1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4,+fma -fp-contract=fast | FileCheck %s
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma4 -fp-contract=fast | FileCheck %s --check-prefix=CHECK_FMA4
5 define <16 x float> @test_x86_fmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
6 ; CHECK-LABEL: test_x86_fmadd_ps_y_wide:
8 ; CHECK-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0
9 ; CHECK-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1
12 ; CHECK_FMA4-LABEL: test_x86_fmadd_ps_y_wide:
14 ; CHECK_FMA4-NEXT: vfmaddps %ymm4, %ymm2, %ymm0, %ymm0
15 ; CHECK_FMA4-NEXT: vfmaddps %ymm5, %ymm3, %ymm1, %ymm1
16 ; CHECK_FMA4-NEXT: retq
17 %x = fmul <16 x float> %a0, %a1
18 %res = fadd <16 x float> %x, %a2
22 define <16 x float> @test_x86_fmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
23 ; CHECK-LABEL: test_x86_fmsub_ps_y_wide:
25 ; CHECK-NEXT: vfmsub213ps %ymm4, %ymm2, %ymm0
26 ; CHECK-NEXT: vfmsub213ps %ymm5, %ymm3, %ymm1
29 ; CHECK_FMA4-LABEL: test_x86_fmsub_ps_y_wide:
31 ; CHECK_FMA4-NEXT: vfmsubps %ymm4, %ymm2, %ymm0, %ymm0
32 ; CHECK_FMA4-NEXT: vfmsubps %ymm5, %ymm3, %ymm1, %ymm1
33 ; CHECK_FMA4-NEXT: retq
34 %x = fmul <16 x float> %a0, %a1
35 %res = fsub <16 x float> %x, %a2
39 define <16 x float> @test_x86_fnmadd_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
40 ; CHECK-LABEL: test_x86_fnmadd_ps_y_wide:
42 ; CHECK-NEXT: vfnmadd213ps %ymm4, %ymm2, %ymm0
43 ; CHECK-NEXT: vfnmadd213ps %ymm5, %ymm3, %ymm1
46 ; CHECK_FMA4-LABEL: test_x86_fnmadd_ps_y_wide:
48 ; CHECK_FMA4-NEXT: vfnmaddps %ymm4, %ymm2, %ymm0, %ymm0
49 ; CHECK_FMA4-NEXT: vfnmaddps %ymm5, %ymm3, %ymm1, %ymm1
50 ; CHECK_FMA4-NEXT: retq
51 %x = fmul <16 x float> %a0, %a1
52 %res = fsub <16 x float> %a2, %x
56 define <16 x float> @test_x86_fnmsub_ps_y_wide(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
57 ; CHECK-LABEL: test_x86_fnmsub_ps_y_wide:
59 ; CHECK-NEXT: vfnmsub213ps %ymm4, %ymm2, %ymm0
60 ; CHECK-NEXT: vfnmsub213ps %ymm5, %ymm3, %ymm1
63 ; CHECK_FMA4-LABEL: test_x86_fnmsub_ps_y_wide:
65 ; CHECK_FMA4-NEXT: vfnmsubps %ymm4, %ymm2, %ymm0, %ymm0
66 ; CHECK_FMA4-NEXT: vfnmsubps %ymm5, %ymm3, %ymm1, %ymm1
67 ; CHECK_FMA4-NEXT: retq
68 %x = fmul <16 x float> %a0, %a1
69 %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
70 %res = fsub <16 x float> %y, %a2
74 define <8 x double> @test_x86_fmadd_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
75 ; CHECK-LABEL: test_x86_fmadd_pd_y_wide:
77 ; CHECK-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0
78 ; CHECK-NEXT: vfmadd213pd %ymm5, %ymm3, %ymm1
81 ; CHECK_FMA4-LABEL: test_x86_fmadd_pd_y_wide:
83 ; CHECK_FMA4-NEXT: vfmaddpd %ymm4, %ymm2, %ymm0, %ymm0
84 ; CHECK_FMA4-NEXT: vfmaddpd %ymm5, %ymm3, %ymm1, %ymm1
85 ; CHECK_FMA4-NEXT: retq
86 %x = fmul <8 x double> %a0, %a1
87 %res = fadd <8 x double> %x, %a2
91 define <8 x double> @test_x86_fmsub_pd_y_wide(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
92 ; CHECK-LABEL: test_x86_fmsub_pd_y_wide:
94 ; CHECK-NEXT: vfmsub213pd %ymm4, %ymm2, %ymm0
95 ; CHECK-NEXT: vfmsub213pd %ymm5, %ymm3, %ymm1
98 ; CHECK_FMA4-LABEL: test_x86_fmsub_pd_y_wide:
100 ; CHECK_FMA4-NEXT: vfmsubpd %ymm4, %ymm2, %ymm0, %ymm0
101 ; CHECK_FMA4-NEXT: vfmsubpd %ymm5, %ymm3, %ymm1, %ymm1
102 ; CHECK_FMA4-NEXT: retq
103 %x = fmul <8 x double> %a0, %a1
104 %res = fsub <8 x double> %x, %a2
105 ret <8 x double> %res