xref: /llvm-project/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-fmax.ll (revision 32f9983c064557883223b585810eac3c6797d500) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
2; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-simplifylib %s | FileCheck %s
3
4target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
5
6declare float @_Z4fmaxff(float, float)
7declare <2 x float> @_Z4fmaxDv2_fS_(<2 x float>, <2 x float>)
8declare <3 x float> @_Z4fmaxDv3_fS_(<3 x float>, <3 x float>)
9declare <4 x float> @_Z4fmaxDv4_fS_(<4 x float>, <4 x float>)
10declare <8 x float> @_Z4fmaxDv8_fS_(<8 x float>, <8 x float>)
11declare <16 x float> @_Z4fmaxDv16_fS_(<16 x float>, <16 x float>)
12declare double @_Z4fmaxdd(double, double)
13declare <2 x double> @_Z4fmaxDv2_dS_(<2 x double>, <2 x double>)
14declare <3 x double> @_Z4fmaxDv3_dS_(<3 x double>, <3 x double>)
15declare <4 x double> @_Z4fmaxDv4_dS_(<4 x double>, <4 x double>)
16declare <8 x double> @_Z4fmaxDv8_dS_(<8 x double>, <8 x double>)
17declare <16 x double> @_Z4fmaxDv16_dS_(<16 x double>, <16 x double>)
18declare half @_Z4fmaxDhDh(half, half)
19declare <2 x half> @_Z4fmaxDv2_DhS_(<2 x half>, <2 x half>)
20declare <3 x half> @_Z4fmaxDv3_DhS_(<3 x half>, <3 x half>)
21declare <4 x half> @_Z4fmaxDv4_DhS_(<4 x half>, <4 x half>)
22declare <8 x half> @_Z4fmaxDv8_DhS_(<8 x half>, <8 x half>)
23declare <16 x half> @_Z4fmaxDv16_DhS_(<16 x half>, <16 x half>)
24
25define float @test_fmax_f32(float %x, float %y) {
26; CHECK-LABEL: define float @test_fmax_f32
27; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
28; CHECK-NEXT:    [[FMAX:%.*]] = tail call float @llvm.maxnum.f32(float [[X]], float [[Y]])
29; CHECK-NEXT:    ret float [[FMAX]]
30;
31  %fmax = tail call float @_Z4fmaxff(float %x, float %y)
32  ret float %fmax
33}
34
35define float @test_fmax_f32_nnan(float %x, float %y) {
36; CHECK-LABEL: define float @test_fmax_f32_nnan
37; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
38; CHECK-NEXT:    [[FMAX:%.*]] = tail call nnan float @llvm.maxnum.f32(float [[X]], float [[Y]])
39; CHECK-NEXT:    ret float [[FMAX]]
40;
41  %fmax = tail call nnan float @_Z4fmaxff(float %x, float %y)
42  ret float %fmax
43}
44
45define <2 x float> @test_fmax_v2f32(<2 x float> %x, <2 x float> %y) {
46; CHECK-LABEL: define <2 x float> @test_fmax_v2f32
47; CHECK-SAME: (<2 x float> [[X:%.*]], <2 x float> [[Y:%.*]]) {
48; CHECK-NEXT:    [[FMAX:%.*]] = tail call <2 x float> @llvm.maxnum.v2f32(<2 x float> [[X]], <2 x float> [[Y]])
49; CHECK-NEXT:    ret <2 x float> [[FMAX]]
50;
51  %fmax = tail call <2 x float> @_Z4fmaxDv2_fS_(<2 x float> %x, <2 x float> %y)
52  ret <2 x float> %fmax
53}
54
55define <3 x float> @test_fmax_v3f32(<3 x float> %x, <3 x float> %y) {
56; CHECK-LABEL: define <3 x float> @test_fmax_v3f32
57; CHECK-SAME: (<3 x float> [[X:%.*]], <3 x float> [[Y:%.*]]) {
58; CHECK-NEXT:    [[FMAX:%.*]] = tail call <3 x float> @llvm.maxnum.v3f32(<3 x float> [[X]], <3 x float> [[Y]])
59; CHECK-NEXT:    ret <3 x float> [[FMAX]]
60;
61  %fmax = tail call <3 x float> @_Z4fmaxDv3_fS_(<3 x float> %x, <3 x float> %y)
62  ret <3 x float> %fmax
63}
64
65define <4 x float> @test_fmax_v4f32(<4 x float> %x, <4 x float> %y) {
66; CHECK-LABEL: define <4 x float> @test_fmax_v4f32
67; CHECK-SAME: (<4 x float> [[X:%.*]], <4 x float> [[Y:%.*]]) {
68; CHECK-NEXT:    [[FMAX:%.*]] = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[X]], <4 x float> [[Y]])
69; CHECK-NEXT:    ret <4 x float> [[FMAX]]
70;
71  %fmax = tail call <4 x float> @_Z4fmaxDv4_fS_(<4 x float> %x, <4 x float> %y)
72  ret <4 x float> %fmax
73}
74
75define <8 x float> @test_fmax_v8f32(<8 x float> %x, <8 x float> %y) {
76; CHECK-LABEL: define <8 x float> @test_fmax_v8f32
77; CHECK-SAME: (<8 x float> [[X:%.*]], <8 x float> [[Y:%.*]]) {
78; CHECK-NEXT:    [[FMAX:%.*]] = tail call <8 x float> @llvm.maxnum.v8f32(<8 x float> [[X]], <8 x float> [[Y]])
79; CHECK-NEXT:    ret <8 x float> [[FMAX]]
80;
81  %fmax = tail call <8 x float> @_Z4fmaxDv8_fS_(<8 x float> %x, <8 x float> %y)
82  ret <8 x float> %fmax
83}
84
85define <16 x float> @test_fmax_v16f32(<16 x float> %x, <16 x float> %y) {
86; CHECK-LABEL: define <16 x float> @test_fmax_v16f32
87; CHECK-SAME: (<16 x float> [[X:%.*]], <16 x float> [[Y:%.*]]) {
88; CHECK-NEXT:    [[FMAX:%.*]] = tail call <16 x float> @llvm.maxnum.v16f32(<16 x float> [[X]], <16 x float> [[Y]])
89; CHECK-NEXT:    ret <16 x float> [[FMAX]]
90;
91  %fmax = tail call <16 x float> @_Z4fmaxDv16_fS_(<16 x float> %x, <16 x float> %y)
92  ret <16 x float> %fmax
93}
94
95define double @test_fmax_f64(double %x, double %y) {
96; CHECK-LABEL: define double @test_fmax_f64
97; CHECK-SAME: (double [[X:%.*]], double [[Y:%.*]]) {
98; CHECK-NEXT:    [[FMAX:%.*]] = tail call double @llvm.maxnum.f64(double [[X]], double [[Y]])
99; CHECK-NEXT:    ret double [[FMAX]]
100;
101  %fmax = tail call double @_Z4fmaxdd(double %x, double %y)
102  ret double %fmax
103}
104
105define <2 x double> @test_fmax_v2f64(<2 x double> %x, <2 x double> %y) {
106; CHECK-LABEL: define <2 x double> @test_fmax_v2f64
107; CHECK-SAME: (<2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
108; CHECK-NEXT:    [[FMAX:%.*]] = tail call <2 x double> @llvm.maxnum.v2f64(<2 x double> [[X]], <2 x double> [[Y]])
109; CHECK-NEXT:    ret <2 x double> [[FMAX]]
110;
111  %fmax = tail call <2 x double> @_Z4fmaxDv2_dS_(<2 x double> %x, <2 x double> %y)
112  ret <2 x double> %fmax
113}
114
115define <3 x double> @test_fmax_v3f64(<3 x double> %x, <3 x double> %y) {
116; CHECK-LABEL: define <3 x double> @test_fmax_v3f64
117; CHECK-SAME: (<3 x double> [[X:%.*]], <3 x double> [[Y:%.*]]) {
118; CHECK-NEXT:    [[FMAX:%.*]] = tail call <3 x double> @llvm.maxnum.v3f64(<3 x double> [[X]], <3 x double> [[Y]])
119; CHECK-NEXT:    ret <3 x double> [[FMAX]]
120;
121  %fmax = tail call <3 x double> @_Z4fmaxDv3_dS_(<3 x double> %x, <3 x double> %y)
122  ret <3 x double> %fmax
123}
124
125define <4 x double> @test_fmax_v4f64(<4 x double> %x, <4 x double> %y) {
126; CHECK-LABEL: define <4 x double> @test_fmax_v4f64
127; CHECK-SAME: (<4 x double> [[X:%.*]], <4 x double> [[Y:%.*]]) {
128; CHECK-NEXT:    [[FMAX:%.*]] = tail call <4 x double> @llvm.maxnum.v4f64(<4 x double> [[X]], <4 x double> [[Y]])
129; CHECK-NEXT:    ret <4 x double> [[FMAX]]
130;
131  %fmax = tail call <4 x double> @_Z4fmaxDv4_dS_(<4 x double> %x, <4 x double> %y)
132  ret <4 x double> %fmax
133}
134
135define <8 x double> @test_fmax_v8f64(<8 x double> %x, <8 x double> %y) {
136; CHECK-LABEL: define <8 x double> @test_fmax_v8f64
137; CHECK-SAME: (<8 x double> [[X:%.*]], <8 x double> [[Y:%.*]]) {
138; CHECK-NEXT:    [[FMAX:%.*]] = tail call <8 x double> @llvm.maxnum.v8f64(<8 x double> [[X]], <8 x double> [[Y]])
139; CHECK-NEXT:    ret <8 x double> [[FMAX]]
140;
141  %fmax = tail call <8 x double> @_Z4fmaxDv8_dS_(<8 x double> %x, <8 x double> %y)
142  ret <8 x double> %fmax
143}
144
145define <16 x double> @test_fmax_v16f64(<16 x double> %x, <16 x double> %y) {
146; CHECK-LABEL: define <16 x double> @test_fmax_v16f64
147; CHECK-SAME: (<16 x double> [[X:%.*]], <16 x double> [[Y:%.*]]) {
148; CHECK-NEXT:    [[FMAX:%.*]] = tail call <16 x double> @llvm.maxnum.v16f64(<16 x double> [[X]], <16 x double> [[Y]])
149; CHECK-NEXT:    ret <16 x double> [[FMAX]]
150;
151  %fmax = tail call <16 x double> @_Z4fmaxDv16_dS_(<16 x double> %x, <16 x double> %y)
152  ret <16 x double> %fmax
153}
154
155define half @test_fmax_f16(half %x, half %y) {
156; CHECK-LABEL: define half @test_fmax_f16
157; CHECK-SAME: (half [[X:%.*]], half [[Y:%.*]]) {
158; CHECK-NEXT:    [[FMAX:%.*]] = tail call half @llvm.maxnum.f16(half [[X]], half [[Y]])
159; CHECK-NEXT:    ret half [[FMAX]]
160;
161  %fmax = tail call half @_Z4fmaxDhDh(half %x, half %y)
162  ret half %fmax
163}
164
165define <2 x half> @test_fmax_v2f16(<2 x half> %x, <2 x half> %y) {
166; CHECK-LABEL: define <2 x half> @test_fmax_v2f16
167; CHECK-SAME: (<2 x half> [[X:%.*]], <2 x half> [[Y:%.*]]) {
168; CHECK-NEXT:    [[FMAX:%.*]] = tail call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[X]], <2 x half> [[Y]])
169; CHECK-NEXT:    ret <2 x half> [[FMAX]]
170;
171  %fmax = tail call <2 x half> @_Z4fmaxDv2_DhS_(<2 x half> %x, <2 x half> %y)
172  ret <2 x half> %fmax
173}
174
175define <3 x half> @test_fmax_v3f16(<3 x half> %x, <3 x half> %y) {
176; CHECK-LABEL: define <3 x half> @test_fmax_v3f16
177; CHECK-SAME: (<3 x half> [[X:%.*]], <3 x half> [[Y:%.*]]) {
178; CHECK-NEXT:    [[FMAX:%.*]] = tail call <3 x half> @llvm.maxnum.v3f16(<3 x half> [[X]], <3 x half> [[Y]])
179; CHECK-NEXT:    ret <3 x half> [[FMAX]]
180;
181  %fmax = tail call <3 x half> @_Z4fmaxDv3_DhS_(<3 x half> %x, <3 x half> %y)
182  ret <3 x half> %fmax
183}
184
185define <4 x half> @test_fmax_v4f16(<4 x half> %x, <4 x half> %y) {
186; CHECK-LABEL: define <4 x half> @test_fmax_v4f16
187; CHECK-SAME: (<4 x half> [[X:%.*]], <4 x half> [[Y:%.*]]) {
188; CHECK-NEXT:    [[FMAX:%.*]] = tail call <4 x half> @llvm.maxnum.v4f16(<4 x half> [[X]], <4 x half> [[Y]])
189; CHECK-NEXT:    ret <4 x half> [[FMAX]]
190;
191  %fmax = tail call <4 x half> @_Z4fmaxDv4_DhS_(<4 x half> %x, <4 x half> %y)
192  ret <4 x half> %fmax
193}
194
195define <8 x half> @test_fmax_v8f16(<8 x half> %x, <8 x half> %y) {
196; CHECK-LABEL: define <8 x half> @test_fmax_v8f16
197; CHECK-SAME: (<8 x half> [[X:%.*]], <8 x half> [[Y:%.*]]) {
198; CHECK-NEXT:    [[FMAX:%.*]] = tail call <8 x half> @llvm.maxnum.v8f16(<8 x half> [[X]], <8 x half> [[Y]])
199; CHECK-NEXT:    ret <8 x half> [[FMAX]]
200;
201  %fmax = tail call <8 x half> @_Z4fmaxDv8_DhS_(<8 x half> %x, <8 x half> %y)
202  ret <8 x half> %fmax
203}
204
205define <16 x half> @test_fmax_v16f16(<16 x half> %x, <16 x half> %y) {
206; CHECK-LABEL: define <16 x half> @test_fmax_v16f16
207; CHECK-SAME: (<16 x half> [[X:%.*]], <16 x half> [[Y:%.*]]) {
208; CHECK-NEXT:    [[FMAX:%.*]] = tail call <16 x half> @llvm.maxnum.v16f16(<16 x half> [[X]], <16 x half> [[Y]])
209; CHECK-NEXT:    ret <16 x half> [[FMAX]]
210;
211  %fmax = tail call <16 x half> @_Z4fmaxDv16_DhS_(<16 x half> %x, <16 x half> %y)
212  ret <16 x half> %fmax
213}
214
215define float @test_fmax_f32_minsize(float %x, float %y) #0 {
216; CHECK-LABEL: define float @test_fmax_f32_minsize
217; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) #[[ATTR0:[0-9]+]] {
218; CHECK-NEXT:    [[FMAX:%.*]] = tail call float @llvm.maxnum.f32(float [[X]], float [[Y]])
219; CHECK-NEXT:    ret float [[FMAX]]
220;
221  %fmax = tail call float @_Z4fmaxff(float %x, float %y)
222  ret float %fmax
223}
224
225define float @test_fmax_f32_nnan_minsize(float %x, float %y) #0 {
226; CHECK-LABEL: define float @test_fmax_f32_nnan_minsize
227; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) #[[ATTR0]] {
228; CHECK-NEXT:    [[FMAX:%.*]] = tail call nnan float @llvm.maxnum.f32(float [[X]], float [[Y]])
229; CHECK-NEXT:    ret float [[FMAX]]
230;
231  %fmax = tail call nnan float @_Z4fmaxff(float %x, float %y)
232  ret float %fmax
233}
234
235define float @test_fmax_f32_noinline(float %x, float %y) {
236; CHECK-LABEL: define float @test_fmax_f32_noinline
237; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
238; CHECK-NEXT:    [[FMAX:%.*]] = tail call float @_Z4fmaxff(float [[X]], float [[Y]]) #[[ATTR3:[0-9]+]]
239; CHECK-NEXT:    ret float [[FMAX]]
240;
241  %fmax = tail call float @_Z4fmaxff(float %x, float %y) #1
242  ret float %fmax
243}
244
245define float @test_fmax_f32_nnan_noinline(float %x, float %y) {
246; CHECK-LABEL: define float @test_fmax_f32_nnan_noinline
247; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
248; CHECK-NEXT:    [[FMAX:%.*]] = tail call nnan float @_Z4fmaxff(float [[X]], float [[Y]]) #[[ATTR3]]
249; CHECK-NEXT:    ret float [[FMAX]]
250;
251  %fmax = tail call nnan float @_Z4fmaxff(float %x, float %y) #1
252  ret float %fmax
253}
254
255define float @test_fmax_f32_strictfp(float %x, float %y) #2 {
256; CHECK-LABEL: define float @test_fmax_f32_strictfp
257; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) #[[ATTR1:[0-9]+]] {
258; CHECK-NEXT:    [[FMAX:%.*]] = tail call nnan nsz float @_Z4fmaxff(float [[X]], float [[Y]]) #[[ATTR1]]
259; CHECK-NEXT:    ret float [[FMAX]]
260;
261  %fmax = tail call nsz nnan float @_Z4fmaxff(float %x, float %y) #2
262  ret float %fmax
263}
264
265define float @test_fmax_f32_fast_nobuiltin(float %x, float %y) {
266; CHECK-LABEL: define float @test_fmax_f32_fast_nobuiltin
267; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
268; CHECK-NEXT:    [[FMAX:%.*]] = tail call fast float @_Z4fmaxff(float [[X]], float [[Y]]) #[[ATTR4:[0-9]+]]
269; CHECK-NEXT:    ret float [[FMAX]]
270;
271  %fmax = tail call fast float @_Z4fmaxff(float %x, float %y) #3
272  ret float %fmax
273}
274
275attributes #0 = { minsize }
276attributes #1 = { noinline }
277attributes #2 = { strictfp }
278attributes #3 = { nobuiltin }
279