Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / 46a804cfd937211b6a8e07e1daa48f104e0c948f / . / llvm / test / CodeGen / X86 / fmf-flags.ll
blob: d7a50f808ac55116c323562c1dbbc1cfe0bf557a [file] [log] [blame]
Jatin Bhateja1a86c382017-09-21 09:53:21 +0000[diff] [blame]1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s -check-prefix=X64
3; RUN: llc < %s -mtriple=i686-unknown | FileCheck %s -check-prefix=X86
4
5declare float @llvm.sqrt.f32(float %x);
6
7define float @fast_recip_sqrt(float %x) {
8; X64-LABEL: fast_recip_sqrt:
9; X64: # BB#0:
10; X64-NEXT: sqrtss %xmm0, %xmm1
11; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
12; X64-NEXT: divss %xmm1, %xmm0
13; X64-NEXT: retq
14;
15; X86-LABEL: fast_recip_sqrt:
16; X86: # BB#0:
17; X86-NEXT: flds {{[0-9]+}}(%esp)
18; X86-NEXT: fsqrt
19; X86-NEXT: fld1
20; X86-NEXT: fdivp %st(1)
21; X86-NEXT: retl
22 %y = call fast float @llvm.sqrt.f32(float %x)
23 %z = fdiv fast float 1.0, %y
24 ret float %z
25}
26
27declare float @llvm.fmuladd.f32(float %a, float %b, float %c);
28
29define float @fast_fmuladd_opts(float %a , float %b , float %c) {
30; X64-LABEL: fast_fmuladd_opts:
31; X64: # BB#0:
32; X64-NEXT: movaps %xmm0, %xmm1
Geoff Berrybfc5fb42017-10-02 22:01:37 +0000[diff] [blame]33; X64-NEXT: addss %xmm0, %xmm1
Jatin Bhateja1a86c382017-09-21 09:53:21 +0000[diff] [blame]34; X64-NEXT: addss %xmm0, %xmm1
35; X64-NEXT: movaps %xmm1, %xmm0
36; X64-NEXT: retq
37;
38; X86-LABEL: fast_fmuladd_opts:
39; X86: # BB#0:
40; X86-NEXT: flds {{[0-9]+}}(%esp)
41; X86-NEXT: fld %st(0)
42; X86-NEXT: fadd %st(1)
43; X86-NEXT: faddp %st(1)
44; X86-NEXT: retl
45 %res = call fast float @llvm.fmuladd.f32(float %a, float 2.0, float %a)
46 ret float %res
47}
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]48
49; The multiply is strict.
50
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]51@mul1 = common global double 0.000000e+00, align 4
52
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]53define double @not_so_fast_mul_add(double %x) {
54; X64-LABEL: not_so_fast_mul_add:
55; X64: # BB#0:
56; X64-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
57; X64-NEXT: mulsd %xmm0, %xmm1
58; X64-NEXT: addsd %xmm1, %xmm0
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]59; X64-NEXT: movsd %xmm1, {{.*}}(%rip)
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]60; X64-NEXT: retq
61;
62; X86-LABEL: not_so_fast_mul_add:
63; X86: # BB#0:
64; X86-NEXT: fldl {{[0-9]+}}(%esp)
65; X86-NEXT: fld %st(0)
66; X86-NEXT: fmull {{\.LCPI.*}}
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]67; X86-NEXT: fadd %st(0), %st(1)
68; X86-NEXT: fstpl mul1
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]69; X86-NEXT: retl
70 %m = fmul double %x, 4.2
71 %a = fadd fast double %m, %x
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]72 store double %m, double* @mul1, align 4
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]73 ret double %a
74}
75
76; The sqrt is strict.
77
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]78@sqrt1 = common global float 0.000000e+00, align 4
79
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]80define float @not_so_fast_recip_sqrt(float %x) {
81; X64-LABEL: not_so_fast_recip_sqrt:
82; X64: # BB#0:
83; X64-NEXT: sqrtss %xmm0, %xmm1
84; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
85; X64-NEXT: divss %xmm1, %xmm0
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]86; X64-NEXT: movss %xmm1, {{.*}}(%rip)
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]87; X64-NEXT: retq
88;
89; X86-LABEL: not_so_fast_recip_sqrt:
90; X86: # BB#0:
91; X86-NEXT: flds {{[0-9]+}}(%esp)
92; X86-NEXT: fsqrt
93; X86-NEXT: fld1
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]94; X86-NEXT: fdiv %st(1)
95; X86-NEXT: fxch %st(1)
96; X86-NEXT: fstps sqrt1
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]97; X86-NEXT: retl
98 %y = call float @llvm.sqrt.f32(float %x)
99 %z = fdiv fast float 1.0, %y
Jatin Bhatejac034d362017-09-22 05:48:20 +0000[diff] [blame]100 store float %y, float* @sqrt1, align 4
101 %ret = fadd float %z , 14.5
Sanjay Patel58f02af2017-09-21 17:40:58 +0000[diff] [blame]102 ret float %z
103}
104