1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
2; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
3
4; GCN-LABEL: {{^}}test_fmax_f32_ieee_mode_on:
5; GCN: v_mul_f32_e64 [[QUIET0:v[0-9]+]], 1.0, s{{[0-9]+}}
6; GCN: v_mul_f32_e64 [[QUIET1:v[0-9]+]], 1.0, s{{[0-9]+}}
7; GCN: v_max_f32_e32 [[RESULT:v[0-9]+]], [[QUIET1]], [[QUIET0]]
8; GCN-NOT: [[RESULT]]
9; GCN: buffer_store_dword [[RESULT]]
10define amdgpu_kernel void @test_fmax_f32_ieee_mode_on(float addrspace(1)* %out, float %a, float %b) #0 {
11  %val = call float @llvm.maxnum.f32(float %a, float %b) #1
12  store float %val, float addrspace(1)* %out, align 4
13  ret void
14}
15
16; GCN-LABEL: {{^}}test_fmax_f32_ieee_mode_off:
17; GCN: v_max_f32_e32 v0, v0, v1
18; GCN-NEXT: ; return
19define amdgpu_ps float @test_fmax_f32_ieee_mode_off(float %a, float %b) #0 {
20  %val = call float @llvm.maxnum.f32(float %a, float %b) #1
21  ret float %val
22}
23
24; GCN-LABEL: {{^}}test_fmax_v2f32:
25; GCN: v_max_f32_e32
26; GCN: v_max_f32_e32
27define amdgpu_kernel void @test_fmax_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) #0 {
28  %val = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %a, <2 x float> %b)
29  store <2 x float> %val, <2 x float> addrspace(1)* %out, align 8
30  ret void
31}
32
33; GCN-LABEL: {{^}}test_fmax_v3f32:
34; GCN: v_max_f32_e32
35; GCN: v_max_f32_e32
36; GCN: v_max_f32_e32
37; GCN-NOT: v_max_f32
38define amdgpu_kernel void @test_fmax_v3f32(<3 x float> addrspace(1)* %out, <3 x float> %a, <3 x float> %b) nounwind {
39  %val = call <3 x float> @llvm.maxnum.v3f32(<3 x float> %a, <3 x float> %b) #0
40  store <3 x float> %val, <3 x float> addrspace(1)* %out, align 16
41  ret void
42}
43
44; GCN-LABEL: {{^}}test_fmax_v4f32:
45; GCN: v_max_f32_e32
46; GCN: v_max_f32_e32
47; GCN: v_max_f32_e32
48; GCN: v_max_f32_e32
49define amdgpu_kernel void @test_fmax_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b) #0 {
50  %val = call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %b)
51  store <4 x float> %val, <4 x float> addrspace(1)* %out, align 16
52  ret void
53}
54
55; GCN-LABEL: {{^}}test_fmax_v8f32:
56; GCN: v_max_f32_e32
57; GCN: v_max_f32_e32
58; GCN: v_max_f32_e32
59; GCN: v_max_f32_e32
60; GCN: v_max_f32_e32
61; GCN: v_max_f32_e32
62; GCN: v_max_f32_e32
63; GCN: v_max_f32_e32
64define amdgpu_kernel void @test_fmax_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b) #0 {
65  %val = call <8 x float> @llvm.maxnum.v8f32(<8 x float> %a, <8 x float> %b)
66  store <8 x float> %val, <8 x float> addrspace(1)* %out, align 32
67  ret void
68}
69
70; GCN-LABEL: {{^}}test_fmax_v16f32:
71; GCN: v_max_f32_e32
72; GCN: v_max_f32_e32
73; GCN: v_max_f32_e32
74; GCN: v_max_f32_e32
75; GCN: v_max_f32_e32
76; GCN: v_max_f32_e32
77; GCN: v_max_f32_e32
78; GCN: v_max_f32_e32
79; GCN: v_max_f32_e32
80; GCN: v_max_f32_e32
81; GCN: v_max_f32_e32
82; GCN: v_max_f32_e32
83; GCN: v_max_f32_e32
84; GCN: v_max_f32_e32
85; GCN: v_max_f32_e32
86; GCN: v_max_f32_e32
87define amdgpu_kernel void @test_fmax_v16f32(<16 x float> addrspace(1)* %out, <16 x float> %a, <16 x float> %b) #0 {
88  %val = call <16 x float> @llvm.maxnum.v16f32(<16 x float> %a, <16 x float> %b)
89  store <16 x float> %val, <16 x float> addrspace(1)* %out, align 64
90  ret void
91}
92
93; GCN-LABEL: {{^}}constant_fold_fmax_f32:
94; GCN-NOT: v_max_f32_e32
95; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 2.0
96; GCN: buffer_store_dword [[REG]]
97define amdgpu_kernel void @constant_fold_fmax_f32(float addrspace(1)* %out) #0 {
98  %val = call float @llvm.maxnum.f32(float 1.0, float 2.0)
99  store float %val, float addrspace(1)* %out, align 4
100  ret void
101}
102
103; GCN-LABEL: {{^}}constant_fold_fmax_f32_nan_nan:
104; GCN-NOT: v_max_f32_e32
105; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7fc00000
106; GCN: buffer_store_dword [[REG]]
107define amdgpu_kernel void @constant_fold_fmax_f32_nan_nan(float addrspace(1)* %out) #0 {
108  %val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 0x7FF8000000000000)
109  store float %val, float addrspace(1)* %out, align 4
110  ret void
111}
112
113; GCN-LABEL: {{^}}constant_fold_fmax_f32_val_nan:
114; GCN-NOT: v_max_f32_e32
115; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
116; GCN: buffer_store_dword [[REG]]
117define amdgpu_kernel void @constant_fold_fmax_f32_val_nan(float addrspace(1)* %out) #0 {
118  %val = call float @llvm.maxnum.f32(float 1.0, float 0x7FF8000000000000)
119  store float %val, float addrspace(1)* %out, align 4
120  ret void
121}
122
123; GCN-LABEL: {{^}}constant_fold_fmax_f32_nan_val:
124; GCN-NOT: v_max_f32_e32
125; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0
126; GCN: buffer_store_dword [[REG]]
127define amdgpu_kernel void @constant_fold_fmax_f32_nan_val(float addrspace(1)* %out) #0 {
128  %val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 1.0)
129  store float %val, float addrspace(1)* %out, align 4
130  ret void
131}
132
133; GCN-LABEL: {{^}}constant_fold_fmax_f32_p0_p0:
134; GCN-NOT: v_max_f32_e32
135; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0
136; GCN: buffer_store_dword [[REG]]
137define amdgpu_kernel void @constant_fold_fmax_f32_p0_p0(float addrspace(1)* %out) #0 {
138  %val = call float @llvm.maxnum.f32(float 0.0, float 0.0)
139  store float %val, float addrspace(1)* %out, align 4
140  ret void
141}
142
143; GCN-LABEL: {{^}}constant_fold_fmax_f32_p0_n0:
144; GCN-NOT: v_max_f32_e32
145; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0
146; GCN: buffer_store_dword [[REG]]
147define amdgpu_kernel void @constant_fold_fmax_f32_p0_n0(float addrspace(1)* %out) #0 {
148  %val = call float @llvm.maxnum.f32(float 0.0, float -0.0)
149  store float %val, float addrspace(1)* %out, align 4
150  ret void
151}
152
153; GCN-LABEL: {{^}}constant_fold_fmax_f32_n0_p0:
154; GCN-NOT: v_max_f32_e32
155; GCN: v_bfrev_b32_e32 [[REG:v[0-9]+]], 1{{$}}
156; GCN: buffer_store_dword [[REG]]
157define amdgpu_kernel void @constant_fold_fmax_f32_n0_p0(float addrspace(1)* %out) #0 {
158  %val = call float @llvm.maxnum.f32(float -0.0, float 0.0)
159  store float %val, float addrspace(1)* %out, align 4
160  ret void
161}
162
163; GCN-LABEL: {{^}}constant_fold_fmax_f32_n0_n0:
164; GCN-NOT: v_max_f32_e32
165; GCN: v_bfrev_b32_e32 [[REG:v[0-9]+]], 1{{$}}
166; GCN: buffer_store_dword [[REG]]
167define amdgpu_kernel void @constant_fold_fmax_f32_n0_n0(float addrspace(1)* %out) #0 {
168  %val = call float @llvm.maxnum.f32(float -0.0, float -0.0)
169  store float %val, float addrspace(1)* %out, align 4
170  ret void
171}
172
173; GCN-LABEL: {{^}}fmax_var_immediate_f32_no_ieee:
174; GCN: v_max_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, 2.0
175define amdgpu_ps float @fmax_var_immediate_f32_no_ieee(float inreg %a) #0 {
176  %val = call float @llvm.maxnum.f32(float %a, float 2.0) #0
177  ret float %val
178}
179
180; GCN-LABEL: {{^}}fmax_immediate_var_f32_no_ieee:
181; GCN: v_max_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, 2.0
182define amdgpu_ps float @fmax_immediate_var_f32_no_ieee(float inreg %a) #0 {
183  %val = call float @llvm.maxnum.f32(float 2.0, float %a) #0
184  ret float %val
185}
186
187; GCN-LABEL: {{^}}fmax_var_literal_f32_no_ieee:
188; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
189; GCN: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
190define amdgpu_ps float @fmax_var_literal_f32_no_ieee(float inreg %a) #0 {
191  %val = call float @llvm.maxnum.f32(float %a, float 99.0) #0
192  ret float %val
193}
194
195; GCN-LABEL: {{^}}fmax_literal_var_f32_no_ieee:
196; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000
197; GCN: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]
198define amdgpu_ps float @fmax_literal_var_f32_no_ieee(float inreg %a) #0 {
199  %val = call float @llvm.maxnum.f32(float 99.0, float %a) #0
200  ret float %val
201}
202
203; GCN-LABEL: {{^}}test_func_fmax_v3f32:
204; GCN: v_max_f32_e32
205; GCN: v_max_f32_e32
206; GCN: v_max_f32_e32
207; GCN-NOT: v_max_f32
208define <3 x float> @test_func_fmax_v3f32(<3 x float> %a, <3 x float> %b) #0 {
209  %val = call <3 x float> @llvm.maxnum.v3f32(<3 x float> %a, <3 x float> %b) #0
210  ret <3 x float> %val
211}
212
213declare float @llvm.maxnum.f32(float, float) #1
214declare <2 x float> @llvm.maxnum.v2f32(<2 x float>, <2 x float>) #1
215declare <3 x float> @llvm.maxnum.v3f32(<3 x float>, <3 x float>) #1
216declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>) #1
217declare <8 x float> @llvm.maxnum.v8f32(<8 x float>, <8 x float>) #1
218declare <16 x float> @llvm.maxnum.v16f32(<16 x float>, <16 x float>) #1
219declare double @llvm.maxnum.f64(double, double)
220
221attributes #0 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
222attributes #1 = { nounwind readnone }
223