xref: /llvm-project/llvm/test/CodeGen/X86/wide-fma-contraction.ll (revision 4aa7b9cc963ae0df6a1411936b879d3c979d7299)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mcpu=bdver2 -mattr=-fma -mtriple=i686-apple-darwin < %s | FileCheck %s
3; RUN: llc -mcpu=bdver2 -mattr=-fma,-fma4 -mtriple=i686-apple-darwin < %s | FileCheck %s --check-prefix=CHECK-NOFMA
4
5; CHECK-LABEL: fmafunc
6; CHECK-NOFMA-LABEL: fmafunc
7define <16 x float> @fmafunc(<16 x float> %a, <16 x float> %b, <16 x float> %c) {
8; CHECK-LABEL: fmafunc:
9; CHECK:       ## %bb.0:
10; CHECK-NEXT:    pushl %ebp
11; CHECK-NEXT:    .cfi_def_cfa_offset 8
12; CHECK-NEXT:    .cfi_offset %ebp, -8
13; CHECK-NEXT:    movl %esp, %ebp
14; CHECK-NEXT:    .cfi_def_cfa_register %ebp
15; CHECK-NEXT:    andl $-32, %esp
16; CHECK-NEXT:    subl $32, %esp
17; CHECK-NEXT:    vfmaddps {{.*#+}} ymm0 = (ymm0 * ymm2) + mem
18; CHECK-NEXT:    vfmaddps {{.*#+}} ymm1 = (ymm1 * ymm3) + mem
19; CHECK-NEXT:    movl %ebp, %esp
20; CHECK-NEXT:    popl %ebp
21; CHECK-NEXT:    retl
22;
23; CHECK-NOFMA-LABEL: fmafunc:
24; CHECK-NOFMA:       ## %bb.0:
25; CHECK-NOFMA-NEXT:    pushl %ebp
26; CHECK-NOFMA-NEXT:    .cfi_def_cfa_offset 8
27; CHECK-NOFMA-NEXT:    .cfi_offset %ebp, -8
28; CHECK-NOFMA-NEXT:    movl %esp, %ebp
29; CHECK-NOFMA-NEXT:    .cfi_def_cfa_register %ebp
30; CHECK-NOFMA-NEXT:    andl $-32, %esp
31; CHECK-NOFMA-NEXT:    subl $32, %esp
32; CHECK-NOFMA-NEXT:    vmulps %ymm2, %ymm0, %ymm0
33; CHECK-NOFMA-NEXT:    vmulps %ymm3, %ymm1, %ymm1
34; CHECK-NOFMA-NEXT:    vaddps 8(%ebp), %ymm0, %ymm0
35; CHECK-NOFMA-NEXT:    vaddps 40(%ebp), %ymm1, %ymm1
36; CHECK-NOFMA-NEXT:    movl %ebp, %esp
37; CHECK-NOFMA-NEXT:    popl %ebp
38; CHECK-NOFMA-NEXT:    retl
39
40
41  %ret = tail call <16 x float> @llvm.fmuladd.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c)
42  ret <16 x float> %ret
43}
44
45declare <16 x float> @llvm.fmuladd.v16f32(<16 x float>, <16 x float>, <16 x float>) nounwind readnone
46