• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=corei7-avx -mattr=+fma4 | FileCheck %s
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -mattr=+avx,-fma | FileCheck %s
4
5; VFMADD
6define < 4 x float > @test_x86_fma4_vfmadd_ss_load(< 4 x float > %a0, < 4 x float > %a1, float* %a2) {
7; CHECK-LABEL: test_x86_fma4_vfmadd_ss_load:
8; CHECK:       # %bb.0:
9; CHECK-NEXT:    vfmaddss (%rdi), %xmm1, %xmm0, %xmm0
10; CHECK-NEXT:    retq
11  %x = load float , float *%a2
12  %y = insertelement <4 x float> undef, float %x, i32 0
13  %res = call < 4 x float > @llvm.x86.fma4.vfmadd.ss(< 4 x float > %a0, < 4 x float > %a1, < 4 x float > %y)
14  ret < 4 x float > %res
15}
16define < 4 x float > @test_x86_fma4_vfmadd_ss_load2(< 4 x float > %a0, float* %a1, < 4 x float > %a2) {
17; CHECK-LABEL: test_x86_fma4_vfmadd_ss_load2:
18; CHECK:       # %bb.0:
19; CHECK-NEXT:    vfmaddss %xmm1, (%rdi), %xmm0, %xmm0
20; CHECK-NEXT:    retq
21  %x = load float , float *%a1
22  %y = insertelement <4 x float> undef, float %x, i32 0
23  %res = call < 4 x float > @llvm.x86.fma4.vfmadd.ss(< 4 x float > %a0, < 4 x float > %y, < 4 x float > %a2)
24  ret < 4 x float > %res
25}
26
27declare < 4 x float > @llvm.x86.fma4.vfmadd.ss(< 4 x float >, < 4 x float >, < 4 x float >) nounwind readnone
28
29define < 2 x double > @test_x86_fma4_vfmadd_sd_load(< 2 x double > %a0, < 2 x double > %a1, double* %a2) {
30; CHECK-LABEL: test_x86_fma4_vfmadd_sd_load:
31; CHECK:       # %bb.0:
32; CHECK-NEXT:    vfmaddsd (%rdi), %xmm1, %xmm0, %xmm0
33; CHECK-NEXT:    retq
34  %x = load double , double *%a2
35  %y = insertelement <2 x double> undef, double %x, i32 0
36  %res = call < 2 x double > @llvm.x86.fma4.vfmadd.sd(< 2 x double > %a0, < 2 x double > %a1, < 2 x double > %y)
37  ret < 2 x double > %res
38}
39define < 2 x double > @test_x86_fma4_vfmadd_sd_load2(< 2 x double > %a0, double* %a1, < 2 x double > %a2) {
40; CHECK-LABEL: test_x86_fma4_vfmadd_sd_load2:
41; CHECK:       # %bb.0:
42; CHECK-NEXT:    vfmaddsd %xmm1, (%rdi), %xmm0, %xmm0
43; CHECK-NEXT:    retq
44  %x = load double , double *%a1
45  %y = insertelement <2 x double> undef, double %x, i32 0
46  %res = call < 2 x double > @llvm.x86.fma4.vfmadd.sd(< 2 x double > %a0, < 2 x double > %y, < 2 x double > %a2)
47  ret < 2 x double > %res
48}
49declare < 2 x double > @llvm.x86.fma4.vfmadd.sd(< 2 x double >, < 2 x double >, < 2 x double >) nounwind readnone
50define < 4 x float > @test_x86_fma_vfmadd_ps_load(< 4 x float > %a0, < 4 x float > %a1, < 4 x float >* %a2) {
51; CHECK-LABEL: test_x86_fma_vfmadd_ps_load:
52; CHECK:       # %bb.0:
53; CHECK-NEXT:    vfmaddps (%rdi), %xmm1, %xmm0, %xmm0
54; CHECK-NEXT:    retq
55  %x = load <4 x float>, <4 x float>* %a2
56  %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %a0, < 4 x float > %a1, < 4 x float > %x)
57  ret < 4 x float > %res
58}
59define < 4 x float > @test_x86_fma_vfmadd_ps_load2(< 4 x float > %a0, < 4 x float >* %a1, < 4 x float > %a2) {
60; CHECK-LABEL: test_x86_fma_vfmadd_ps_load2:
61; CHECK:       # %bb.0:
62; CHECK-NEXT:    vfmaddps %xmm1, (%rdi), %xmm0, %xmm0
63; CHECK-NEXT:    retq
64  %x = load <4 x float>, <4 x float>* %a1
65  %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %a0, < 4 x float > %x, < 4 x float > %a2)
66  ret < 4 x float > %res
67}
68declare < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float >, < 4 x float >, < 4 x float >) nounwind readnone
69
70; To test execution dependency
71define < 4 x float > @test_x86_fma_vfmadd_ps_load3(< 4 x float >* %a0, < 4 x float >* %a1, < 4 x float > %a2) {
72; CHECK-LABEL: test_x86_fma_vfmadd_ps_load3:
73; CHECK:       # %bb.0:
74; CHECK-NEXT:    vmovaps (%rdi), %xmm1
75; CHECK-NEXT:    vfmaddps %xmm0, (%rsi), %xmm1, %xmm0
76; CHECK-NEXT:    retq
77  %x = load <4 x float>, <4 x float>* %a0
78  %y = load <4 x float>, <4 x float>* %a1
79  %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %x, < 4 x float > %y, < 4 x float > %a2)
80  ret < 4 x float > %res
81}
82
83define < 2 x double > @test_x86_fma_vfmadd_pd_load(< 2 x double > %a0, < 2 x double > %a1, < 2 x double >* %a2) {
84; CHECK-LABEL: test_x86_fma_vfmadd_pd_load:
85; CHECK:       # %bb.0:
86; CHECK-NEXT:    vfmaddpd (%rdi), %xmm1, %xmm0, %xmm0
87; CHECK-NEXT:    retq
88  %x = load <2 x double>, <2 x double>* %a2
89  %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %a0, < 2 x double > %a1, < 2 x double > %x)
90  ret < 2 x double > %res
91}
92define < 2 x double > @test_x86_fma_vfmadd_pd_load2(< 2 x double > %a0, < 2 x double >* %a1, < 2 x double > %a2) {
93; CHECK-LABEL: test_x86_fma_vfmadd_pd_load2:
94; CHECK:       # %bb.0:
95; CHECK-NEXT:    vfmaddpd %xmm1, (%rdi), %xmm0, %xmm0
96; CHECK-NEXT:    retq
97  %x = load <2 x double>, <2 x double>* %a1
98  %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %a0, < 2 x double > %x, < 2 x double > %a2)
99  ret < 2 x double > %res
100}
101declare < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double >, < 2 x double >, < 2 x double >) nounwind readnone
102
103; To test execution dependency
104define < 2 x double > @test_x86_fma_vfmadd_pd_load3(< 2 x double >* %a0, < 2 x double >* %a1, < 2 x double > %a2) {
105; CHECK-LABEL: test_x86_fma_vfmadd_pd_load3:
106; CHECK:       # %bb.0:
107; CHECK-NEXT:    vmovapd (%rdi), %xmm1
108; CHECK-NEXT:    vfmaddpd %xmm0, (%rsi), %xmm1, %xmm0
109; CHECK-NEXT:    retq
110  %x = load <2 x double>, <2 x double>* %a0
111  %y = load <2 x double>, <2 x double>* %a1
112  %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %x, < 2 x double > %y, < 2 x double > %a2)
113  ret < 2 x double > %res
114}
115
116