1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
3; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
4
5define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 {
6; GCN-LABEL: v_constained_fadd_f64_fpexcept_strict:
7; GCN:       ; %bb.0:
8; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
9; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
10; GCN-NEXT:    s_setpc_b64 s[30:31]
11;
12; GFX10-LABEL: v_constained_fadd_f64_fpexcept_strict:
13; GFX10:       ; %bb.0:
14; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
15; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
16; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
17; GFX10-NEXT:    s_setpc_b64 s[30:31]
18  %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
19  ret double %val
20}
21
22define double @v_constained_fadd_f64_fpexcept_ignore(double %x, double %y) #0 {
23; GCN-LABEL: v_constained_fadd_f64_fpexcept_ignore:
24; GCN:       ; %bb.0:
25; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
26; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
27; GCN-NEXT:    s_setpc_b64 s[30:31]
28;
29; GFX10-LABEL: v_constained_fadd_f64_fpexcept_ignore:
30; GFX10:       ; %bb.0:
31; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
32; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
33; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
34; GFX10-NEXT:    s_setpc_b64 s[30:31]
35  %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
36  ret double %val
37}
38
39define double @v_constained_fadd_f64_fpexcept_maytrap(double %x, double %y) #0 {
40; GCN-LABEL: v_constained_fadd_f64_fpexcept_maytrap:
41; GCN:       ; %bb.0:
42; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
43; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
44; GCN-NEXT:    s_setpc_b64 s[30:31]
45;
46; GFX10-LABEL: v_constained_fadd_f64_fpexcept_maytrap:
47; GFX10:       ; %bb.0:
48; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
49; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
50; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
51; GFX10-NEXT:    s_setpc_b64 s[30:31]
52  %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
53  ret double %val
54}
55
56define <2 x double> @v_constained_fadd_v2f64_fpexcept_strict(<2 x double> %x, <2 x double> %y) #0 {
57; GCN-LABEL: v_constained_fadd_v2f64_fpexcept_strict:
58; GCN:       ; %bb.0:
59; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
60; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
61; GCN-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
62; GCN-NEXT:    s_setpc_b64 s[30:31]
63;
64; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_strict:
65; GFX10:       ; %bb.0:
66; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
67; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
68; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
69; GFX10-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
70; GFX10-NEXT:    s_setpc_b64 s[30:31]
71  %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
72  ret <2 x double> %val
73}
74
75define <2 x double> @v_constained_fadd_v2f64_fpexcept_ignore(<2 x double> %x, <2 x double> %y) #0 {
76; GCN-LABEL: v_constained_fadd_v2f64_fpexcept_ignore:
77; GCN:       ; %bb.0:
78; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
79; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
80; GCN-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
81; GCN-NEXT:    s_setpc_b64 s[30:31]
82;
83; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_ignore:
84; GFX10:       ; %bb.0:
85; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
86; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
87; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
88; GFX10-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
89; GFX10-NEXT:    s_setpc_b64 s[30:31]
90  %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
91  ret <2 x double> %val
92}
93
94define <2 x double> @v_constained_fadd_v2f64_fpexcept_maytrap(<2 x double> %x, <2 x double> %y) #0 {
95; GCN-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap:
96; GCN:       ; %bb.0:
97; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
98; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
99; GCN-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
100; GCN-NEXT:    s_setpc_b64 s[30:31]
101;
102; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap:
103; GFX10:       ; %bb.0:
104; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
105; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
106; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
107; GFX10-NEXT:    v_add_f64 v[2:3], v[2:3], v[6:7]
108; GFX10-NEXT:    s_setpc_b64 s[30:31]
109  %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
110  ret <2 x double> %val
111}
112
113define <3 x double> @v_constained_fadd_v3f64_fpexcept_strict(<3 x double> %x, <3 x double> %y) #0 {
114; GCN-LABEL: v_constained_fadd_v3f64_fpexcept_strict:
115; GCN:       ; %bb.0:
116; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
117; GCN-NEXT:    v_add_f64 v[0:1], v[0:1], v[6:7]
118; GCN-NEXT:    v_add_f64 v[2:3], v[2:3], v[8:9]
119; GCN-NEXT:    v_add_f64 v[4:5], v[4:5], v[10:11]
120; GCN-NEXT:    s_setpc_b64 s[30:31]
121;
122; GFX10-LABEL: v_constained_fadd_v3f64_fpexcept_strict:
123; GFX10:       ; %bb.0:
124; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
125; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
126; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[6:7]
127; GFX10-NEXT:    v_add_f64 v[2:3], v[2:3], v[8:9]
128; GFX10-NEXT:    v_add_f64 v[4:5], v[4:5], v[10:11]
129; GFX10-NEXT:    s_setpc_b64 s[30:31]
130  %val = call <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
131  ret <3 x double> %val
132}
133
134define amdgpu_ps <2 x float> @s_constained_fadd_f64_fpexcept_strict(double inreg %x, double inreg %y) #0 {
135; GCN-LABEL: s_constained_fadd_f64_fpexcept_strict:
136; GCN:       ; %bb.0:
137; GCN-NEXT:    v_mov_b32_e32 v0, s4
138; GCN-NEXT:    v_mov_b32_e32 v1, s5
139; GCN-NEXT:    v_add_f64 v[0:1], s[2:3], v[0:1]
140; GCN-NEXT:    ; return to shader part epilog
141;
142; GFX10-LABEL: s_constained_fadd_f64_fpexcept_strict:
143; GFX10:       ; %bb.0:
144; GFX10-NEXT:    v_add_f64 v[0:1], s[2:3], s[4:5]
145; GFX10-NEXT:    ; return to shader part epilog
146  %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
147  %cast = bitcast double %val to <2 x float>
148  ret <2 x float> %cast
149}
150
151declare double @llvm.experimental.constrained.fadd.f64(double, double, metadata, metadata) #1
152declare <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double>, <2 x double>, metadata, metadata) #1
153declare <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double>, <3 x double>, metadata, metadata) #1
154
155attributes #0 = { strictfp }
156attributes #1 = { inaccessiblememonly nounwind willreturn }
157