1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4 -fp-contract=fast | FileCheck %s --check-prefix=FMA4
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fp-contract=fast | FileCheck %s --check-prefix=AVX512
7 define <16 x float> @test_x86_fmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
8 ; FMA-LABEL: test_x86_fmadd_ps_y_wide:
10 ; FMA-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0
11 ; FMA-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1
14 ; FMA4-LABEL: test_x86_fmadd_ps_y_wide:
16 ; FMA4-NEXT: vfmaddps %ymm4, %ymm2, %ymm0, %ymm0
17 ; FMA4-NEXT: vfmaddps %ymm5, %ymm3, %ymm1, %ymm1
20 ; AVX512-LABEL: test_x86_fmadd_ps_y_wide:
22 ; AVX512-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0
24 %x = fmul <16 x float> %a0, %a1
25 %res = fadd <16 x float> %x, %a2
29 define <16 x float> @test_x86_fmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
30 ; FMA-LABEL: test_x86_fmsub_ps_y_wide:
32 ; FMA-NEXT: vfmsub213ps %ymm4, %ymm2, %ymm0
33 ; FMA-NEXT: vfmsub213ps %ymm5, %ymm3, %ymm1
36 ; FMA4-LABEL: test_x86_fmsub_ps_y_wide:
38 ; FMA4-NEXT: vfmsubps %ymm4, %ymm2, %ymm0, %ymm0
39 ; FMA4-NEXT: vfmsubps %ymm5, %ymm3, %ymm1, %ymm1
42 ; AVX512-LABEL: test_x86_fmsub_ps_y_wide:
44 ; AVX512-NEXT: vfmsub213ps %zmm2, %zmm1, %zmm0
46 %x = fmul <16 x float> %a0, %a1
47 %res = fsub <16 x float> %x, %a2
51 define <16 x float> @test_x86_fnmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
52 ; FMA-LABEL: test_x86_fnmadd_ps_y_wide:
54 ; FMA-NEXT: vfnmadd213ps %ymm4, %ymm2, %ymm0
55 ; FMA-NEXT: vfnmadd213ps %ymm5, %ymm3, %ymm1
58 ; FMA4-LABEL: test_x86_fnmadd_ps_y_wide:
60 ; FMA4-NEXT: vfnmaddps %ymm4, %ymm2, %ymm0, %ymm0
61 ; FMA4-NEXT: vfnmaddps %ymm5, %ymm3, %ymm1, %ymm1
64 ; AVX512-LABEL: test_x86_fnmadd_ps_y_wide:
66 ; AVX512-NEXT: vfnmadd213ps %zmm2, %zmm1, %zmm0
68 %x = fmul <16 x float> %a0, %a1
69 %res = fsub <16 x float> %a2, %x
73 define <16 x float> @test_x86_fnmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
74 ; FMA-LABEL: test_x86_fnmsub_ps_y_wide:
76 ; FMA-NEXT: vfnmsub213ps %ymm4, %ymm2, %ymm0
77 ; FMA-NEXT: vfnmsub213ps %ymm5, %ymm3, %ymm1
80 ; FMA4-LABEL: test_x86_fnmsub_ps_y_wide:
82 ; FMA4-NEXT: vfnmsubps %ymm4, %ymm2, %ymm0, %ymm0
83 ; FMA4-NEXT: vfnmsubps %ymm5, %ymm3, %ymm1, %ymm1
86 ; AVX512-LABEL: test_x86_fnmsub_ps_y_wide:
88 ; AVX512-NEXT: vfnmsub213ps %zmm2, %zmm1, %zmm0
90 %x = fmul <16 x float> %a0, %a1
91 %y = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x
92 %res = fsub <16 x float> %y, %a2
96 define <8 x double> @test_x86_fmadd_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
97 ; FMA-LABEL: test_x86_fmadd_pd_y_wide:
99 ; FMA-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0
100 ; FMA-NEXT: vfmadd213pd %ymm5, %ymm3, %ymm1
103 ; FMA4-LABEL: test_x86_fmadd_pd_y_wide:
105 ; FMA4-NEXT: vfmaddpd %ymm4, %ymm2, %ymm0, %ymm0
106 ; FMA4-NEXT: vfmaddpd %ymm5, %ymm3, %ymm1, %ymm1
109 ; AVX512-LABEL: test_x86_fmadd_pd_y_wide:
111 ; AVX512-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0
113 %x = fmul <8 x double> %a0, %a1
114 %res = fadd <8 x double> %x, %a2
115 ret <8 x double> %res
118 define <8 x double> @test_x86_fmsub_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
119 ; FMA-LABEL: test_x86_fmsub_pd_y_wide:
121 ; FMA-NEXT: vfmsub213pd %ymm4, %ymm2, %ymm0
122 ; FMA-NEXT: vfmsub213pd %ymm5, %ymm3, %ymm1
125 ; FMA4-LABEL: test_x86_fmsub_pd_y_wide:
127 ; FMA4-NEXT: vfmsubpd %ymm4, %ymm2, %ymm0, %ymm0
128 ; FMA4-NEXT: vfmsubpd %ymm5, %ymm3, %ymm1, %ymm1
131 ; AVX512-LABEL: test_x86_fmsub_pd_y_wide:
133 ; AVX512-NEXT: vfmsub213pd %zmm2, %zmm1, %zmm0
135 %x = fmul <8 x double> %a0, %a1
136 %res = fsub <8 x double> %x, %a2
137 ret <8 x double> %res