af1e59aea2
Need to correctly track reduced values with multiple uses in the same reduction emission attempt. Otherwise, the number of the reuses might be calculated incorrectly, and may cause compiler crash. Fixes https://github.com/llvm/llvm-project/issues/107037
48 lines
1.5 KiB
LLVM
48 lines
1.5 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
|
|
|
|
define i8 @test() {
|
|
; CHECK-LABEL: define i8 @test() {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 0 to i8
|
|
; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 0 to i8
|
|
; CHECK-NEXT: [[TMP2:%.*]] = trunc i32 0 to i8
|
|
; CHECK-NEXT: [[TMP3:%.*]] = trunc i32 0 to i8
|
|
; CHECK-NEXT: [[TMP4:%.*]] = call i8 @llvm.vector.reduce.or.v8i8(<8 x i8> zeroinitializer)
|
|
; CHECK-NEXT: [[OP_RDX:%.*]] = or i8 [[TMP4]], [[TMP0]]
|
|
; CHECK-NEXT: [[OP_RDX1:%.*]] = or i8 [[OP_RDX]], [[TMP2]]
|
|
; CHECK-NEXT: [[OP_RDX2:%.*]] = or i8 [[OP_RDX1]], [[TMP0]]
|
|
; CHECK-NEXT: [[OP_RDX3:%.*]] = or i8 [[OP_RDX2]], [[TMP1]]
|
|
; CHECK-NEXT: [[OP_RDX4:%.*]] = or i8 [[OP_RDX3]], [[TMP3]]
|
|
; CHECK-NEXT: ret i8 [[OP_RDX4]]
|
|
;
|
|
entry:
|
|
%0 = trunc i32 0 to i8
|
|
%1 = add i8 %0, 0
|
|
%2 = add i8 %0, 0
|
|
%3 = add i8 %0, 0
|
|
%4 = add i8 %0, 0
|
|
%5 = trunc i32 0 to i8
|
|
%6 = or i8 %5, %0
|
|
%7 = or i8 %6, %2
|
|
%8 = or i8 %7, %3
|
|
%9 = or i8 %8, %0
|
|
%10 = or i8 %9, %4
|
|
%conv4 = or i8 %10, %1
|
|
%11 = trunc i32 0 to i8
|
|
%12 = add i8 %11, 0
|
|
%conv7 = or i8 %conv4, %12
|
|
%13 = add i8 %11, 0
|
|
%14 = add i8 %11, 0
|
|
%15 = add i8 %11, 0
|
|
%16 = trunc i32 0 to i8
|
|
%17 = or i8 %13, %16
|
|
%18 = or i8 %17, %14
|
|
%19 = or i8 %18, %11
|
|
%20 = or i8 %19, %15
|
|
%conv5 = or i8 %20, %conv7
|
|
%21 = trunc i32 0 to i8
|
|
%conv6 = or i8 %21, %conv5
|
|
ret i8 %conv6
|
|
}
|