xref: /llvm-project/llvm/test/Transforms/VectorCombine/X86/load-extractelement-scalarization.ll (revision 5ed6229019de43df0ff4b3e73097781e0f1a6651)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=sse2 | FileCheck %s
3; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=avx2 | FileCheck %s
4
5; infinite loop if we add the erased instructions to the work list in the wrong order.
6define void @multiple_extract(ptr %p) {
7; CHECK-LABEL: @multiple_extract(
8; CHECK-NEXT:    [[VP:%.*]] = load ptr, ptr [[P:%.*]], align 8
9; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <2 x i32>, ptr [[VP]], i32 0, i64 0
10; CHECK-NEXT:    [[E0:%.*]] = load i32, ptr [[TMP1]], align 16
11; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds <2 x i32>, ptr [[VP]], i32 0, i64 1
12; CHECK-NEXT:    [[E1:%.*]] = load i32, ptr [[TMP2]], align 4
13; CHECK-NEXT:    store i32 [[E0]], ptr [[P]], align 4
14; CHECK-NEXT:    [[P1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i64 4
15; CHECK-NEXT:    store i32 [[E1]], ptr [[P1]], align 4
16; CHECK-NEXT:    ret void
17;
18  %vp = load ptr, ptr %p, align 8
19  %v = load <2 x i32>, ptr %vp, align 16
20  %e0 = extractelement <2 x i32> %v, i64 0
21  %e1 = extractelement <2 x i32> %v, i64 1
22  store i32 %e0, ptr %p, align 4
23  %p1 = getelementptr inbounds nuw i8, ptr %p, i64 4
24  store i32 %e1, ptr %p1, align 4
25  ret void
26}
27