xref: /llvm-project/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-fma.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 @_Z3fmafff(float, float, float)
7declare <2 x float> @_Z3fmaDv2_fS_S_(<2 x float>, <2 x float>, <2 x float>)
8declare <3 x float> @_Z3fmaDv3_fS_S_(<3 x float>, <3 x float>, <3 x float>)
9declare <4 x float> @_Z3fmaDv4_fS_S_(<4 x float>, <4 x float>, <4 x float>)
10declare <8 x float> @_Z3fmaDv8_fS_S_(<8 x float>, <8 x float>, <8 x float>)
11declare <16 x float> @_Z3fmaDv16_fS_S_(<16 x float>, <16 x float>, <16 x float>)
12declare double @_Z3fmaddd(double, double, double)
13declare <2 x double> @_Z3fmaDv2_dS_S_(<2 x double>, <2 x double>, <2 x double>)
14declare <3 x double> @_Z3fmaDv3_dS_S_(<3 x double>, <3 x double>, <3 x double>)
15declare <4 x double> @_Z3fmaDv4_dS_S_(<4 x double>, <4 x double>, <4 x double>)
16declare <8 x double> @_Z3fmaDv8_dS_S_(<8 x double>, <8 x double>, <8 x double>)
17declare <16 x double> @_Z3fmaDv16_dS_S_(<16 x double>, <16 x double>, <16 x double>)
18declare half @_Z3fmaDhDhDh(half, half, half)
19declare <2 x half> @_Z3fmaDv2_DhS_S_(<2 x half>, <2 x half>, <2 x half>)
20declare <3 x half> @_Z3fmaDv3_DhS_S_(<3 x half>, <3 x half>, <3 x half>)
21declare <4 x half> @_Z3fmaDv4_DhS_S_(<4 x half>, <4 x half>, <4 x half>)
22declare <8 x half> @_Z3fmaDv8_DhS_S_(<8 x half>, <8 x half>, <8 x half>)
23declare <16 x half> @_Z3fmaDv16_DhS_S_(<16 x half>, <16 x half>, <16 x half>)
24
25define float @test_fma_f32(float %x, float %y, float %z) {
26; CHECK-LABEL: define float @test_fma_f32
27; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
28; CHECK-NEXT:    [[FMA:%.*]] = tail call float @llvm.fma.f32(float [[X]], float [[Y]], float [[Z]])
29; CHECK-NEXT:    ret float [[FMA]]
30;
31  %fma = tail call float @_Z3fmafff(float %x, float %y, float %z)
32  ret float %fma
33}
34
35define <2 x float> @test_fma_v2f32(<2 x float> %x, <2 x float> %y, <2 x float> %z) {
36; CHECK-LABEL: define <2 x float> @test_fma_v2f32
37; CHECK-SAME: (<2 x float> [[X:%.*]], <2 x float> [[Y:%.*]], <2 x float> [[Z:%.*]]) {
38; CHECK-NEXT:    [[FMA:%.*]] = tail call <2 x float> @llvm.fma.v2f32(<2 x float> [[X]], <2 x float> [[Y]], <2 x float> [[Z]])
39; CHECK-NEXT:    ret <2 x float> [[FMA]]
40;
41  %fma = tail call <2 x float> @_Z3fmaDv2_fS_S_(<2 x float> %x, <2 x float> %y, <2 x float> %z)
42  ret <2 x float> %fma
43}
44
45define <3 x float> @test_fma_v3f32(<3 x float> %x, <3 x float> %y, <3 x float> %z) {
46; CHECK-LABEL: define <3 x float> @test_fma_v3f32
47; CHECK-SAME: (<3 x float> [[X:%.*]], <3 x float> [[Y:%.*]], <3 x float> [[Z:%.*]]) {
48; CHECK-NEXT:    [[FMA:%.*]] = tail call <3 x float> @llvm.fma.v3f32(<3 x float> [[X]], <3 x float> [[Y]], <3 x float> [[Z]])
49; CHECK-NEXT:    ret <3 x float> [[FMA]]
50;
51  %fma = tail call <3 x float> @_Z3fmaDv3_fS_S_(<3 x float> %x, <3 x float> %y, <3 x float> %z)
52  ret <3 x float> %fma
53}
54
55define <4 x float> @test_fma_v4f32(<4 x float> %x, <4 x float> %y, <4 x float> %z) {
56; CHECK-LABEL: define <4 x float> @test_fma_v4f32
57; CHECK-SAME: (<4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x float> [[Z:%.*]]) {
58; CHECK-NEXT:    [[FMA:%.*]] = tail call <4 x float> @llvm.fma.v4f32(<4 x float> [[X]], <4 x float> [[Y]], <4 x float> [[Z]])
59; CHECK-NEXT:    ret <4 x float> [[FMA]]
60;
61  %fma = tail call <4 x float> @_Z3fmaDv4_fS_S_(<4 x float> %x, <4 x float> %y, <4 x float> %z)
62  ret <4 x float> %fma
63}
64
65define <8 x float> @test_fma_v8f32(<8 x float> %x, <8 x float> %y, <8 x float> %z) {
66; CHECK-LABEL: define <8 x float> @test_fma_v8f32
67; CHECK-SAME: (<8 x float> [[X:%.*]], <8 x float> [[Y:%.*]], <8 x float> [[Z:%.*]]) {
68; CHECK-NEXT:    [[FMA:%.*]] = tail call <8 x float> @llvm.fma.v8f32(<8 x float> [[X]], <8 x float> [[Y]], <8 x float> [[Z]])
69; CHECK-NEXT:    ret <8 x float> [[FMA]]
70;
71  %fma = tail call <8 x float> @_Z3fmaDv8_fS_S_(<8 x float> %x, <8 x float> %y, <8 x float> %z)
72  ret <8 x float> %fma
73}
74
75define <16 x float> @test_fma_v16f32(<16 x float> %x, <16 x float> %y, <16 x float> %z) {
76; CHECK-LABEL: define <16 x float> @test_fma_v16f32
77; CHECK-SAME: (<16 x float> [[X:%.*]], <16 x float> [[Y:%.*]], <16 x float> [[Z:%.*]]) {
78; CHECK-NEXT:    [[FMA:%.*]] = tail call <16 x float> @llvm.fma.v16f32(<16 x float> [[X]], <16 x float> [[Y]], <16 x float> [[Z]])
79; CHECK-NEXT:    ret <16 x float> [[FMA]]
80;
81  %fma = tail call <16 x float> @_Z3fmaDv16_fS_S_(<16 x float> %x, <16 x float> %y, <16 x float> %z)
82  ret <16 x float> %fma
83}
84
85define double @test_fma_f64(double %x, double %y, double %z) {
86; CHECK-LABEL: define double @test_fma_f64
87; CHECK-SAME: (double [[X:%.*]], double [[Y:%.*]], double [[Z:%.*]]) {
88; CHECK-NEXT:    [[FMA:%.*]] = tail call double @llvm.fma.f64(double [[X]], double [[Y]], double [[Z]])
89; CHECK-NEXT:    ret double [[FMA]]
90;
91  %fma = tail call double @_Z3fmaddd(double %x, double %y, double %z)
92  ret double %fma
93}
94
95define <2 x double> @test_fma_v2f64(<2 x double> %x, <2 x double> %y, <2 x double> %z) {
96; CHECK-LABEL: define <2 x double> @test_fma_v2f64
97; CHECK-SAME: (<2 x double> [[X:%.*]], <2 x double> [[Y:%.*]], <2 x double> [[Z:%.*]]) {
98; CHECK-NEXT:    [[FMA:%.*]] = tail call <2 x double> @llvm.fma.v2f64(<2 x double> [[X]], <2 x double> [[Y]], <2 x double> [[Z]])
99; CHECK-NEXT:    ret <2 x double> [[FMA]]
100;
101  %fma = tail call <2 x double> @_Z3fmaDv2_dS_S_(<2 x double> %x, <2 x double> %y, <2 x double> %z)
102  ret <2 x double> %fma
103}
104
105define <3 x double> @test_fma_v3f64(<3 x double> %x, <3 x double> %y, <3 x double> %z) {
106; CHECK-LABEL: define <3 x double> @test_fma_v3f64
107; CHECK-SAME: (<3 x double> [[X:%.*]], <3 x double> [[Y:%.*]], <3 x double> [[Z:%.*]]) {
108; CHECK-NEXT:    [[FMA:%.*]] = tail call <3 x double> @llvm.fma.v3f64(<3 x double> [[X]], <3 x double> [[Y]], <3 x double> [[Z]])
109; CHECK-NEXT:    ret <3 x double> [[FMA]]
110;
111  %fma = tail call <3 x double> @_Z3fmaDv3_dS_S_(<3 x double> %x, <3 x double> %y, <3 x double> %z)
112  ret <3 x double> %fma
113}
114
115define <4 x double> @test_fma_v4f64(<4 x double> %x, <4 x double> %y, <4 x double> %z) {
116; CHECK-LABEL: define <4 x double> @test_fma_v4f64
117; CHECK-SAME: (<4 x double> [[X:%.*]], <4 x double> [[Y:%.*]], <4 x double> [[Z:%.*]]) {
118; CHECK-NEXT:    [[FMA:%.*]] = tail call <4 x double> @llvm.fma.v4f64(<4 x double> [[X]], <4 x double> [[Y]], <4 x double> [[Z]])
119; CHECK-NEXT:    ret <4 x double> [[FMA]]
120;
121  %fma = tail call <4 x double> @_Z3fmaDv4_dS_S_(<4 x double> %x, <4 x double> %y, <4 x double> %z)
122  ret <4 x double> %fma
123}
124
125define <8 x double> @test_fma_v8f64(<8 x double> %x, <8 x double> %y, <8 x double> %z) {
126; CHECK-LABEL: define <8 x double> @test_fma_v8f64
127; CHECK-SAME: (<8 x double> [[X:%.*]], <8 x double> [[Y:%.*]], <8 x double> [[Z:%.*]]) {
128; CHECK-NEXT:    [[FMA:%.*]] = tail call <8 x double> @llvm.fma.v8f64(<8 x double> [[X]], <8 x double> [[Y]], <8 x double> [[Z]])
129; CHECK-NEXT:    ret <8 x double> [[FMA]]
130;
131  %fma = tail call <8 x double> @_Z3fmaDv8_dS_S_(<8 x double> %x, <8 x double> %y, <8 x double> %z)
132  ret <8 x double> %fma
133}
134
135define <16 x double> @test_fma_v16f64(<16 x double> %x, <16 x double> %y, <16 x double> %z) {
136; CHECK-LABEL: define <16 x double> @test_fma_v16f64
137; CHECK-SAME: (<16 x double> [[X:%.*]], <16 x double> [[Y:%.*]], <16 x double> [[Z:%.*]]) {
138; CHECK-NEXT:    [[FMA:%.*]] = tail call <16 x double> @llvm.fma.v16f64(<16 x double> [[X]], <16 x double> [[Y]], <16 x double> [[Z]])
139; CHECK-NEXT:    ret <16 x double> [[FMA]]
140;
141  %fma = tail call <16 x double> @_Z3fmaDv16_dS_S_(<16 x double> %x, <16 x double> %y, <16 x double> %z)
142  ret <16 x double> %fma
143}
144
145define half @test_fma_f16(half %x, half %y, half %z) {
146; CHECK-LABEL: define half @test_fma_f16
147; CHECK-SAME: (half [[X:%.*]], half [[Y:%.*]], half [[Z:%.*]]) {
148; CHECK-NEXT:    [[FMA:%.*]] = tail call half @llvm.fma.f16(half [[X]], half [[Y]], half [[Z]])
149; CHECK-NEXT:    ret half [[FMA]]
150;
151  %fma = tail call half @_Z3fmaDhDhDh(half %x, half %y, half %z)
152  ret half %fma
153}
154
155define <2 x half> @test_fma_v2f16(<2 x half> %x, <2 x half> %y, <2 x half> %z) {
156; CHECK-LABEL: define <2 x half> @test_fma_v2f16
157; CHECK-SAME: (<2 x half> [[X:%.*]], <2 x half> [[Y:%.*]], <2 x half> [[Z:%.*]]) {
158; CHECK-NEXT:    [[FMA:%.*]] = tail call <2 x half> @llvm.fma.v2f16(<2 x half> [[X]], <2 x half> [[Y]], <2 x half> [[Z]])
159; CHECK-NEXT:    ret <2 x half> [[FMA]]
160;
161  %fma = tail call <2 x half> @_Z3fmaDv2_DhS_S_(<2 x half> %x, <2 x half> %y, <2 x half> %z)
162  ret <2 x half> %fma
163}
164
165define <3 x half> @test_fma_v3f16(<3 x half> %x, <3 x half> %y, <3 x half> %z) {
166; CHECK-LABEL: define <3 x half> @test_fma_v3f16
167; CHECK-SAME: (<3 x half> [[X:%.*]], <3 x half> [[Y:%.*]], <3 x half> [[Z:%.*]]) {
168; CHECK-NEXT:    [[FMA:%.*]] = tail call <3 x half> @llvm.fma.v3f16(<3 x half> [[X]], <3 x half> [[Y]], <3 x half> [[Z]])
169; CHECK-NEXT:    ret <3 x half> [[FMA]]
170;
171  %fma = tail call <3 x half> @_Z3fmaDv3_DhS_S_(<3 x half> %x, <3 x half> %y, <3 x half> %z)
172  ret <3 x half> %fma
173}
174
175define <4 x half> @test_fma_v4f16(<4 x half> %x, <4 x half> %y, <4 x half> %z) {
176; CHECK-LABEL: define <4 x half> @test_fma_v4f16
177; CHECK-SAME: (<4 x half> [[X:%.*]], <4 x half> [[Y:%.*]], <4 x half> [[Z:%.*]]) {
178; CHECK-NEXT:    [[FMA:%.*]] = tail call <4 x half> @llvm.fma.v4f16(<4 x half> [[X]], <4 x half> [[Y]], <4 x half> [[Z]])
179; CHECK-NEXT:    ret <4 x half> [[FMA]]
180;
181  %fma = tail call <4 x half> @_Z3fmaDv4_DhS_S_(<4 x half> %x, <4 x half> %y, <4 x half> %z)
182  ret <4 x half> %fma
183}
184
185define <8 x half> @test_fma_v8f16(<8 x half> %x, <8 x half> %y, <8 x half> %z) {
186; CHECK-LABEL: define <8 x half> @test_fma_v8f16
187; CHECK-SAME: (<8 x half> [[X:%.*]], <8 x half> [[Y:%.*]], <8 x half> [[Z:%.*]]) {
188; CHECK-NEXT:    [[FMA:%.*]] = tail call <8 x half> @llvm.fma.v8f16(<8 x half> [[X]], <8 x half> [[Y]], <8 x half> [[Z]])
189; CHECK-NEXT:    ret <8 x half> [[FMA]]
190;
191  %fma = tail call <8 x half> @_Z3fmaDv8_DhS_S_(<8 x half> %x, <8 x half> %y, <8 x half> %z)
192  ret <8 x half> %fma
193}
194
195define <16 x half> @test_fma_v16f16(<16 x half> %x, <16 x half> %y, <16 x half> %z) {
196; CHECK-LABEL: define <16 x half> @test_fma_v16f16
197; CHECK-SAME: (<16 x half> [[X:%.*]], <16 x half> [[Y:%.*]], <16 x half> [[Z:%.*]]) {
198; CHECK-NEXT:    [[FMA:%.*]] = tail call <16 x half> @llvm.fma.v16f16(<16 x half> [[X]], <16 x half> [[Y]], <16 x half> [[Z]])
199; CHECK-NEXT:    ret <16 x half> [[FMA]]
200;
201  %fma = tail call <16 x half> @_Z3fmaDv16_DhS_S_(<16 x half> %x, <16 x half> %y, <16 x half> %z)
202  ret <16 x half> %fma
203}
204
205define float @test_fma_f32_fast(float %x, float %y, float %z) {
206; CHECK-LABEL: define float @test_fma_f32_fast
207; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
208; CHECK-NEXT:    [[FMA:%.*]] = tail call fast float @llvm.fma.f32(float [[X]], float [[Y]], float [[Z]])
209; CHECK-NEXT:    ret float [[FMA]]
210;
211  %fma = tail call fast float @_Z3fmafff(float %x, float %y, float %z)
212  ret float %fma
213}
214
215define float @test_fma_f32_noinline(float %x, float %y, float %z) {
216; CHECK-LABEL: define float @test_fma_f32_noinline
217; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
218; CHECK-NEXT:    [[FMA:%.*]] = tail call fast float @_Z3fmafff(float [[X]], float [[Y]], float [[Z]]) #[[ATTR3:[0-9]+]]
219; CHECK-NEXT:    ret float [[FMA]]
220;
221  %fma = tail call fast float @_Z3fmafff(float %x, float %y, float %z) #1
222  ret float %fma
223}
224
225define float @test_fma_f32_fast_minsize(float %x, float %y, float %z) #0 {
226; CHECK-LABEL: define float @test_fma_f32_fast_minsize
227; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) #[[ATTR0:[0-9]+]] {
228; CHECK-NEXT:    [[FMA:%.*]] = tail call fast float @llvm.fma.f32(float [[X]], float [[Y]], float [[Z]])
229; CHECK-NEXT:    ret float [[FMA]]
230;
231  %fma = tail call fast float @_Z3fmafff(float %x, float %y, float %z)
232  ret float %fma
233}
234
235define float @test_fma_f32_fast_strictfp(float %x, float %y, float %z) #2 {
236; CHECK-LABEL: define float @test_fma_f32_fast_strictfp
237; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) #[[ATTR1:[0-9]+]] {
238; CHECK-NEXT:    [[FMA:%.*]] = tail call nnan nsz float @_Z3fmafff(float [[X]], float [[Y]], float [[Z]]) #[[ATTR1]]
239; CHECK-NEXT:    ret float [[FMA]]
240;
241  %fma = tail call nsz nnan float @_Z3fmafff(float %x, float %y, float %z) #2
242  ret float %fma
243}
244
245define float @test_fma_f32_fast_nobuiltin(float %x, float %y, float %z) {
246; CHECK-LABEL: define float @test_fma_f32_fast_nobuiltin
247; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
248; CHECK-NEXT:    [[FMA:%.*]] = tail call fast float @_Z3fmafff(float [[X]], float [[Y]], float [[Z]]) #[[ATTR4:[0-9]+]]
249; CHECK-NEXT:    ret float [[FMA]]
250;
251  %fma = tail call fast float @_Z3fmafff(float %x, float %y, float %z) #3
252  ret float %fma
253}
254
255attributes #0 = { minsize }
256attributes #1 = { noinline }
257attributes #2 = { strictfp }
258attributes #3 = { nobuiltin }
259