Lines Matching refs:va
258 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_sse() local
271 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_sse()
272 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
274 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_sse()
275 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
277 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_sse()
278 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
280 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_sse()
281 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
284 va += 8; in conv_im2col_sgemm_int8_sse()
292 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_sse()
293 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_sse()
294 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_sse()
295 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_sse()
298 va += 4; in conv_im2col_sgemm_int8_sse()
323 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_sse() local
329 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
330 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_sse()
332 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_sse()
333 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_sse()
335 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_sse()
336 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_sse()
338 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_sse()
339 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_sse()
341 va += 8; in conv_im2col_sgemm_int8_sse()
347 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
348 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_sse()
349 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_sse()
350 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_sse()
352 va += 4; in conv_im2col_sgemm_int8_sse()
377 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_sse() local
385 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_sse()
386 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_sse()
388 va += 2; in conv_im2col_sgemm_int8_sse()
396 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_sse()
398 va += 1; in conv_im2col_sgemm_int8_sse()
414 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_sse() local
418 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_sse()
420 va += 1; in conv_im2col_sgemm_int8_sse()
704 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_dequant_sse() local
717 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_dequant_sse()
718 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
720 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_dequant_sse()
721 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
723 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_dequant_sse()
724 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
726 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_dequant_sse()
727 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
730 va += 8; in conv_im2col_sgemm_int8_dequant_sse()
738 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
739 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
740 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
741 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
744 va += 4; in conv_im2col_sgemm_int8_dequant_sse()
769 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_dequant_sse() local
775 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
776 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
778 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
779 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
781 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
782 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
784 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
785 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_dequant_sse()
787 va += 8; in conv_im2col_sgemm_int8_dequant_sse()
793 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
794 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
795 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
796 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
798 va += 4; in conv_im2col_sgemm_int8_dequant_sse()
826 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_dequant_sse() local
834 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_dequant_sse()
835 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_dequant_sse()
837 va += 2; in conv_im2col_sgemm_int8_dequant_sse()
845 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_dequant_sse()
847 va += 1; in conv_im2col_sgemm_int8_dequant_sse()
863 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_dequant_sse() local
867 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_dequant_sse()
869 va += 1; in conv_im2col_sgemm_int8_dequant_sse()
1157 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_requant_sse() local
1170 sum0[n] += (int)va[0] * vb[2 * n]; // k0 in conv_im2col_sgemm_int8_requant_sse()
1171 sum0[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1173 sum1[n] += (int)va[2] * vb[2 * n]; // k1 in conv_im2col_sgemm_int8_requant_sse()
1174 sum1[n] += (int)va[3] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1176 sum2[n] += (int)va[4] * vb[2 * n]; // k2 in conv_im2col_sgemm_int8_requant_sse()
1177 sum2[n] += (int)va[5] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1179 sum3[n] += (int)va[6] * vb[2 * n]; // k3 in conv_im2col_sgemm_int8_requant_sse()
1180 sum3[n] += (int)va[7] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1183 va += 8; in conv_im2col_sgemm_int8_requant_sse()
1191 sum0[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1192 sum1[n] += (int)va[1] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1193 sum2[n] += (int)va[2] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1194 sum3[n] += (int)va[3] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1197 va += 4; in conv_im2col_sgemm_int8_requant_sse()
1222 signed char* va = kernel_tm.channel(i / 4); in conv_im2col_sgemm_int8_requant_sse() local
1228 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1229 sum0 += (int)va[1] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1231 sum1 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1232 sum1 += (int)va[3] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1234 sum2 += (int)va[4] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1235 sum2 += (int)va[5] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1237 sum3 += (int)va[6] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1238 sum3 += (int)va[7] * vb[1]; in conv_im2col_sgemm_int8_requant_sse()
1240 va += 8; in conv_im2col_sgemm_int8_requant_sse()
1246 sum0 += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1247 sum1 += (int)va[1] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1248 sum2 += (int)va[2] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1249 sum3 += (int)va[3] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1251 va += 4; in conv_im2col_sgemm_int8_requant_sse()
1281 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_requant_sse() local
1289 sum[n] += (int)va[0] * vb[2 * n]; in conv_im2col_sgemm_int8_requant_sse()
1290 sum[n] += (int)va[1] * vb[2 * n + 1]; in conv_im2col_sgemm_int8_requant_sse()
1292 va += 2; in conv_im2col_sgemm_int8_requant_sse()
1300 sum[n] += (int)va[0] * vb[n]; in conv_im2col_sgemm_int8_requant_sse()
1302 va += 1; in conv_im2col_sgemm_int8_requant_sse()
1318 signed char* va = kernel_tm.channel(i / 4 + i % 4); in conv_im2col_sgemm_int8_requant_sse() local
1322 sum += (int)va[0] * vb[0]; in conv_im2col_sgemm_int8_requant_sse()
1324 va += 1; in conv_im2col_sgemm_int8_requant_sse()