xref: /llvm-project/llvm/test/CodeGen/X86/strict-fsub-combines.ll (revision 8ccf1c117b0dc08f7e9c24fe98f45ebe32e95cd1)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X86
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X64
4
5; FIXME: Missing fsub(x,fneg(y)) -> fadd(x,y) fold
6define float @fneg_strict_fsub_to_strict_fadd(float %x, float %y) nounwind strictfp {
7; X86-LABEL: fneg_strict_fsub_to_strict_fadd:
8; X86:       # %bb.0:
9; X86-NEXT:    pushl %eax
10; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
11; X86-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
12; X86-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
13; X86-NEXT:    subss %xmm1, %xmm0
14; X86-NEXT:    movss %xmm0, (%esp)
15; X86-NEXT:    flds (%esp)
16; X86-NEXT:    wait
17; X86-NEXT:    popl %eax
18; X86-NEXT:    retl
19;
20; X64-LABEL: fneg_strict_fsub_to_strict_fadd:
21; X64:       # %bb.0:
22; X64-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
23; X64-NEXT:    subss %xmm1, %xmm0
24; X64-NEXT:    retq
25  %neg = fneg float %y
26  %sub = call float @llvm.experimental.constrained.fsub.f32(float %x, float %neg, metadata!"round.dynamic", metadata!"fpexcept.strict")
27  ret float %sub
28}
29
30; FIXME: Missing fsub(x,fneg(y)) -> fadd(x,y) fold
31define double @fneg_strict_fsub_to_strict_fadd_d(double %x, double %y) nounwind strictfp {
32; X86-LABEL: fneg_strict_fsub_to_strict_fadd_d:
33; X86:       # %bb.0:
34; X86-NEXT:    pushl %ebp
35; X86-NEXT:    movl %esp, %ebp
36; X86-NEXT:    andl $-8, %esp
37; X86-NEXT:    subl $8, %esp
38; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
39; X86-NEXT:    movsd {{.*#+}} xmm1 = mem[0],zero
40; X86-NEXT:    xorpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
41; X86-NEXT:    subsd %xmm1, %xmm0
42; X86-NEXT:    movsd %xmm0, (%esp)
43; X86-NEXT:    fldl (%esp)
44; X86-NEXT:    wait
45; X86-NEXT:    movl %ebp, %esp
46; X86-NEXT:    popl %ebp
47; X86-NEXT:    retl
48;
49; X64-LABEL: fneg_strict_fsub_to_strict_fadd_d:
50; X64:       # %bb.0:
51; X64-NEXT:    xorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
52; X64-NEXT:    subsd %xmm1, %xmm0
53; X64-NEXT:    retq
54  %neg = fneg double %y
55  %sub = call double @llvm.experimental.constrained.fsub.f64(double %x, double %neg, metadata!"round.dynamic", metadata!"fpexcept.strict")
56  ret double %sub
57}
58
59; FIXME: Missing fneg(fsub(x,y)) -> fsub(y,x) fold
60define float @strict_fsub_fneg_to_strict_fsub(float %x, float %y) nounwind strictfp {
61; X86-LABEL: strict_fsub_fneg_to_strict_fsub:
62; X86:       # %bb.0:
63; X86-NEXT:    pushl %eax
64; X86-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
65; X86-NEXT:    subss {{[0-9]+}}(%esp), %xmm0
66; X86-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
67; X86-NEXT:    movss %xmm0, (%esp)
68; X86-NEXT:    flds (%esp)
69; X86-NEXT:    wait
70; X86-NEXT:    popl %eax
71; X86-NEXT:    retl
72;
73; X64-LABEL: strict_fsub_fneg_to_strict_fsub:
74; X64:       # %bb.0:
75; X64-NEXT:    subss %xmm1, %xmm0
76; X64-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
77; X64-NEXT:    retq
78  %sub = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata!"round.dynamic", metadata!"fpexcept.strict")
79  %neg = fneg float %sub
80  ret float %neg
81}
82
83; FIXME: Missing fneg(fsub(x,y)) -> fsub(y,x) fold
84define double @strict_fsub_fneg_to_strict_fsub_d(double %x, double %y) nounwind strictfp {
85; X86-LABEL: strict_fsub_fneg_to_strict_fsub_d:
86; X86:       # %bb.0:
87; X86-NEXT:    pushl %ebp
88; X86-NEXT:    movl %esp, %ebp
89; X86-NEXT:    andl $-8, %esp
90; X86-NEXT:    subl $8, %esp
91; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
92; X86-NEXT:    subsd 16(%ebp), %xmm0
93; X86-NEXT:    xorpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
94; X86-NEXT:    movlpd %xmm0, (%esp)
95; X86-NEXT:    fldl (%esp)
96; X86-NEXT:    wait
97; X86-NEXT:    movl %ebp, %esp
98; X86-NEXT:    popl %ebp
99; X86-NEXT:    retl
100;
101; X64-LABEL: strict_fsub_fneg_to_strict_fsub_d:
102; X64:       # %bb.0:
103; X64-NEXT:    subsd %xmm1, %xmm0
104; X64-NEXT:    xorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
105; X64-NEXT:    retq
106  %sub = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata!"round.dynamic", metadata!"fpexcept.strict")
107  %neg = fneg double %sub
108  ret double %neg
109}
110
111
112declare float @llvm.experimental.constrained.fsub.f32(float, float, metadata, metadata)
113declare double @llvm.experimental.constrained.fsub.f64(double, double, metadata, metadata)
114