Lines Matching refs:wrapper
67 wrapper::vstore(reinterpret_cast<uint8_t *>(out.ptr()) + x, in u8_neon_scale_nearest()
68 wrapper::vloadq(in_ptr + offset + offset_row + x)); in u8_neon_scale_nearest()
122 …using ExactTagType = typename wrapper::traits::neon_bitvector_tag_t<float, wrapper::traits::BitWid… in u8_neon_scale_bilinear()
192 const auto s00 = wrapper::vdup_n(s00_s, ExactTagType{}); in u8_neon_scale_bilinear()
193 const auto s01 = wrapper::vdup_n(s01_s, ExactTagType{}); in u8_neon_scale_bilinear()
194 const auto s10 = wrapper::vdup_n(s10_s, ExactTagType{}); in u8_neon_scale_bilinear()
195 const auto s11 = wrapper::vdup_n(s11_s, ExactTagType{}); in u8_neon_scale_bilinear()
210 const auto in00 = wrapper::vloadq(in_ptr_xi0_yi0 + cout * sizeof(uint8_t)); in u8_neon_scale_bilinear()
211 const auto in01 = wrapper::vloadq(in_ptr_xi1_yi0 + cout * sizeof(uint8_t)); in u8_neon_scale_bilinear()
212 const auto in10 = wrapper::vloadq(in_ptr_xi0_yi1 + cout * sizeof(uint8_t)); in u8_neon_scale_bilinear()
213 const auto in11 = wrapper::vloadq(in_ptr_xi1_yi1 + cout * sizeof(uint8_t)); in u8_neon_scale_bilinear()
215 const uint16x8_t in00_low = wrapper::vmovl(wrapper::vgetlow(in00)); in u8_neon_scale_bilinear()
216 const uint16x8_t in00_high = wrapper::vmovl(wrapper::vgethigh(in00)); in u8_neon_scale_bilinear()
218 … const auto in00_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in00_low))); in u8_neon_scale_bilinear()
219 … const auto in00_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in00_low))); in u8_neon_scale_bilinear()
220 … const auto in00_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in00_high))); in u8_neon_scale_bilinear()
221 … const auto in00_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in00_high))); in u8_neon_scale_bilinear()
223 const uint16x8_t in01_low = wrapper::vmovl(wrapper::vgetlow(in01)); in u8_neon_scale_bilinear()
224 const uint16x8_t in01_high = wrapper::vmovl(wrapper::vgethigh(in01)); in u8_neon_scale_bilinear()
226 … const auto in01_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in01_low))); in u8_neon_scale_bilinear()
227 … const auto in01_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in01_low))); in u8_neon_scale_bilinear()
228 … const auto in01_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in01_high))); in u8_neon_scale_bilinear()
229 … const auto in01_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in01_high))); in u8_neon_scale_bilinear()
231 const uint16x8_t in10_low = wrapper::vmovl(wrapper::vgetlow(in10)); in u8_neon_scale_bilinear()
232 const uint16x8_t in10_high = wrapper::vmovl(wrapper::vgethigh(in10)); in u8_neon_scale_bilinear()
234 … const auto in10_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in10_low))); in u8_neon_scale_bilinear()
235 … const auto in10_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in10_low))); in u8_neon_scale_bilinear()
236 … const auto in10_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in10_high))); in u8_neon_scale_bilinear()
237 … const auto in10_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in10_high))); in u8_neon_scale_bilinear()
239 const uint16x8_t in11_low = wrapper::vmovl(wrapper::vgetlow(in11)); in u8_neon_scale_bilinear()
240 const uint16x8_t in11_high = wrapper::vmovl(wrapper::vgethigh(in11)); in u8_neon_scale_bilinear()
242 … const auto in11_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in11_low))); in u8_neon_scale_bilinear()
243 … const auto in11_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in11_low))); in u8_neon_scale_bilinear()
244 … const auto in11_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in11_high))); in u8_neon_scale_bilinear()
245 … const auto in11_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in11_high))); in u8_neon_scale_bilinear()
247 auto out_0 = wrapper::vmul(in00_0, s00); in u8_neon_scale_bilinear()
248 out_0 = wrapper::vmla(out_0, in01_0, s01); in u8_neon_scale_bilinear()
249 out_0 = wrapper::vmla(out_0, in10_0, s10); in u8_neon_scale_bilinear()
250 out_0 = wrapper::vmla(out_0, in11_0, s11); in u8_neon_scale_bilinear()
252 auto out_1 = wrapper::vmul(in00_1, s00); in u8_neon_scale_bilinear()
253 out_1 = wrapper::vmla(out_1, in01_1, s01); in u8_neon_scale_bilinear()
254 out_1 = wrapper::vmla(out_1, in10_1, s10); in u8_neon_scale_bilinear()
255 out_1 = wrapper::vmla(out_1, in11_1, s11); in u8_neon_scale_bilinear()
257 auto out_2 = wrapper::vmul(in00_2, s00); in u8_neon_scale_bilinear()
258 out_2 = wrapper::vmla(out_2, in01_2, s01); in u8_neon_scale_bilinear()
259 out_2 = wrapper::vmla(out_2, in10_2, s10); in u8_neon_scale_bilinear()
260 out_2 = wrapper::vmla(out_2, in11_2, s11); in u8_neon_scale_bilinear()
262 auto out_3 = wrapper::vmul(in00_3, s00); in u8_neon_scale_bilinear()
263 out_3 = wrapper::vmla(out_3, in01_3, s01); in u8_neon_scale_bilinear()
264 out_3 = wrapper::vmla(out_3, in10_3, s10); in u8_neon_scale_bilinear()
265 out_3 = wrapper::vmla(out_3, in11_3, s11); in u8_neon_scale_bilinear()
268 const auto out_0_int = wrapper::vcvta<uint32_t>(out_0); in u8_neon_scale_bilinear()
269 const auto out_1_int = wrapper::vcvta<uint32_t>(out_1); in u8_neon_scale_bilinear()
270 const auto out_2_int = wrapper::vcvta<uint32_t>(out_2); in u8_neon_scale_bilinear()
271 const auto out_3_int = wrapper::vcvta<uint32_t>(out_3); in u8_neon_scale_bilinear()
273 const auto out_0_int = wrapper::vcvt<uint32_t>(out_0); in u8_neon_scale_bilinear()
274 const auto out_1_int = wrapper::vcvt<uint32_t>(out_1); in u8_neon_scale_bilinear()
275 const auto out_2_int = wrapper::vcvt<uint32_t>(out_2); in u8_neon_scale_bilinear()
276 const auto out_3_int = wrapper::vcvt<uint32_t>(out_3); in u8_neon_scale_bilinear()
278 …const auto low_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_0_int), wrapper::vqmo… in u8_neon_scale_bilinear()
279 …const auto high_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_2_int), wrapper::vqmo… in u8_neon_scale_bilinear()
280 const auto out = wrapper::vcombine(low_part, high_part); in u8_neon_scale_bilinear()
282 wrapper::vstore(out_ptr_xo_yo + cout * sizeof(uint8_t), out); in u8_neon_scale_bilinear()
321 …using ExactTagType = typename wrapper::traits::neon_bitvector_tag_t<float, wrapper::traits::BitWid… in s8_neon_scale_bilinear()
396 const auto s00 = wrapper::vdup_n(s00_s, ExactTagType{}); in s8_neon_scale_bilinear()
397 const auto s01 = wrapper::vdup_n(s01_s, ExactTagType{}); in s8_neon_scale_bilinear()
398 const auto s10 = wrapper::vdup_n(s10_s, ExactTagType{}); in s8_neon_scale_bilinear()
399 const auto s11 = wrapper::vdup_n(s11_s, ExactTagType{}); in s8_neon_scale_bilinear()
414 const auto in00 = wrapper::vloadq(in_ptr_xi0_yi0 + cout * sizeof(int8_t)); in s8_neon_scale_bilinear()
415 const auto in01 = wrapper::vloadq(in_ptr_xi1_yi0 + cout * sizeof(int8_t)); in s8_neon_scale_bilinear()
416 const auto in10 = wrapper::vloadq(in_ptr_xi0_yi1 + cout * sizeof(int8_t)); in s8_neon_scale_bilinear()
417 const auto in11 = wrapper::vloadq(in_ptr_xi1_yi1 + cout * sizeof(int8_t)); in s8_neon_scale_bilinear()
419 const int16x8_t in00_low = wrapper::vmovl(wrapper::vgetlow(in00)); in s8_neon_scale_bilinear()
420 const int16x8_t in00_high = wrapper::vmovl(wrapper::vgethigh(in00)); in s8_neon_scale_bilinear()
422 … const auto in00_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in00_low))); in s8_neon_scale_bilinear()
423 … const auto in00_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in00_low))); in s8_neon_scale_bilinear()
424 … const auto in00_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in00_high))); in s8_neon_scale_bilinear()
425 … const auto in00_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in00_high))); in s8_neon_scale_bilinear()
427 const int16x8_t in01_low = wrapper::vmovl(wrapper::vgetlow(in01)); in s8_neon_scale_bilinear()
428 const int16x8_t in01_high = wrapper::vmovl(wrapper::vgethigh(in01)); in s8_neon_scale_bilinear()
430 … const auto in01_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in01_low))); in s8_neon_scale_bilinear()
431 … const auto in01_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in01_low))); in s8_neon_scale_bilinear()
432 … const auto in01_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in01_high))); in s8_neon_scale_bilinear()
433 … const auto in01_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in01_high))); in s8_neon_scale_bilinear()
435 const int16x8_t in10_low = wrapper::vmovl(wrapper::vgetlow(in10)); in s8_neon_scale_bilinear()
436 const int16x8_t in10_high = wrapper::vmovl(wrapper::vgethigh(in10)); in s8_neon_scale_bilinear()
438 … const auto in10_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in10_low))); in s8_neon_scale_bilinear()
439 … const auto in10_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in10_low))); in s8_neon_scale_bilinear()
440 … const auto in10_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in10_high))); in s8_neon_scale_bilinear()
441 … const auto in10_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in10_high))); in s8_neon_scale_bilinear()
443 const int16x8_t in11_low = wrapper::vmovl(wrapper::vgetlow(in11)); in s8_neon_scale_bilinear()
444 const int16x8_t in11_high = wrapper::vmovl(wrapper::vgethigh(in11)); in s8_neon_scale_bilinear()
446 … const auto in11_0 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in11_low))); in s8_neon_scale_bilinear()
447 … const auto in11_1 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in11_low))); in s8_neon_scale_bilinear()
448 … const auto in11_2 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgetlow(in11_high))); in s8_neon_scale_bilinear()
449 … const auto in11_3 = wrapper::vcvt<float>(wrapper::vmovl(wrapper::vgethigh(in11_high))); in s8_neon_scale_bilinear()
451 auto out_0 = wrapper::vmul(in00_0, s00); in s8_neon_scale_bilinear()
452 out_0 = wrapper::vmla(out_0, in01_0, s01); in s8_neon_scale_bilinear()
453 out_0 = wrapper::vmla(out_0, in10_0, s10); in s8_neon_scale_bilinear()
454 out_0 = wrapper::vmla(out_0, in11_0, s11); in s8_neon_scale_bilinear()
456 auto out_1 = wrapper::vmul(in00_1, s00); in s8_neon_scale_bilinear()
457 out_1 = wrapper::vmla(out_1, in01_1, s01); in s8_neon_scale_bilinear()
458 out_1 = wrapper::vmla(out_1, in10_1, s10); in s8_neon_scale_bilinear()
459 out_1 = wrapper::vmla(out_1, in11_1, s11); in s8_neon_scale_bilinear()
461 auto out_2 = wrapper::vmul(in00_2, s00); in s8_neon_scale_bilinear()
462 out_2 = wrapper::vmla(out_2, in01_2, s01); in s8_neon_scale_bilinear()
463 out_2 = wrapper::vmla(out_2, in10_2, s10); in s8_neon_scale_bilinear()
464 out_2 = wrapper::vmla(out_2, in11_2, s11); in s8_neon_scale_bilinear()
466 auto out_3 = wrapper::vmul(in00_3, s00); in s8_neon_scale_bilinear()
467 out_3 = wrapper::vmla(out_3, in01_3, s01); in s8_neon_scale_bilinear()
468 out_3 = wrapper::vmla(out_3, in10_3, s10); in s8_neon_scale_bilinear()
469 out_3 = wrapper::vmla(out_3, in11_3, s11); in s8_neon_scale_bilinear()
472 const auto out_0_int = wrapper::vcvta<int32_t>(out_0); in s8_neon_scale_bilinear()
473 const auto out_1_int = wrapper::vcvta<int32_t>(out_1); in s8_neon_scale_bilinear()
474 const auto out_2_int = wrapper::vcvta<int32_t>(out_2); in s8_neon_scale_bilinear()
475 const auto out_3_int = wrapper::vcvta<int32_t>(out_3); in s8_neon_scale_bilinear()
477 const auto out_0_int = wrapper::vcvt<int32_t>(out_0); in s8_neon_scale_bilinear()
478 const auto out_1_int = wrapper::vcvt<int32_t>(out_1); in s8_neon_scale_bilinear()
479 const auto out_2_int = wrapper::vcvt<int32_t>(out_2); in s8_neon_scale_bilinear()
480 const auto out_3_int = wrapper::vcvt<int32_t>(out_3); in s8_neon_scale_bilinear()
482 …const auto low_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_0_int), wrapper::vqmo… in s8_neon_scale_bilinear()
483 …const auto high_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_2_int), wrapper::vqmo… in s8_neon_scale_bilinear()
484 const auto out = wrapper::vcombine(low_part, high_part); in s8_neon_scale_bilinear()
486 wrapper::vstore(out_ptr_xo_yo + cout * sizeof(int8_t), out); in s8_neon_scale_bilinear()
549 wrapper::vstore(reinterpret_cast<int16_t *>(out.ptr()) + x, in s16_neon_scale_nearest()
550 wrapper::vloadq(in_ptr + offset + offset_row + x)); in s16_neon_scale_nearest()