1 /* Test vmov_n works correctly. */
2 /* { dg-do run } */
3 /* { dg-options "-O3 --save-temps" } */
4
5 #include <arm_neon.h>
6
7 extern void abort (void);
8
9 #define INHIB_OPTIMIZATION asm volatile ("" : : : "memory")
10
11 #define CONCAT(a, b) a##b
12 #define CONCAT1(a, b) CONCAT (a, b)
13 #define REG_INFEX64 _
14 #define REG_INFEX128 q_
15 #define REG_INFEX(reg_len) REG_INFEX##reg_len
16 #define POSTFIX_N(reg_len, data_len, data_type) \
17 CONCAT1 (REG_INFEX (reg_len), n_##data_type##data_len)
18 #define LANE_POSTFIX(reg_len, data_len, data_type) \
19 CONCAT1 (REG_INFEX (reg_len),lane_##data_type##data_len)
20
21 /* Test values consist of bytes with following hex values.
22 For example:
23 TEST1 for int16_t will be 0xaaaa
24 TEST1 for int32_t will be 0xaaaaaaaa
25 etc. */
26
27 #define TEST1h aa
28 #define TEST2h 55
29 #define TEST3h ff
30 #define TEST4h 00
31 #define TEST5h cc
32 #define TEST6h 33
33
34 #define TESTh_8(x) TEST##x##h
35 #define TESTh_16(x) CONCAT1 (TESTh_8 (x), TESTh_8 (x))
36 #define TESTh_32(x) CONCAT1 (TESTh_16 (x), TESTh_16 (x))
37 #define TESTh_64(x) CONCAT1 (TESTh_32 (x), TESTh_32 (x))
38
39 #define TEST_8(x) CONCAT1 (0x, TESTh_8 (x))
40 #define TEST_16(x) CONCAT1 (0x, TESTh_16 (x))
41 #define TEST_32(x) CONCAT1 (0x, TESTh_32 (x))
42 #define TEST_64(x) CONCAT1 (0x, TESTh_64 (x))
43
44 #define TEST(test, data_len) \
45 CONCAT1 (TEST, _##data_len) (test)
46
47 #define GET_ELEMENT(reg_len, data_len, data_type) \
48 CONCAT1 (vget, LANE_POSTFIX (reg_len, data_len, data_type))
49
50 #define VMOV_INST(reg_len, data_len, data_type) \
51 CONCAT1 (vmov, POSTFIX_N (reg_len, data_len, data_type))
52
53 #define VMOV_OBSCURE_INST(reg_len, data_len, data_type) \
54 CONCAT1 (VMOV_INST (reg_len, data_len, data_type), _obscure)
55
56 #define RUN_TEST(reg_len, data_len, data_type, \
57 test, n, a, b, c) \
58 { \
59 int i; \
60 INHIB_OPTIMIZATION; \
61 (a) = TEST (test, data_len); \
62 INHIB_OPTIMIZATION; \
63 (b) = VMOV_OBSCURE_INST (reg_len, data_len, data_type) (&(a)); \
64 (c) = TEST (test, data_len); \
65 for (i = 0; i < n; i++) \
66 { \
67 INHIB_OPTIMIZATION; \
68 a = GET_ELEMENT (reg_len, data_len, data_type) (b, i); \
69 if ((a) != (c)) \
70 return 1; \
71 } \
72 }
73
74 #define TYPE_f32 float32_t
75 #define TYPE_64_f32 float32x2_t
76 #define TYPE_128_f32 float32x4_t
77
78 #define TYPE_f64 float64_t
79 #define TYPE_64_f64 float64x1_t
80 #define TYPE_128_f64 float64x2_t
81
82 #define TYPE_s8 int8_t
83 #define TYPE_64_s8 int8x8_t
84 #define TYPE_128_s8 int8x16_t
85
86 #define TYPE_s16 int16_t
87 #define TYPE_64_s16 int16x4_t
88 #define TYPE_128_s16 int16x8_t
89
90 #define TYPE_s32 int32_t
91 #define TYPE_64_s32 int32x2_t
92 #define TYPE_128_s32 int32x4_t
93
94 #define TYPE_s64 int64_t
95 #define TYPE_64_s64 int64x1_t
96 #define TYPE_128_s64 int64x2_t
97
98 #define TYPE_u8 uint8_t
99 #define TYPE_64_u8 uint8x8_t
100 #define TYPE_128_u8 uint8x16_t
101
102 #define TYPE_u16 uint16_t
103 #define TYPE_64_u16 uint16x4_t
104 #define TYPE_128_u16 uint16x8_t
105
106 #define TYPE_u32 uint32_t
107 #define TYPE_64_u32 uint32x2_t
108 #define TYPE_128_u32 uint32x4_t
109
110 #define TYPE_u64 uint64_t
111 #define TYPE_64_u64 uint64x1_t
112 #define TYPE_128_u64 uint64x2_t
113
114 #define TYPE_p8 poly8_t
115 #define TYPE_64_p8 poly8x8_t
116 #define TYPE_128_p8 poly8x16_t
117
118 #define TYPE_p16 poly16_t
119 #define TYPE_64_p16 poly16x4_t
120 #define TYPE_128_p16 poly16x8_t
121
122 #define DIV64_8 8
123 #define DIV64_16 4
124 #define DIV64_32 2
125 #define DIV64_64 1
126
127 #define DIV128_8 16
128 #define DIV128_16 8
129 #define DIV128_32 4
130 #define DIV128_64 2
131
132 #define DIV(reg_len, data_len) \
133 CONCAT1 (CONCAT1 (DIV, reg_len), \
134 CONCAT1 (_, data_len))
135
136 #define VECTOR_TYPE(reg_len, data_len, data_type) \
137 CONCAT1 (CONCAT1 (CONCAT1 (TYPE_,reg_len), \
138 CONCAT1 (_,data_type)), \
139 data_len)
140
141 #define SIMPLE_TYPE(data_len, data_type) \
142 CONCAT1 (TYPE_, \
143 CONCAT1 (data_type, \
144 data_len))
145
146 #define OBSCURE_FUNC_NAME(reg_len, data_type, data_len) \
147 CONCAT1 (CONCAT1 (vmov, \
148 POSTFIX_N (reg_len, data_len, data_type)), \
149 _obscure)
150
151 #define OBSCURE_FUNC(reg_len, data_len, data_type) \
152 VECTOR_TYPE (reg_len, data_len, data_type) \
153 __attribute__ ((noinline)) \
154 OBSCURE_FUNC_NAME (reg_len, data_type, data_len) \
155 (SIMPLE_TYPE (data_len, data_type) *ap) \
156 { \
157 SIMPLE_TYPE (data_len, data_type) register a; \
158 INHIB_OPTIMIZATION; \
159 a = *ap; \
160 INHIB_OPTIMIZATION; \
161 return VMOV_INST (reg_len, data_len, data_type) (a); \
162 }
163
164 #define TESTFUNC_NAME(reg_len, data_type, data_len) \
165 CONCAT1 (test_vmov, \
166 POSTFIX_N (reg_len, data_len, data_type))
167
168 #define TESTFUNC(reg_len, data_len, data_type) \
169 int \
170 TESTFUNC_NAME (reg_len, data_type, data_len) () \
171 { \
172 SIMPLE_TYPE (data_len, data_type) a; \
173 VECTOR_TYPE (reg_len, data_len, data_type) b; \
174 SIMPLE_TYPE (data_len, data_type) c; \
175 \
176 RUN_TEST (reg_len, data_len, data_type, 1, \
177 DIV (reg_len, data_len), a, b, c); \
178 RUN_TEST (reg_len, data_len, data_type, 2, \
179 DIV (reg_len, data_len), a, b, c); \
180 RUN_TEST (reg_len, data_len, data_type, 3, \
181 DIV (reg_len, data_len), a, b, c); \
182 RUN_TEST (reg_len, data_len, data_type, 4, \
183 DIV (reg_len, data_len), a, b, c); \
184 RUN_TEST (reg_len, data_len, data_type, 5, \
185 DIV (reg_len, data_len), a, b, c); \
186 RUN_TEST (reg_len, data_len, data_type, 6, \
187 DIV (reg_len, data_len), a, b, c); \
188 return 0; \
189 }
190
191 OBSCURE_FUNC (64, 32, f)
192 TESTFUNC (64, 32, f)
193 /* "dup Vd.2s, Rn" is less preferable than "dup Vd.2s, Vn.s[lane]". */
194 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.2s, w\[0-9\]+" } } */
195 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.2s, v\[0-9\]+\.s\\\[\[0-9\]+\\\]" 3 } } */
196
197 OBSCURE_FUNC (64, 64, f)
198 TESTFUNC (64, 64, f)
199 /* "fmov Dd, Rn" is generated instead of "dup Dd, Rn".
200 No assembley scan included. */
201
202 OBSCURE_FUNC (64, 8, p)
203 TESTFUNC (64, 8, p)
204 /* Generates "dup Vd.8b, Rn". Scan found near s8 version. */
205
206 OBSCURE_FUNC (64, 16, p)
207 TESTFUNC (64, 16, p)
208 /* Generates "dup Vd.4h, Rn". Scan found near s16 version. */
209
210 OBSCURE_FUNC (64, 8, s)
211 TESTFUNC (64, 8, s)
212 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.8b, w\[0-9\]+" 1 } } */
213
214 OBSCURE_FUNC (64, 16, s)
215 TESTFUNC (64, 16, s)
216 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.4h, w\[0-9\]+" 1 } } */
217
218 OBSCURE_FUNC (64, 32, s)
219 TESTFUNC (64, 32, s)
220 /* "dup Vd.2s, Rn" is less preferable than "dup Vd.2s, Vn.s[lane]". */
221 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.2s, w\[0-9\]+" } } */
222 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.2s, v\[0-9\]+\.s\\\[\[0-9\]+\\\]" 3 } } */
223
224 OBSCURE_FUNC (64, 64, s)
225 TESTFUNC (64, 64, s)
226 /* "fmov Dd, Rn" is generated instead of "dup Dd, Rn".
227 No assembley scan included. */
228
229 OBSCURE_FUNC (64, 8, u)
230 TESTFUNC (64, 8, u)
231 /* Generates "dup Vd.8b, Rn". Scan found near s8 version. */
232
233 OBSCURE_FUNC (64, 16, u)
234 TESTFUNC (64, 16, u)
235 /* Generates "dup Vd.4h, Rn". Scan found near s16 version. */
236
237 OBSCURE_FUNC (64, 32, u)
238 TESTFUNC (64, 32, u)
239 /* Generates "dup Vd.2s, Rn". Scan found near s32 version. */
240
241 OBSCURE_FUNC (64, 64, u)
242 TESTFUNC (64, 64, u)
243 /* "fmov Dd, Rn" is generated instead of "dup Dd, Rn".
244 No assembley scan included. */
245
246 OBSCURE_FUNC (128, 32, f)
247 TESTFUNC (128, 32, f)
248 /* "dup Vd.4s, Rn" is less preferable than "dup Vd.4s, Vn.s[lane]". */
249 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.4s, w\[0-9\]+" } } */
250 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.4s, v\[0-9\]+\.s\\\[\[0-9\]+\\\]" 3 } } */
251
252 OBSCURE_FUNC (128, 64, f)
253 TESTFUNC (128, 64, f)
254 /* "dup Vd.2d, Rn" is less preferable than "dup Vd.2d, Vn.d[lane]". */
255 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.2d, x\[0-9\]+" } } */
256 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.2d, v\[0-9\]+\.d\\\[\[0-9\]+\\\]" 3 } } */
257
258 OBSCURE_FUNC (128, 8, p)
259 TESTFUNC (128, 8, p)
260 /* Generates "dup Vd.16b, Rn". Scan found near s8 version. */
261
262 OBSCURE_FUNC (128, 16, p)
263 TESTFUNC (128, 16, p)
264 /* Generates "dup Vd.8h, Rn". Scan found near s16 version. */
265
266 OBSCURE_FUNC (128, 8, s)
267 TESTFUNC (128, 8, s)
268 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.16b, w\[0-9\]+" 1 } } */
269
270 OBSCURE_FUNC (128, 16, s)
271 TESTFUNC (128, 16, s)
272 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.8h, w\[0-9\]+" 1 } } */
273
274 OBSCURE_FUNC (128, 32, s)
275 TESTFUNC (128, 32, s)
276 /* "dup Vd.4s, Rn" is less preferable than "dup Vd.4s, Vn.s[lane]". */
277 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.4s, w\[0-9\]+" } } */
278 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.4s, v\[0-9\]+\.s\\\[\[0-9\]+\\\]" 3 } } */
279
280 OBSCURE_FUNC (128, 64, s)
281 TESTFUNC (128, 64, s)
282 /* "dup Vd.2d, Rn" is less preferable than "dup Vd.2d, Vn.d[lane]". */
283 /* { dg-final { scan-assembler-not "dup\\tv\[0-9\]+\.2d, x\[0-9\]+" } } */
284 /* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.2d, v\[0-9\]+\.d\\\[\[0-9\]+\\\]" 3 } } */
285
286 OBSCURE_FUNC (128, 8, u)
287 TESTFUNC (128, 8, u)
288 /* Generates "dup Vd.16b, Rn". Scan found near s8 version. */
289
290 OBSCURE_FUNC (128, 16, u)
291 TESTFUNC (128, 16, u)
292 /* Generates "dup Vd.8h, Rn". Scan found near s16 version. */
293
294 OBSCURE_FUNC (128, 32, u)
295 TESTFUNC (128, 32, u)
296 /* Generates "dup Vd.4s, Rn". Scan found near s32 version. */
297
298 OBSCURE_FUNC (128, 64, u)
299 TESTFUNC (128, 64, u)
300 /* Generates "dup Vd.2d, Rn". Scan found near s64 version. */
301
302 int
303 main (int argc, char **argv)
304 {
305 if (test_vmov_n_f32 ())
306 abort ();
307 if (test_vmov_n_f64 ())
308 abort ();
309 if (test_vmov_n_p8 ())
310 abort ();
311 if (test_vmov_n_p16 ())
312 abort ();
313 if (test_vmov_n_s8 ())
314 abort ();
315 if (test_vmov_n_s16 ())
316 abort ();
317 if (test_vmov_n_s32 ())
318 abort ();
319 if (test_vmov_n_s64 ())
320 abort ();
321 if (test_vmov_n_u8 ())
322 abort ();
323 if (test_vmov_n_u16 ())
324 abort ();
325 if (test_vmov_n_u32 ())
326 abort ();
327 if (test_vmov_n_u64 ())
328 abort ();
329
330 if (test_vmovq_n_f32 ())
331 abort ();
332 if (test_vmovq_n_f64 ())
333 abort ();
334 if (test_vmovq_n_p8 ())
335 abort ();
336 if (test_vmovq_n_p16 ())
337 abort ();
338 if (test_vmovq_n_s8 ())
339 abort ();
340 if (test_vmovq_n_s16 ())
341 abort ();
342 if (test_vmovq_n_s32 ())
343 abort ();
344 if (test_vmovq_n_s64 ())
345 abort ();
346 if (test_vmovq_n_u8 ())
347 abort ();
348 if (test_vmovq_n_u16 ())
349 abort ();
350 if (test_vmovq_n_u32 ())
351 abort ();
352 if (test_vmovq_n_u64 ())
353 abort ();
354
355 return 0;
356 }
357