24 #ifdef ARM_COMPUTE_ENABLE_SVE
27 #include "../../utils.hpp"
34 void sve_hybrid_fp32_mla_6x4VL_a64fx (
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);
105 "mov x12, %x[bias]\n"
106 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
107 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
108 "mov x9, %x[output_ptr]\n"
111 "whilelt p3.s, x20, x11\n"
113 "whilelt p2.s, x20, x11\n"
115 "whilelt p1.s, x20, x11\n"
117 "whilelt p0.s, x20, x11\n"
119 "ld1w { z8.s }, p4/Z, [x12]\n"
120 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
121 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
122 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
123 "addvl x12, x12, #4\n"
126 "tbz %x[flags], #0, 4f\n"
127 "ld1w { z8.s }, p3/Z, [x9]\n"
128 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
129 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
130 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
140 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
141 "ldr w27, [x20, x28, LSL #0x2]\n"
142 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
143 "tbz %x[flags], #3, 7f\n"
144 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
145 "add x20, x20, x21, LSL #3\n"
146 "ldr x26, [x20, #0x0]\n"
148 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
149 "add x26, x26, x20, LSL #2\n"
152 "mov x26, %x[input_ptr]\n"
154 "subs x27, x27, #0x1\n"
155 "ld1rw { z0.s }, p4/Z, [x26]\n"
156 "ld1w { z6.s }, p4/Z, [x10]\n"
157 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
160 "fmla z8.s, p4/M, z6.s, z0.s\n"
161 "fmla z9.s, p4/M, z7.s, z0.s\n"
162 "ld1w { z17.s }, p4/Z, [x10, #2, MUL VL]\n"
163 "ld1w { z16.s }, p4/Z, [x10, #3, MUL VL]\n"
164 "addvl x10, x10, #4\n"
165 "add x26, x26, #0x4\n"
166 "fmla z10.s, p4/M, z17.s, z0.s\n"
167 "fmla z11.s, p4/M, z16.s, z0.s\n"
168 "subs x27, x27, #0x1\n"
169 "ld1rw { z0.s }, p4/Z, [x26]\n"
170 "ld1w { z6.s }, p4/Z, [x10]\n"
171 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
174 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
175 "fmla z8.s, p4/M, z6.s, z0.s\n"
176 "fmla z9.s, p4/M, z7.s, z0.s\n"
177 "ld1w { z17.s }, p4/Z, [x10, #2, MUL VL]\n"
178 "ld1w { z16.s }, p4/Z, [x10, #3, MUL VL]\n"
179 "add x28, x28, #0x1\n"
181 "fmla z10.s, p4/M, z17.s, z0.s\n"
182 "fmla z11.s, p4/M, z16.s, z0.s\n"
183 "addvl x10, x10, #4\n"
185 "tbz %x[flags], #1, 11f\n"
186 "add x20, %x[args_ptr], %[offset_max]\n"
187 "ld1rw { z17.s }, p4/Z, [x20]\n"
188 "add x20, %x[args_ptr], %[offset_min]\n"
189 "ld1rw { z16.s }, p4/Z, [x20]\n"
190 "fmin z8.s, p4/M, z8.s, z17.s\n"
191 "fmin z9.s, p4/M, z9.s, z17.s\n"
192 "fmin z10.s, p4/M, z10.s, z17.s\n"
193 "fmin z11.s, p4/M, z11.s, z17.s\n"
194 "fmax z8.s, p4/M, z8.s, z16.s\n"
195 "fmax z9.s, p4/M, z9.s, z16.s\n"
196 "fmax z10.s, p4/M, z10.s, z16.s\n"
197 "fmax z11.s, p4/M, z11.s, z16.s\n"
199 "st1w { z8.s }, p3, [x9]\n"
200 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
201 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
202 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
205 "decw x11, ALL, MUL #4\n"
210 "mov x12, %x[bias]\n"
211 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
212 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
213 "mov x9, %x[output_ptr]\n"
216 "whilelt p3.s, x20, x11\n"
218 "whilelt p2.s, x20, x11\n"
220 "whilelt p1.s, x20, x11\n"
222 "whilelt p0.s, x20, x11\n"
224 "ld1w { z8.s }, p4/Z, [x12]\n"
225 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
228 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
229 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
232 "addvl x12, x12, #4\n"
235 "tbz %x[flags], #0, 16f\n"
236 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
237 "add x20, x9, x20, LSL #2\n"
238 "ld1w { z8.s }, p3/Z, [x9]\n"
239 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
240 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
241 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
242 "ld1w { z12.s }, p3/Z, [x20]\n"
243 "ld1w { z13.s }, p2/Z, [x20, #1, MUL VL]\n"
244 "ld1w { z14.s }, p1/Z, [x20, #2, MUL VL]\n"
245 "ld1w { z15.s }, p0/Z, [x20, #3, MUL VL]\n"
259 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
260 "ldr w27, [x20, x28, LSL #0x2]\n"
261 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
262 "tbz %x[flags], #3, 19f\n"
263 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
264 "add x20, x20, x21, LSL #3\n"
265 "ldr x26, [x20, #0x0]\n"
266 "ldr x25, [x20, #0x8]\n"
268 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
269 "add x26, x26, x20, LSL #2\n"
270 "add x25, x25, x20, LSL #2\n"
273 "mov x26, %x[input_ptr]\n"
274 "add x25, x26, x21, LSL #2\n"
276 "subs x27, x27, #0x1\n"
277 "ld1rw { z0.s }, p4/Z, [x26]\n"
278 "ld1rw { z1.s }, p4/Z, [x25]\n"
279 "ld1w { z6.s }, p4/Z, [x10]\n"
280 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
283 "fmla z8.s, p4/M, z6.s, z0.s\n"
284 "fmla z12.s, p4/M, z6.s, z1.s\n"
285 "ld1w { z17.s }, p4/Z, [x10, #2, MUL VL]\n"
286 "add x26, x26, #0x4\n"
287 "fmla z9.s, p4/M, z7.s, z0.s\n"
288 "fmla z13.s, p4/M, z7.s, z1.s\n"
289 "ld1w { z16.s }, p4/Z, [x10, #3, MUL VL]\n"
290 "addvl x10, x10, #4\n"
291 "subs x27, x27, #0x1\n"
292 "add x25, x25, #0x4\n"
293 "fmla z10.s, p4/M, z17.s, z0.s\n"
294 "fmla z14.s, p4/M, z17.s, z1.s\n"
295 "fmla z11.s, p4/M, z16.s, z0.s\n"
296 "fmla z15.s, p4/M, z16.s, z1.s\n"
297 "ld1rw { z0.s }, p4/Z, [x26]\n"
298 "ld1rw { z1.s }, p4/Z, [x25]\n"
299 "ld1w { z6.s }, p4/Z, [x10]\n"
300 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
303 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
304 "fmla z8.s, p4/M, z6.s, z0.s\n"
305 "fmla z12.s, p4/M, z6.s, z1.s\n"
306 "ld1w { z17.s }, p4/Z, [x10, #2, MUL VL]\n"
307 "fmla z9.s, p4/M, z7.s, z0.s\n"
308 "fmla z13.s, p4/M, z7.s, z1.s\n"
309 "ld1w { z16.s }, p4/Z, [x10, #3, MUL VL]\n"
310 "add x28, x28, #0x1\n"
312 "fmla z10.s, p4/M, z17.s, z0.s\n"
313 "fmla z14.s, p4/M, z17.s, z1.s\n"
314 "addvl x10, x10, #4\n"
315 "fmla z11.s, p4/M, z16.s, z0.s\n"
316 "fmla z15.s, p4/M, z16.s, z1.s\n"
318 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
319 "add x25, x9, x20, LSL #2\n"
320 "tbz %x[flags], #1, 23f\n"
321 "add x20, %x[args_ptr], %[offset_max]\n"
322 "ld1rw { z17.s }, p4/Z, [x20]\n"
323 "add x20, %x[args_ptr], %[offset_min]\n"
324 "ld1rw { z16.s }, p4/Z, [x20]\n"
325 "fmin z8.s, p4/M, z8.s, z17.s\n"
326 "fmin z9.s, p4/M, z9.s, z17.s\n"
327 "fmin z10.s, p4/M, z10.s, z17.s\n"
328 "fmin z11.s, p4/M, z11.s, z17.s\n"
329 "fmin z12.s, p4/M, z12.s, z17.s\n"
330 "fmin z13.s, p4/M, z13.s, z17.s\n"
331 "fmin z14.s, p4/M, z14.s, z17.s\n"
332 "fmin z15.s, p4/M, z15.s, z17.s\n"
333 "fmax z8.s, p4/M, z8.s, z16.s\n"
334 "fmax z9.s, p4/M, z9.s, z16.s\n"
335 "fmax z10.s, p4/M, z10.s, z16.s\n"
336 "fmax z11.s, p4/M, z11.s, z16.s\n"
337 "fmax z12.s, p4/M, z12.s, z16.s\n"
338 "fmax z13.s, p4/M, z13.s, z16.s\n"
339 "fmax z14.s, p4/M, z14.s, z16.s\n"
340 "fmax z15.s, p4/M, z15.s, z16.s\n"
342 "st1w { z8.s }, p3, [x9]\n"
343 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
344 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
345 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
347 "st1w { z12.s }, p3, [x25]\n"
348 "st1w { z13.s }, p2, [x25, #1, MUL VL]\n"
349 "st1w { z14.s }, p1, [x25, #2, MUL VL]\n"
350 "st1w { z15.s }, p0, [x25, #3, MUL VL]\n"
352 "decw x11, ALL, MUL #4\n"
357 "mov x12, %x[bias]\n"
358 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
359 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
360 "mov x9, %x[output_ptr]\n"
363 "whilelt p3.s, x20, x11\n"
365 "whilelt p2.s, x20, x11\n"
367 "whilelt p1.s, x20, x11\n"
369 "whilelt p0.s, x20, x11\n"
371 "ld1w { z8.s }, p4/Z, [x12]\n"
372 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
375 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
376 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
381 "addvl x12, x12, #4\n"
386 "tbz %x[flags], #0, 28f\n"
387 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
388 "add x21, x9, x20, LSL #2\n"
389 "add x20, x21, x20, LSL #2\n"
390 "ld1w { z8.s }, p3/Z, [x9]\n"
391 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
392 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
393 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
394 "ld1w { z12.s }, p3/Z, [x21]\n"
395 "ld1w { z13.s }, p2/Z, [x21, #1, MUL VL]\n"
396 "ld1w { z14.s }, p1/Z, [x21, #2, MUL VL]\n"
397 "ld1w { z15.s }, p0/Z, [x21, #3, MUL VL]\n"
398 "ld1w { z16.s }, p3/Z, [x20]\n"
399 "ld1w { z17.s }, p2/Z, [x20, #1, MUL VL]\n"
400 "ld1w { z18.s }, p1/Z, [x20, #2, MUL VL]\n"
401 "ld1w { z19.s }, p0/Z, [x20, #3, MUL VL]\n"
419 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
420 "ldr w27, [x20, x28, LSL #0x2]\n"
421 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
422 "tbz %x[flags], #3, 31f\n"
423 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
424 "add x20, x20, x21, LSL #3\n"
425 "ldr x26, [x20, #0x0]\n"
426 "ldr x25, [x20, #0x8]\n"
427 "ldr x24, [x20, #0x10]\n"
429 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
430 "add x26, x26, x20, LSL #2\n"
431 "add x25, x25, x20, LSL #2\n"
432 "add x24, x24, x20, LSL #2\n"
435 "mov x26, %x[input_ptr]\n"
436 "add x25, x26, x21, LSL #2\n"
437 "add x24, x25, x21, LSL #2\n"
439 "subs x27, x27, #0x1\n"
440 "ld1rw { z0.s }, p4/Z, [x26]\n"
441 "ld1rw { z1.s }, p4/Z, [x25]\n"
442 "ld1rw { z2.s }, p4/Z, [x24]\n"
443 "ld1w { z6.s }, p4/Z, [x10]\n"
444 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
447 "fmla z8.s, p4/M, z6.s, z0.s\n"
448 "fmla z12.s, p4/M, z6.s, z1.s\n"
449 "add x26, x26, #0x4\n"
450 "subs x27, x27, #0x1\n"
451 "fmla z16.s, p4/M, z6.s, z2.s\n"
452 "fmla z9.s, p4/M, z7.s, z0.s\n"
453 "ld1w { z21.s }, p4/Z, [x10, #2, MUL VL]\n"
454 "add x25, x25, #0x4\n"
455 "fmla z13.s, p4/M, z7.s, z1.s\n"
456 "fmla z17.s, p4/M, z7.s, z2.s\n"
457 "ld1w { z20.s }, p4/Z, [x10, #3, MUL VL]\n"
458 "addvl x10, x10, #4\n"
459 "add x24, x24, #0x4\n"
460 "fmla z10.s, p4/M, z21.s, z0.s\n"
461 "fmla z14.s, p4/M, z21.s, z1.s\n"
462 "fmla z18.s, p4/M, z21.s, z2.s\n"
463 "fmla z11.s, p4/M, z20.s, z0.s\n"
464 "ld1rw { z0.s }, p4/Z, [x26]\n"
465 "ld1w { z6.s }, p4/Z, [x10]\n"
466 "fmla z15.s, p4/M, z20.s, z1.s\n"
467 "fmla z19.s, p4/M, z20.s, z2.s\n"
468 "ld1rw { z1.s }, p4/Z, [x25]\n"
469 "ld1rw { z2.s }, p4/Z, [x24]\n"
470 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
473 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
474 "fmla z8.s, p4/M, z6.s, z0.s\n"
475 "fmla z12.s, p4/M, z6.s, z1.s\n"
476 "add x28, x28, #0x1\n"
477 "fmla z16.s, p4/M, z6.s, z2.s\n"
478 "fmla z9.s, p4/M, z7.s, z0.s\n"
479 "ld1w { z21.s }, p4/Z, [x10, #2, MUL VL]\n"
481 "fmla z13.s, p4/M, z7.s, z1.s\n"
482 "fmla z17.s, p4/M, z7.s, z2.s\n"
483 "ld1w { z20.s }, p4/Z, [x10, #3, MUL VL]\n"
484 "addvl x10, x10, #4\n"
485 "fmla z10.s, p4/M, z21.s, z0.s\n"
486 "fmla z14.s, p4/M, z21.s, z1.s\n"
487 "fmla z18.s, p4/M, z21.s, z2.s\n"
488 "fmla z11.s, p4/M, z20.s, z0.s\n"
489 "fmla z15.s, p4/M, z20.s, z1.s\n"
490 "fmla z19.s, p4/M, z20.s, z2.s\n"
492 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
493 "add x25, x9, x20, LSL #2\n"
494 "add x24, x25, x20, LSL #2\n"
495 "tbz %x[flags], #1, 35f\n"
496 "add x20, %x[args_ptr], %[offset_max]\n"
497 "ld1rw { z21.s }, p4/Z, [x20]\n"
498 "add x20, %x[args_ptr], %[offset_min]\n"
499 "ld1rw { z20.s }, p4/Z, [x20]\n"
500 "fmin z8.s, p4/M, z8.s, z21.s\n"
501 "fmin z9.s, p4/M, z9.s, z21.s\n"
502 "fmin z10.s, p4/M, z10.s, z21.s\n"
503 "fmin z11.s, p4/M, z11.s, z21.s\n"
504 "fmin z12.s, p4/M, z12.s, z21.s\n"
505 "fmin z13.s, p4/M, z13.s, z21.s\n"
506 "fmin z14.s, p4/M, z14.s, z21.s\n"
507 "fmin z15.s, p4/M, z15.s, z21.s\n"
508 "fmin z16.s, p4/M, z16.s, z21.s\n"
509 "fmin z17.s, p4/M, z17.s, z21.s\n"
510 "fmin z18.s, p4/M, z18.s, z21.s\n"
511 "fmin z19.s, p4/M, z19.s, z21.s\n"
512 "fmax z8.s, p4/M, z8.s, z20.s\n"
513 "fmax z9.s, p4/M, z9.s, z20.s\n"
514 "fmax z10.s, p4/M, z10.s, z20.s\n"
515 "fmax z11.s, p4/M, z11.s, z20.s\n"
516 "fmax z12.s, p4/M, z12.s, z20.s\n"
517 "fmax z13.s, p4/M, z13.s, z20.s\n"
518 "fmax z14.s, p4/M, z14.s, z20.s\n"
519 "fmax z15.s, p4/M, z15.s, z20.s\n"
520 "fmax z16.s, p4/M, z16.s, z20.s\n"
521 "fmax z17.s, p4/M, z17.s, z20.s\n"
522 "fmax z18.s, p4/M, z18.s, z20.s\n"
523 "fmax z19.s, p4/M, z19.s, z20.s\n"
525 "st1w { z8.s }, p3, [x9]\n"
526 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
527 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
528 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
530 "st1w { z12.s }, p3, [x25]\n"
531 "st1w { z13.s }, p2, [x25, #1, MUL VL]\n"
532 "st1w { z14.s }, p1, [x25, #2, MUL VL]\n"
533 "st1w { z15.s }, p0, [x25, #3, MUL VL]\n"
534 "st1w { z16.s }, p3, [x24]\n"
535 "st1w { z17.s }, p2, [x24, #1, MUL VL]\n"
536 "st1w { z18.s }, p1, [x24, #2, MUL VL]\n"
537 "st1w { z19.s }, p0, [x24, #3, MUL VL]\n"
539 "decw x11, ALL, MUL #4\n"
544 "mov x12, %x[bias]\n"
545 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
546 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
547 "mov x9, %x[output_ptr]\n"
550 "whilelt p3.s, x20, x11\n"
552 "whilelt p2.s, x20, x11\n"
554 "whilelt p1.s, x20, x11\n"
556 "whilelt p0.s, x20, x11\n"
558 "ld1w { z8.s }, p4/Z, [x12]\n"
559 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
562 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
563 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
568 "addvl x12, x12, #4\n"
577 "tbz %x[flags], #0, 40f\n"
578 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
579 "add x22, x9, x20, LSL #2\n"
580 "add x21, x22, x20, LSL #2\n"
581 "ld1w { z8.s }, p3/Z, [x9]\n"
582 "add x20, x21, x20, LSL #2\n"
583 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
584 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
585 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
586 "ld1w { z12.s }, p3/Z, [x22]\n"
587 "ld1w { z13.s }, p2/Z, [x22, #1, MUL VL]\n"
588 "ld1w { z14.s }, p1/Z, [x22, #2, MUL VL]\n"
589 "ld1w { z15.s }, p0/Z, [x22, #3, MUL VL]\n"
590 "ld1w { z16.s }, p3/Z, [x21]\n"
591 "ld1w { z17.s }, p2/Z, [x21, #1, MUL VL]\n"
592 "ld1w { z18.s }, p1/Z, [x21, #2, MUL VL]\n"
593 "ld1w { z19.s }, p0/Z, [x21, #3, MUL VL]\n"
594 "ld1w { z20.s }, p3/Z, [x20]\n"
595 "ld1w { z21.s }, p2/Z, [x20, #1, MUL VL]\n"
596 "ld1w { z22.s }, p1/Z, [x20, #2, MUL VL]\n"
597 "ld1w { z23.s }, p0/Z, [x20, #3, MUL VL]\n"
619 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
620 "ldr w27, [x20, x28, LSL #0x2]\n"
621 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
622 "tbz %x[flags], #3, 43f\n"
623 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
624 "add x20, x20, x21, LSL #3\n"
625 "ldr x26, [x20, #0x0]\n"
626 "ldr x25, [x20, #0x8]\n"
627 "ldr x24, [x20, #0x10]\n"
628 "ldr x23, [x20, #0x18]\n"
630 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
631 "add x26, x26, x20, LSL #2\n"
632 "add x25, x25, x20, LSL #2\n"
633 "add x24, x24, x20, LSL #2\n"
634 "add x23, x23, x20, LSL #2\n"
637 "mov x26, %x[input_ptr]\n"
638 "add x25, x26, x21, LSL #2\n"
639 "add x24, x25, x21, LSL #2\n"
640 "add x23, x24, x21, LSL #2\n"
642 "subs x27, x27, #0x1\n"
643 "ld1rw { z0.s }, p4/Z, [x26]\n"
644 "ld1rw { z1.s }, p4/Z, [x25]\n"
645 "ld1rw { z2.s }, p4/Z, [x24]\n"
646 "ld1rw { z3.s }, p4/Z, [x23]\n"
647 "ld1w { z6.s }, p4/Z, [x10]\n"
648 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
651 "fmla z8.s, p4/M, z6.s, z0.s\n"
652 "fmla z12.s, p4/M, z6.s, z1.s\n"
653 "add x26, x26, #0x4\n"
654 "subs x27, x27, #0x1\n"
655 "fmla z16.s, p4/M, z6.s, z2.s\n"
656 "fmla z20.s, p4/M, z6.s, z3.s\n"
657 "ld1w { z25.s }, p4/Z, [x10, #2, MUL VL]\n"
658 "add x25, x25, #0x4\n"
659 "fmla z9.s, p4/M, z7.s, z0.s\n"
660 "fmla z13.s, p4/M, z7.s, z1.s\n"
661 "add x24, x24, #0x4\n"
662 "add x23, x23, #0x4\n"
663 "fmla z17.s, p4/M, z7.s, z2.s\n"
664 "fmla z21.s, p4/M, z7.s, z3.s\n"
665 "ld1w { z24.s }, p4/Z, [x10, #3, MUL VL]\n"
666 "addvl x10, x10, #4\n"
667 "fmla z10.s, p4/M, z25.s, z0.s\n"
668 "fmla z14.s, p4/M, z25.s, z1.s\n"
669 "fmla z18.s, p4/M, z25.s, z2.s\n"
670 "fmla z22.s, p4/M, z25.s, z3.s\n"
671 "ld1w { z6.s }, p4/Z, [x10]\n"
672 "fmla z11.s, p4/M, z24.s, z0.s\n"
673 "fmla z15.s, p4/M, z24.s, z1.s\n"
674 "ld1rw { z0.s }, p4/Z, [x26]\n"
675 "ld1rw { z1.s }, p4/Z, [x25]\n"
676 "fmla z19.s, p4/M, z24.s, z2.s\n"
677 "fmla z23.s, p4/M, z24.s, z3.s\n"
678 "ld1rw { z2.s }, p4/Z, [x24]\n"
679 "ld1rw { z3.s }, p4/Z, [x23]\n"
680 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
683 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
684 "fmla z8.s, p4/M, z6.s, z0.s\n"
685 "fmla z12.s, p4/M, z6.s, z1.s\n"
686 "add x28, x28, #0x1\n"
687 "fmla z16.s, p4/M, z6.s, z2.s\n"
688 "fmla z20.s, p4/M, z6.s, z3.s\n"
689 "ld1w { z25.s }, p4/Z, [x10, #2, MUL VL]\n"
691 "fmla z9.s, p4/M, z7.s, z0.s\n"
692 "fmla z13.s, p4/M, z7.s, z1.s\n"
693 "fmla z17.s, p4/M, z7.s, z2.s\n"
694 "fmla z21.s, p4/M, z7.s, z3.s\n"
695 "ld1w { z24.s }, p4/Z, [x10, #3, MUL VL]\n"
696 "addvl x10, x10, #4\n"
697 "fmla z10.s, p4/M, z25.s, z0.s\n"
698 "fmla z14.s, p4/M, z25.s, z1.s\n"
699 "fmla z18.s, p4/M, z25.s, z2.s\n"
700 "fmla z22.s, p4/M, z25.s, z3.s\n"
701 "fmla z11.s, p4/M, z24.s, z0.s\n"
702 "fmla z15.s, p4/M, z24.s, z1.s\n"
703 "fmla z19.s, p4/M, z24.s, z2.s\n"
704 "fmla z23.s, p4/M, z24.s, z3.s\n"
706 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
707 "add x25, x9, x20, LSL #2\n"
708 "add x24, x25, x20, LSL #2\n"
709 "add x23, x24, x20, LSL #2\n"
710 "tbz %x[flags], #1, 47f\n"
711 "add x20, %x[args_ptr], %[offset_max]\n"
712 "ld1rw { z25.s }, p4/Z, [x20]\n"
713 "add x20, %x[args_ptr], %[offset_min]\n"
714 "ld1rw { z24.s }, p4/Z, [x20]\n"
715 "fmin z8.s, p4/M, z8.s, z25.s\n"
716 "fmin z9.s, p4/M, z9.s, z25.s\n"
717 "fmin z10.s, p4/M, z10.s, z25.s\n"
718 "fmin z11.s, p4/M, z11.s, z25.s\n"
719 "fmin z12.s, p4/M, z12.s, z25.s\n"
720 "fmin z13.s, p4/M, z13.s, z25.s\n"
721 "fmin z14.s, p4/M, z14.s, z25.s\n"
722 "fmin z15.s, p4/M, z15.s, z25.s\n"
723 "fmin z16.s, p4/M, z16.s, z25.s\n"
724 "fmin z17.s, p4/M, z17.s, z25.s\n"
725 "fmin z18.s, p4/M, z18.s, z25.s\n"
726 "fmin z19.s, p4/M, z19.s, z25.s\n"
727 "fmin z20.s, p4/M, z20.s, z25.s\n"
728 "fmin z21.s, p4/M, z21.s, z25.s\n"
729 "fmin z22.s, p4/M, z22.s, z25.s\n"
730 "fmin z23.s, p4/M, z23.s, z25.s\n"
731 "fmax z8.s, p4/M, z8.s, z24.s\n"
732 "fmax z9.s, p4/M, z9.s, z24.s\n"
733 "fmax z10.s, p4/M, z10.s, z24.s\n"
734 "fmax z11.s, p4/M, z11.s, z24.s\n"
735 "fmax z12.s, p4/M, z12.s, z24.s\n"
736 "fmax z13.s, p4/M, z13.s, z24.s\n"
737 "fmax z14.s, p4/M, z14.s, z24.s\n"
738 "fmax z15.s, p4/M, z15.s, z24.s\n"
739 "fmax z16.s, p4/M, z16.s, z24.s\n"
740 "fmax z17.s, p4/M, z17.s, z24.s\n"
741 "fmax z18.s, p4/M, z18.s, z24.s\n"
742 "fmax z19.s, p4/M, z19.s, z24.s\n"
743 "fmax z20.s, p4/M, z20.s, z24.s\n"
744 "fmax z21.s, p4/M, z21.s, z24.s\n"
745 "fmax z22.s, p4/M, z22.s, z24.s\n"
746 "fmax z23.s, p4/M, z23.s, z24.s\n"
748 "st1w { z8.s }, p3, [x9]\n"
749 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
750 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
751 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
753 "st1w { z12.s }, p3, [x25]\n"
754 "st1w { z13.s }, p2, [x25, #1, MUL VL]\n"
755 "st1w { z14.s }, p1, [x25, #2, MUL VL]\n"
756 "st1w { z15.s }, p0, [x25, #3, MUL VL]\n"
757 "st1w { z16.s }, p3, [x24]\n"
758 "st1w { z17.s }, p2, [x24, #1, MUL VL]\n"
759 "st1w { z18.s }, p1, [x24, #2, MUL VL]\n"
760 "st1w { z19.s }, p0, [x24, #3, MUL VL]\n"
761 "st1w { z20.s }, p3, [x23]\n"
762 "st1w { z21.s }, p2, [x23, #1, MUL VL]\n"
763 "st1w { z22.s }, p1, [x23, #2, MUL VL]\n"
764 "st1w { z23.s }, p0, [x23, #3, MUL VL]\n"
766 "decw x11, ALL, MUL #4\n"
771 "mov x12, %x[bias]\n"
772 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
773 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
774 "mov x9, %x[output_ptr]\n"
777 "whilelt p3.s, x20, x11\n"
779 "whilelt p2.s, x20, x11\n"
781 "whilelt p1.s, x20, x11\n"
783 "whilelt p0.s, x20, x11\n"
785 "ld1w { z8.s }, p4/Z, [x12]\n"
786 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
789 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
790 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
795 "addvl x12, x12, #4\n"
808 "tbz %x[flags], #0, 52f\n"
809 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
810 "add x23, x9, x20, LSL #2\n"
811 "add x22, x23, x20, LSL #2\n"
812 "ld1w { z8.s }, p3/Z, [x9]\n"
813 "add x21, x22, x20, LSL #2\n"
814 "add x20, x21, x20, LSL #2\n"
815 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
816 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
817 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
818 "ld1w { z12.s }, p3/Z, [x23]\n"
819 "ld1w { z13.s }, p2/Z, [x23, #1, MUL VL]\n"
820 "ld1w { z14.s }, p1/Z, [x23, #2, MUL VL]\n"
821 "ld1w { z15.s }, p0/Z, [x23, #3, MUL VL]\n"
822 "ld1w { z16.s }, p3/Z, [x22]\n"
823 "ld1w { z17.s }, p2/Z, [x22, #1, MUL VL]\n"
824 "ld1w { z18.s }, p1/Z, [x22, #2, MUL VL]\n"
825 "ld1w { z19.s }, p0/Z, [x22, #3, MUL VL]\n"
826 "ld1w { z20.s }, p3/Z, [x21]\n"
827 "ld1w { z21.s }, p2/Z, [x21, #1, MUL VL]\n"
828 "ld1w { z22.s }, p1/Z, [x21, #2, MUL VL]\n"
829 "ld1w { z23.s }, p0/Z, [x21, #3, MUL VL]\n"
830 "ld1w { z24.s }, p3/Z, [x20]\n"
831 "ld1w { z25.s }, p2/Z, [x20, #1, MUL VL]\n"
832 "ld1w { z26.s }, p1/Z, [x20, #2, MUL VL]\n"
833 "ld1w { z27.s }, p0/Z, [x20, #3, MUL VL]\n"
859 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
860 "ldr w27, [x20, x28, LSL #0x2]\n"
861 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
862 "tbz %x[flags], #3, 55f\n"
863 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
864 "add x20, x20, x21, LSL #3\n"
865 "ldr x26, [x20, #0x0]\n"
866 "ldr x25, [x20, #0x8]\n"
867 "ldr x24, [x20, #0x10]\n"
868 "ldr x23, [x20, #0x18]\n"
869 "ldr x22, [x20, #0x20]\n"
871 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
872 "add x26, x26, x20, LSL #2\n"
873 "add x25, x25, x20, LSL #2\n"
874 "add x24, x24, x20, LSL #2\n"
875 "add x23, x23, x20, LSL #2\n"
876 "add x22, x22, x20, LSL #2\n"
879 "mov x26, %x[input_ptr]\n"
880 "add x25, x26, x21, LSL #2\n"
881 "add x24, x25, x21, LSL #2\n"
882 "add x23, x24, x21, LSL #2\n"
883 "add x22, x23, x21, LSL #2\n"
885 "subs x27, x27, #0x1\n"
886 "ld1rw { z0.s }, p4/Z, [x26]\n"
887 "ld1rw { z1.s }, p4/Z, [x25]\n"
888 "ld1rw { z2.s }, p4/Z, [x24]\n"
889 "ld1rw { z3.s }, p4/Z, [x23]\n"
890 "ld1rw { z4.s }, p4/Z, [x22]\n"
891 "ld1w { z6.s }, p4/Z, [x10]\n"
892 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
895 "fmla z8.s, p4/M, z6.s, z0.s\n"
896 "fmla z12.s, p4/M, z6.s, z1.s\n"
897 "add x26, x26, #0x4\n"
898 "subs x27, x27, #0x1\n"
899 "fmla z16.s, p4/M, z6.s, z2.s\n"
900 "fmla z20.s, p4/M, z6.s, z3.s\n"
901 "add x25, x25, #0x4\n"
902 "add x24, x24, #0x4\n"
903 "fmla z24.s, p4/M, z6.s, z4.s\n"
904 "fmla z9.s, p4/M, z7.s, z0.s\n"
905 "ld1w { z29.s }, p4/Z, [x10, #2, MUL VL]\n"
906 "add x23, x23, #0x4\n"
907 "fmla z13.s, p4/M, z7.s, z1.s\n"
908 "fmla z17.s, p4/M, z7.s, z2.s\n"
909 "add x22, x22, #0x4\n"
910 "fmla z21.s, p4/M, z7.s, z3.s\n"
911 "fmla z25.s, p4/M, z7.s, z4.s\n"
912 "ld1w { z28.s }, p4/Z, [x10, #3, MUL VL]\n"
913 "addvl x10, x10, #4\n"
914 "fmla z10.s, p4/M, z29.s, z0.s\n"
915 "fmla z14.s, p4/M, z29.s, z1.s\n"
916 "fmla z18.s, p4/M, z29.s, z2.s\n"
917 "fmla z22.s, p4/M, z29.s, z3.s\n"
918 "fmla z26.s, p4/M, z29.s, z4.s\n"
919 "fmla z11.s, p4/M, z28.s, z0.s\n"
920 "ld1rw { z0.s }, p4/Z, [x26]\n"
921 "ld1w { z6.s }, p4/Z, [x10]\n"
922 "fmla z15.s, p4/M, z28.s, z1.s\n"
923 "fmla z19.s, p4/M, z28.s, z2.s\n"
924 "ld1rw { z1.s }, p4/Z, [x25]\n"
925 "ld1rw { z2.s }, p4/Z, [x24]\n"
926 "fmla z23.s, p4/M, z28.s, z3.s\n"
927 "fmla z27.s, p4/M, z28.s, z4.s\n"
928 "ld1rw { z3.s }, p4/Z, [x23]\n"
929 "ld1rw { z4.s }, p4/Z, [x22]\n"
930 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
933 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
934 "fmla z8.s, p4/M, z6.s, z0.s\n"
935 "fmla z12.s, p4/M, z6.s, z1.s\n"
936 "add x28, x28, #0x1\n"
937 "fmla z16.s, p4/M, z6.s, z2.s\n"
938 "fmla z20.s, p4/M, z6.s, z3.s\n"
940 "fmla z24.s, p4/M, z6.s, z4.s\n"
941 "fmla z9.s, p4/M, z7.s, z0.s\n"
942 "ld1w { z29.s }, p4/Z, [x10, #2, MUL VL]\n"
943 "fmla z13.s, p4/M, z7.s, z1.s\n"
944 "fmla z17.s, p4/M, z7.s, z2.s\n"
945 "fmla z21.s, p4/M, z7.s, z3.s\n"
946 "fmla z25.s, p4/M, z7.s, z4.s\n"
947 "ld1w { z28.s }, p4/Z, [x10, #3, MUL VL]\n"
948 "addvl x10, x10, #4\n"
949 "fmla z10.s, p4/M, z29.s, z0.s\n"
950 "fmla z14.s, p4/M, z29.s, z1.s\n"
951 "fmla z18.s, p4/M, z29.s, z2.s\n"
952 "fmla z22.s, p4/M, z29.s, z3.s\n"
953 "fmla z26.s, p4/M, z29.s, z4.s\n"
954 "fmla z11.s, p4/M, z28.s, z0.s\n"
955 "fmla z15.s, p4/M, z28.s, z1.s\n"
956 "fmla z19.s, p4/M, z28.s, z2.s\n"
957 "fmla z23.s, p4/M, z28.s, z3.s\n"
958 "fmla z27.s, p4/M, z28.s, z4.s\n"
960 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
961 "add x25, x9, x20, LSL #2\n"
962 "add x24, x25, x20, LSL #2\n"
963 "add x23, x24, x20, LSL #2\n"
964 "add x22, x23, x20, LSL #2\n"
965 "tbz %x[flags], #1, 59f\n"
966 "add x20, %x[args_ptr], %[offset_max]\n"
967 "ld1rw { z29.s }, p4/Z, [x20]\n"
968 "add x20, %x[args_ptr], %[offset_min]\n"
969 "ld1rw { z28.s }, p4/Z, [x20]\n"
970 "fmin z8.s, p4/M, z8.s, z29.s\n"
971 "fmin z9.s, p4/M, z9.s, z29.s\n"
972 "fmin z10.s, p4/M, z10.s, z29.s\n"
973 "fmin z11.s, p4/M, z11.s, z29.s\n"
974 "fmin z12.s, p4/M, z12.s, z29.s\n"
975 "fmin z13.s, p4/M, z13.s, z29.s\n"
976 "fmin z14.s, p4/M, z14.s, z29.s\n"
977 "fmin z15.s, p4/M, z15.s, z29.s\n"
978 "fmin z16.s, p4/M, z16.s, z29.s\n"
979 "fmin z17.s, p4/M, z17.s, z29.s\n"
980 "fmin z18.s, p4/M, z18.s, z29.s\n"
981 "fmin z19.s, p4/M, z19.s, z29.s\n"
982 "fmin z20.s, p4/M, z20.s, z29.s\n"
983 "fmin z21.s, p4/M, z21.s, z29.s\n"
984 "fmin z22.s, p4/M, z22.s, z29.s\n"
985 "fmin z23.s, p4/M, z23.s, z29.s\n"
986 "fmin z24.s, p4/M, z24.s, z29.s\n"
987 "fmin z25.s, p4/M, z25.s, z29.s\n"
988 "fmin z26.s, p4/M, z26.s, z29.s\n"
989 "fmin z27.s, p4/M, z27.s, z29.s\n"
990 "fmax z8.s, p4/M, z8.s, z28.s\n"
991 "fmax z9.s, p4/M, z9.s, z28.s\n"
992 "fmax z10.s, p4/M, z10.s, z28.s\n"
993 "fmax z11.s, p4/M, z11.s, z28.s\n"
994 "fmax z12.s, p4/M, z12.s, z28.s\n"
995 "fmax z13.s, p4/M, z13.s, z28.s\n"
996 "fmax z14.s, p4/M, z14.s, z28.s\n"
997 "fmax z15.s, p4/M, z15.s, z28.s\n"
998 "fmax z16.s, p4/M, z16.s, z28.s\n"
999 "fmax z17.s, p4/M, z17.s, z28.s\n"
1000 "fmax z18.s, p4/M, z18.s, z28.s\n"
1001 "fmax z19.s, p4/M, z19.s, z28.s\n"
1002 "fmax z20.s, p4/M, z20.s, z28.s\n"
1003 "fmax z21.s, p4/M, z21.s, z28.s\n"
1004 "fmax z22.s, p4/M, z22.s, z28.s\n"
1005 "fmax z23.s, p4/M, z23.s, z28.s\n"
1006 "fmax z24.s, p4/M, z24.s, z28.s\n"
1007 "fmax z25.s, p4/M, z25.s, z28.s\n"
1008 "fmax z26.s, p4/M, z26.s, z28.s\n"
1009 "fmax z27.s, p4/M, z27.s, z28.s\n"
1011 "st1w { z8.s }, p3, [x9]\n"
1012 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
1013 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
1014 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
1015 "addvl x9, x9, #4\n"
1016 "st1w { z12.s }, p3, [x25]\n"
1017 "st1w { z13.s }, p2, [x25, #1, MUL VL]\n"
1018 "st1w { z14.s }, p1, [x25, #2, MUL VL]\n"
1019 "st1w { z15.s }, p0, [x25, #3, MUL VL]\n"
1020 "st1w { z16.s }, p3, [x24]\n"
1021 "st1w { z17.s }, p2, [x24, #1, MUL VL]\n"
1022 "st1w { z18.s }, p1, [x24, #2, MUL VL]\n"
1023 "st1w { z19.s }, p0, [x24, #3, MUL VL]\n"
1024 "st1w { z20.s }, p3, [x23]\n"
1025 "st1w { z21.s }, p2, [x23, #1, MUL VL]\n"
1026 "st1w { z22.s }, p1, [x23, #2, MUL VL]\n"
1027 "st1w { z23.s }, p0, [x23, #3, MUL VL]\n"
1028 "st1w { z24.s }, p3, [x22]\n"
1029 "st1w { z25.s }, p2, [x22, #1, MUL VL]\n"
1030 "st1w { z26.s }, p1, [x22, #2, MUL VL]\n"
1031 "st1w { z27.s }, p0, [x22, #3, MUL VL]\n"
1033 "decw x11, ALL, MUL #4\n"
1038 "ldr x21, [%x[args_ptr], %[offsetof_output_offset]]\n"
1040 "mov x12, %x[bias]\n"
1041 "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
1042 "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
1043 "mov x9, %x[output_ptr]\n"
1044 "madd %x[output_ptr], x21, x20, %x[output_ptr]\n"
1047 "whilelt p3.s, x20, x11\n"
1049 "whilelt p2.s, x20, x11\n"
1051 "whilelt p1.s, x20, x11\n"
1053 "whilelt p0.s, x20, x11\n"
1055 "ld1w { z8.s }, p4/Z, [x12]\n"
1056 "ld1w { z9.s }, p4/Z, [x12, #1, MUL VL]\n"
1059 "ld1w { z10.s }, p4/Z, [x12, #2, MUL VL]\n"
1060 "ld1w { z11.s }, p4/Z, [x12, #3, MUL VL]\n"
1061 "mov z14.d, z10.d\n"
1062 "mov z15.d, z11.d\n"
1065 "addvl x12, x12, #4\n"
1066 "mov z18.d, z10.d\n"
1067 "mov z19.d, z11.d\n"
1070 "mov z22.d, z10.d\n"
1071 "mov z23.d, z11.d\n"
1074 "mov z26.d, z10.d\n"
1075 "mov z27.d, z11.d\n"
1078 "mov z30.d, z10.d\n"
1079 "mov z31.d, z11.d\n"
1082 "tbz %x[flags], #0, 64f\n"
1083 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
1084 "add x24, x9, x20, LSL #2\n"
1085 "add x23, x24, x20, LSL #2\n"
1086 "ld1w { z8.s }, p3/Z, [x9]\n"
1087 "add x22, x23, x20, LSL #2\n"
1088 "add x21, x22, x20, LSL #2\n"
1089 "ld1w { z9.s }, p2/Z, [x9, #1, MUL VL]\n"
1090 "ld1w { z10.s }, p1/Z, [x9, #2, MUL VL]\n"
1091 "add x20, x21, x20, LSL #2\n"
1092 "ld1w { z11.s }, p0/Z, [x9, #3, MUL VL]\n"
1093 "ld1w { z12.s }, p3/Z, [x24]\n"
1094 "ld1w { z13.s }, p2/Z, [x24, #1, MUL VL]\n"
1095 "ld1w { z14.s }, p1/Z, [x24, #2, MUL VL]\n"
1096 "ld1w { z15.s }, p0/Z, [x24, #3, MUL VL]\n"
1097 "ld1w { z16.s }, p3/Z, [x23]\n"
1098 "ld1w { z17.s }, p2/Z, [x23, #1, MUL VL]\n"
1099 "ld1w { z18.s }, p1/Z, [x23, #2, MUL VL]\n"
1100 "ld1w { z19.s }, p0/Z, [x23, #3, MUL VL]\n"
1101 "ld1w { z20.s }, p3/Z, [x22]\n"
1102 "ld1w { z21.s }, p2/Z, [x22, #1, MUL VL]\n"
1103 "ld1w { z22.s }, p1/Z, [x22, #2, MUL VL]\n"
1104 "ld1w { z23.s }, p0/Z, [x22, #3, MUL VL]\n"
1105 "ld1w { z24.s }, p3/Z, [x21]\n"
1106 "ld1w { z25.s }, p2/Z, [x21, #1, MUL VL]\n"
1107 "ld1w { z26.s }, p1/Z, [x21, #2, MUL VL]\n"
1108 "ld1w { z27.s }, p0/Z, [x21, #3, MUL VL]\n"
1109 "ld1w { z28.s }, p3/Z, [x20]\n"
1110 "ld1w { z29.s }, p2/Z, [x20, #1, MUL VL]\n"
1111 "ld1w { z30.s }, p1/Z, [x20, #2, MUL VL]\n"
1112 "ld1w { z31.s }, p0/Z, [x20, #3, MUL VL]\n"
1142 "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n"
1143 "ldr w27, [x20, x28, LSL #0x2]\n"
1144 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1145 "tbz %x[flags], #3, 67f\n"
1146 "ldr x20, [%x[input_ptr], x28, LSL #0x3]\n"
1147 "add x20, x20, x21, LSL #3\n"
1148 "ldr x26, [x20, #0x0]\n"
1149 "ldr x25, [x20, #0x8]\n"
1150 "ldr x24, [x20, #0x10]\n"
1151 "ldr x23, [x20, #0x18]\n"
1152 "ldr x22, [x20, #0x20]\n"
1153 "ldr x21, [x20, #0x28]\n"
1155 "ldr x20, [%x[args_ptr], %[offsetof_input_initial_col]]\n"
1156 "add x26, x26, x20, LSL #2\n"
1157 "add x25, x25, x20, LSL #2\n"
1158 "add x24, x24, x20, LSL #2\n"
1159 "add x23, x23, x20, LSL #2\n"
1160 "add x22, x22, x20, LSL #2\n"
1161 "add x21, x21, x20, LSL #2\n"
1164 "mov x26, %x[input_ptr]\n"
1165 "add x25, x26, x21, LSL #2\n"
1166 "add x24, x25, x21, LSL #2\n"
1167 "add x23, x24, x21, LSL #2\n"
1168 "add x22, x23, x21, LSL #2\n"
1169 "add x21, x22, x21, LSL #2\n"
1171 "subs x27, x27, #0x1\n"
1172 "ld1rw { z0.s }, p4/Z, [x26]\n"
1173 "ld1rw { z1.s }, p4/Z, [x25]\n"
1174 "ld1rw { z2.s }, p4/Z, [x24]\n"
1175 "ld1rw { z3.s }, p4/Z, [x23]\n"
1176 "ld1rw { z4.s }, p4/Z, [x22]\n"
1177 "ld1rw { z5.s }, p4/Z, [x21]\n"
1178 "ld1w { z6.s }, p4/Z, [x10]\n"
1179 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
1182 "fmla z8.s, p4/M, z6.s, z0.s\n"
1183 "fmla z12.s, p4/M, z6.s, z1.s\n"
1184 "add x26, x26, #0x4\n"
1185 "subs x27, x27, #0x1\n"
1186 "fmla z16.s, p4/M, z6.s, z2.s\n"
1187 "fmla z20.s, p4/M, z6.s, z3.s\n"
1188 "add x25, x25, #0x4\n"
1189 "add x24, x24, #0x4\n"
1190 "fmla z24.s, p4/M, z6.s, z4.s\n"
1191 "fmla z28.s, p4/M, z6.s, z5.s\n"
1192 "ld1w { z6.s }, p4/Z, [x10, #2, MUL VL]\n"
1193 "add x23, x23, #0x4\n"
1194 "fmla z9.s, p4/M, z7.s, z0.s\n"
1195 "fmla z13.s, p4/M, z7.s, z1.s\n"
1196 "add x22, x22, #0x4\n"
1197 "add x21, x21, #0x4\n"
1198 "fmla z17.s, p4/M, z7.s, z2.s\n"
1199 "fmla z21.s, p4/M, z7.s, z3.s\n"
1200 "fmla z25.s, p4/M, z7.s, z4.s\n"
1201 "fmla z29.s, p4/M, z7.s, z5.s\n"
1202 "ld1w { z7.s }, p4/Z, [x10, #3, MUL VL]\n"
1203 "addvl x10, x10, #4\n"
1204 "fmla z10.s, p4/M, z6.s, z0.s\n"
1205 "fmla z14.s, p4/M, z6.s, z1.s\n"
1206 "fmla z18.s, p4/M, z6.s, z2.s\n"
1207 "fmla z22.s, p4/M, z6.s, z3.s\n"
1208 "fmla z26.s, p4/M, z6.s, z4.s\n"
1209 "fmla z30.s, p4/M, z6.s, z5.s\n"
1210 "ld1w { z6.s }, p4/Z, [x10]\n"
1211 "fmla z11.s, p4/M, z7.s, z0.s\n"
1212 "fmla z15.s, p4/M, z7.s, z1.s\n"
1213 "ld1rw { z0.s }, p4/Z, [x26]\n"
1214 "ld1rw { z1.s }, p4/Z, [x25]\n"
1215 "fmla z19.s, p4/M, z7.s, z2.s\n"
1216 "fmla z23.s, p4/M, z7.s, z3.s\n"
1217 "ld1rw { z2.s }, p4/Z, [x24]\n"
1218 "ld1rw { z3.s }, p4/Z, [x23]\n"
1219 "fmla z27.s, p4/M, z7.s, z4.s\n"
1220 "fmla z31.s, p4/M, z7.s, z5.s\n"
1221 "ld1rw { z4.s }, p4/Z, [x22]\n"
1222 "ld1rw { z5.s }, p4/Z, [x21]\n"
1223 "ld1w { z7.s }, p4/Z, [x10, #1, MUL VL]\n"
1226 "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n"
1227 "fmla z8.s, p4/M, z6.s, z0.s\n"
1228 "fmla z12.s, p4/M, z6.s, z1.s\n"
1229 "add x28, x28, #0x1\n"
1230 "fmla z16.s, p4/M, z6.s, z2.s\n"
1231 "fmla z20.s, p4/M, z6.s, z3.s\n"
1233 "fmla z24.s, p4/M, z6.s, z4.s\n"
1234 "fmla z28.s, p4/M, z6.s, z5.s\n"
1235 "ld1w { z6.s }, p4/Z, [x10, #2, MUL VL]\n"
1236 "fmla z9.s, p4/M, z7.s, z0.s\n"
1237 "fmla z13.s, p4/M, z7.s, z1.s\n"
1238 "fmla z17.s, p4/M, z7.s, z2.s\n"
1239 "fmla z21.s, p4/M, z7.s, z3.s\n"
1240 "fmla z25.s, p4/M, z7.s, z4.s\n"
1241 "fmla z29.s, p4/M, z7.s, z5.s\n"
1242 "ld1w { z7.s }, p4/Z, [x10, #3, MUL VL]\n"
1243 "addvl x10, x10, #4\n"
1244 "fmla z10.s, p4/M, z6.s, z0.s\n"
1245 "fmla z14.s, p4/M, z6.s, z1.s\n"
1246 "fmla z18.s, p4/M, z6.s, z2.s\n"
1247 "fmla z22.s, p4/M, z6.s, z3.s\n"
1248 "fmla z26.s, p4/M, z6.s, z4.s\n"
1249 "fmla z30.s, p4/M, z6.s, z5.s\n"
1250 "fmla z11.s, p4/M, z7.s, z0.s\n"
1251 "fmla z15.s, p4/M, z7.s, z1.s\n"
1252 "fmla z19.s, p4/M, z7.s, z2.s\n"
1253 "fmla z23.s, p4/M, z7.s, z3.s\n"
1254 "fmla z27.s, p4/M, z7.s, z4.s\n"
1255 "fmla z31.s, p4/M, z7.s, z5.s\n"
1257 "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
1258 "add x25, x9, x20, LSL #2\n"
1259 "add x24, x25, x20, LSL #2\n"
1260 "add x23, x24, x20, LSL #2\n"
1261 "add x22, x23, x20, LSL #2\n"
1262 "add x21, x22, x20, LSL #2\n"
1263 "tbz %x[flags], #1, 71f\n"
1264 "add x20, %x[args_ptr], %[offset_max]\n"
1265 "ld1rw { z1.s }, p4/Z, [x20]\n"
1266 "add x20, %x[args_ptr], %[offset_min]\n"
1267 "ld1rw { z0.s }, p4/Z, [x20]\n"
1268 "fmin z8.s, p4/M, z8.s, z1.s\n"
1269 "fmin z9.s, p4/M, z9.s, z1.s\n"
1270 "fmin z10.s, p4/M, z10.s, z1.s\n"
1271 "fmin z11.s, p4/M, z11.s, z1.s\n"
1272 "fmin z12.s, p4/M, z12.s, z1.s\n"
1273 "fmin z13.s, p4/M, z13.s, z1.s\n"
1274 "fmin z14.s, p4/M, z14.s, z1.s\n"
1275 "fmin z15.s, p4/M, z15.s, z1.s\n"
1276 "fmin z16.s, p4/M, z16.s, z1.s\n"
1277 "fmin z17.s, p4/M, z17.s, z1.s\n"
1278 "fmin z18.s, p4/M, z18.s, z1.s\n"
1279 "fmin z19.s, p4/M, z19.s, z1.s\n"
1280 "fmin z20.s, p4/M, z20.s, z1.s\n"
1281 "fmin z21.s, p4/M, z21.s, z1.s\n"
1282 "fmin z22.s, p4/M, z22.s, z1.s\n"
1283 "fmin z23.s, p4/M, z23.s, z1.s\n"
1284 "fmin z24.s, p4/M, z24.s, z1.s\n"
1285 "fmin z25.s, p4/M, z25.s, z1.s\n"
1286 "fmin z26.s, p4/M, z26.s, z1.s\n"
1287 "fmin z27.s, p4/M, z27.s, z1.s\n"
1288 "fmin z28.s, p4/M, z28.s, z1.s\n"
1289 "fmin z29.s, p4/M, z29.s, z1.s\n"
1290 "fmin z30.s, p4/M, z30.s, z1.s\n"
1291 "fmin z31.s, p4/M, z31.s, z1.s\n"
1292 "fmax z8.s, p4/M, z8.s, z0.s\n"
1293 "fmax z9.s, p4/M, z9.s, z0.s\n"
1294 "fmax z10.s, p4/M, z10.s, z0.s\n"
1295 "fmax z11.s, p4/M, z11.s, z0.s\n"
1296 "fmax z12.s, p4/M, z12.s, z0.s\n"
1297 "fmax z13.s, p4/M, z13.s, z0.s\n"
1298 "fmax z14.s, p4/M, z14.s, z0.s\n"
1299 "fmax z15.s, p4/M, z15.s, z0.s\n"
1300 "fmax z16.s, p4/M, z16.s, z0.s\n"
1301 "fmax z17.s, p4/M, z17.s, z0.s\n"
1302 "fmax z18.s, p4/M, z18.s, z0.s\n"
1303 "fmax z19.s, p4/M, z19.s, z0.s\n"
1304 "fmax z20.s, p4/M, z20.s, z0.s\n"
1305 "fmax z21.s, p4/M, z21.s, z0.s\n"
1306 "fmax z22.s, p4/M, z22.s, z0.s\n"
1307 "fmax z23.s, p4/M, z23.s, z0.s\n"
1308 "fmax z24.s, p4/M, z24.s, z0.s\n"
1309 "fmax z25.s, p4/M, z25.s, z0.s\n"
1310 "fmax z26.s, p4/M, z26.s, z0.s\n"
1311 "fmax z27.s, p4/M, z27.s, z0.s\n"
1312 "fmax z28.s, p4/M, z28.s, z0.s\n"
1313 "fmax z29.s, p4/M, z29.s, z0.s\n"
1314 "fmax z30.s, p4/M, z30.s, z0.s\n"
1315 "fmax z31.s, p4/M, z31.s, z0.s\n"
1317 "st1w { z8.s }, p3, [x9]\n"
1318 "st1w { z9.s }, p2, [x9, #1, MUL VL]\n"
1319 "st1w { z10.s }, p1, [x9, #2, MUL VL]\n"
1320 "st1w { z11.s }, p0, [x9, #3, MUL VL]\n"
1321 "addvl x9, x9, #4\n"
1322 "st1w { z12.s }, p3, [x25]\n"
1323 "st1w { z13.s }, p2, [x25, #1, MUL VL]\n"
1324 "st1w { z14.s }, p1, [x25, #2, MUL VL]\n"
1325 "st1w { z15.s }, p0, [x25, #3, MUL VL]\n"
1326 "st1w { z16.s }, p3, [x24]\n"
1327 "st1w { z17.s }, p2, [x24, #1, MUL VL]\n"
1328 "st1w { z18.s }, p1, [x24, #2, MUL VL]\n"
1329 "st1w { z19.s }, p0, [x24, #3, MUL VL]\n"
1330 "st1w { z20.s }, p3, [x23]\n"
1331 "st1w { z21.s }, p2, [x23, #1, MUL VL]\n"
1332 "st1w { z22.s }, p1, [x23, #2, MUL VL]\n"
1333 "st1w { z23.s }, p0, [x23, #3, MUL VL]\n"
1334 "st1w { z24.s }, p3, [x22]\n"
1335 "st1w { z25.s }, p2, [x22, #1, MUL VL]\n"
1336 "st1w { z26.s }, p1, [x22, #2, MUL VL]\n"
1337 "st1w { z27.s }, p0, [x22, #3, MUL VL]\n"
1338 "st1w { z28.s }, p3, [x21]\n"
1339 "st1w { z29.s }, p2, [x21, #1, MUL VL]\n"
1340 "st1w { z30.s }, p1, [x21, #2, MUL VL]\n"
1341 "st1w { z31.s }, p0, [x21, #3, MUL VL]\n"
1343 "decw x11, ALL, MUL #4\n"
1346 "subs %x[M], %x[M], #0x6\n"
1348 "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1349 "tbz %x[flags], #3, 73f\n"
1350 "add x21, x21, #0x6\n"
1351 "str x21, [%x[args_ptr], %[offsetof_input_offset]]\n"
1355 "madd %x[input_ptr], x20, x21, %x[input_ptr]\n"
1358 : [
M]
"+&r" (
M), [input_ptr]
"+&r" (input_ptr), [output_ptr]
"+&r" (output_ptr)
1359 : [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))
1360 :
"cc",
"memory",
"p0",
"p1",
"p2",
"p3",
"p4",
"x9",
"x10",
"x11",
"x12",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25",
"x26",
"x27",
"x28",
"z0",
"z1",
"z2",
"z3",
"z4",
"z5",
"z6",
"z7",
"z8",
"z9",
"z10",
"z11",
"z12",
"z13",
"z14",
"z15",
"z16",
"z17",
"z18",
"z19",
"z20",
"z21",
"z22",
"z23",
"z24",
"z25",
"z26",
"z27",
"z28",
"z29",
"z30",
"z31"
1365 #endif // ARM_COMPUTE_ENABLE_SVE