Lines Matching refs:va
250 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_sse() local
263 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_sse()
264 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
266 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_sse()
267 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
269 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_sse()
270 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
272 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_sse()
273 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
276 va += 8; in conv_im2col_sgemm_int8_sse()
284 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_sse()
285 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_sse()
286 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_sse()
287 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_sse()
290 va += 4; in conv_im2col_sgemm_int8_sse()
315 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_sse() local
321 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
322 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_sse()
324 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_sse()
325 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_sse()
327 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_sse()
328 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_sse()
330 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_sse()
331 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_sse()
333 va += 8; in conv_im2col_sgemm_int8_sse()
339 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
340 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_sse()
341 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_sse()
342 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_sse()
344 va += 4; in conv_im2col_sgemm_int8_sse()
369 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_sse() local
377 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_sse()
378 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
380 va += 2; in conv_im2col_sgemm_int8_sse()
388 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_sse()
390 va += 1; in conv_im2col_sgemm_int8_sse()
406 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_sse() local
410 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
412 va += 1; in conv_im2col_sgemm_int8_sse()
696 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_dequant_sse() local
709 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_dequant_sse()
710 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
712 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_dequant_sse()
713 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
715 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_dequant_sse()
716 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
718 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_dequant_sse()
719 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
722 va += 8; in conv_im2col_sgemm_int8_dequant_sse()
730 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
731 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
732 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
733 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
736 va += 4; in conv_im2col_sgemm_int8_dequant_sse()
761 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_dequant_sse() local
767 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
768 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
770 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
771 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
773 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
774 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
776 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
777 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
779 va += 8; in conv_im2col_sgemm_int8_dequant_sse()
785 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
786 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
787 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
788 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
790 va += 4; in conv_im2col_sgemm_int8_dequant_sse()
818 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_dequant_sse() local
826 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_dequant_sse()
827 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
829 va += 2; in conv_im2col_sgemm_int8_dequant_sse()
837 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
839 va += 1; in conv_im2col_sgemm_int8_dequant_sse()
855 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_dequant_sse() local
859 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
861 va += 1; in conv_im2col_sgemm_int8_dequant_sse()
1149 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_requant_sse() local
1162 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_requant_sse()
1163 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1165 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_requant_sse()
1166 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1168 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_requant_sse()
1169 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1171 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_requant_sse()
1172 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1175 va += 8; in conv_im2col_sgemm_int8_requant_sse()
1183 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1184 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1185 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1186 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1189 va += 4; in conv_im2col_sgemm_int8_requant_sse()
1214 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_requant_sse() local
1220 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1221 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1223 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1224 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1226 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1227 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1229 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1230 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1232 va += 8; in conv_im2col_sgemm_int8_requant_sse()
1238 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1239 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1240 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1241 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1243 va += 4; in conv_im2col_sgemm_int8_requant_sse()
1273 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_requant_sse() local
1281 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_requant_sse()
1282 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1284 va += 2; in conv_im2col_sgemm_int8_requant_sse()
1292 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1294 va += 1; in conv_im2col_sgemm_int8_requant_sse()
1310 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_requant_sse() local
1314 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1316 va += 1; in conv_im2col_sgemm_int8_requant_sse()