xref: /llvm-project/llvm/test/CodeGen/AArch64/sve-fp-reciprocal.ll (revision cc82f1290a1e2157a6c0530d78d8cc84d2b8553d) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
3
4; FDIV
5
6define <vscale x 8 x half> @fdiv_8f16(<vscale x 8 x half> %a, <vscale x 8 x half> %b) {
7; CHECK-LABEL: fdiv_8f16:
8; CHECK:       // %bb.0:
9; CHECK-NEXT:    ptrue p0.h
10; CHECK-NEXT:    fdiv z0.h, p0/m, z0.h, z1.h
11; CHECK-NEXT:    ret
12  %fdiv = fdiv fast <vscale x 8 x half> %a, %b
13  ret <vscale x 8 x half> %fdiv
14}
15
16define <vscale x 8 x half> @fdiv_recip_8f16(<vscale x 8 x half> %a, <vscale x 8 x half> %b) #0 {
17; CHECK-LABEL: fdiv_recip_8f16:
18; CHECK:       // %bb.0:
19; CHECK-NEXT:    frecpe z2.h, z1.h
20; CHECK-NEXT:    frecps z1.h, z1.h, z2.h
21; CHECK-NEXT:    fmul z1.h, z2.h, z1.h
22; CHECK-NEXT:    fmul z0.h, z1.h, z0.h
23; CHECK-NEXT:    ret
24  %fdiv = fdiv fast <vscale x 8 x half> %a, %b
25  ret <vscale x 8 x half> %fdiv
26}
27
28define <vscale x 4 x float> @fdiv_4f32(<vscale x 4 x float> %a, <vscale x 4 x float> %b) {
29; CHECK-LABEL: fdiv_4f32:
30; CHECK:       // %bb.0:
31; CHECK-NEXT:    ptrue p0.s
32; CHECK-NEXT:    fdiv z0.s, p0/m, z0.s, z1.s
33; CHECK-NEXT:    ret
34  %fdiv = fdiv fast <vscale x 4 x float> %a, %b
35  ret <vscale x 4 x float> %fdiv
36}
37
38define <vscale x 4 x float> @fdiv_recip_4f32(<vscale x 4 x float> %a, <vscale x 4 x float> %b) #0 {
39; CHECK-LABEL: fdiv_recip_4f32:
40; CHECK:       // %bb.0:
41; CHECK-NEXT:    frecpe z2.s, z1.s
42; CHECK-NEXT:    frecps z3.s, z1.s, z2.s
43; CHECK-NEXT:    fmul z2.s, z2.s, z3.s
44; CHECK-NEXT:    frecps z1.s, z1.s, z2.s
45; CHECK-NEXT:    fmul z1.s, z2.s, z1.s
46; CHECK-NEXT:    fmul z0.s, z1.s, z0.s
47; CHECK-NEXT:    ret
48  %fdiv = fdiv fast <vscale x 4 x float> %a, %b
49  ret <vscale x 4 x float> %fdiv
50}
51
52define <vscale x 2 x double> @fdiv_2f64(<vscale x 2 x double> %a, <vscale x 2 x double> %b) {
53; CHECK-LABEL: fdiv_2f64:
54; CHECK:       // %bb.0:
55; CHECK-NEXT:    ptrue p0.d
56; CHECK-NEXT:    fdiv z0.d, p0/m, z0.d, z1.d
57; CHECK-NEXT:    ret
58  %fdiv = fdiv fast <vscale x 2 x double> %a, %b
59  ret <vscale x 2 x double> %fdiv
60}
61
62define <vscale x 2 x double> @fdiv_recip_2f64(<vscale x 2 x double> %a, <vscale x 2 x double> %b) #0 {
63; CHECK-LABEL: fdiv_recip_2f64:
64; CHECK:       // %bb.0:
65; CHECK-NEXT:    frecpe z2.d, z1.d
66; CHECK-NEXT:    frecps z3.d, z1.d, z2.d
67; CHECK-NEXT:    fmul z2.d, z2.d, z3.d
68; CHECK-NEXT:    frecps z3.d, z1.d, z2.d
69; CHECK-NEXT:    fmul z2.d, z2.d, z3.d
70; CHECK-NEXT:    frecps z1.d, z1.d, z2.d
71; CHECK-NEXT:    fmul z1.d, z2.d, z1.d
72; CHECK-NEXT:    fmul z0.d, z1.d, z0.d
73; CHECK-NEXT:    ret
74  %fdiv = fdiv fast <vscale x 2 x double> %a, %b
75  ret <vscale x 2 x double> %fdiv
76}
77
78; FSQRT
79
80define <vscale x 8 x half> @fsqrt_8f16(<vscale x 8 x half> %a) {
81; CHECK-LABEL: fsqrt_8f16:
82; CHECK:       // %bb.0:
83; CHECK-NEXT:    ptrue p0.h
84; CHECK-NEXT:    fsqrt z0.h, p0/m, z0.h
85; CHECK-NEXT:    ret
86  %fsqrt = call fast <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> %a)
87  ret <vscale x 8 x half> %fsqrt
88}
89
90define <vscale x 8 x half> @fsqrt_recip_8f16(<vscale x 8 x half> %a) #0 {
91; CHECK-LABEL: fsqrt_recip_8f16:
92; CHECK:       // %bb.0:
93; CHECK-NEXT:    frsqrte z1.h, z0.h
94; CHECK-NEXT:    ptrue p0.h
95; CHECK-NEXT:    fcmne p0.h, p0/z, z0.h, #0.0
96; CHECK-NEXT:    fmul z2.h, z1.h, z1.h
97; CHECK-NEXT:    frsqrts z2.h, z0.h, z2.h
98; CHECK-NEXT:    fmul z1.h, z1.h, z2.h
99; CHECK-NEXT:    fmul z0.h, p0/m, z0.h, z1.h
100; CHECK-NEXT:    ret
101  %fsqrt = call fast <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> %a)
102  ret <vscale x 8 x half> %fsqrt
103}
104
105define <vscale x 4 x float> @fsqrt_4f32(<vscale x 4 x float> %a) {
106; CHECK-LABEL: fsqrt_4f32:
107; CHECK:       // %bb.0:
108; CHECK-NEXT:    ptrue p0.s
109; CHECK-NEXT:    fsqrt z0.s, p0/m, z0.s
110; CHECK-NEXT:    ret
111  %fsqrt = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %a)
112  ret <vscale x 4 x float> %fsqrt
113}
114
115define <vscale x 4 x float> @fsqrt_recip_4f32(<vscale x 4 x float> %a) #0 {
116; CHECK-LABEL: fsqrt_recip_4f32:
117; CHECK:       // %bb.0:
118; CHECK-NEXT:    frsqrte z1.s, z0.s
119; CHECK-NEXT:    ptrue p0.s
120; CHECK-NEXT:    fcmne p0.s, p0/z, z0.s, #0.0
121; CHECK-NEXT:    fmul z2.s, z1.s, z1.s
122; CHECK-NEXT:    frsqrts z2.s, z0.s, z2.s
123; CHECK-NEXT:    fmul z1.s, z1.s, z2.s
124; CHECK-NEXT:    fmul z2.s, z1.s, z1.s
125; CHECK-NEXT:    frsqrts z2.s, z0.s, z2.s
126; CHECK-NEXT:    fmul z1.s, z1.s, z2.s
127; CHECK-NEXT:    fmul z0.s, p0/m, z0.s, z1.s
128; CHECK-NEXT:    ret
129  %fsqrt = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %a)
130  ret <vscale x 4 x float> %fsqrt
131}
132
133define <vscale x 2 x double> @fsqrt_2f64(<vscale x 2 x double> %a) {
134; CHECK-LABEL: fsqrt_2f64:
135; CHECK:       // %bb.0:
136; CHECK-NEXT:    ptrue p0.d
137; CHECK-NEXT:    fsqrt z0.d, p0/m, z0.d
138; CHECK-NEXT:    ret
139  %fsqrt = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %a)
140  ret <vscale x 2 x double> %fsqrt
141}
142
143define <vscale x 2 x double> @fsqrt_recip_2f64(<vscale x 2 x double> %a) #0 {
144; CHECK-LABEL: fsqrt_recip_2f64:
145; CHECK:       // %bb.0:
146; CHECK-NEXT:    frsqrte z1.d, z0.d
147; CHECK-NEXT:    ptrue p0.d
148; CHECK-NEXT:    fcmne p0.d, p0/z, z0.d, #0.0
149; CHECK-NEXT:    fmul z2.d, z1.d, z1.d
150; CHECK-NEXT:    frsqrts z2.d, z0.d, z2.d
151; CHECK-NEXT:    fmul z1.d, z1.d, z2.d
152; CHECK-NEXT:    fmul z2.d, z1.d, z1.d
153; CHECK-NEXT:    frsqrts z2.d, z0.d, z2.d
154; CHECK-NEXT:    fmul z1.d, z1.d, z2.d
155; CHECK-NEXT:    fmul z2.d, z1.d, z1.d
156; CHECK-NEXT:    frsqrts z2.d, z0.d, z2.d
157; CHECK-NEXT:    fmul z1.d, z1.d, z2.d
158; CHECK-NEXT:    fmul z0.d, p0/m, z0.d, z1.d
159; CHECK-NEXT:    ret
160  %fsqrt = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %a)
161  ret <vscale x 2 x double> %fsqrt
162}
163
164declare <vscale x 2 x half> @llvm.sqrt.nxv2f16(<vscale x 2 x half>)
165declare <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half>)
166declare <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half>)
167declare <vscale x 2 x float> @llvm.sqrt.nxv2f32(<vscale x 2 x float>)
168declare <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float>)
169declare <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double>)
170
171attributes #0 = { "reciprocal-estimates"="all" }
172