xref: /llvm-project/llvm/test/CodeGen/X86/combine-sext-in-reg.ll (revision 25528d6de70e98683722e28655d8568d5f09b5c7)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX
4
5; fold sextinreg(zext) -> sext
6define <4 x i64> @sextinreg_zext_v16i8_4i64(<16 x i8> %a0) {
7; SSE-LABEL: sextinreg_zext_v16i8_4i64:
8; SSE:       # %bb.0:
9; SSE-NEXT:    pmovsxbq %xmm0, %xmm2
10; SSE-NEXT:    psrld $16, %xmm0
11; SSE-NEXT:    pmovsxbq %xmm0, %xmm1
12; SSE-NEXT:    movdqa %xmm2, %xmm0
13; SSE-NEXT:    retq
14;
15; AVX-LABEL: sextinreg_zext_v16i8_4i64:
16; AVX:       # %bb.0:
17; AVX-NEXT:    vpmovsxbq %xmm0, %ymm0
18; AVX-NEXT:    retq
19  %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
20  %2 = zext <4 x i8> %1 to <4 x i64>
21  %3 = shl <4 x i64> %2, <i64 56, i64 56, i64 56, i64 56>
22  %4 = ashr <4 x i64> %3, <i64 56, i64 56, i64 56, i64 56>
23  ret <4 x i64> %4
24}
25
26; fold sextinreg(zext(sext)) -> sext
27define <4 x i64> @sextinreg_zext_sext_v16i8_4i64(<16 x i8> %a0) {
28; SSE-LABEL: sextinreg_zext_sext_v16i8_4i64:
29; SSE:       # %bb.0:
30; SSE-NEXT:    pmovsxbq %xmm0, %xmm2
31; SSE-NEXT:    psrld $16, %xmm0
32; SSE-NEXT:    pmovsxbq %xmm0, %xmm1
33; SSE-NEXT:    movdqa %xmm2, %xmm0
34; SSE-NEXT:    retq
35;
36; AVX-LABEL: sextinreg_zext_sext_v16i8_4i64:
37; AVX:       # %bb.0:
38; AVX-NEXT:    vpmovsxbq %xmm0, %ymm0
39; AVX-NEXT:    retq
40  %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
41  %2 = sext <4 x i8> %1 to <4 x i32>
42  %3 = zext <4 x i32> %2 to <4 x i64>
43  %4 = shl <4 x i64> %3, <i64 32, i64 32, i64 32, i64 32>
44  %5 = ashr <4 x i64> %4, <i64 32, i64 32, i64 32, i64 32>
45  ret <4 x i64> %5
46}
47