Lines Matching refs:wrapper
102 …using FloatTagType = typename wrapper::traits::neon_bitvector_tag_t<float, wrapper::traits::BitWid… in qasymm8_neon_scale_bilinear()
103 …using Int32TagType = typename wrapper::traits::neon_bitvector_tag_t<int32_t, wrapper::traits::BitW… in qasymm8_neon_scale_bilinear()
135 const float32x4_t vscale_in = wrapper::vdup_n(iq_info.scale, FloatTagType{}); in qasymm8_neon_scale_bilinear()
136 …const int32x4_t voffset_in = wrapper::vdup_n(iq_info.offset, Int32TagType{}); // Offsets will be… in qasymm8_neon_scale_bilinear()
138 const float32x4_t invvscale_o = wrapper::vdup_n(1.f / oq_info.scale, FloatTagType{}); in qasymm8_neon_scale_bilinear()
181 const auto s00 = wrapper::vdup_n(s00_s, FloatTagType{}); in qasymm8_neon_scale_bilinear()
182 const auto s01 = wrapper::vdup_n(s01_s, FloatTagType{}); in qasymm8_neon_scale_bilinear()
183 const auto s10 = wrapper::vdup_n(s10_s, FloatTagType{}); in qasymm8_neon_scale_bilinear()
184 const auto s11 = wrapper::vdup_n(s11_s, FloatTagType{}); in qasymm8_neon_scale_bilinear()
199 const auto in00 = wrapper::vloadq(in_ptr_xi0_yi0 + cout * sizeof(uint8_t)); in qasymm8_neon_scale_bilinear()
200 const auto in01 = wrapper::vloadq(in_ptr_xi1_yi0 + cout * sizeof(uint8_t)); in qasymm8_neon_scale_bilinear()
201 const auto in10 = wrapper::vloadq(in_ptr_xi0_yi1 + cout * sizeof(uint8_t)); in qasymm8_neon_scale_bilinear()
202 const auto in11 = wrapper::vloadq(in_ptr_xi1_yi1 + cout * sizeof(uint8_t)); in qasymm8_neon_scale_bilinear()
204 const uint16x8_t in00_low = wrapper::vmovl(wrapper::vgetlow(in00)); in qasymm8_neon_scale_bilinear()
205 const uint16x8_t in00_high = wrapper::vmovl(wrapper::vgethigh(in00)); in qasymm8_neon_scale_bilinear()
207 …const auto in00_0 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
208 …const auto in00_1 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
209 …const auto in00_2 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
210 …const auto in00_3 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
212 const uint16x8_t in01_low = wrapper::vmovl(wrapper::vgetlow(in01)); in qasymm8_neon_scale_bilinear()
213 const uint16x8_t in01_high = wrapper::vmovl(wrapper::vgethigh(in01)); in qasymm8_neon_scale_bilinear()
215 …const auto in01_0 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
216 …const auto in01_1 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
217 …const auto in01_2 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
218 …const auto in01_3 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
220 const uint16x8_t in10_low = wrapper::vmovl(wrapper::vgetlow(in10)); in qasymm8_neon_scale_bilinear()
221 const uint16x8_t in10_high = wrapper::vmovl(wrapper::vgethigh(in10)); in qasymm8_neon_scale_bilinear()
223 …const auto in10_0 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
224 …const auto in10_1 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
225 …const auto in10_2 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
226 …const auto in10_3 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
228 const uint16x8_t in11_low = wrapper::vmovl(wrapper::vgetlow(in11)); in qasymm8_neon_scale_bilinear()
229 const uint16x8_t in11_high = wrapper::vmovl(wrapper::vgethigh(in11)); in qasymm8_neon_scale_bilinear()
231 …const auto in11_0 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
232 …const auto in11_1 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
233 …const auto in11_2 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
234 …const auto in11_3 = wrapper::vmul(wrapper::vcvt<float>(wrapper::vsub(wrapper::vreinterpret(wrapper… in qasymm8_neon_scale_bilinear()
236 auto out_0 = wrapper::vmul(in00_0, s00); in qasymm8_neon_scale_bilinear()
237 out_0 = wrapper::vmla(out_0, in01_0, s01); in qasymm8_neon_scale_bilinear()
238 out_0 = wrapper::vmla(out_0, in10_0, s10); in qasymm8_neon_scale_bilinear()
239 out_0 = wrapper::vmla(out_0, in11_0, s11); in qasymm8_neon_scale_bilinear()
241 auto out_1 = wrapper::vmul(in00_1, s00); in qasymm8_neon_scale_bilinear()
242 out_1 = wrapper::vmla(out_1, in01_1, s01); in qasymm8_neon_scale_bilinear()
243 out_1 = wrapper::vmla(out_1, in10_1, s10); in qasymm8_neon_scale_bilinear()
244 out_1 = wrapper::vmla(out_1, in11_1, s11); in qasymm8_neon_scale_bilinear()
246 auto out_2 = wrapper::vmul(in00_2, s00); in qasymm8_neon_scale_bilinear()
247 out_2 = wrapper::vmla(out_2, in01_2, s01); in qasymm8_neon_scale_bilinear()
248 out_2 = wrapper::vmla(out_2, in10_2, s10); in qasymm8_neon_scale_bilinear()
249 out_2 = wrapper::vmla(out_2, in11_2, s11); in qasymm8_neon_scale_bilinear()
251 auto out_3 = wrapper::vmul(in00_3, s00); in qasymm8_neon_scale_bilinear()
252 out_3 = wrapper::vmla(out_3, in01_3, s01); in qasymm8_neon_scale_bilinear()
253 out_3 = wrapper::vmla(out_3, in10_3, s10); in qasymm8_neon_scale_bilinear()
254 out_3 = wrapper::vmla(out_3, in11_3, s11); in qasymm8_neon_scale_bilinear()
257 … const auto out_0_int = wrapper::vcvta<uint32_t>(wrapper::vmla(voffset_o, out_0, invvscale_o)); in qasymm8_neon_scale_bilinear()
258 … const auto out_1_int = wrapper::vcvta<uint32_t>(wrapper::vmla(voffset_o, out_1, invvscale_o)); in qasymm8_neon_scale_bilinear()
259 … const auto out_2_int = wrapper::vcvta<uint32_t>(wrapper::vmla(voffset_o, out_2, invvscale_o)); in qasymm8_neon_scale_bilinear()
260 … const auto out_3_int = wrapper::vcvta<uint32_t>(wrapper::vmla(voffset_o, out_3, invvscale_o)); in qasymm8_neon_scale_bilinear()
262 … const auto out_0_int = wrapper::vcvt<uint32_t>(wrapper::vmla(voffset_o, out_0, invvscale_o)); in qasymm8_neon_scale_bilinear()
263 … const auto out_1_int = wrapper::vcvt<uint32_t>(wrapper::vmla(voffset_o, out_1, invvscale_o)); in qasymm8_neon_scale_bilinear()
264 … const auto out_2_int = wrapper::vcvt<uint32_t>(wrapper::vmla(voffset_o, out_2, invvscale_o)); in qasymm8_neon_scale_bilinear()
265 … const auto out_3_int = wrapper::vcvt<uint32_t>(wrapper::vmla(voffset_o, out_3, invvscale_o)); in qasymm8_neon_scale_bilinear()
267 …const auto low_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_0_int), wrapper::vqmo… in qasymm8_neon_scale_bilinear()
268 …const auto high_part = wrapper::vqmovn(wrapper::vcombine(wrapper::vqmovn(out_2_int), wrapper::vqmo… in qasymm8_neon_scale_bilinear()
269 const auto out = wrapper::vcombine(low_part, high_part); in qasymm8_neon_scale_bilinear()
271 wrapper::vstore(out_ptr_xo_yo + cout * sizeof(uint8_t), out); in qasymm8_neon_scale_bilinear()