Debian Codegraph @

gcc/testsuite/gcc.target/i386/part-vect-roundhf.c

raw
1 /* { dg-do run { target avx512fp16 } } */
2 /* { dg-options "-O1 -mavx512fp16 -mavx512vl -fdump-tree-slp-details -fdump-tree-optimized" } */
3
4 extern void abort ();
5
6 static void do_test (void);
7
8 #define DO_TEST do_test
9 #define AVX512FP16
10 #include "avx512-check.h"
11
12 #define N 16
13 _Float16 b[N] = {-1.2f, 3.4f, -5.6f, 7.8f,
14 -9.0f, 1.0f, -2.0f, 3.0f,
15 -4.0f, -5.0f, 6.0f, 7.0f,
16 -8.0f, -9.0f, 10.0f, 11.0f};
17 _Float16 r[N];
18
19 void
20 __attribute__((noipa,noinline,optimize("Ofast")))
21 round_32 (void)
22 {
23 r[0] = __builtin_roundf16 (b[0]);
24 r[1] = __builtin_roundf16 (b[1]);
25 }
26
27 void
28 __attribute__((noipa,noinline,optimize("Ofast")))
29 round_64 (void)
30 {
31 r[0] = __builtin_roundf16 (b[0]);
32 r[1] = __builtin_roundf16 (b[1]);
33 r[2] = __builtin_roundf16 (b[2]);
34 r[3] = __builtin_roundf16 (b[3]);
35 }
36
37 void
38 __attribute__((noipa,noinline,optimize("O2")))
39 rint_32 (void)
40 {
41 r[0] = __builtin_rintf16 (b[0]);
42 r[1] = __builtin_rintf16 (b[1]);
43 }
44
45 void
46 __attribute__((noipa,noinline,optimize("O2")))
47 rint_64 (void)
48 {
49 r[0] = __builtin_rintf16 (b[0]);
50 r[1] = __builtin_rintf16 (b[1]);
51 r[2] = __builtin_rintf16 (b[2]);
52 r[3] = __builtin_rintf16 (b[3]);
53 }
54
55 void
56 __attribute__((noipa,noinline,optimize("O2")))
57 nearbyint_32 (void)
58 {
59 r[0] = __builtin_nearbyintf16 (b[0]);
60 r[1] = __builtin_nearbyintf16 (b[1]);
61 }
62
63 void
64 __attribute__((noipa,noinline,optimize("O2")))
65 nearbyint_64 (void)
66 {
67 r[0] = __builtin_nearbyintf16 (b[0]);
68 r[1] = __builtin_nearbyintf16 (b[1]);
69 r[2] = __builtin_nearbyintf16 (b[2]);
70 r[3] = __builtin_nearbyintf16 (b[3]);
71 }
72
73 void
74 __attribute__((noipa,noinline,optimize("Ofast")))
75 trunc_32 (void)
76 {
77 r[0] = __builtin_truncf16 (b[0]);
78 r[1] = __builtin_truncf16 (b[1]);
79 }
80
81 void
82 __attribute__((noipa,noinline,optimize("Ofast")))
83 trunc_64 (void)
84 {
85 r[0] = __builtin_truncf16 (b[0]);
86 r[1] = __builtin_truncf16 (b[1]);
87 r[2] = __builtin_truncf16 (b[2]);
88 r[3] = __builtin_truncf16 (b[3]);
89 }
90
91 void
92 __attribute__((noipa,noinline,optimize("Ofast")))
93 floor_32 (void)
94 {
95 r[0] = __builtin_floorf16 (b[0]);
96 r[1] = __builtin_floorf16 (b[1]);
97 }
98
99 void
100 __attribute__((noipa,noinline,optimize("Ofast")))
101 floor_64 (void)
102 {
103 r[0] = __builtin_floorf16 (b[0]);
104 r[1] = __builtin_floorf16 (b[1]);
105 r[2] = __builtin_floorf16 (b[2]);
106 r[3] = __builtin_floorf16 (b[3]);
107 }
108
109 void
110 __attribute__((noipa,noinline,optimize("Ofast")))
111 ceil_32 (void)
112 {
113 r[0] = __builtin_ceilf16 (b[0]);
114 r[1] = __builtin_ceilf16 (b[1]);
115 }
116
117 void
118 __attribute__((noipa,noinline,optimize("Ofast")))
119 ceil_64 (void)
120 {
121 r[0] = __builtin_ceilf16 (b[0]);
122 r[1] = __builtin_ceilf16 (b[1]);
123 r[2] = __builtin_ceilf16 (b[2]);
124 r[3] = __builtin_ceilf16 (b[3]);
125 }
126
127 _Float16
128 __attribute__((noipa,noinline,optimize("Ofast")))
129 dummy_roundf16 (_Float16 a)
130 {
131 return __builtin_roundf16 (a);
132 }
133 static void
134 __attribute__ ((noinline, noclone))
135 do_test (void)
136 {
137 round_32 ();
138 /* check results: */
139 for (int i = 0; i != 2; i++)
140 if (r[i] != dummy_roundf16 (b[i]))
141 abort ();
142
143 round_64 ();
144 /* check results: */
145 for (int i = 0; i != 4; i++)
146 if (r[i] != dummy_roundf16 (b[i]))
147 abort ();
148
149 rint_32 ();
150 /* check results: */
151 for (int i = 0; i != 2; i++)
152 if (r[i] != __builtin_rintf16 (b[i]))
153 abort ();
154
155 rint_64 ();
156 /* check results: */
157 for (int i = 0; i != 4; i++)
158 if (r[i] != __builtin_rintf16 (b[i]))
159 abort ();
160
161 nearbyint_32 ();
162 /* check results: */
163 for (int i = 0; i != 2; i++)
164 if (r[i] != __builtin_nearbyintf16 (b[i]))
165 abort ();
166
167 nearbyint_64 ();
168 /* check results: */
169 for (int i = 0; i != 4; i++)
170 if (r[i] != __builtin_nearbyintf16 (b[i]))
171 abort ();
172
173 trunc_32 ();
174 /* check results: */
175 for (int i = 0; i != 2; i++)
176 if (r[i] != __builtin_truncf16 (b[i]))
177 abort ();
178
179 trunc_64 ();
180 /* check results: */
181 for (int i = 0; i != 4; i++)
182 if (r[i] != __builtin_truncf16 (b[i]))
183 abort ();
184
185 floor_32 ();
186 /* check results: */
187 for (int i = 0; i != 2; i++)
188 if (r[i] != __builtin_floorf16 (b[i]))
189 abort ();
190
191 floor_64 ();
192 /* check results: */
193 for (int i = 0; i != 4; i++)
194 if (r[i] != __builtin_floorf16 (b[i]))
195 abort ();
196
197 ceil_32 ();
198 /* check results: */
199 for (int i = 0; i != 2; i++)
200 if (r[i] != __builtin_ceilf16 (b[i]))
201 abort ();
202
203 ceil_64 ();
204 /* check results: */
205 for (int i = 0; i != 4; i++)
206 if (r[i] != __builtin_ceilf16 (b[i]))
207 abort ();
208 }
209
210 /* { dg-final { scan-tree-dump-times "vectorized using 8 byte vectors" 6 "slp2" { target { ! ia32 } } } } */
211 /* { dg-final { scan-tree-dump-times "vectorized using 4 byte vectors" 6 "slp2" { target { ! ia32 } } } } */
212 /* { dg-final { scan-tree-dump-times {(?n).CEIL \(vect} 2 "optimized" { target { ! ia32 } } } } */
213 /* { dg-final { scan-tree-dump-times {(?n).FLOOR \(vect} 2 "optimized" { target { ! ia32 } } } } */
214 /* { dg-final { scan-tree-dump-times {(?n).ROUND \(vect} 2 "optimized" { target { ! ia32 } } } } */
215 /* { dg-final { scan-tree-dump-times {(?n).RINT \(vect} 2 "optimized" { target { ! ia32 } } } } */
216 /* { dg-final { scan-tree-dump-times {(?n).NEARBYINT \(vect} 2 "optimized" { target { ! ia32 } } } } */
217 /* { dg-final { scan-tree-dump-times {(?n).TRUNC \(vect} 2 "optimized" { target { ! ia32 } } } } */