llvm/test/Transforms/SLPVectorizer/X86/lhs-constant-non-cummutative.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S --passes=slp-vectorizer -slp-threshold=-50 -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s

define void @or_lhs_to_shl(ptr %p, ptr %s) {
; CHECK-LABEL: define void @or_lhs_to_shl(
; CHECK-SAME: ptr [[P:%.*]], ptr [[S:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x i16>, ptr [[P]], align 2
; CHECK-NEXT:    [[TMP1:%.*]] = shl <4 x i16> [[TMP0]], <i16 1, i16 2, i16 3, i16 0>
; CHECK-NEXT:    store <4 x i16> [[TMP1]], ptr [[S]], align 2
; CHECK-NEXT:    ret void
;
entry:
  %p1 = getelementptr i16, ptr %p, i64 1
  %p2 = getelementptr i16, ptr %p, i64 2
  %p3 = getelementptr i16, ptr %p, i64 3

  %l0 = load i16, ptr %p, align 2
  %l1 = load i16, ptr %p1, align 2
  %l2 = load i16, ptr %p2, align 2
  %l3 = load i16, ptr %p3, align 2

  ; 3 shl instructions force the vectorizer to choose shl as the target
  %op0 = shl i16 %l0, 1
  %op1 = shl i16 %l1, 2
  %op2 = shl i16 %l2, 3
  %op3 = or i16 0, %l3 ; The buggy instruction

  %s1 = getelementptr i16, ptr %s, i64 1
  %s2 = getelementptr i16, ptr %s, i64 2
  %s3 = getelementptr i16, ptr %s, i64 3

  store i16 %op0, ptr %s, align 2
  store i16 %op1, ptr %s1, align 2
  store i16 %op2, ptr %s2, align 2
  store i16 %op3, ptr %s3, align 2

  ret void
}