24 #ifdef ARM_COMPUTE_ENABLE_SVE
27 #include "../../utils.hpp"
34 void sve_hybrid_fp32_mla_8x1VL (
35 unsigned int num_strings,
const unsigned int *string_lengths, IndirectInputArg<float> A_arg,
36 size_t M,
size_t N,
const float *B_ptr, IndirectOutputArg<float> output_arg,
41 float maxval =
static_cast<float>(std::numeric_limits<float>::infinity());
42 float minval = -
static_cast<float>(std::numeric_limits<float>::infinity());
43 unsigned int num_strings = {};
44 const unsigned int *string_lengths = {};
46 const float *B_ptr = {};
47 size_t output_offset = {};
48 size_t input_initial_col = {};
49 size_t input_offset = {};
52 unsigned long flags=0;
56 if (output_arg.is_indirect) {
57 output_ptr=(
void *)(output_arg.indirect.ptr);
58 ka.output_offset=output_arg.indirect.offset;
61 output_ptr=(
void *)(output_arg.direct.base);
62 ka.output_offset=output_arg.direct.stride;
65 if (A_arg.is_indirect) {
66 input_ptr=(
void *)(A_arg.indirect.ptr);
67 ka.input_offset=A_arg.indirect.start_row;
68 ka.input_initial_col=A_arg.indirect.start_col;
71 assert(num_strings==1);
72 input_ptr=(
void *)(A_arg.direct.base);
73 ka.input_offset=A_arg.direct.stride;
78 ka.num_strings = num_strings;
79 ka.string_lengths = string_lengths;
87 ka.maxval =
static_cast<float>(act.param1);
108 "mov x14, %x[bias]\n"
109 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
110 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
111 "mov x11, %x[output_ptr]\n"
114 "whilelt p1.s, x20, x13\n"
116 "ld1w { z24.s }, p2/Z, [x14]\n"
117 "addvl x14, x14, #1\n"
120 "tbz %x[flags], #0, 4f\n"
121 "ld1w { z24.s }, p1/Z, [x11]\n"
128 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
129 "ldr w9, [x20, x10, LSL #0x2]\n"
130 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
131 "tbz %x[flags], #3, 7f\n"
132 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
133 "add x20, x20, x21, LSL #3\n"
134 "ldr x28, [x20, #0x0]\n"
136 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
137 "add x28, x28, x20, LSL #2\n"
140 "mov x28, %x[input_ptr]\n"
145 "whilelt p0.s, XZR, x9\n"
146 "ld1rqw { z0.s }, p0/Z, [x28]\n"
147 "ld1w { z16.s }, p2/Z, [x12]\n"
148 "fmla z24.s, z16.s, z0.s[0]\n"
149 "ld1w { z16.s }, p2/Z, [x12, #1, MUL VL]\n"
150 "fmla z24.s, z16.s, z0.s[1]\n"
151 "ld1w { z16.s }, p2/Z, [x12, #2, MUL VL]\n"
152 "fmla z24.s, z16.s, z0.s[2]\n"
153 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
156 "fmla z24.s, z16.s, z0.s[3]\n"
157 "add x28, x28, #0x10\n"
158 "addvl x12, x12, #4\n"
161 "whilelt p0.s, XZR, x9\n"
162 "ld1rqw { z0.s }, p0/Z, [x28]\n"
163 "ld1w { z16.s }, p2/Z, [x12]\n"
164 "subs x9, x9, #0x1\n"
165 "fmla z24.s, z16.s, z0.s[0]\n"
166 "addvl x12, x12, #1\n"
168 "ld1w { z16.s }, p2/Z, [x12]\n"
169 "subs x9, x9, #0x1\n"
170 "fmla z24.s, z16.s, z0.s[1]\n"
171 "addvl x12, x12, #1\n"
173 "ld1w { z16.s }, p2/Z, [x12]\n"
174 "subs x9, x9, #0x1\n"
175 "fmla z24.s, z16.s, z0.s[2]\n"
176 "addvl x12, x12, #1\n"
178 "ld1w { z16.s }, p2/Z, [x12]\n"
179 "fmla z24.s, z16.s, z0.s[3]\n"
180 "addvl x12, x12, #1\n"
182 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
183 "add x10, x10, #0x1\n"
186 "tbz %x[flags], #1, 12f\n"
187 "add x20, %x[args_ptr], %[offset_max]\n"
188 "ld1rw { z17.s }, p2/Z, [x20]\n"
189 "add x20, %x[args_ptr], %[offset_min]\n"
190 "ld1rw { z16.s }, p2/Z, [x20]\n"
191 "fmin z24.s, p2/M, z24.s, z17.s\n"
192 "fmax z24.s, p2/M, z24.s, z16.s\n"
194 "st1w { z24.s }, p1, [x11]\n"
195 "addvl x11, x11, #1\n"
202 "mov x14, %x[bias]\n"
203 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
204 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
205 "mov x11, %x[output_ptr]\n"
208 "whilelt p1.s, x20, x13\n"
210 "ld1w { z24.s }, p2/Z, [x14]\n"
212 "addvl x14, x14, #1\n"
215 "tbz %x[flags], #0, 17f\n"
216 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
217 "add x20, x11, x20, LSL #2\n"
218 "ld1w { z24.s }, p1/Z, [x11]\n"
219 "ld1w { z25.s }, p1/Z, [x20]\n"
227 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
228 "ldr w9, [x20, x10, LSL #0x2]\n"
229 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
230 "tbz %x[flags], #3, 20f\n"
231 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
232 "add x20, x20, x21, LSL #3\n"
233 "ldr x28, [x20, #0x0]\n"
234 "ldr x27, [x20, #0x8]\n"
236 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
237 "add x28, x28, x20, LSL #2\n"
238 "add x27, x27, x20, LSL #2\n"
241 "mov x28, %x[input_ptr]\n"
242 "add x27, x28, x21, LSL #2\n"
247 "whilelt p0.s, XZR, x9\n"
248 "ld1rqw { z1.s }, p0/Z, [x28]\n"
249 "ld1rqw { z0.s }, p0/Z, [x27]\n"
251 "ld1w { z16.s }, p2/Z, [x12]\n"
252 "fmla z24.s, z16.s, z1.s[0]\n"
253 "fmla z25.s, z16.s, z0.s[0]\n"
254 "ld1w { z16.s }, p2/Z, [x12, #1, MUL VL]\n"
255 "fmla z24.s, z16.s, z1.s[1]\n"
256 "fmla z25.s, z16.s, z0.s[1]\n"
257 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
258 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
259 "fmla z24.s, z17.s, z1.s[2]\n"
260 "fmla z25.s, z17.s, z0.s[2]\n"
262 "add x28, x28, #0x10\n"
263 "fmla z24.s, z16.s, z1.s[3]\n"
264 "fmla z25.s, z16.s, z0.s[3]\n"
265 "add x27, x27, #0x10\n"
266 "addvl x12, x12, #4\n"
269 "whilelt p0.s, XZR, x9\n"
270 "ld1rqw { z0.s }, p0/Z, [x28]\n"
271 "ld1rqw { z1.s }, p0/Z, [x27]\n"
272 "subs x9, x9, #0x1\n"
273 "ld1w { z16.s }, p2/Z, [x12]\n"
274 "fmla z24.s, z16.s, z0.s[0]\n"
275 "fmla z25.s, z16.s, z1.s[0]\n"
276 "addvl x12, x12, #1\n"
278 "ld1w { z16.s }, p2/Z, [x12]\n"
279 "subs x9, x9, #0x1\n"
280 "fmla z24.s, z16.s, z0.s[1]\n"
281 "fmla z25.s, z16.s, z1.s[1]\n"
282 "addvl x12, x12, #1\n"
284 "ld1w { z16.s }, p2/Z, [x12]\n"
285 "subs x9, x9, #0x1\n"
286 "fmla z24.s, z16.s, z0.s[2]\n"
287 "fmla z25.s, z16.s, z1.s[2]\n"
288 "addvl x12, x12, #1\n"
290 "ld1w { z16.s }, p2/Z, [x12]\n"
291 "fmla z24.s, z16.s, z0.s[3]\n"
292 "fmla z25.s, z16.s, z1.s[3]\n"
293 "addvl x12, x12, #1\n"
295 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
296 "add x10, x10, #0x1\n"
299 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
300 "add x27, x11, x20, LSL #2\n"
301 "tbz %x[flags], #1, 25f\n"
302 "add x20, %x[args_ptr], %[offset_max]\n"
303 "ld1rw { z17.s }, p2/Z, [x20]\n"
304 "add x20, %x[args_ptr], %[offset_min]\n"
305 "ld1rw { z16.s }, p2/Z, [x20]\n"
306 "fmin z24.s, p2/M, z24.s, z17.s\n"
307 "fmin z25.s, p2/M, z25.s, z17.s\n"
308 "fmax z24.s, p2/M, z24.s, z16.s\n"
309 "fmax z25.s, p2/M, z25.s, z16.s\n"
311 "st1w { z24.s }, p1, [x11]\n"
312 "addvl x11, x11, #1\n"
313 "st1w { z25.s }, p1, [x27]\n"
320 "mov x14, %x[bias]\n"
321 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
322 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
323 "mov x11, %x[output_ptr]\n"
326 "whilelt p1.s, x20, x13\n"
328 "ld1w { z24.s }, p2/Z, [x14]\n"
331 "addvl x14, x14, #1\n"
334 "tbz %x[flags], #0, 30f\n"
335 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
336 "add x21, x11, x20, LSL #2\n"
337 "add x20, x21, x20, LSL #2\n"
338 "ld1w { z24.s }, p1/Z, [x11]\n"
339 "ld1w { z25.s }, p1/Z, [x21]\n"
340 "ld1w { z26.s }, p1/Z, [x20]\n"
349 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
350 "ldr w9, [x20, x10, LSL #0x2]\n"
351 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
352 "tbz %x[flags], #3, 33f\n"
353 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
354 "add x20, x20, x21, LSL #3\n"
355 "ldr x28, [x20, #0x0]\n"
356 "ldr x27, [x20, #0x8]\n"
357 "ldr x26, [x20, #0x10]\n"
359 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
360 "add x28, x28, x20, LSL #2\n"
361 "add x27, x27, x20, LSL #2\n"
362 "add x26, x26, x20, LSL #2\n"
365 "mov x28, %x[input_ptr]\n"
366 "add x27, x28, x21, LSL #2\n"
367 "add x26, x27, x21, LSL #2\n"
372 "whilelt p0.s, XZR, x9\n"
373 "ld1rqw { z2.s }, p0/Z, [x28]\n"
374 "ld1rqw { z1.s }, p0/Z, [x27]\n"
376 "ld1rqw { z0.s }, p0/Z, [x26]\n"
377 "ld1w { z16.s }, p2/Z, [x12]\n"
378 "fmla z24.s, z16.s, z2.s[0]\n"
379 "fmla z25.s, z16.s, z1.s[0]\n"
380 "fmla z26.s, z16.s, z0.s[0]\n"
381 "ld1w { z16.s }, p2/Z, [x12, #1, MUL VL]\n"
382 "fmla z24.s, z16.s, z2.s[1]\n"
383 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
384 "fmla z25.s, z16.s, z1.s[1]\n"
385 "fmla z26.s, z16.s, z0.s[1]\n"
386 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
388 "fmla z24.s, z17.s, z2.s[2]\n"
389 "fmla z25.s, z17.s, z1.s[2]\n"
390 "add x28, x28, #0x10\n"
391 "add x27, x27, #0x10\n"
392 "fmla z26.s, z17.s, z0.s[2]\n"
393 "fmla z24.s, z16.s, z2.s[3]\n"
394 "add x26, x26, #0x10\n"
395 "addvl x12, x12, #4\n"
396 "fmla z25.s, z16.s, z1.s[3]\n"
397 "fmla z26.s, z16.s, z0.s[3]\n"
400 "whilelt p0.s, XZR, x9\n"
401 "ld1rqw { z0.s }, p0/Z, [x28]\n"
402 "ld1rqw { z1.s }, p0/Z, [x27]\n"
403 "subs x9, x9, #0x1\n"
404 "ld1rqw { z2.s }, p0/Z, [x26]\n"
405 "ld1w { z16.s }, p2/Z, [x12]\n"
406 "fmla z24.s, z16.s, z0.s[0]\n"
407 "fmla z25.s, z16.s, z1.s[0]\n"
408 "fmla z26.s, z16.s, z2.s[0]\n"
409 "addvl x12, x12, #1\n"
411 "ld1w { z16.s }, p2/Z, [x12]\n"
412 "subs x9, x9, #0x1\n"
413 "fmla z24.s, z16.s, z0.s[1]\n"
414 "fmla z25.s, z16.s, z1.s[1]\n"
415 "fmla z26.s, z16.s, z2.s[1]\n"
416 "addvl x12, x12, #1\n"
418 "ld1w { z16.s }, p2/Z, [x12]\n"
419 "subs x9, x9, #0x1\n"
420 "fmla z24.s, z16.s, z0.s[2]\n"
421 "fmla z25.s, z16.s, z1.s[2]\n"
422 "fmla z26.s, z16.s, z2.s[2]\n"
423 "addvl x12, x12, #1\n"
425 "ld1w { z16.s }, p2/Z, [x12]\n"
426 "fmla z24.s, z16.s, z0.s[3]\n"
427 "fmla z25.s, z16.s, z1.s[3]\n"
428 "addvl x12, x12, #1\n"
429 "fmla z26.s, z16.s, z2.s[3]\n"
431 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
432 "add x10, x10, #0x1\n"
435 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
436 "add x27, x11, x20, LSL #2\n"
437 "add x26, x27, x20, LSL #2\n"
438 "tbz %x[flags], #1, 38f\n"
439 "add x20, %x[args_ptr], %[offset_max]\n"
440 "ld1rw { z17.s }, p2/Z, [x20]\n"
441 "add x20, %x[args_ptr], %[offset_min]\n"
442 "ld1rw { z16.s }, p2/Z, [x20]\n"
443 "fmin z24.s, p2/M, z24.s, z17.s\n"
444 "fmin z25.s, p2/M, z25.s, z17.s\n"
445 "fmin z26.s, p2/M, z26.s, z17.s\n"
446 "fmax z24.s, p2/M, z24.s, z16.s\n"
447 "fmax z25.s, p2/M, z25.s, z16.s\n"
448 "fmax z26.s, p2/M, z26.s, z16.s\n"
450 "st1w { z24.s }, p1, [x11]\n"
451 "addvl x11, x11, #1\n"
452 "st1w { z25.s }, p1, [x27]\n"
453 "st1w { z26.s }, p1, [x26]\n"
460 "mov x14, %x[bias]\n"
461 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
462 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
463 "mov x11, %x[output_ptr]\n"
466 "whilelt p1.s, x20, x13\n"
468 "ld1w { z24.s }, p2/Z, [x14]\n"
471 "addvl x14, x14, #1\n"
475 "tbz %x[flags], #0, 43f\n"
476 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
477 "add x22, x11, x20, LSL #2\n"
478 "add x21, x22, x20, LSL #2\n"
479 "ld1w { z24.s }, p1/Z, [x11]\n"
480 "add x20, x21, x20, LSL #2\n"
481 "ld1w { z25.s }, p1/Z, [x22]\n"
482 "ld1w { z26.s }, p1/Z, [x21]\n"
483 "ld1w { z27.s }, p1/Z, [x20]\n"
493 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
494 "ldr w9, [x20, x10, LSL #0x2]\n"
495 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
496 "tbz %x[flags], #3, 46f\n"
497 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
498 "add x20, x20, x21, LSL #3\n"
499 "ldr x28, [x20, #0x0]\n"
500 "ldr x27, [x20, #0x8]\n"
501 "ldr x26, [x20, #0x10]\n"
502 "ldr x25, [x20, #0x18]\n"
504 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
505 "add x28, x28, x20, LSL #2\n"
506 "add x27, x27, x20, LSL #2\n"
507 "add x26, x26, x20, LSL #2\n"
508 "add x25, x25, x20, LSL #2\n"
511 "mov x28, %x[input_ptr]\n"
512 "add x27, x28, x21, LSL #2\n"
513 "add x26, x27, x21, LSL #2\n"
514 "add x25, x26, x21, LSL #2\n"
519 "whilelt p0.s, XZR, x9\n"
520 "ld1rqw { z3.s }, p0/Z, [x28]\n"
521 "ld1rqw { z2.s }, p0/Z, [x27]\n"
523 "ld1rqw { z1.s }, p0/Z, [x26]\n"
524 "ld1rqw { z0.s }, p0/Z, [x25]\n"
526 "add x28, x28, #0x10\n"
527 "ld1w { z16.s }, p2/Z, [x12]\n"
528 "fmla z24.s, z16.s, z3.s[0]\n"
529 "fmla z25.s, z16.s, z2.s[0]\n"
530 "ld1w { z18.s }, p2/Z, [x12, #1, MUL VL]\n"
531 "fmla z26.s, z16.s, z1.s[0]\n"
532 "fmla z27.s, z16.s, z0.s[0]\n"
533 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
534 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
535 "fmla z24.s, z18.s, z3.s[1]\n"
536 "fmla z25.s, z18.s, z2.s[1]\n"
537 "add x27, x27, #0x10\n"
538 "add x26, x26, #0x10\n"
539 "fmla z26.s, z18.s, z1.s[1]\n"
540 "fmla z27.s, z18.s, z0.s[1]\n"
541 "add x25, x25, #0x10\n"
542 "addvl x12, x12, #4\n"
543 "fmla z24.s, z17.s, z3.s[2]\n"
544 "fmla z25.s, z17.s, z2.s[2]\n"
545 "fmla z26.s, z17.s, z1.s[2]\n"
546 "fmla z27.s, z17.s, z0.s[2]\n"
547 "fmla z24.s, z16.s, z3.s[3]\n"
548 "fmla z25.s, z16.s, z2.s[3]\n"
549 "fmla z26.s, z16.s, z1.s[3]\n"
550 "fmla z27.s, z16.s, z0.s[3]\n"
553 "whilelt p0.s, XZR, x9\n"
554 "ld1rqw { z0.s }, p0/Z, [x28]\n"
555 "ld1rqw { z1.s }, p0/Z, [x27]\n"
556 "subs x9, x9, #0x1\n"
557 "ld1rqw { z2.s }, p0/Z, [x26]\n"
558 "ld1rqw { z3.s }, p0/Z, [x25]\n"
559 "ld1w { z16.s }, p2/Z, [x12]\n"
560 "fmla z24.s, z16.s, z0.s[0]\n"
561 "fmla z25.s, z16.s, z1.s[0]\n"
562 "addvl x12, x12, #1\n"
563 "fmla z26.s, z16.s, z2.s[0]\n"
564 "fmla z27.s, z16.s, z3.s[0]\n"
566 "ld1w { z16.s }, p2/Z, [x12]\n"
567 "subs x9, x9, #0x1\n"
568 "fmla z24.s, z16.s, z0.s[1]\n"
569 "fmla z25.s, z16.s, z1.s[1]\n"
570 "fmla z26.s, z16.s, z2.s[1]\n"
571 "fmla z27.s, z16.s, z3.s[1]\n"
572 "addvl x12, x12, #1\n"
574 "ld1w { z16.s }, p2/Z, [x12]\n"
575 "subs x9, x9, #0x1\n"
576 "fmla z24.s, z16.s, z0.s[2]\n"
577 "fmla z25.s, z16.s, z1.s[2]\n"
578 "fmla z26.s, z16.s, z2.s[2]\n"
579 "fmla z27.s, z16.s, z3.s[2]\n"
580 "addvl x12, x12, #1\n"
582 "ld1w { z16.s }, p2/Z, [x12]\n"
583 "fmla z24.s, z16.s, z0.s[3]\n"
584 "fmla z25.s, z16.s, z1.s[3]\n"
585 "addvl x12, x12, #1\n"
586 "fmla z26.s, z16.s, z2.s[3]\n"
587 "fmla z27.s, z16.s, z3.s[3]\n"
589 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
590 "add x10, x10, #0x1\n"
593 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
594 "add x27, x11, x20, LSL #2\n"
595 "add x26, x27, x20, LSL #2\n"
596 "add x25, x26, x20, LSL #2\n"
597 "tbz %x[flags], #1, 51f\n"
598 "add x20, %x[args_ptr], %[offset_max]\n"
599 "ld1rw { z17.s }, p2/Z, [x20]\n"
600 "add x20, %x[args_ptr], %[offset_min]\n"
601 "ld1rw { z16.s }, p2/Z, [x20]\n"
602 "fmin z24.s, p2/M, z24.s, z17.s\n"
603 "fmin z25.s, p2/M, z25.s, z17.s\n"
604 "fmin z26.s, p2/M, z26.s, z17.s\n"
605 "fmin z27.s, p2/M, z27.s, z17.s\n"
606 "fmax z24.s, p2/M, z24.s, z16.s\n"
607 "fmax z25.s, p2/M, z25.s, z16.s\n"
608 "fmax z26.s, p2/M, z26.s, z16.s\n"
609 "fmax z27.s, p2/M, z27.s, z16.s\n"
611 "st1w { z24.s }, p1, [x11]\n"
612 "addvl x11, x11, #1\n"
613 "st1w { z25.s }, p1, [x27]\n"
614 "st1w { z26.s }, p1, [x26]\n"
615 "st1w { z27.s }, p1, [x25]\n"
622 "mov x14, %x[bias]\n"
623 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
624 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
625 "mov x11, %x[output_ptr]\n"
628 "whilelt p1.s, x20, x13\n"
630 "ld1w { z24.s }, p2/Z, [x14]\n"
633 "addvl x14, x14, #1\n"
638 "tbz %x[flags], #0, 56f\n"
639 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
640 "add x23, x11, x20, LSL #2\n"
641 "add x22, x23, x20, LSL #2\n"
642 "ld1w { z24.s }, p1/Z, [x11]\n"
643 "add x21, x22, x20, LSL #2\n"
644 "add x20, x21, x20, LSL #2\n"
645 "ld1w { z25.s }, p1/Z, [x23]\n"
646 "ld1w { z26.s }, p1/Z, [x22]\n"
647 "ld1w { z27.s }, p1/Z, [x21]\n"
648 "ld1w { z28.s }, p1/Z, [x20]\n"
659 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
660 "ldr w9, [x20, x10, LSL #0x2]\n"
661 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
662 "tbz %x[flags], #3, 59f\n"
663 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
664 "add x20, x20, x21, LSL #3\n"
665 "ldr x28, [x20, #0x0]\n"
666 "ldr x27, [x20, #0x8]\n"
667 "ldr x26, [x20, #0x10]\n"
668 "ldr x25, [x20, #0x18]\n"
669 "ldr x24, [x20, #0x20]\n"
671 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
672 "add x28, x28, x20, LSL #2\n"
673 "add x27, x27, x20, LSL #2\n"
674 "add x26, x26, x20, LSL #2\n"
675 "add x25, x25, x20, LSL #2\n"
676 "add x24, x24, x20, LSL #2\n"
679 "mov x28, %x[input_ptr]\n"
680 "add x27, x28, x21, LSL #2\n"
681 "add x26, x27, x21, LSL #2\n"
682 "add x25, x26, x21, LSL #2\n"
683 "add x24, x25, x21, LSL #2\n"
688 "whilelt p0.s, XZR, x9\n"
689 "ld1rqw { z4.s }, p0/Z, [x28]\n"
690 "ld1rqw { z3.s }, p0/Z, [x27]\n"
692 "ld1rqw { z2.s }, p0/Z, [x26]\n"
693 "ld1rqw { z1.s }, p0/Z, [x25]\n"
695 "add x28, x28, #0x10\n"
696 "ld1rqw { z0.s }, p0/Z, [x24]\n"
697 "ld1w { z16.s }, p2/Z, [x12]\n"
698 "fmla z24.s, z16.s, z4.s[0]\n"
699 "fmla z25.s, z16.s, z3.s[0]\n"
700 "fmla z26.s, z16.s, z2.s[0]\n"
701 "fmla z27.s, z16.s, z1.s[0]\n"
702 "ld1w { z18.s }, p2/Z, [x12, #1, MUL VL]\n"
703 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
704 "fmla z28.s, z16.s, z0.s[0]\n"
705 "fmla z24.s, z18.s, z4.s[1]\n"
706 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
707 "add x27, x27, #0x10\n"
708 "fmla z25.s, z18.s, z3.s[1]\n"
709 "fmla z26.s, z18.s, z2.s[1]\n"
710 "add x26, x26, #0x10\n"
711 "add x25, x25, #0x10\n"
712 "fmla z27.s, z18.s, z1.s[1]\n"
713 "fmla z28.s, z18.s, z0.s[1]\n"
714 "add x24, x24, #0x10\n"
715 "addvl x12, x12, #4\n"
716 "fmla z24.s, z17.s, z4.s[2]\n"
717 "fmla z25.s, z17.s, z3.s[2]\n"
718 "fmla z26.s, z17.s, z2.s[2]\n"
719 "fmla z27.s, z17.s, z1.s[2]\n"
720 "fmla z28.s, z17.s, z0.s[2]\n"
721 "fmla z24.s, z16.s, z4.s[3]\n"
722 "fmla z25.s, z16.s, z3.s[3]\n"
723 "fmla z26.s, z16.s, z2.s[3]\n"
724 "fmla z27.s, z16.s, z1.s[3]\n"
725 "fmla z28.s, z16.s, z0.s[3]\n"
728 "whilelt p0.s, XZR, x9\n"
729 "ld1rqw { z0.s }, p0/Z, [x28]\n"
730 "ld1rqw { z1.s }, p0/Z, [x27]\n"
731 "subs x9, x9, #0x1\n"
732 "ld1rqw { z2.s }, p0/Z, [x26]\n"
733 "ld1rqw { z3.s }, p0/Z, [x25]\n"
734 "ld1rqw { z4.s }, p0/Z, [x24]\n"
735 "ld1w { z16.s }, p2/Z, [x12]\n"
736 "fmla z24.s, z16.s, z0.s[0]\n"
737 "fmla z25.s, z16.s, z1.s[0]\n"
738 "fmla z26.s, z16.s, z2.s[0]\n"
739 "fmla z27.s, z16.s, z3.s[0]\n"
740 "addvl x12, x12, #1\n"
741 "fmla z28.s, z16.s, z4.s[0]\n"
743 "ld1w { z16.s }, p2/Z, [x12]\n"
744 "subs x9, x9, #0x1\n"
745 "fmla z24.s, z16.s, z0.s[1]\n"
746 "fmla z25.s, z16.s, z1.s[1]\n"
747 "fmla z26.s, z16.s, z2.s[1]\n"
748 "fmla z27.s, z16.s, z3.s[1]\n"
749 "addvl x12, x12, #1\n"
750 "fmla z28.s, z16.s, z4.s[1]\n"
752 "ld1w { z16.s }, p2/Z, [x12]\n"
753 "subs x9, x9, #0x1\n"
754 "fmla z24.s, z16.s, z0.s[2]\n"
755 "fmla z25.s, z16.s, z1.s[2]\n"
756 "fmla z26.s, z16.s, z2.s[2]\n"
757 "fmla z27.s, z16.s, z3.s[2]\n"
758 "addvl x12, x12, #1\n"
759 "fmla z28.s, z16.s, z4.s[2]\n"
761 "ld1w { z16.s }, p2/Z, [x12]\n"
762 "fmla z24.s, z16.s, z0.s[3]\n"
763 "fmla z25.s, z16.s, z1.s[3]\n"
764 "addvl x12, x12, #1\n"
765 "fmla z26.s, z16.s, z2.s[3]\n"
766 "fmla z27.s, z16.s, z3.s[3]\n"
767 "fmla z28.s, z16.s, z4.s[3]\n"
769 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
770 "add x10, x10, #0x1\n"
773 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
774 "add x27, x11, x20, LSL #2\n"
775 "add x26, x27, x20, LSL #2\n"
776 "add x25, x26, x20, LSL #2\n"
777 "add x24, x25, x20, LSL #2\n"
778 "tbz %x[flags], #1, 64f\n"
779 "add x20, %x[args_ptr], %[offset_max]\n"
780 "ld1rw { z17.s }, p2/Z, [x20]\n"
781 "add x20, %x[args_ptr], %[offset_min]\n"
782 "ld1rw { z16.s }, p2/Z, [x20]\n"
783 "fmin z24.s, p2/M, z24.s, z17.s\n"
784 "fmin z25.s, p2/M, z25.s, z17.s\n"
785 "fmin z26.s, p2/M, z26.s, z17.s\n"
786 "fmin z27.s, p2/M, z27.s, z17.s\n"
787 "fmin z28.s, p2/M, z28.s, z17.s\n"
788 "fmax z24.s, p2/M, z24.s, z16.s\n"
789 "fmax z25.s, p2/M, z25.s, z16.s\n"
790 "fmax z26.s, p2/M, z26.s, z16.s\n"
791 "fmax z27.s, p2/M, z27.s, z16.s\n"
792 "fmax z28.s, p2/M, z28.s, z16.s\n"
794 "st1w { z24.s }, p1, [x11]\n"
795 "addvl x11, x11, #1\n"
796 "st1w { z25.s }, p1, [x27]\n"
797 "st1w { z26.s }, p1, [x26]\n"
798 "st1w { z27.s }, p1, [x25]\n"
799 "st1w { z28.s }, p1, [x24]\n"
806 "mov x14, %x[bias]\n"
807 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
808 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
809 "mov x11, %x[output_ptr]\n"
812 "whilelt p1.s, x20, x13\n"
814 "ld1w { z24.s }, p2/Z, [x14]\n"
817 "addvl x14, x14, #1\n"
823 "tbz %x[flags], #0, 69f\n"
824 "ldr x24, [%x[args_ptr], %[offsetof_output_offset]]\n"
825 "add x23, x11, x24, LSL #2\n"
826 "add x20, x23, x24, LSL #2\n"
827 "ld1w { z24.s }, p1/Z, [x11]\n"
828 "add x22, x20, x24, LSL #2\n"
829 "add x21, x22, x24, LSL #2\n"
830 "ld1w { z25.s }, p1/Z, [x23]\n"
831 "ld1w { z26.s }, p1/Z, [x20]\n"
832 "add x20, x21, x24, LSL #2\n"
833 "ld1w { z27.s }, p1/Z, [x22]\n"
834 "ld1w { z28.s }, p1/Z, [x21]\n"
835 "ld1w { z29.s }, p1/Z, [x20]\n"
847 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
848 "ldr w9, [x20, x10, LSL #0x2]\n"
849 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
850 "tbz %x[flags], #3, 72f\n"
851 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
852 "add x20, x20, x21, LSL #3\n"
853 "ldr x28, [x20, #0x0]\n"
854 "ldr x27, [x20, #0x8]\n"
855 "ldr x26, [x20, #0x10]\n"
856 "ldr x25, [x20, #0x18]\n"
857 "ldr x24, [x20, #0x20]\n"
858 "ldr x23, [x20, #0x28]\n"
860 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
861 "add x28, x28, x20, LSL #2\n"
862 "add x27, x27, x20, LSL #2\n"
863 "add x26, x26, x20, LSL #2\n"
864 "add x25, x25, x20, LSL #2\n"
865 "add x24, x24, x20, LSL #2\n"
866 "add x23, x23, x20, LSL #2\n"
869 "mov x28, %x[input_ptr]\n"
870 "add x27, x28, x21, LSL #2\n"
871 "add x26, x27, x21, LSL #2\n"
872 "add x25, x26, x21, LSL #2\n"
873 "add x24, x25, x21, LSL #2\n"
874 "add x23, x24, x21, LSL #2\n"
879 "whilelt p0.s, XZR, x9\n"
880 "ld1rqw { z5.s }, p0/Z, [x28]\n"
881 "ld1rqw { z4.s }, p0/Z, [x27]\n"
883 "ld1rqw { z3.s }, p0/Z, [x26]\n"
884 "ld1rqw { z2.s }, p0/Z, [x25]\n"
886 "add x28, x28, #0x10\n"
887 "ld1rqw { z1.s }, p0/Z, [x24]\n"
888 "ld1rqw { z0.s }, p0/Z, [x23]\n"
889 "add x27, x27, #0x10\n"
890 "add x26, x26, #0x10\n"
891 "ld1w { z19.s }, p2/Z, [x12]\n"
892 "fmla z24.s, z19.s, z5.s[0]\n"
893 "fmla z25.s, z19.s, z4.s[0]\n"
894 "ld1w { z18.s }, p2/Z, [x12, #1, MUL VL]\n"
895 "fmla z26.s, z19.s, z3.s[0]\n"
896 "fmla z27.s, z19.s, z2.s[0]\n"
897 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
898 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
899 "fmla z28.s, z19.s, z1.s[0]\n"
900 "fmla z29.s, z19.s, z0.s[0]\n"
901 "add x25, x25, #0x10\n"
902 "add x24, x24, #0x10\n"
903 "fmla z24.s, z18.s, z5.s[1]\n"
904 "fmla z25.s, z18.s, z4.s[1]\n"
905 "add x23, x23, #0x10\n"
906 "addvl x12, x12, #4\n"
907 "fmla z26.s, z18.s, z3.s[1]\n"
908 "fmla z27.s, z18.s, z2.s[1]\n"
909 "fmla z28.s, z18.s, z1.s[1]\n"
910 "fmla z29.s, z18.s, z0.s[1]\n"
911 "fmla z24.s, z17.s, z5.s[2]\n"
912 "fmla z25.s, z17.s, z4.s[2]\n"
913 "fmla z26.s, z17.s, z3.s[2]\n"
914 "fmla z27.s, z17.s, z2.s[2]\n"
915 "fmla z28.s, z17.s, z1.s[2]\n"
916 "fmla z29.s, z17.s, z0.s[2]\n"
917 "fmla z24.s, z16.s, z5.s[3]\n"
918 "fmla z25.s, z16.s, z4.s[3]\n"
919 "fmla z26.s, z16.s, z3.s[3]\n"
920 "fmla z27.s, z16.s, z2.s[3]\n"
921 "fmla z28.s, z16.s, z1.s[3]\n"
922 "fmla z29.s, z16.s, z0.s[3]\n"
925 "whilelt p0.s, XZR, x9\n"
926 "ld1rqw { z0.s }, p0/Z, [x28]\n"
927 "ld1rqw { z1.s }, p0/Z, [x27]\n"
928 "subs x9, x9, #0x1\n"
929 "ld1rqw { z2.s }, p0/Z, [x26]\n"
930 "ld1rqw { z3.s }, p0/Z, [x25]\n"
931 "ld1rqw { z4.s }, p0/Z, [x24]\n"
932 "ld1rqw { z5.s }, p0/Z, [x23]\n"
933 "ld1w { z16.s }, p2/Z, [x12]\n"
934 "fmla z24.s, z16.s, z0.s[0]\n"
935 "fmla z25.s, z16.s, z1.s[0]\n"
936 "addvl x12, x12, #1\n"
937 "fmla z26.s, z16.s, z2.s[0]\n"
938 "fmla z27.s, z16.s, z3.s[0]\n"
939 "fmla z28.s, z16.s, z4.s[0]\n"
940 "fmla z29.s, z16.s, z5.s[0]\n"
942 "ld1w { z16.s }, p2/Z, [x12]\n"
943 "subs x9, x9, #0x1\n"
944 "fmla z24.s, z16.s, z0.s[1]\n"
945 "fmla z25.s, z16.s, z1.s[1]\n"
946 "fmla z26.s, z16.s, z2.s[1]\n"
947 "fmla z27.s, z16.s, z3.s[1]\n"
948 "addvl x12, x12, #1\n"
949 "fmla z28.s, z16.s, z4.s[1]\n"
950 "fmla z29.s, z16.s, z5.s[1]\n"
952 "ld1w { z16.s }, p2/Z, [x12]\n"
953 "subs x9, x9, #0x1\n"
954 "fmla z24.s, z16.s, z0.s[2]\n"
955 "fmla z25.s, z16.s, z1.s[2]\n"
956 "fmla z26.s, z16.s, z2.s[2]\n"
957 "fmla z27.s, z16.s, z3.s[2]\n"
958 "addvl x12, x12, #1\n"
959 "fmla z28.s, z16.s, z4.s[2]\n"
960 "fmla z29.s, z16.s, z5.s[2]\n"
962 "ld1w { z16.s }, p2/Z, [x12]\n"
963 "fmla z24.s, z16.s, z0.s[3]\n"
964 "fmla z25.s, z16.s, z1.s[3]\n"
965 "addvl x12, x12, #1\n"
966 "fmla z26.s, z16.s, z2.s[3]\n"
967 "fmla z27.s, z16.s, z3.s[3]\n"
968 "fmla z28.s, z16.s, z4.s[3]\n"
969 "fmla z29.s, z16.s, z5.s[3]\n"
971 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
972 "add x10, x10, #0x1\n"
975 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
976 "add x27, x11, x20, LSL #2\n"
977 "add x26, x27, x20, LSL #2\n"
978 "add x25, x26, x20, LSL #2\n"
979 "add x24, x25, x20, LSL #2\n"
980 "add x23, x24, x20, LSL #2\n"
981 "tbz %x[flags], #1, 77f\n"
982 "add x20, %x[args_ptr], %[offset_max]\n"
983 "ld1rw { z17.s }, p2/Z, [x20]\n"
984 "add x20, %x[args_ptr], %[offset_min]\n"
985 "ld1rw { z16.s }, p2/Z, [x20]\n"
986 "fmin z24.s, p2/M, z24.s, z17.s\n"
987 "fmin z25.s, p2/M, z25.s, z17.s\n"
988 "fmin z26.s, p2/M, z26.s, z17.s\n"
989 "fmin z27.s, p2/M, z27.s, z17.s\n"
990 "fmin z28.s, p2/M, z28.s, z17.s\n"
991 "fmin z29.s, p2/M, z29.s, z17.s\n"
992 "fmax z24.s, p2/M, z24.s, z16.s\n"
993 "fmax z25.s, p2/M, z25.s, z16.s\n"
994 "fmax z26.s, p2/M, z26.s, z16.s\n"
995 "fmax z27.s, p2/M, z27.s, z16.s\n"
996 "fmax z28.s, p2/M, z28.s, z16.s\n"
997 "fmax z29.s, p2/M, z29.s, z16.s\n"
999 "st1w { z24.s }, p1, [x11]\n"
1000 "addvl x11, x11, #1\n"
1001 "st1w { z25.s }, p1, [x27]\n"
1002 "st1w { z26.s }, p1, [x26]\n"
1003 "st1w { z27.s }, p1, [x25]\n"
1004 "st1w { z28.s }, p1, [x24]\n"
1005 "st1w { z29.s }, p1, [x23]\n"
1012 "mov x14, %x[bias]\n"
1013 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
1014 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
1015 "mov x11, %x[output_ptr]\n"
1018 "whilelt p1.s, x20, x13\n"
1020 "ld1w { z24.s }, p2/Z, [x14]\n"
1021 "mov z25.d, z24.d\n"
1022 "mov z26.d, z24.d\n"
1023 "addvl x14, x14, #1\n"
1024 "mov z27.d, z24.d\n"
1025 "mov z28.d, z24.d\n"
1026 "mov z29.d, z24.d\n"
1027 "mov z30.d, z24.d\n"
1030 "tbz %x[flags], #0, 82f\n"
1031 "ldr x24, [%x[args_ptr], %[offsetof_output_offset]]\n"
1032 "add x21, x11, x24, LSL #2\n"
1033 "add x20, x21, x24, LSL #2\n"
1034 "ld1w { z24.s }, p1/Z, [x11]\n"
1035 "add x23, x20, x24, LSL #2\n"
1036 "add x22, x23, x24, LSL #2\n"
1037 "ld1w { z25.s }, p1/Z, [x21]\n"
1038 "ld1w { z26.s }, p1/Z, [x20]\n"
1039 "add x21, x22, x24, LSL #2\n"
1040 "add x20, x21, x24, LSL #2\n"
1041 "ld1w { z27.s }, p1/Z, [x23]\n"
1042 "ld1w { z28.s }, p1/Z, [x22]\n"
1043 "ld1w { z29.s }, p1/Z, [x21]\n"
1044 "ld1w { z30.s }, p1/Z, [x20]\n"
1057 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
1058 "ldr w9, [x20, x10, LSL #0x2]\n"
1059 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1060 "tbz %x[flags], #3, 85f\n"
1061 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
1062 "add x20, x20, x21, LSL #3\n"
1063 "ldr x28, [x20, #0x0]\n"
1064 "ldr x27, [x20, #0x8]\n"
1065 "ldr x26, [x20, #0x10]\n"
1066 "ldr x25, [x20, #0x18]\n"
1067 "ldr x24, [x20, #0x20]\n"
1068 "ldr x23, [x20, #0x28]\n"
1069 "ldr x22, [x20, #0x30]\n"
1071 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
1072 "add x28, x28, x20, LSL #2\n"
1073 "add x27, x27, x20, LSL #2\n"
1074 "add x26, x26, x20, LSL #2\n"
1075 "add x25, x25, x20, LSL #2\n"
1076 "add x24, x24, x20, LSL #2\n"
1077 "add x23, x23, x20, LSL #2\n"
1078 "add x22, x22, x20, LSL #2\n"
1081 "mov x28, %x[input_ptr]\n"
1082 "add x27, x28, x21, LSL #2\n"
1083 "add x26, x27, x21, LSL #2\n"
1084 "add x25, x26, x21, LSL #2\n"
1085 "add x24, x25, x21, LSL #2\n"
1086 "add x23, x24, x21, LSL #2\n"
1087 "add x22, x23, x21, LSL #2\n"
1092 "whilelt p0.s, XZR, x9\n"
1093 "ld1rqw { z6.s }, p0/Z, [x28]\n"
1094 "ld1rqw { z5.s }, p0/Z, [x27]\n"
1095 "sub x9, x9, #0x4\n"
1096 "ld1rqw { z4.s }, p0/Z, [x26]\n"
1097 "ld1rqw { z3.s }, p0/Z, [x25]\n"
1099 "add x28, x28, #0x10\n"
1100 "ld1rqw { z2.s }, p0/Z, [x24]\n"
1101 "ld1rqw { z1.s }, p0/Z, [x23]\n"
1102 "add x27, x27, #0x10\n"
1103 "add x26, x26, #0x10\n"
1104 "ld1rqw { z0.s }, p0/Z, [x22]\n"
1105 "ld1w { z19.s }, p2/Z, [x12]\n"
1106 "fmla z24.s, z19.s, z6.s[0]\n"
1107 "fmla z25.s, z19.s, z5.s[0]\n"
1108 "fmla z26.s, z19.s, z4.s[0]\n"
1109 "fmla z27.s, z19.s, z3.s[0]\n"
1110 "ld1w { z18.s }, p2/Z, [x12, #1, MUL VL]\n"
1111 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
1112 "fmla z28.s, z19.s, z2.s[0]\n"
1113 "fmla z29.s, z19.s, z1.s[0]\n"
1114 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
1115 "add x25, x25, #0x10\n"
1116 "fmla z30.s, z19.s, z0.s[0]\n"
1117 "fmla z24.s, z18.s, z6.s[1]\n"
1118 "add x24, x24, #0x10\n"
1119 "add x23, x23, #0x10\n"
1120 "fmla z25.s, z18.s, z5.s[1]\n"
1121 "fmla z26.s, z18.s, z4.s[1]\n"
1122 "add x22, x22, #0x10\n"
1123 "addvl x12, x12, #4\n"
1124 "fmla z27.s, z18.s, z3.s[1]\n"
1125 "fmla z28.s, z18.s, z2.s[1]\n"
1126 "fmla z29.s, z18.s, z1.s[1]\n"
1127 "fmla z30.s, z18.s, z0.s[1]\n"
1128 "fmla z24.s, z17.s, z6.s[2]\n"
1129 "fmla z25.s, z17.s, z5.s[2]\n"
1130 "fmla z26.s, z17.s, z4.s[2]\n"
1131 "fmla z27.s, z17.s, z3.s[2]\n"
1132 "fmla z28.s, z17.s, z2.s[2]\n"
1133 "fmla z29.s, z17.s, z1.s[2]\n"
1134 "fmla z30.s, z17.s, z0.s[2]\n"
1135 "fmla z24.s, z16.s, z6.s[3]\n"
1136 "fmla z25.s, z16.s, z5.s[3]\n"
1137 "fmla z26.s, z16.s, z4.s[3]\n"
1138 "fmla z27.s, z16.s, z3.s[3]\n"
1139 "fmla z28.s, z16.s, z2.s[3]\n"
1140 "fmla z29.s, z16.s, z1.s[3]\n"
1141 "fmla z30.s, z16.s, z0.s[3]\n"
1144 "whilelt p0.s, XZR, x9\n"
1145 "ld1rqw { z0.s }, p0/Z, [x28]\n"
1146 "ld1rqw { z1.s }, p0/Z, [x27]\n"
1147 "subs x9, x9, #0x1\n"
1148 "ld1rqw { z2.s }, p0/Z, [x26]\n"
1149 "ld1rqw { z3.s }, p0/Z, [x25]\n"
1150 "ld1rqw { z4.s }, p0/Z, [x24]\n"
1151 "ld1rqw { z5.s }, p0/Z, [x23]\n"
1152 "ld1rqw { z6.s }, p0/Z, [x22]\n"
1153 "ld1w { z16.s }, p2/Z, [x12]\n"
1154 "fmla z24.s, z16.s, z0.s[0]\n"
1155 "fmla z25.s, z16.s, z1.s[0]\n"
1156 "fmla z26.s, z16.s, z2.s[0]\n"
1157 "fmla z27.s, z16.s, z3.s[0]\n"
1158 "addvl x12, x12, #1\n"
1159 "fmla z28.s, z16.s, z4.s[0]\n"
1160 "fmla z29.s, z16.s, z5.s[0]\n"
1161 "fmla z30.s, z16.s, z6.s[0]\n"
1163 "ld1w { z16.s }, p2/Z, [x12]\n"
1164 "subs x9, x9, #0x1\n"
1165 "fmla z24.s, z16.s, z0.s[1]\n"
1166 "fmla z25.s, z16.s, z1.s[1]\n"
1167 "fmla z26.s, z16.s, z2.s[1]\n"
1168 "fmla z27.s, z16.s, z3.s[1]\n"
1169 "addvl x12, x12, #1\n"
1170 "fmla z28.s, z16.s, z4.s[1]\n"
1171 "fmla z29.s, z16.s, z5.s[1]\n"
1172 "fmla z30.s, z16.s, z6.s[1]\n"
1174 "ld1w { z16.s }, p2/Z, [x12]\n"
1175 "subs x9, x9, #0x1\n"
1176 "fmla z24.s, z16.s, z0.s[2]\n"
1177 "fmla z25.s, z16.s, z1.s[2]\n"
1178 "fmla z26.s, z16.s, z2.s[2]\n"
1179 "fmla z27.s, z16.s, z3.s[2]\n"
1180 "addvl x12, x12, #1\n"
1181 "fmla z28.s, z16.s, z4.s[2]\n"
1182 "fmla z29.s, z16.s, z5.s[2]\n"
1183 "fmla z30.s, z16.s, z6.s[2]\n"
1185 "ld1w { z16.s }, p2/Z, [x12]\n"
1186 "fmla z24.s, z16.s, z0.s[3]\n"
1187 "fmla z25.s, z16.s, z1.s[3]\n"
1188 "addvl x12, x12, #1\n"
1189 "fmla z26.s, z16.s, z2.s[3]\n"
1190 "fmla z27.s, z16.s, z3.s[3]\n"
1191 "fmla z28.s, z16.s, z4.s[3]\n"
1192 "fmla z29.s, z16.s, z5.s[3]\n"
1193 "fmla z30.s, z16.s, z6.s[3]\n"
1195 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
1196 "add x10, x10, #0x1\n"
1199 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
1200 "add x27, x11, x20, LSL #2\n"
1201 "add x26, x27, x20, LSL #2\n"
1202 "add x25, x26, x20, LSL #2\n"
1203 "add x24, x25, x20, LSL #2\n"
1204 "add x23, x24, x20, LSL #2\n"
1205 "add x22, x23, x20, LSL #2\n"
1206 "tbz %x[flags], #1, 90f\n"
1207 "add x20, %x[args_ptr], %[offset_max]\n"
1208 "ld1rw { z17.s }, p2/Z, [x20]\n"
1209 "add x20, %x[args_ptr], %[offset_min]\n"
1210 "ld1rw { z16.s }, p2/Z, [x20]\n"
1211 "fmin z24.s, p2/M, z24.s, z17.s\n"
1212 "fmin z25.s, p2/M, z25.s, z17.s\n"
1213 "fmin z26.s, p2/M, z26.s, z17.s\n"
1214 "fmin z27.s, p2/M, z27.s, z17.s\n"
1215 "fmin z28.s, p2/M, z28.s, z17.s\n"
1216 "fmin z29.s, p2/M, z29.s, z17.s\n"
1217 "fmin z30.s, p2/M, z30.s, z17.s\n"
1218 "fmax z24.s, p2/M, z24.s, z16.s\n"
1219 "fmax z25.s, p2/M, z25.s, z16.s\n"
1220 "fmax z26.s, p2/M, z26.s, z16.s\n"
1221 "fmax z27.s, p2/M, z27.s, z16.s\n"
1222 "fmax z28.s, p2/M, z28.s, z16.s\n"
1223 "fmax z29.s, p2/M, z29.s, z16.s\n"
1224 "fmax z30.s, p2/M, z30.s, z16.s\n"
1226 "st1w { z24.s }, p1, [x11]\n"
1227 "addvl x11, x11, #1\n"
1228 "st1w { z25.s }, p1, [x27]\n"
1229 "st1w { z26.s }, p1, [x26]\n"
1230 "st1w { z27.s }, p1, [x25]\n"
1231 "st1w { z28.s }, p1, [x24]\n"
1232 "st1w { z29.s }, p1, [x23]\n"
1233 "st1w { z30.s }, p1, [x22]\n"
1240 "ldr x21, [%x[args_ptr], %[offsetof_output_offset]]\n"
1242 "mov x14, %x[bias]\n"
1243 "ldr x13, [%x[args_ptr], %[offsetof_N]]\n"
1244 "ldr x12, [%x[args_ptr], %[offsetof_B_ptr]]\n"
1245 "mov x11, %x[output_ptr]\n"
1246 "madd %x[output_ptr], x21, x20, %x[output_ptr]\n"
1249 "whilelt p1.s, x20, x13\n"
1251 "ld1w { z24.s }, p2/Z, [x14]\n"
1252 "mov z25.d, z24.d\n"
1253 "mov z26.d, z24.d\n"
1254 "addvl x14, x14, #1\n"
1255 "mov z27.d, z24.d\n"
1256 "mov z28.d, z24.d\n"
1257 "mov z29.d, z24.d\n"
1258 "mov z30.d, z24.d\n"
1259 "mov z31.d, z24.d\n"
1262 "tbz %x[flags], #0, 95f\n"
1263 "ldr x24, [%x[args_ptr], %[offsetof_output_offset]]\n"
1264 "add x22, x11, x24, LSL #2\n"
1265 "add x21, x22, x24, LSL #2\n"
1266 "ld1w { z24.s }, p1/Z, [x11]\n"
1267 "add x23, x21, x24, LSL #2\n"
1268 "add x20, x23, x24, LSL #2\n"
1269 "ld1w { z25.s }, p1/Z, [x22]\n"
1270 "ld1w { z26.s }, p1/Z, [x21]\n"
1271 "add x22, x20, x24, LSL #2\n"
1272 "add x21, x22, x24, LSL #2\n"
1273 "ld1w { z27.s }, p1/Z, [x23]\n"
1274 "ld1w { z28.s }, p1/Z, [x20]\n"
1275 "add x20, x21, x24, LSL #2\n"
1276 "ld1w { z29.s }, p1/Z, [x22]\n"
1277 "ld1w { z30.s }, p1/Z, [x21]\n"
1278 "ld1w { z31.s }, p1/Z, [x20]\n"
1292 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
1293 "ldr w9, [x20, x10, LSL #0x2]\n"
1294 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1295 "tbz %x[flags], #3, 98f\n"
1296 "ldr x20, [%x[input_ptr], x10, LSL #0x3]\n"
1297 "add x20, x20, x21, LSL #3\n"
1298 "ldr x28, [x20, #0x0]\n"
1299 "ldr x27, [x20, #0x8]\n"
1300 "ldr x26, [x20, #0x10]\n"
1301 "ldr x25, [x20, #0x18]\n"
1302 "ldr x24, [x20, #0x20]\n"
1303 "ldr x23, [x20, #0x28]\n"
1304 "ldr x22, [x20, #0x30]\n"
1305 "ldr x21, [x20, #0x38]\n"
1307 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
1308 "add x28, x28, x20, LSL #2\n"
1309 "add x27, x27, x20, LSL #2\n"
1310 "add x26, x26, x20, LSL #2\n"
1311 "add x25, x25, x20, LSL #2\n"
1312 "add x24, x24, x20, LSL #2\n"
1313 "add x23, x23, x20, LSL #2\n"
1314 "add x22, x22, x20, LSL #2\n"
1315 "add x21, x21, x20, LSL #2\n"
1318 "mov x28, %x[input_ptr]\n"
1319 "add x27, x28, x21, LSL #2\n"
1320 "add x26, x27, x21, LSL #2\n"
1321 "add x25, x26, x21, LSL #2\n"
1322 "add x24, x25, x21, LSL #2\n"
1323 "add x23, x24, x21, LSL #2\n"
1324 "add x22, x23, x21, LSL #2\n"
1325 "add x21, x22, x21, LSL #2\n"
1330 "whilelt p0.s, XZR, x9\n"
1331 "ld1rqw { z7.s }, p0/Z, [x28]\n"
1332 "ld1rqw { z6.s }, p0/Z, [x27]\n"
1333 "sub x9, x9, #0x4\n"
1334 "ld1rqw { z5.s }, p0/Z, [x26]\n"
1335 "ld1rqw { z4.s }, p0/Z, [x25]\n"
1337 "add x28, x28, #0x10\n"
1338 "ld1rqw { z3.s }, p0/Z, [x24]\n"
1339 "ld1rqw { z2.s }, p0/Z, [x23]\n"
1340 "add x27, x27, #0x10\n"
1341 "add x26, x26, #0x10\n"
1342 "ld1rqw { z1.s }, p0/Z, [x22]\n"
1343 "ld1rqw { z0.s }, p0/Z, [x21]\n"
1344 "add x25, x25, #0x10\n"
1345 "add x24, x24, #0x10\n"
1346 "ld1w { z19.s }, p2/Z, [x12]\n"
1347 "fmla z24.s, z19.s, z7.s[0]\n"
1348 "fmla z25.s, z19.s, z6.s[0]\n"
1349 "ld1w { z18.s }, p2/Z, [x12, #1, MUL VL]\n"
1350 "fmla z26.s, z19.s, z5.s[0]\n"
1351 "fmla z27.s, z19.s, z4.s[0]\n"
1352 "ld1w { z17.s }, p2/Z, [x12, #2, MUL VL]\n"
1353 "ld1w { z16.s }, p2/Z, [x12, #3, MUL VL]\n"
1354 "fmla z28.s, z19.s, z3.s[0]\n"
1355 "fmla z29.s, z19.s, z2.s[0]\n"
1356 "add x23, x23, #0x10\n"
1357 "add x22, x22, #0x10\n"
1358 "fmla z30.s, z19.s, z1.s[0]\n"
1359 "fmla z31.s, z19.s, z0.s[0]\n"
1360 "add x21, x21, #0x10\n"
1361 "addvl x12, x12, #4\n"
1362 "fmla z24.s, z18.s, z7.s[1]\n"
1363 "fmla z25.s, z18.s, z6.s[1]\n"
1364 "fmla z26.s, z18.s, z5.s[1]\n"
1365 "fmla z27.s, z18.s, z4.s[1]\n"
1366 "fmla z28.s, z18.s, z3.s[1]\n"
1367 "fmla z29.s, z18.s, z2.s[1]\n"
1368 "fmla z30.s, z18.s, z1.s[1]\n"
1369 "fmla z31.s, z18.s, z0.s[1]\n"
1370 "fmla z24.s, z17.s, z7.s[2]\n"
1371 "fmla z25.s, z17.s, z6.s[2]\n"
1372 "fmla z26.s, z17.s, z5.s[2]\n"
1373 "fmla z27.s, z17.s, z4.s[2]\n"
1374 "fmla z28.s, z17.s, z3.s[2]\n"
1375 "fmla z29.s, z17.s, z2.s[2]\n"
1376 "fmla z30.s, z17.s, z1.s[2]\n"
1377 "fmla z31.s, z17.s, z0.s[2]\n"
1378 "fmla z24.s, z16.s, z7.s[3]\n"
1379 "fmla z25.s, z16.s, z6.s[3]\n"
1380 "fmla z26.s, z16.s, z5.s[3]\n"
1381 "fmla z27.s, z16.s, z4.s[3]\n"
1382 "fmla z28.s, z16.s, z3.s[3]\n"
1383 "fmla z29.s, z16.s, z2.s[3]\n"
1384 "fmla z30.s, z16.s, z1.s[3]\n"
1385 "fmla z31.s, z16.s, z0.s[3]\n"
1388 "whilelt p0.s, XZR, x9\n"
1389 "ld1rqw { z0.s }, p0/Z, [x28]\n"
1390 "ld1rqw { z1.s }, p0/Z, [x27]\n"
1391 "subs x9, x9, #0x1\n"
1392 "ld1rqw { z2.s }, p0/Z, [x26]\n"
1393 "ld1rqw { z3.s }, p0/Z, [x25]\n"
1394 "ld1rqw { z4.s }, p0/Z, [x24]\n"
1395 "ld1rqw { z5.s }, p0/Z, [x23]\n"
1396 "ld1rqw { z6.s }, p0/Z, [x22]\n"
1397 "ld1rqw { z7.s }, p0/Z, [x21]\n"
1398 "ld1w { z16.s }, p2/Z, [x12]\n"
1399 "fmla z24.s, z16.s, z0.s[0]\n"
1400 "fmla z25.s, z16.s, z1.s[0]\n"
1401 "addvl x12, x12, #1\n"
1402 "fmla z26.s, z16.s, z2.s[0]\n"
1403 "fmla z27.s, z16.s, z3.s[0]\n"
1404 "fmla z28.s, z16.s, z4.s[0]\n"
1405 "fmla z29.s, z16.s, z5.s[0]\n"
1406 "fmla z30.s, z16.s, z6.s[0]\n"
1407 "fmla z31.s, z16.s, z7.s[0]\n"
1409 "ld1w { z16.s }, p2/Z, [x12]\n"
1410 "subs x9, x9, #0x1\n"
1411 "fmla z24.s, z16.s, z0.s[1]\n"
1412 "fmla z25.s, z16.s, z1.s[1]\n"
1413 "fmla z26.s, z16.s, z2.s[1]\n"
1414 "fmla z27.s, z16.s, z3.s[1]\n"
1415 "addvl x12, x12, #1\n"
1416 "fmla z28.s, z16.s, z4.s[1]\n"
1417 "fmla z29.s, z16.s, z5.s[1]\n"
1418 "fmla z30.s, z16.s, z6.s[1]\n"
1419 "fmla z31.s, z16.s, z7.s[1]\n"
1421 "ld1w { z16.s }, p2/Z, [x12]\n"
1422 "subs x9, x9, #0x1\n"
1423 "fmla z24.s, z16.s, z0.s[2]\n"
1424 "fmla z25.s, z16.s, z1.s[2]\n"
1425 "fmla z26.s, z16.s, z2.s[2]\n"
1426 "fmla z27.s, z16.s, z3.s[2]\n"
1427 "addvl x12, x12, #1\n"
1428 "fmla z28.s, z16.s, z4.s[2]\n"
1429 "fmla z29.s, z16.s, z5.s[2]\n"
1430 "fmla z30.s, z16.s, z6.s[2]\n"
1431 "fmla z31.s, z16.s, z7.s[2]\n"
1433 "ld1w { z16.s }, p2/Z, [x12]\n"
1434 "fmla z24.s, z16.s, z0.s[3]\n"
1435 "fmla z25.s, z16.s, z1.s[3]\n"
1436 "addvl x12, x12, #1\n"
1437 "fmla z26.s, z16.s, z2.s[3]\n"
1438 "fmla z27.s, z16.s, z3.s[3]\n"
1439 "fmla z28.s, z16.s, z4.s[3]\n"
1440 "fmla z29.s, z16.s, z5.s[3]\n"
1441 "fmla z30.s, z16.s, z6.s[3]\n"
1442 "fmla z31.s, z16.s, z7.s[3]\n"
1444 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
1445 "add x10, x10, #0x1\n"
1448 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
1449 "add x27, x11, x20, LSL #2\n"
1450 "add x26, x27, x20, LSL #2\n"
1451 "add x25, x26, x20, LSL #2\n"
1452 "add x24, x25, x20, LSL #2\n"
1453 "add x23, x24, x20, LSL #2\n"
1454 "add x22, x23, x20, LSL #2\n"
1455 "add x21, x22, x20, LSL #2\n"
1456 "tbz %x[flags], #1, 103f\n"
1457 "add x20, %x[args_ptr], %[offset_max]\n"
1458 "ld1rw { z17.s }, p2/Z, [x20]\n"
1459 "add x20, %x[args_ptr], %[offset_min]\n"
1460 "ld1rw { z16.s }, p2/Z, [x20]\n"
1461 "fmin z24.s, p2/M, z24.s, z17.s\n"
1462 "fmin z25.s, p2/M, z25.s, z17.s\n"
1463 "fmin z26.s, p2/M, z26.s, z17.s\n"
1464 "fmin z27.s, p2/M, z27.s, z17.s\n"
1465 "fmin z28.s, p2/M, z28.s, z17.s\n"
1466 "fmin z29.s, p2/M, z29.s, z17.s\n"
1467 "fmin z30.s, p2/M, z30.s, z17.s\n"
1468 "fmin z31.s, p2/M, z31.s, z17.s\n"
1469 "fmax z24.s, p2/M, z24.s, z16.s\n"
1470 "fmax z25.s, p2/M, z25.s, z16.s\n"
1471 "fmax z26.s, p2/M, z26.s, z16.s\n"
1472 "fmax z27.s, p2/M, z27.s, z16.s\n"
1473 "fmax z28.s, p2/M, z28.s, z16.s\n"
1474 "fmax z29.s, p2/M, z29.s, z16.s\n"
1475 "fmax z30.s, p2/M, z30.s, z16.s\n"
1476 "fmax z31.s, p2/M, z31.s, z16.s\n"
1478 "st1w { z24.s }, p1, [x11]\n"
1479 "addvl x11, x11, #1\n"
1480 "st1w { z25.s }, p1, [x27]\n"
1481 "st1w { z26.s }, p1, [x26]\n"
1482 "st1w { z27.s }, p1, [x25]\n"
1483 "st1w { z28.s }, p1, [x24]\n"
1484 "st1w { z29.s }, p1, [x23]\n"
1485 "st1w { z30.s }, p1, [x22]\n"
1486 "st1w { z31.s }, p1, [x21]\n"
1491 "subs %x[M], %x[M], #0x8\n"
1493 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1494 "tbz %x[flags], #3, 105f\n"
1495 "add x21, x21, #0x8\n"
1496 "str x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1500 "madd %x[input_ptr], x20, x21, %x[input_ptr]\n"
1503 : [
M]
"+&r" (
M), [input_ptr]
"+&r" (input_ptr), [output_ptr]
"+&r" (output_ptr)
1504 : [args_ptr]
"r" (&ka), [
bias]
"r" (
bias), [flags]
"r" (flags), [offset_max]
"I" (offsetof(KernelArgs, maxval)), [offset_min]
"I" (offsetof(KernelArgs, minval)), [offsetof_B_ptr]
"I" (offsetof(KernelArgs, B_ptr)), [offsetof_N]
"I" (offsetof(KernelArgs,
N)), [offsetof_input_initial_col]
"I" (offsetof(KernelArgs, input_initial_col)), [offsetof_input_offset]
"I" (offsetof(KernelArgs, input_offset)), [offsetof_num_strings]
"I" (offsetof(KernelArgs, num_strings)), [offsetof_output_offset]
"I" (offsetof(KernelArgs, output_offset)), [offsetof_string_lengths]
"I" (offsetof(KernelArgs, string_lengths))
1505 :
"cc",
"memory",
"p0",
"p1",
"p2",
"x9",
"x10",
"x11",
"x12",
"x13",
"x14",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25",
"x26",
"x27",
"x28",
"z0",
"z1",
"z2",
"z3",
"z4",
"z5",
"z6",
"z7",
"z16",
"z17",
"z18",
"z19",
"z24",
"z25",
"z26",
"z27",
"z28",
"z29",
"z30",
"z31"
1510 #endif // ARM_COMPUTE_ENABLE_SVE