]>
Commit | Line | Data |
---|---|---|
e677137d PB |
1 | /* |
2 | * ARM NEON vector operations. | |
3 | * | |
4 | * Copyright (c) 2007, 2008 CodeSourcery. | |
5 | * Written by Paul Brook | |
6 | * | |
7 | * This code is licenced under the GNU GPL v2. | |
8 | */ | |
ad69471c PB |
9 | #include <stdlib.h> |
10 | #include <stdio.h> | |
11 | ||
12 | #include "cpu.h" | |
13 | #include "exec-all.h" | |
14 | #include "helpers.h" | |
15 | ||
16 | #define SIGNBIT (uint32_t)0x80000000 | |
17 | #define SIGNBIT64 ((uint64_t)1 << 63) | |
18 | ||
19 | #define SET_QC() env->vfp.xregs[ARM_VFP_FPSCR] = CPSR_Q | |
20 | ||
21 | static float_status neon_float_status; | |
22 | #define NFS &neon_float_status | |
23 | ||
24 | /* Helper routines to perform bitwise copies between float and int. */ | |
25 | static inline float32 vfp_itos(uint32_t i) | |
26 | { | |
27 | union { | |
28 | uint32_t i; | |
29 | float32 s; | |
30 | } v; | |
31 | ||
32 | v.i = i; | |
33 | return v.s; | |
34 | } | |
35 | ||
36 | static inline uint32_t vfp_stoi(float32 s) | |
37 | { | |
38 | union { | |
39 | uint32_t i; | |
40 | float32 s; | |
41 | } v; | |
42 | ||
43 | v.s = s; | |
44 | return v.i; | |
45 | } | |
46 | ||
47 | #define NEON_TYPE1(name, type) \ | |
48 | typedef struct \ | |
49 | { \ | |
50 | type v1; \ | |
51 | } neon_##name; | |
e2542fe2 | 52 | #ifdef HOST_WORDS_BIGENDIAN |
ad69471c PB |
53 | #define NEON_TYPE2(name, type) \ |
54 | typedef struct \ | |
55 | { \ | |
56 | type v2; \ | |
57 | type v1; \ | |
58 | } neon_##name; | |
59 | #define NEON_TYPE4(name, type) \ | |
60 | typedef struct \ | |
61 | { \ | |
62 | type v4; \ | |
63 | type v3; \ | |
64 | type v2; \ | |
65 | type v1; \ | |
66 | } neon_##name; | |
67 | #else | |
68 | #define NEON_TYPE2(name, type) \ | |
69 | typedef struct \ | |
70 | { \ | |
71 | type v1; \ | |
72 | type v2; \ | |
73 | } neon_##name; | |
74 | #define NEON_TYPE4(name, type) \ | |
75 | typedef struct \ | |
76 | { \ | |
77 | type v1; \ | |
78 | type v2; \ | |
79 | type v3; \ | |
80 | type v4; \ | |
81 | } neon_##name; | |
82 | #endif | |
83 | ||
84 | NEON_TYPE4(s8, int8_t) | |
85 | NEON_TYPE4(u8, uint8_t) | |
86 | NEON_TYPE2(s16, int16_t) | |
87 | NEON_TYPE2(u16, uint16_t) | |
88 | NEON_TYPE1(s32, int32_t) | |
89 | NEON_TYPE1(u32, uint32_t) | |
90 | #undef NEON_TYPE4 | |
91 | #undef NEON_TYPE2 | |
92 | #undef NEON_TYPE1 | |
93 | ||
94 | /* Copy from a uint32_t to a vector structure type. */ | |
95 | #define NEON_UNPACK(vtype, dest, val) do { \ | |
96 | union { \ | |
97 | vtype v; \ | |
98 | uint32_t i; \ | |
99 | } conv_u; \ | |
100 | conv_u.i = (val); \ | |
101 | dest = conv_u.v; \ | |
102 | } while(0) | |
103 | ||
104 | /* Copy from a vector structure type to a uint32_t. */ | |
105 | #define NEON_PACK(vtype, dest, val) do { \ | |
106 | union { \ | |
107 | vtype v; \ | |
108 | uint32_t i; \ | |
109 | } conv_u; \ | |
110 | conv_u.v = (val); \ | |
111 | dest = conv_u.i; \ | |
112 | } while(0) | |
113 | ||
114 | #define NEON_DO1 \ | |
115 | NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); | |
116 | #define NEON_DO2 \ | |
117 | NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \ | |
118 | NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2); | |
119 | #define NEON_DO4 \ | |
120 | NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \ | |
121 | NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2); \ | |
122 | NEON_FN(vdest.v3, vsrc1.v3, vsrc2.v3); \ | |
123 | NEON_FN(vdest.v4, vsrc1.v4, vsrc2.v4); | |
124 | ||
125 | #define NEON_VOP_BODY(vtype, n) \ | |
126 | { \ | |
127 | uint32_t res; \ | |
128 | vtype vsrc1; \ | |
129 | vtype vsrc2; \ | |
130 | vtype vdest; \ | |
131 | NEON_UNPACK(vtype, vsrc1, arg1); \ | |
132 | NEON_UNPACK(vtype, vsrc2, arg2); \ | |
133 | NEON_DO##n; \ | |
134 | NEON_PACK(vtype, res, vdest); \ | |
135 | return res; \ | |
136 | } | |
137 | ||
138 | #define NEON_VOP(name, vtype, n) \ | |
139 | uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \ | |
140 | NEON_VOP_BODY(vtype, n) | |
141 | ||
142 | #define NEON_VOP_ENV(name, vtype, n) \ | |
143 | uint32_t HELPER(glue(neon_,name))(CPUState *env, uint32_t arg1, uint32_t arg2) \ | |
144 | NEON_VOP_BODY(vtype, n) | |
145 | ||
146 | /* Pairwise operations. */ | |
147 | /* For 32-bit elements each segment only contains a single element, so | |
148 | the elementwise and pairwise operations are the same. */ | |
149 | #define NEON_PDO2 \ | |
150 | NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \ | |
151 | NEON_FN(vdest.v2, vsrc2.v1, vsrc2.v2); | |
152 | #define NEON_PDO4 \ | |
153 | NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \ | |
154 | NEON_FN(vdest.v2, vsrc1.v3, vsrc1.v4); \ | |
155 | NEON_FN(vdest.v3, vsrc2.v1, vsrc2.v2); \ | |
156 | NEON_FN(vdest.v4, vsrc2.v3, vsrc2.v4); \ | |
157 | ||
158 | #define NEON_POP(name, vtype, n) \ | |
159 | uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \ | |
160 | { \ | |
161 | uint32_t res; \ | |
162 | vtype vsrc1; \ | |
163 | vtype vsrc2; \ | |
164 | vtype vdest; \ | |
165 | NEON_UNPACK(vtype, vsrc1, arg1); \ | |
166 | NEON_UNPACK(vtype, vsrc2, arg2); \ | |
167 | NEON_PDO##n; \ | |
168 | NEON_PACK(vtype, res, vdest); \ | |
169 | return res; \ | |
170 | } | |
171 | ||
172 | /* Unary operators. */ | |
173 | #define NEON_VOP1(name, vtype, n) \ | |
174 | uint32_t HELPER(glue(neon_,name))(uint32_t arg) \ | |
175 | { \ | |
176 | vtype vsrc1; \ | |
177 | vtype vdest; \ | |
178 | NEON_UNPACK(vtype, vsrc1, arg); \ | |
179 | NEON_DO##n; \ | |
180 | NEON_PACK(vtype, arg, vdest); \ | |
181 | return arg; \ | |
182 | } | |
183 | ||
184 | ||
185 | #define NEON_USAT(dest, src1, src2, type) do { \ | |
186 | uint32_t tmp = (uint32_t)src1 + (uint32_t)src2; \ | |
187 | if (tmp != (type)tmp) { \ | |
188 | SET_QC(); \ | |
189 | dest = ~0; \ | |
190 | } else { \ | |
191 | dest = tmp; \ | |
192 | }} while(0) | |
193 | #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t) | |
194 | NEON_VOP_ENV(qadd_u8, neon_u8, 4) | |
195 | #undef NEON_FN | |
196 | #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t) | |
197 | NEON_VOP_ENV(qadd_u16, neon_u16, 2) | |
198 | #undef NEON_FN | |
199 | #undef NEON_USAT | |
200 | ||
201 | #define NEON_SSAT(dest, src1, src2, type) do { \ | |
202 | int32_t tmp = (uint32_t)src1 + (uint32_t)src2; \ | |
203 | if (tmp != (type)tmp) { \ | |
204 | SET_QC(); \ | |
205 | if (src2 > 0) { \ | |
206 | tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \ | |
207 | } else { \ | |
208 | tmp = 1 << (sizeof(type) * 8 - 1); \ | |
209 | } \ | |
210 | } \ | |
211 | dest = tmp; \ | |
212 | } while(0) | |
213 | #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t) | |
214 | NEON_VOP_ENV(qadd_s8, neon_s8, 4) | |
215 | #undef NEON_FN | |
216 | #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t) | |
217 | NEON_VOP_ENV(qadd_s16, neon_s16, 2) | |
218 | #undef NEON_FN | |
219 | #undef NEON_SSAT | |
220 | ||
221 | #define NEON_USAT(dest, src1, src2, type) do { \ | |
222 | uint32_t tmp = (uint32_t)src1 - (uint32_t)src2; \ | |
223 | if (tmp != (type)tmp) { \ | |
224 | SET_QC(); \ | |
225 | dest = 0; \ | |
226 | } else { \ | |
227 | dest = tmp; \ | |
228 | }} while(0) | |
229 | #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t) | |
230 | NEON_VOP_ENV(qsub_u8, neon_u8, 4) | |
231 | #undef NEON_FN | |
232 | #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t) | |
233 | NEON_VOP_ENV(qsub_u16, neon_u16, 2) | |
234 | #undef NEON_FN | |
235 | #undef NEON_USAT | |
236 | ||
237 | #define NEON_SSAT(dest, src1, src2, type) do { \ | |
238 | int32_t tmp = (uint32_t)src1 - (uint32_t)src2; \ | |
239 | if (tmp != (type)tmp) { \ | |
240 | SET_QC(); \ | |
241 | if (src2 < 0) { \ | |
242 | tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \ | |
243 | } else { \ | |
244 | tmp = 1 << (sizeof(type) * 8 - 1); \ | |
245 | } \ | |
246 | } \ | |
247 | dest = tmp; \ | |
248 | } while(0) | |
249 | #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t) | |
250 | NEON_VOP_ENV(qsub_s8, neon_s8, 4) | |
251 | #undef NEON_FN | |
252 | #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t) | |
253 | NEON_VOP_ENV(qsub_s16, neon_s16, 2) | |
254 | #undef NEON_FN | |
255 | #undef NEON_SSAT | |
256 | ||
257 | #define NEON_FN(dest, src1, src2) dest = (src1 + src2) >> 1 | |
258 | NEON_VOP(hadd_s8, neon_s8, 4) | |
259 | NEON_VOP(hadd_u8, neon_u8, 4) | |
260 | NEON_VOP(hadd_s16, neon_s16, 2) | |
261 | NEON_VOP(hadd_u16, neon_u16, 2) | |
262 | #undef NEON_FN | |
263 | ||
264 | int32_t HELPER(neon_hadd_s32)(int32_t src1, int32_t src2) | |
265 | { | |
266 | int32_t dest; | |
267 | ||
268 | dest = (src1 >> 1) + (src2 >> 1); | |
269 | if (src1 & src2 & 1) | |
270 | dest++; | |
271 | return dest; | |
272 | } | |
273 | ||
274 | uint32_t HELPER(neon_hadd_u32)(uint32_t src1, uint32_t src2) | |
275 | { | |
276 | uint32_t dest; | |
277 | ||
278 | dest = (src1 >> 1) + (src2 >> 1); | |
279 | if (src1 & src2 & 1) | |
280 | dest++; | |
281 | return dest; | |
282 | } | |
283 | ||
284 | #define NEON_FN(dest, src1, src2) dest = (src1 + src2 + 1) >> 1 | |
285 | NEON_VOP(rhadd_s8, neon_s8, 4) | |
286 | NEON_VOP(rhadd_u8, neon_u8, 4) | |
287 | NEON_VOP(rhadd_s16, neon_s16, 2) | |
288 | NEON_VOP(rhadd_u16, neon_u16, 2) | |
289 | #undef NEON_FN | |
290 | ||
291 | int32_t HELPER(neon_rhadd_s32)(int32_t src1, int32_t src2) | |
292 | { | |
293 | int32_t dest; | |
294 | ||
295 | dest = (src1 >> 1) + (src2 >> 1); | |
296 | if ((src1 | src2) & 1) | |
297 | dest++; | |
298 | return dest; | |
299 | } | |
300 | ||
301 | uint32_t HELPER(neon_rhadd_u32)(uint32_t src1, uint32_t src2) | |
302 | { | |
303 | uint32_t dest; | |
304 | ||
305 | dest = (src1 >> 1) + (src2 >> 1); | |
306 | if ((src1 | src2) & 1) | |
307 | dest++; | |
308 | return dest; | |
309 | } | |
310 | ||
311 | #define NEON_FN(dest, src1, src2) dest = (src1 - src2) >> 1 | |
312 | NEON_VOP(hsub_s8, neon_s8, 4) | |
313 | NEON_VOP(hsub_u8, neon_u8, 4) | |
314 | NEON_VOP(hsub_s16, neon_s16, 2) | |
315 | NEON_VOP(hsub_u16, neon_u16, 2) | |
316 | #undef NEON_FN | |
317 | ||
318 | int32_t HELPER(neon_hsub_s32)(int32_t src1, int32_t src2) | |
319 | { | |
320 | int32_t dest; | |
321 | ||
322 | dest = (src1 >> 1) - (src2 >> 1); | |
323 | if ((~src1) & src2 & 1) | |
324 | dest--; | |
325 | return dest; | |
326 | } | |
327 | ||
328 | uint32_t HELPER(neon_hsub_u32)(uint32_t src1, uint32_t src2) | |
329 | { | |
330 | uint32_t dest; | |
331 | ||
332 | dest = (src1 >> 1) - (src2 >> 1); | |
333 | if ((~src1) & src2 & 1) | |
334 | dest--; | |
335 | return dest; | |
336 | } | |
337 | ||
338 | #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? ~0 : 0 | |
339 | NEON_VOP(cgt_s8, neon_s8, 4) | |
340 | NEON_VOP(cgt_u8, neon_u8, 4) | |
341 | NEON_VOP(cgt_s16, neon_s16, 2) | |
342 | NEON_VOP(cgt_u16, neon_u16, 2) | |
343 | NEON_VOP(cgt_s32, neon_s32, 1) | |
344 | NEON_VOP(cgt_u32, neon_u32, 1) | |
345 | #undef NEON_FN | |
346 | ||
347 | #define NEON_FN(dest, src1, src2) dest = (src1 >= src2) ? ~0 : 0 | |
348 | NEON_VOP(cge_s8, neon_s8, 4) | |
349 | NEON_VOP(cge_u8, neon_u8, 4) | |
350 | NEON_VOP(cge_s16, neon_s16, 2) | |
351 | NEON_VOP(cge_u16, neon_u16, 2) | |
352 | NEON_VOP(cge_s32, neon_s32, 1) | |
353 | NEON_VOP(cge_u32, neon_u32, 1) | |
354 | #undef NEON_FN | |
355 | ||
356 | #define NEON_FN(dest, src1, src2) dest = (src1 < src2) ? src1 : src2 | |
357 | NEON_VOP(min_s8, neon_s8, 4) | |
358 | NEON_VOP(min_u8, neon_u8, 4) | |
359 | NEON_VOP(min_s16, neon_s16, 2) | |
360 | NEON_VOP(min_u16, neon_u16, 2) | |
361 | NEON_VOP(min_s32, neon_s32, 1) | |
362 | NEON_VOP(min_u32, neon_u32, 1) | |
363 | NEON_POP(pmin_s8, neon_s8, 4) | |
364 | NEON_POP(pmin_u8, neon_u8, 4) | |
365 | NEON_POP(pmin_s16, neon_s16, 2) | |
366 | NEON_POP(pmin_u16, neon_u16, 2) | |
367 | #undef NEON_FN | |
368 | ||
369 | #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? src1 : src2 | |
370 | NEON_VOP(max_s8, neon_s8, 4) | |
371 | NEON_VOP(max_u8, neon_u8, 4) | |
372 | NEON_VOP(max_s16, neon_s16, 2) | |
373 | NEON_VOP(max_u16, neon_u16, 2) | |
374 | NEON_VOP(max_s32, neon_s32, 1) | |
375 | NEON_VOP(max_u32, neon_u32, 1) | |
376 | NEON_POP(pmax_s8, neon_s8, 4) | |
377 | NEON_POP(pmax_u8, neon_u8, 4) | |
378 | NEON_POP(pmax_s16, neon_s16, 2) | |
379 | NEON_POP(pmax_u16, neon_u16, 2) | |
380 | #undef NEON_FN | |
381 | ||
382 | #define NEON_FN(dest, src1, src2) \ | |
383 | dest = (src1 > src2) ? (src1 - src2) : (src2 - src1) | |
384 | NEON_VOP(abd_s8, neon_s8, 4) | |
385 | NEON_VOP(abd_u8, neon_u8, 4) | |
386 | NEON_VOP(abd_s16, neon_s16, 2) | |
387 | NEON_VOP(abd_u16, neon_u16, 2) | |
388 | NEON_VOP(abd_s32, neon_s32, 1) | |
389 | NEON_VOP(abd_u32, neon_u32, 1) | |
390 | #undef NEON_FN | |
391 | ||
392 | #define NEON_FN(dest, src1, src2) do { \ | |
393 | int8_t tmp; \ | |
394 | tmp = (int8_t)src2; \ | |
50f67e95 JR |
395 | if (tmp >= (ssize_t)sizeof(src1) * 8 || \ |
396 | tmp <= -(ssize_t)sizeof(src1) * 8) { \ | |
ad69471c PB |
397 | dest = 0; \ |
398 | } else if (tmp < 0) { \ | |
399 | dest = src1 >> -tmp; \ | |
400 | } else { \ | |
401 | dest = src1 << tmp; \ | |
402 | }} while (0) | |
403 | NEON_VOP(shl_u8, neon_u8, 4) | |
404 | NEON_VOP(shl_u16, neon_u16, 2) | |
405 | NEON_VOP(shl_u32, neon_u32, 1) | |
406 | #undef NEON_FN | |
407 | ||
408 | uint64_t HELPER(neon_shl_u64)(uint64_t val, uint64_t shiftop) | |
409 | { | |
410 | int8_t shift = (int8_t)shiftop; | |
411 | if (shift >= 64 || shift <= -64) { | |
412 | val = 0; | |
413 | } else if (shift < 0) { | |
414 | val >>= -shift; | |
415 | } else { | |
416 | val <<= shift; | |
417 | } | |
418 | return val; | |
419 | } | |
420 | ||
421 | #define NEON_FN(dest, src1, src2) do { \ | |
422 | int8_t tmp; \ | |
423 | tmp = (int8_t)src2; \ | |
50f67e95 | 424 | if (tmp >= (ssize_t)sizeof(src1) * 8) { \ |
ad69471c | 425 | dest = 0; \ |
50f67e95 | 426 | } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
427 | dest = src1 >> (sizeof(src1) * 8 - 1); \ |
428 | } else if (tmp < 0) { \ | |
429 | dest = src1 >> -tmp; \ | |
430 | } else { \ | |
431 | dest = src1 << tmp; \ | |
432 | }} while (0) | |
433 | NEON_VOP(shl_s8, neon_s8, 4) | |
434 | NEON_VOP(shl_s16, neon_s16, 2) | |
435 | NEON_VOP(shl_s32, neon_s32, 1) | |
436 | #undef NEON_FN | |
437 | ||
438 | uint64_t HELPER(neon_shl_s64)(uint64_t valop, uint64_t shiftop) | |
439 | { | |
440 | int8_t shift = (int8_t)shiftop; | |
441 | int64_t val = valop; | |
442 | if (shift >= 64) { | |
443 | val = 0; | |
444 | } else if (shift <= -64) { | |
445 | val >>= 63; | |
446 | } else if (shift < 0) { | |
447 | val >>= -shift; | |
448 | } else { | |
449 | val <<= shift; | |
450 | } | |
451 | return val; | |
452 | } | |
453 | ||
454 | #define NEON_FN(dest, src1, src2) do { \ | |
455 | int8_t tmp; \ | |
456 | tmp = (int8_t)src2; \ | |
50f67e95 | 457 | if (tmp >= (ssize_t)sizeof(src1) * 8) { \ |
ad69471c | 458 | dest = 0; \ |
50f67e95 | 459 | } else if (tmp < -(ssize_t)sizeof(src1) * 8) { \ |
cb76e138 | 460 | dest = src1 >> (sizeof(src1) * 8 - 1); \ |
50f67e95 | 461 | } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
462 | dest = src1 >> (tmp - 1); \ |
463 | dest++; \ | |
cb76e138 | 464 | dest >>= 1; \ |
ad69471c PB |
465 | } else if (tmp < 0) { \ |
466 | dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \ | |
467 | } else { \ | |
468 | dest = src1 << tmp; \ | |
469 | }} while (0) | |
470 | NEON_VOP(rshl_s8, neon_s8, 4) | |
471 | NEON_VOP(rshl_s16, neon_s16, 2) | |
472 | NEON_VOP(rshl_s32, neon_s32, 1) | |
473 | #undef NEON_FN | |
474 | ||
475 | uint64_t HELPER(neon_rshl_s64)(uint64_t valop, uint64_t shiftop) | |
476 | { | |
477 | int8_t shift = (int8_t)shiftop; | |
478 | int64_t val = valop; | |
479 | if (shift >= 64) { | |
480 | val = 0; | |
481 | } else if (shift < -64) { | |
482 | val >>= 63; | |
483 | } else if (shift == -63) { | |
484 | val >>= 63; | |
485 | val++; | |
486 | val >>= 1; | |
487 | } else if (shift < 0) { | |
488 | val = (val + ((int64_t)1 << (-1 - shift))) >> -shift; | |
489 | } else { | |
490 | val <<= shift; | |
491 | } | |
492 | return val; | |
493 | } | |
494 | ||
495 | #define NEON_FN(dest, src1, src2) do { \ | |
496 | int8_t tmp; \ | |
497 | tmp = (int8_t)src2; \ | |
50f67e95 JR |
498 | if (tmp >= (ssize_t)sizeof(src1) * 8 || \ |
499 | tmp < -(ssize_t)sizeof(src1) * 8) { \ | |
ad69471c | 500 | dest = 0; \ |
50f67e95 | 501 | } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
502 | dest = src1 >> (tmp - 1); \ |
503 | } else if (tmp < 0) { \ | |
504 | dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \ | |
505 | } else { \ | |
506 | dest = src1 << tmp; \ | |
507 | }} while (0) | |
508 | NEON_VOP(rshl_u8, neon_u8, 4) | |
509 | NEON_VOP(rshl_u16, neon_u16, 2) | |
510 | NEON_VOP(rshl_u32, neon_u32, 1) | |
511 | #undef NEON_FN | |
512 | ||
513 | uint64_t HELPER(neon_rshl_u64)(uint64_t val, uint64_t shiftop) | |
514 | { | |
515 | int8_t shift = (uint8_t)shiftop; | |
516 | if (shift >= 64 || shift < 64) { | |
517 | val = 0; | |
518 | } else if (shift == -64) { | |
519 | /* Rounding a 1-bit result just preserves that bit. */ | |
520 | val >>= 63; | |
521 | } if (shift < 0) { | |
522 | val = (val + ((uint64_t)1 << (-1 - shift))) >> -shift; | |
523 | val >>= -shift; | |
524 | } else { | |
525 | val <<= shift; | |
526 | } | |
527 | return val; | |
528 | } | |
529 | ||
530 | #define NEON_FN(dest, src1, src2) do { \ | |
531 | int8_t tmp; \ | |
532 | tmp = (int8_t)src2; \ | |
50f67e95 | 533 | if (tmp >= (ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
534 | if (src1) { \ |
535 | SET_QC(); \ | |
536 | dest = ~0; \ | |
537 | } else { \ | |
538 | dest = 0; \ | |
539 | } \ | |
50f67e95 | 540 | } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
541 | dest = 0; \ |
542 | } else if (tmp < 0) { \ | |
543 | dest = src1 >> -tmp; \ | |
544 | } else { \ | |
545 | dest = src1 << tmp; \ | |
546 | if ((dest >> tmp) != src1) { \ | |
547 | SET_QC(); \ | |
548 | dest = ~0; \ | |
549 | } \ | |
550 | }} while (0) | |
551 | NEON_VOP_ENV(qshl_u8, neon_u8, 4) | |
552 | NEON_VOP_ENV(qshl_u16, neon_u16, 2) | |
553 | NEON_VOP_ENV(qshl_u32, neon_u32, 1) | |
554 | #undef NEON_FN | |
555 | ||
556 | uint64_t HELPER(neon_qshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop) | |
557 | { | |
558 | int8_t shift = (int8_t)shiftop; | |
559 | if (shift >= 64) { | |
560 | if (val) { | |
561 | val = ~(uint64_t)0; | |
562 | SET_QC(); | |
563 | } else { | |
564 | val = 0; | |
565 | } | |
566 | } else if (shift <= -64) { | |
567 | val = 0; | |
568 | } else if (shift < 0) { | |
569 | val >>= -shift; | |
570 | } else { | |
571 | uint64_t tmp = val; | |
572 | val <<= shift; | |
573 | if ((val >> shift) != tmp) { | |
574 | SET_QC(); | |
575 | val = ~(uint64_t)0; | |
576 | } | |
577 | } | |
578 | return val; | |
579 | } | |
580 | ||
581 | #define NEON_FN(dest, src1, src2) do { \ | |
582 | int8_t tmp; \ | |
583 | tmp = (int8_t)src2; \ | |
50f67e95 | 584 | if (tmp >= (ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
585 | if (src1) \ |
586 | SET_QC(); \ | |
587 | dest = src1 >> 31; \ | |
50f67e95 | 588 | } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \ |
ad69471c PB |
589 | dest = src1 >> 31; \ |
590 | } else if (tmp < 0) { \ | |
591 | dest = src1 >> -tmp; \ | |
592 | } else { \ | |
593 | dest = src1 << tmp; \ | |
594 | if ((dest >> tmp) != src1) { \ | |
595 | SET_QC(); \ | |
596 | dest = src2 >> 31; \ | |
597 | } \ | |
598 | }} while (0) | |
599 | NEON_VOP_ENV(qshl_s8, neon_s8, 4) | |
600 | NEON_VOP_ENV(qshl_s16, neon_s16, 2) | |
601 | NEON_VOP_ENV(qshl_s32, neon_s32, 1) | |
602 | #undef NEON_FN | |
603 | ||
604 | uint64_t HELPER(neon_qshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop) | |
605 | { | |
606 | int8_t shift = (uint8_t)shiftop; | |
607 | int64_t val = valop; | |
608 | if (shift >= 64) { | |
609 | if (val) { | |
610 | SET_QC(); | |
611 | val = (val >> 63) & ~SIGNBIT64; | |
612 | } | |
613 | } else if (shift <= 64) { | |
614 | val >>= 63; | |
615 | } else if (shift < 0) { | |
616 | val >>= -shift; | |
617 | } else { | |
618 | int64_t tmp = val; | |
619 | val <<= shift; | |
620 | if ((val >> shift) != tmp) { | |
621 | SET_QC(); | |
622 | val = (tmp >> 63) ^ ~SIGNBIT64; | |
623 | } | |
624 | } | |
625 | return val; | |
626 | } | |
627 | ||
628 | ||
629 | /* FIXME: This is wrong. */ | |
630 | #define NEON_FN(dest, src1, src2) do { \ | |
631 | int8_t tmp; \ | |
632 | tmp = (int8_t)src2; \ | |
633 | if (tmp < 0) { \ | |
634 | dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \ | |
635 | } else { \ | |
636 | dest = src1 << tmp; \ | |
637 | if ((dest >> tmp) != src1) { \ | |
638 | SET_QC(); \ | |
639 | dest = ~0; \ | |
640 | } \ | |
641 | }} while (0) | |
642 | NEON_VOP_ENV(qrshl_u8, neon_u8, 4) | |
643 | NEON_VOP_ENV(qrshl_u16, neon_u16, 2) | |
644 | NEON_VOP_ENV(qrshl_u32, neon_u32, 1) | |
645 | #undef NEON_FN | |
646 | ||
647 | uint64_t HELPER(neon_qrshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop) | |
648 | { | |
649 | int8_t shift = (int8_t)shiftop; | |
650 | if (shift < 0) { | |
651 | val = (val + (1 << (-1 - shift))) >> -shift; | |
652 | } else { \ | |
653 | uint64_t tmp = val; | |
654 | val <<= shift; | |
655 | if ((val >> shift) != tmp) { | |
656 | SET_QC(); | |
657 | val = ~0; | |
658 | } | |
659 | } | |
660 | return val; | |
661 | } | |
662 | ||
663 | #define NEON_FN(dest, src1, src2) do { \ | |
664 | int8_t tmp; \ | |
665 | tmp = (int8_t)src2; \ | |
666 | if (tmp < 0) { \ | |
667 | dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \ | |
668 | } else { \ | |
669 | dest = src1 << tmp; \ | |
670 | if ((dest >> tmp) != src1) { \ | |
671 | SET_QC(); \ | |
672 | dest = src1 >> 31; \ | |
673 | } \ | |
674 | }} while (0) | |
675 | NEON_VOP_ENV(qrshl_s8, neon_s8, 4) | |
676 | NEON_VOP_ENV(qrshl_s16, neon_s16, 2) | |
677 | NEON_VOP_ENV(qrshl_s32, neon_s32, 1) | |
678 | #undef NEON_FN | |
679 | ||
680 | uint64_t HELPER(neon_qrshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop) | |
681 | { | |
682 | int8_t shift = (uint8_t)shiftop; | |
683 | int64_t val = valop; | |
684 | ||
685 | if (shift < 0) { | |
686 | val = (val + (1 << (-1 - shift))) >> -shift; | |
687 | } else { | |
688 | int64_t tmp = val;; | |
689 | val <<= shift; | |
690 | if ((val >> shift) != tmp) { | |
691 | SET_QC(); | |
692 | val = tmp >> 31; | |
693 | } | |
694 | } | |
695 | return val; | |
696 | } | |
697 | ||
698 | uint32_t HELPER(neon_add_u8)(uint32_t a, uint32_t b) | |
699 | { | |
700 | uint32_t mask; | |
701 | mask = (a ^ b) & 0x80808080u; | |
702 | a &= ~0x80808080u; | |
703 | b &= ~0x80808080u; | |
704 | return (a + b) ^ mask; | |
705 | } | |
706 | ||
707 | uint32_t HELPER(neon_add_u16)(uint32_t a, uint32_t b) | |
708 | { | |
709 | uint32_t mask; | |
710 | mask = (a ^ b) & 0x80008000u; | |
711 | a &= ~0x80008000u; | |
712 | b &= ~0x80008000u; | |
713 | return (a + b) ^ mask; | |
714 | } | |
715 | ||
716 | #define NEON_FN(dest, src1, src2) dest = src1 + src2 | |
717 | NEON_POP(padd_u8, neon_u8, 4) | |
718 | NEON_POP(padd_u16, neon_u16, 2) | |
719 | #undef NEON_FN | |
720 | ||
721 | #define NEON_FN(dest, src1, src2) dest = src1 - src2 | |
722 | NEON_VOP(sub_u8, neon_u8, 4) | |
723 | NEON_VOP(sub_u16, neon_u16, 2) | |
724 | #undef NEON_FN | |
725 | ||
726 | #define NEON_FN(dest, src1, src2) dest = src1 * src2 | |
727 | NEON_VOP(mul_u8, neon_u8, 4) | |
728 | NEON_VOP(mul_u16, neon_u16, 2) | |
729 | #undef NEON_FN | |
730 | ||
1654b2d6 | 731 | /* Polynomial multiplication is like integer multiplication except the |
ad69471c PB |
732 | partial products are XORed, not added. */ |
733 | uint32_t HELPER(neon_mul_p8)(uint32_t op1, uint32_t op2) | |
734 | { | |
735 | uint32_t mask; | |
736 | uint32_t result; | |
737 | result = 0; | |
738 | while (op1) { | |
739 | mask = 0; | |
740 | if (op1 & 1) | |
741 | mask |= 0xff; | |
742 | if (op1 & (1 << 8)) | |
743 | mask |= (0xff << 8); | |
744 | if (op1 & (1 << 16)) | |
745 | mask |= (0xff << 16); | |
746 | if (op1 & (1 << 24)) | |
747 | mask |= (0xff << 24); | |
748 | result ^= op2 & mask; | |
749 | op1 = (op1 >> 1) & 0x7f7f7f7f; | |
750 | op2 = (op2 << 1) & 0xfefefefe; | |
751 | } | |
752 | return result; | |
753 | } | |
754 | ||
755 | #define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0 | |
756 | NEON_VOP(tst_u8, neon_u8, 4) | |
757 | NEON_VOP(tst_u16, neon_u16, 2) | |
758 | NEON_VOP(tst_u32, neon_u32, 1) | |
759 | #undef NEON_FN | |
760 | ||
761 | #define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0 | |
762 | NEON_VOP(ceq_u8, neon_u8, 4) | |
763 | NEON_VOP(ceq_u16, neon_u16, 2) | |
764 | NEON_VOP(ceq_u32, neon_u32, 1) | |
765 | #undef NEON_FN | |
766 | ||
767 | #define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src | |
768 | NEON_VOP1(abs_s8, neon_s8, 4) | |
769 | NEON_VOP1(abs_s16, neon_s16, 2) | |
770 | #undef NEON_FN | |
771 | ||
772 | /* Count Leading Sign/Zero Bits. */ | |
773 | static inline int do_clz8(uint8_t x) | |
774 | { | |
775 | int n; | |
776 | for (n = 8; x; n--) | |
777 | x >>= 1; | |
778 | return n; | |
779 | } | |
780 | ||
781 | static inline int do_clz16(uint16_t x) | |
782 | { | |
783 | int n; | |
784 | for (n = 16; x; n--) | |
785 | x >>= 1; | |
786 | return n; | |
787 | } | |
788 | ||
789 | #define NEON_FN(dest, src, dummy) dest = do_clz8(src) | |
790 | NEON_VOP1(clz_u8, neon_u8, 4) | |
791 | #undef NEON_FN | |
792 | ||
793 | #define NEON_FN(dest, src, dummy) dest = do_clz16(src) | |
794 | NEON_VOP1(clz_u16, neon_u16, 2) | |
795 | #undef NEON_FN | |
796 | ||
797 | #define NEON_FN(dest, src, dummy) dest = do_clz8((src < 0) ? ~src : src) - 1 | |
798 | NEON_VOP1(cls_s8, neon_s8, 4) | |
799 | #undef NEON_FN | |
800 | ||
801 | #define NEON_FN(dest, src, dummy) dest = do_clz16((src < 0) ? ~src : src) - 1 | |
802 | NEON_VOP1(cls_s16, neon_s16, 2) | |
803 | #undef NEON_FN | |
804 | ||
805 | uint32_t HELPER(neon_cls_s32)(uint32_t x) | |
806 | { | |
807 | int count; | |
808 | if ((int32_t)x < 0) | |
809 | x = ~x; | |
810 | for (count = 32; x; count--) | |
811 | x = x >> 1; | |
812 | return count - 1; | |
813 | } | |
814 | ||
815 | /* Bit count. */ | |
816 | uint32_t HELPER(neon_cnt_u8)(uint32_t x) | |
817 | { | |
818 | x = (x & 0x55555555) + ((x >> 1) & 0x55555555); | |
819 | x = (x & 0x33333333) + ((x >> 2) & 0x33333333); | |
820 | x = (x & 0x0f0f0f0f) + ((x >> 4) & 0x0f0f0f0f); | |
821 | return x; | |
822 | } | |
823 | ||
824 | #define NEON_QDMULH16(dest, src1, src2, round) do { \ | |
825 | uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \ | |
826 | if ((tmp ^ (tmp << 1)) & SIGNBIT) { \ | |
827 | SET_QC(); \ | |
828 | tmp = (tmp >> 31) ^ ~SIGNBIT; \ | |
829 | } \ | |
830 | tmp <<= 1; \ | |
831 | if (round) { \ | |
832 | int32_t old = tmp; \ | |
833 | tmp += 1 << 15; \ | |
834 | if ((int32_t)tmp < old) { \ | |
835 | SET_QC(); \ | |
836 | tmp = SIGNBIT - 1; \ | |
837 | } \ | |
838 | } \ | |
839 | dest = tmp >> 16; \ | |
840 | } while(0) | |
841 | #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0) | |
842 | NEON_VOP_ENV(qdmulh_s16, neon_s16, 2) | |
843 | #undef NEON_FN | |
844 | #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1) | |
845 | NEON_VOP_ENV(qrdmulh_s16, neon_s16, 2) | |
846 | #undef NEON_FN | |
847 | #undef NEON_QDMULH16 | |
848 | ||
849 | #define NEON_QDMULH32(dest, src1, src2, round) do { \ | |
850 | uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \ | |
851 | if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \ | |
852 | SET_QC(); \ | |
853 | tmp = (tmp >> 63) ^ ~SIGNBIT64; \ | |
854 | } else { \ | |
855 | tmp <<= 1; \ | |
856 | } \ | |
857 | if (round) { \ | |
858 | int64_t old = tmp; \ | |
859 | tmp += (int64_t)1 << 31; \ | |
860 | if ((int64_t)tmp < old) { \ | |
861 | SET_QC(); \ | |
862 | tmp = SIGNBIT64 - 1; \ | |
863 | } \ | |
864 | } \ | |
865 | dest = tmp >> 32; \ | |
866 | } while(0) | |
867 | #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0) | |
868 | NEON_VOP_ENV(qdmulh_s32, neon_s32, 1) | |
869 | #undef NEON_FN | |
870 | #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1) | |
871 | NEON_VOP_ENV(qrdmulh_s32, neon_s32, 1) | |
872 | #undef NEON_FN | |
873 | #undef NEON_QDMULH32 | |
874 | ||
875 | uint32_t HELPER(neon_narrow_u8)(uint64_t x) | |
876 | { | |
877 | return (x & 0xffu) | ((x >> 8) & 0xff00u) | ((x >> 16) & 0xff0000u) | |
878 | | ((x >> 24) & 0xff000000u); | |
879 | } | |
880 | ||
881 | uint32_t HELPER(neon_narrow_u16)(uint64_t x) | |
882 | { | |
883 | return (x & 0xffffu) | ((x >> 16) & 0xffff0000u); | |
884 | } | |
885 | ||
886 | uint32_t HELPER(neon_narrow_high_u8)(uint64_t x) | |
887 | { | |
888 | return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00) | |
889 | | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000); | |
890 | } | |
891 | ||
892 | uint32_t HELPER(neon_narrow_high_u16)(uint64_t x) | |
893 | { | |
894 | return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000); | |
895 | } | |
896 | ||
897 | uint32_t HELPER(neon_narrow_round_high_u8)(uint64_t x) | |
898 | { | |
899 | x &= 0xff80ff80ff80ff80ull; | |
900 | x += 0x0080008000800080ull; | |
901 | return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00) | |
902 | | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000); | |
903 | } | |
904 | ||
905 | uint32_t HELPER(neon_narrow_round_high_u16)(uint64_t x) | |
906 | { | |
907 | x &= 0xffff8000ffff8000ull; | |
908 | x += 0x0000800000008000ull; | |
909 | return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000); | |
910 | } | |
911 | ||
912 | uint32_t HELPER(neon_narrow_sat_u8)(CPUState *env, uint64_t x) | |
913 | { | |
914 | uint16_t s; | |
915 | uint8_t d; | |
916 | uint32_t res = 0; | |
917 | #define SAT8(n) \ | |
918 | s = x >> n; \ | |
919 | if (s > 0xff) { \ | |
920 | d = 0xff; \ | |
921 | SET_QC(); \ | |
922 | } else { \ | |
923 | d = s; \ | |
924 | } \ | |
925 | res |= (uint32_t)d << (n / 2); | |
926 | ||
927 | SAT8(0); | |
928 | SAT8(16); | |
929 | SAT8(32); | |
930 | SAT8(48); | |
931 | #undef SAT8 | |
932 | return res; | |
933 | } | |
934 | ||
935 | uint32_t HELPER(neon_narrow_sat_s8)(CPUState *env, uint64_t x) | |
936 | { | |
937 | int16_t s; | |
938 | uint8_t d; | |
939 | uint32_t res = 0; | |
940 | #define SAT8(n) \ | |
941 | s = x >> n; \ | |
942 | if (s != (int8_t)s) { \ | |
943 | d = (s >> 15) ^ 0x7f; \ | |
944 | SET_QC(); \ | |
945 | } else { \ | |
946 | d = s; \ | |
947 | } \ | |
948 | res |= (uint32_t)d << (n / 2); | |
949 | ||
950 | SAT8(0); | |
951 | SAT8(16); | |
952 | SAT8(32); | |
953 | SAT8(48); | |
954 | #undef SAT8 | |
955 | return res; | |
956 | } | |
957 | ||
958 | uint32_t HELPER(neon_narrow_sat_u16)(CPUState *env, uint64_t x) | |
959 | { | |
960 | uint32_t high; | |
961 | uint32_t low; | |
962 | low = x; | |
963 | if (low > 0xffff) { | |
964 | low = 0xffff; | |
965 | SET_QC(); | |
966 | } | |
967 | high = x >> 32; | |
968 | if (high > 0xffff) { | |
969 | high = 0xffff; | |
970 | SET_QC(); | |
971 | } | |
972 | return low | (high << 16); | |
973 | } | |
974 | ||
975 | uint32_t HELPER(neon_narrow_sat_s16)(CPUState *env, uint64_t x) | |
976 | { | |
977 | int32_t low; | |
978 | int32_t high; | |
979 | low = x; | |
980 | if (low != (int16_t)low) { | |
981 | low = (low >> 31) ^ 0x7fff; | |
982 | SET_QC(); | |
983 | } | |
984 | high = x >> 32; | |
985 | if (high != (int16_t)high) { | |
986 | high = (high >> 31) ^ 0x7fff; | |
987 | SET_QC(); | |
988 | } | |
989 | return (uint16_t)low | (high << 16); | |
990 | } | |
991 | ||
992 | uint32_t HELPER(neon_narrow_sat_u32)(CPUState *env, uint64_t x) | |
993 | { | |
994 | if (x > 0xffffffffu) { | |
995 | SET_QC(); | |
996 | return 0xffffffffu; | |
997 | } | |
998 | return x; | |
999 | } | |
1000 | ||
1001 | uint32_t HELPER(neon_narrow_sat_s32)(CPUState *env, uint64_t x) | |
1002 | { | |
1003 | if ((int64_t)x != (int32_t)x) { | |
1004 | SET_QC(); | |
1005 | return (x >> 63) ^ 0x7fffffff; | |
1006 | } | |
1007 | return x; | |
1008 | } | |
1009 | ||
1010 | uint64_t HELPER(neon_widen_u8)(uint32_t x) | |
1011 | { | |
1012 | uint64_t tmp; | |
1013 | uint64_t ret; | |
1014 | ret = (uint8_t)x; | |
1015 | tmp = (uint8_t)(x >> 8); | |
1016 | ret |= tmp << 16; | |
1017 | tmp = (uint8_t)(x >> 16); | |
1018 | ret |= tmp << 32; | |
1019 | tmp = (uint8_t)(x >> 24); | |
1020 | ret |= tmp << 48; | |
1021 | return ret; | |
1022 | } | |
1023 | ||
1024 | uint64_t HELPER(neon_widen_s8)(uint32_t x) | |
1025 | { | |
1026 | uint64_t tmp; | |
1027 | uint64_t ret; | |
1028 | ret = (uint16_t)(int8_t)x; | |
1029 | tmp = (uint16_t)(int8_t)(x >> 8); | |
1030 | ret |= tmp << 16; | |
1031 | tmp = (uint16_t)(int8_t)(x >> 16); | |
1032 | ret |= tmp << 32; | |
1033 | tmp = (uint16_t)(int8_t)(x >> 24); | |
1034 | ret |= tmp << 48; | |
1035 | return ret; | |
1036 | } | |
1037 | ||
1038 | uint64_t HELPER(neon_widen_u16)(uint32_t x) | |
1039 | { | |
1040 | uint64_t high = (uint16_t)(x >> 16); | |
1041 | return ((uint16_t)x) | (high << 32); | |
1042 | } | |
1043 | ||
1044 | uint64_t HELPER(neon_widen_s16)(uint32_t x) | |
1045 | { | |
1046 | uint64_t high = (int16_t)(x >> 16); | |
1047 | return ((uint32_t)(int16_t)x) | (high << 32); | |
1048 | } | |
1049 | ||
1050 | uint64_t HELPER(neon_addl_u16)(uint64_t a, uint64_t b) | |
1051 | { | |
1052 | uint64_t mask; | |
1053 | mask = (a ^ b) & 0x8000800080008000ull; | |
1054 | a &= ~0x8000800080008000ull; | |
1055 | b &= ~0x8000800080008000ull; | |
1056 | return (a + b) ^ mask; | |
1057 | } | |
1058 | ||
1059 | uint64_t HELPER(neon_addl_u32)(uint64_t a, uint64_t b) | |
1060 | { | |
1061 | uint64_t mask; | |
1062 | mask = (a ^ b) & 0x8000000080000000ull; | |
1063 | a &= ~0x8000000080000000ull; | |
1064 | b &= ~0x8000000080000000ull; | |
1065 | return (a + b) ^ mask; | |
1066 | } | |
1067 | ||
1068 | uint64_t HELPER(neon_paddl_u16)(uint64_t a, uint64_t b) | |
1069 | { | |
1070 | uint64_t tmp; | |
1071 | uint64_t tmp2; | |
1072 | ||
1073 | tmp = a & 0x0000ffff0000ffffull; | |
1074 | tmp += (a >> 16) & 0x0000ffff0000ffffull; | |
1075 | tmp2 = b & 0xffff0000ffff0000ull; | |
1076 | tmp2 += (b << 16) & 0xffff0000ffff0000ull; | |
1077 | return ( tmp & 0xffff) | |
1078 | | ((tmp >> 16) & 0xffff0000ull) | |
1079 | | ((tmp2 << 16) & 0xffff00000000ull) | |
1080 | | ( tmp2 & 0xffff000000000000ull); | |
1081 | } | |
1082 | ||
1083 | uint64_t HELPER(neon_paddl_u32)(uint64_t a, uint64_t b) | |
1084 | { | |
1085 | uint32_t low = a + (a >> 32); | |
1086 | uint32_t high = b + (b >> 32); | |
1087 | return low + ((uint64_t)high << 32); | |
1088 | } | |
1089 | ||
1090 | uint64_t HELPER(neon_subl_u16)(uint64_t a, uint64_t b) | |
1091 | { | |
1092 | uint64_t mask; | |
1093 | mask = (a ^ ~b) & 0x8000800080008000ull; | |
1094 | a |= 0x8000800080008000ull; | |
1095 | b &= ~0x8000800080008000ull; | |
1096 | return (a - b) ^ mask; | |
1097 | } | |
1098 | ||
1099 | uint64_t HELPER(neon_subl_u32)(uint64_t a, uint64_t b) | |
1100 | { | |
1101 | uint64_t mask; | |
1102 | mask = (a ^ ~b) & 0x8000000080000000ull; | |
1103 | a |= 0x8000000080000000ull; | |
1104 | b &= ~0x8000000080000000ull; | |
1105 | return (a - b) ^ mask; | |
1106 | } | |
1107 | ||
1108 | uint64_t HELPER(neon_addl_saturate_s32)(CPUState *env, uint64_t a, uint64_t b) | |
1109 | { | |
1110 | uint32_t x, y; | |
1111 | uint32_t low, high; | |
1112 | ||
1113 | x = a; | |
1114 | y = b; | |
1115 | low = x + y; | |
1116 | if (((low ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) { | |
1117 | SET_QC(); | |
1118 | low = ((int32_t)x >> 31) ^ ~SIGNBIT; | |
1119 | } | |
1120 | x = a >> 32; | |
1121 | y = b >> 32; | |
1122 | high = x + y; | |
1123 | if (((high ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) { | |
1124 | SET_QC(); | |
1125 | high = ((int32_t)x >> 31) ^ ~SIGNBIT; | |
1126 | } | |
1127 | return low | ((uint64_t)high << 32); | |
1128 | } | |
1129 | ||
1130 | uint64_t HELPER(neon_addl_saturate_s64)(CPUState *env, uint64_t a, uint64_t b) | |
1131 | { | |
1132 | uint64_t result; | |
1133 | ||
1134 | result = a + b; | |
1135 | if (((result ^ a) & SIGNBIT64) && !((a ^ b) & SIGNBIT64)) { | |
1136 | SET_QC(); | |
1137 | result = ((int64_t)a >> 63) ^ ~SIGNBIT64; | |
1138 | } | |
1139 | return result; | |
1140 | } | |
1141 | ||
1142 | #define DO_ABD(dest, x, y, type) do { \ | |
1143 | type tmp_x = x; \ | |
1144 | type tmp_y = y; \ | |
1145 | dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \ | |
1146 | } while(0) | |
1147 | ||
1148 | uint64_t HELPER(neon_abdl_u16)(uint32_t a, uint32_t b) | |
1149 | { | |
1150 | uint64_t tmp; | |
1151 | uint64_t result; | |
1152 | DO_ABD(result, a, b, uint8_t); | |
1153 | DO_ABD(tmp, a >> 8, b >> 8, uint8_t); | |
1154 | result |= tmp << 16; | |
1155 | DO_ABD(tmp, a >> 16, b >> 16, uint8_t); | |
1156 | result |= tmp << 32; | |
1157 | DO_ABD(tmp, a >> 24, b >> 24, uint8_t); | |
1158 | result |= tmp << 48; | |
1159 | return result; | |
1160 | } | |
1161 | ||
1162 | uint64_t HELPER(neon_abdl_s16)(uint32_t a, uint32_t b) | |
1163 | { | |
1164 | uint64_t tmp; | |
1165 | uint64_t result; | |
1166 | DO_ABD(result, a, b, int8_t); | |
1167 | DO_ABD(tmp, a >> 8, b >> 8, int8_t); | |
1168 | result |= tmp << 16; | |
1169 | DO_ABD(tmp, a >> 16, b >> 16, int8_t); | |
1170 | result |= tmp << 32; | |
1171 | DO_ABD(tmp, a >> 24, b >> 24, int8_t); | |
1172 | result |= tmp << 48; | |
1173 | return result; | |
1174 | } | |
1175 | ||
1176 | uint64_t HELPER(neon_abdl_u32)(uint32_t a, uint32_t b) | |
1177 | { | |
1178 | uint64_t tmp; | |
1179 | uint64_t result; | |
1180 | DO_ABD(result, a, b, uint16_t); | |
1181 | DO_ABD(tmp, a >> 16, b >> 16, uint16_t); | |
1182 | return result | (tmp << 32); | |
1183 | } | |
1184 | ||
1185 | uint64_t HELPER(neon_abdl_s32)(uint32_t a, uint32_t b) | |
1186 | { | |
1187 | uint64_t tmp; | |
1188 | uint64_t result; | |
1189 | DO_ABD(result, a, b, int16_t); | |
1190 | DO_ABD(tmp, a >> 16, b >> 16, int16_t); | |
1191 | return result | (tmp << 32); | |
1192 | } | |
1193 | ||
1194 | uint64_t HELPER(neon_abdl_u64)(uint32_t a, uint32_t b) | |
1195 | { | |
1196 | uint64_t result; | |
1197 | DO_ABD(result, a, b, uint32_t); | |
1198 | return result; | |
1199 | } | |
1200 | ||
1201 | uint64_t HELPER(neon_abdl_s64)(uint32_t a, uint32_t b) | |
1202 | { | |
1203 | uint64_t result; | |
1204 | DO_ABD(result, a, b, int32_t); | |
1205 | return result; | |
1206 | } | |
1207 | #undef DO_ABD | |
1208 | ||
1209 | /* Widening multiply. Named type is the source type. */ | |
1210 | #define DO_MULL(dest, x, y, type1, type2) do { \ | |
1211 | type1 tmp_x = x; \ | |
1212 | type1 tmp_y = y; \ | |
1213 | dest = (type2)((type2)tmp_x * (type2)tmp_y); \ | |
1214 | } while(0) | |
1215 | ||
1216 | uint64_t HELPER(neon_mull_u8)(uint32_t a, uint32_t b) | |
1217 | { | |
1218 | uint64_t tmp; | |
1219 | uint64_t result; | |
1220 | ||
1221 | DO_MULL(result, a, b, uint8_t, uint16_t); | |
1222 | DO_MULL(tmp, a >> 8, b >> 8, uint8_t, uint16_t); | |
1223 | result |= tmp << 16; | |
1224 | DO_MULL(tmp, a >> 16, b >> 16, uint8_t, uint16_t); | |
1225 | result |= tmp << 32; | |
1226 | DO_MULL(tmp, a >> 24, b >> 24, uint8_t, uint16_t); | |
1227 | result |= tmp << 48; | |
1228 | return result; | |
1229 | } | |
1230 | ||
1231 | uint64_t HELPER(neon_mull_s8)(uint32_t a, uint32_t b) | |
1232 | { | |
1233 | uint64_t tmp; | |
1234 | uint64_t result; | |
1235 | ||
1236 | DO_MULL(result, a, b, int8_t, uint16_t); | |
1237 | DO_MULL(tmp, a >> 8, b >> 8, int8_t, uint16_t); | |
1238 | result |= tmp << 16; | |
1239 | DO_MULL(tmp, a >> 16, b >> 16, int8_t, uint16_t); | |
1240 | result |= tmp << 32; | |
1241 | DO_MULL(tmp, a >> 24, b >> 24, int8_t, uint16_t); | |
1242 | result |= tmp << 48; | |
1243 | return result; | |
1244 | } | |
1245 | ||
1246 | uint64_t HELPER(neon_mull_u16)(uint32_t a, uint32_t b) | |
1247 | { | |
1248 | uint64_t tmp; | |
1249 | uint64_t result; | |
1250 | ||
1251 | DO_MULL(result, a, b, uint16_t, uint32_t); | |
1252 | DO_MULL(tmp, a >> 16, b >> 16, uint16_t, uint32_t); | |
1253 | return result | (tmp << 32); | |
1254 | } | |
1255 | ||
1256 | uint64_t HELPER(neon_mull_s16)(uint32_t a, uint32_t b) | |
1257 | { | |
1258 | uint64_t tmp; | |
1259 | uint64_t result; | |
1260 | ||
1261 | DO_MULL(result, a, b, int16_t, uint32_t); | |
1262 | DO_MULL(tmp, a >> 16, b >> 16, int16_t, uint32_t); | |
1263 | return result | (tmp << 32); | |
1264 | } | |
1265 | ||
1266 | uint64_t HELPER(neon_negl_u16)(uint64_t x) | |
1267 | { | |
1268 | uint16_t tmp; | |
1269 | uint64_t result; | |
1270 | result = (uint16_t)-x; | |
1271 | tmp = -(x >> 16); | |
1272 | result |= (uint64_t)tmp << 16; | |
1273 | tmp = -(x >> 32); | |
1274 | result |= (uint64_t)tmp << 32; | |
1275 | tmp = -(x >> 48); | |
1276 | result |= (uint64_t)tmp << 48; | |
1277 | return result; | |
1278 | } | |
1279 | ||
1280 | #include <stdio.h> | |
1281 | uint64_t HELPER(neon_negl_u32)(uint64_t x) | |
1282 | { | |
1283 | uint32_t low = -x; | |
1284 | uint32_t high = -(x >> 32); | |
1285 | return low | ((uint64_t)high << 32); | |
1286 | } | |
1287 | ||
1288 | /* FIXME: There should be a native op for this. */ | |
1289 | uint64_t HELPER(neon_negl_u64)(uint64_t x) | |
1290 | { | |
1291 | return -x; | |
1292 | } | |
1293 | ||
1294 | /* Saturnating sign manuipulation. */ | |
1295 | /* ??? Make these use NEON_VOP1 */ | |
1296 | #define DO_QABS8(x) do { \ | |
1297 | if (x == (int8_t)0x80) { \ | |
1298 | x = 0x7f; \ | |
1299 | SET_QC(); \ | |
1300 | } else if (x < 0) { \ | |
1301 | x = -x; \ | |
1302 | }} while (0) | |
1303 | uint32_t HELPER(neon_qabs_s8)(CPUState *env, uint32_t x) | |
1304 | { | |
1305 | neon_s8 vec; | |
1306 | NEON_UNPACK(neon_s8, vec, x); | |
1307 | DO_QABS8(vec.v1); | |
1308 | DO_QABS8(vec.v2); | |
1309 | DO_QABS8(vec.v3); | |
1310 | DO_QABS8(vec.v4); | |
1311 | NEON_PACK(neon_s8, x, vec); | |
1312 | return x; | |
1313 | } | |
1314 | #undef DO_QABS8 | |
1315 | ||
1316 | #define DO_QNEG8(x) do { \ | |
1317 | if (x == (int8_t)0x80) { \ | |
1318 | x = 0x7f; \ | |
1319 | SET_QC(); \ | |
1320 | } else { \ | |
1321 | x = -x; \ | |
1322 | }} while (0) | |
1323 | uint32_t HELPER(neon_qneg_s8)(CPUState *env, uint32_t x) | |
1324 | { | |
1325 | neon_s8 vec; | |
1326 | NEON_UNPACK(neon_s8, vec, x); | |
1327 | DO_QNEG8(vec.v1); | |
1328 | DO_QNEG8(vec.v2); | |
1329 | DO_QNEG8(vec.v3); | |
1330 | DO_QNEG8(vec.v4); | |
1331 | NEON_PACK(neon_s8, x, vec); | |
1332 | return x; | |
1333 | } | |
1334 | #undef DO_QNEG8 | |
1335 | ||
1336 | #define DO_QABS16(x) do { \ | |
1337 | if (x == (int16_t)0x8000) { \ | |
1338 | x = 0x7fff; \ | |
1339 | SET_QC(); \ | |
1340 | } else if (x < 0) { \ | |
1341 | x = -x; \ | |
1342 | }} while (0) | |
1343 | uint32_t HELPER(neon_qabs_s16)(CPUState *env, uint32_t x) | |
1344 | { | |
1345 | neon_s16 vec; | |
1346 | NEON_UNPACK(neon_s16, vec, x); | |
1347 | DO_QABS16(vec.v1); | |
1348 | DO_QABS16(vec.v2); | |
1349 | NEON_PACK(neon_s16, x, vec); | |
1350 | return x; | |
1351 | } | |
1352 | #undef DO_QABS16 | |
1353 | ||
1354 | #define DO_QNEG16(x) do { \ | |
1355 | if (x == (int16_t)0x8000) { \ | |
1356 | x = 0x7fff; \ | |
1357 | SET_QC(); \ | |
1358 | } else { \ | |
1359 | x = -x; \ | |
1360 | }} while (0) | |
1361 | uint32_t HELPER(neon_qneg_s16)(CPUState *env, uint32_t x) | |
1362 | { | |
1363 | neon_s16 vec; | |
1364 | NEON_UNPACK(neon_s16, vec, x); | |
1365 | DO_QNEG16(vec.v1); | |
1366 | DO_QNEG16(vec.v2); | |
1367 | NEON_PACK(neon_s16, x, vec); | |
1368 | return x; | |
1369 | } | |
1370 | #undef DO_QNEG16 | |
1371 | ||
1372 | uint32_t HELPER(neon_qabs_s32)(CPUState *env, uint32_t x) | |
1373 | { | |
1374 | if (x == SIGNBIT) { | |
1375 | SET_QC(); | |
1376 | x = ~SIGNBIT; | |
1377 | } else if ((int32_t)x < 0) { | |
1378 | x = -x; | |
1379 | } | |
1380 | return x; | |
1381 | } | |
1382 | ||
1383 | uint32_t HELPER(neon_qneg_s32)(CPUState *env, uint32_t x) | |
1384 | { | |
1385 | if (x == SIGNBIT) { | |
1386 | SET_QC(); | |
1387 | x = ~SIGNBIT; | |
1388 | } else { | |
1389 | x = -x; | |
1390 | } | |
1391 | return x; | |
1392 | } | |
1393 | ||
1394 | /* NEON Float helpers. */ | |
1395 | uint32_t HELPER(neon_min_f32)(uint32_t a, uint32_t b) | |
1396 | { | |
1397 | float32 f0 = vfp_itos(a); | |
1398 | float32 f1 = vfp_itos(b); | |
1399 | return (float32_compare_quiet(f0, f1, NFS) == -1) ? a : b; | |
1400 | } | |
1401 | ||
1402 | uint32_t HELPER(neon_max_f32)(uint32_t a, uint32_t b) | |
1403 | { | |
1404 | float32 f0 = vfp_itos(a); | |
1405 | float32 f1 = vfp_itos(b); | |
1406 | return (float32_compare_quiet(f0, f1, NFS) == 1) ? a : b; | |
1407 | } | |
1408 | ||
1409 | uint32_t HELPER(neon_abd_f32)(uint32_t a, uint32_t b) | |
1410 | { | |
1411 | float32 f0 = vfp_itos(a); | |
1412 | float32 f1 = vfp_itos(b); | |
1413 | return vfp_stoi((float32_compare_quiet(f0, f1, NFS) == 1) | |
1414 | ? float32_sub(f0, f1, NFS) | |
1415 | : float32_sub(f1, f0, NFS)); | |
1416 | } | |
1417 | ||
1418 | uint32_t HELPER(neon_add_f32)(uint32_t a, uint32_t b) | |
1419 | { | |
1420 | return vfp_stoi(float32_add(vfp_itos(a), vfp_itos(b), NFS)); | |
1421 | } | |
1422 | ||
1423 | uint32_t HELPER(neon_sub_f32)(uint32_t a, uint32_t b) | |
1424 | { | |
1425 | return vfp_stoi(float32_sub(vfp_itos(a), vfp_itos(b), NFS)); | |
1426 | } | |
1427 | ||
1428 | uint32_t HELPER(neon_mul_f32)(uint32_t a, uint32_t b) | |
1429 | { | |
1430 | return vfp_stoi(float32_mul(vfp_itos(a), vfp_itos(b), NFS)); | |
1431 | } | |
1432 | ||
1433 | /* Floating point comparisons produce an integer result. */ | |
1434 | #define NEON_VOP_FCMP(name, cmp) \ | |
1435 | uint32_t HELPER(neon_##name)(uint32_t a, uint32_t b) \ | |
1436 | { \ | |
1437 | if (float32_compare_quiet(vfp_itos(a), vfp_itos(b), NFS) cmp 0) \ | |
1438 | return ~0; \ | |
1439 | else \ | |
1440 | return 0; \ | |
1441 | } | |
1442 | ||
1443 | NEON_VOP_FCMP(ceq_f32, ==) | |
1444 | NEON_VOP_FCMP(cge_f32, >=) | |
1445 | NEON_VOP_FCMP(cgt_f32, >) | |
1446 | ||
1447 | uint32_t HELPER(neon_acge_f32)(uint32_t a, uint32_t b) | |
1448 | { | |
1449 | float32 f0 = float32_abs(vfp_itos(a)); | |
1450 | float32 f1 = float32_abs(vfp_itos(b)); | |
1451 | return (float32_compare_quiet(f0, f1,NFS) >= 0) ? ~0 : 0; | |
1452 | } | |
1453 | ||
1454 | uint32_t HELPER(neon_acgt_f32)(uint32_t a, uint32_t b) | |
1455 | { | |
1456 | float32 f0 = float32_abs(vfp_itos(a)); | |
1457 | float32 f1 = float32_abs(vfp_itos(b)); | |
1458 | return (float32_compare_quiet(f0, f1, NFS) > 0) ? ~0 : 0; | |
1459 | } |