts.cpp b/llvm/lib/Target/AArch64/SVEIntrinsicOpts.cpp
index 74fe0cdd1ea7..0245dd1d611a 100644
--- a/llvm/lib/Target/AArch64/SVEIntrinsicOpts.cpp
+++ b/llvm/lib/Target/AArch64/SVEIntrinsicOpts.cpp
@@ -160,7 +160,7 @@ bool SVEIntrinsicOpts::optimizePTest(IntrinsicInst *I) {
I->eraseFromParent();
if (Op1->use_empty())
Op1->eraseFromParent();
- if (Op2->use_empty())
+ if (Op1 != Op2 && Op2->use_empty())
Op2->eraseFromParent();
return true;
diff --git a/llvm/test/CodeGen/AArch64/sve-intrinsic-opts-ptest.ll b/llvm/test/CodeGen/AArch64/sve-intrinsic-opts-ptest.ll
index 191fddacffd1..9af34676b342 100644
--- a/llvm/test/CodeGen/AArch64/sve-intrinsic-opts-ptest.ll
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsic-opts-ptest.ll
@@ -44,6 +44,16 @@ define i1 @ptest_first(<vscale x 4 x i1> %a) {
ret i1 %out
}
+define i1 @ptest_first_same_ops(<vscale x 2 x i1> %a) {
+; OPT-LABEL: ptest_first_same_ops
+; OPT: %[[OUT:.*]] = call i1 @llvm.aarch64.sve.ptest.first.nxv2i1(<vscale x 2 x i1> %a, <vscale x 2 x i1> %a)
+; OPT-NOT: convert
+; OPT-NEXT: ret i1 %[[OUT]]
+ %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %a)
+ %2 = tail call i1 @llvm.aarch64.sve.ptest.first.nxv16i1(<vscale x 16 x i1> %1, <vscale x 16 x i1> %1)
+ ret i1 %2
+}
+
define i1 @ptest_last(<vscale x 8 x i1> %a) {
; OPT-LABEL: ptest_last
; OPT: %mask = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 0)
More information about the llvm-branch-commits
mailing list