30     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_q8_gemm_goi_w()  local
31       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_q8_gemm_goi_w()
35           *((int32_t*) packed_w) = b[nr_block_start + nr_block_offset] + boff;  in xnn_pack_q8_gemm_goi_w()
51 …const uint8_t kv = k[(nr_block_start + nr_block_offset) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_q8_gemm_goi_w()
81   for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_q8_gemm_io_w()  local
82     const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_q8_gemm_io_w()
86         *((int32_t*) packed_w) = b[nr_block_start + nr_block_offset] + boff;  in xnn_pack_q8_gemm_io_w()
102 …const uint8_t kv = k[(kr_block_start + kr_block_offset) * nc + (nr_block_start + nr_block_offset)];  in xnn_pack_q8_gemm_io_w()
130     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_q8_conv_goki_w()  local
131       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_q8_conv_goki_w()
135           *((int32_t*) packed_w) = b[nr_block_start + nr_block_offset] + boff;  in xnn_pack_q8_conv_goki_w()
153 …       k[((nr_block_start + nr_block_offset) * ks + ki) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_q8_conv_goki_w()
186     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_q8_conv_kgo_w()  local
187       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_q8_conv_kgo_w()
191           *((int32_t*) packed_w) = b[nr_block_start + nr_block_offset] + boff;  in xnn_pack_q8_conv_kgo_w()
205             k[ki * g * nc + (nr_block_start + nr_block_offset)];  in xnn_pack_q8_conv_kgo_w()
244         for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_q8_deconv_goki_w()  local
245           const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_q8_deconv_goki_w()
249               *((int32_t*) packed_w) = b[nr_block_start + nr_block_offset] + boff;  in xnn_pack_q8_deconv_goki_w()
268 …k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + (kr_block_start + kr_block_off…  in xnn_pack_q8_deconv_goki_w()
385     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f16_gemm_goi_w()  local
386       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f16_gemm_goi_w()
389           packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f16_gemm_goi_w()
398               k[(nr_block_start + nr_block_offset) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_f16_gemm_goi_w()
421   for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f16_gemm_io_w()  local
422     const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f16_gemm_io_w()
425         packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f16_gemm_io_w()
434             k[(kr_block_start + kr_block_offset) * nc + (nr_block_start + nr_block_offset)];  in xnn_pack_f16_gemm_io_w()
458     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_gemm_goi_w()  local
459       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_gemm_goi_w()
462           packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f32_gemm_goi_w()
471 …k[(nr_block_start + nr_block_offset) * kc + round_down_po2(kr_block_start, skr) + ((kr_block_start…  in xnn_pack_f32_gemm_goi_w()
482               k[(nr_block_start + nr_block_offset) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_f32_gemm_goi_w()
509   for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_gemm_io_w()  local
510     const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_gemm_io_w()
513         packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f32_gemm_io_w()
522 …_start + nr_block_offset * kr) & sr_mask) + kr_block_offset) * nc + (nr_block_start + nr_block_off…  in xnn_pack_f32_gemm_io_w()
533             k[(kr_block_start + kr_block_offset) * nc + (nr_block_start + nr_block_offset)];  in xnn_pack_f32_gemm_io_w()
556     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_gemminc_goi_w()  local
557       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_gemminc_goi_w()
563 …k[(nr_block_start + nr_block_offset) * kc + round_down_po2(kr_block_start, skr) + ((kr_block_start…  in xnn_pack_f32_gemminc_goi_w()
574               k[(nr_block_start + nr_block_offset) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_f32_gemminc_goi_w()
601     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_conv_goki_w()  local
602       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_conv_goki_w()
605           packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f32_conv_goki_w()
615 …k[((nr_block_start + nr_block_offset) * ks + ki) * kc + round_down_po2(kr_block_start, skr) + ((kr…  in xnn_pack_f32_conv_goki_w()
626 …       k[((nr_block_start + nr_block_offset) * ks + ki) * kc + (kr_block_start + kr_block_offset)];  in xnn_pack_f32_conv_goki_w()
652     for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_conv_kgo_w()  local
653       const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_conv_kgo_w()
656           packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f32_conv_kgo_w()
663             k[ki * g * nc + (nr_block_start + nr_block_offset)];  in xnn_pack_f32_conv_kgo_w()
686   for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_dconv_oki_w()  local
687     const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_dconv_oki_w()
703 …*packed_w++ = k[(((nr_block_start + min(nr_block_offset, nr_block_size - 1)) * kh + ky) * kw + kx)…  in xnn_pack_f32_dconv_oki_w()
739         for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {  in xnn_pack_f32_deconv_goki_w()  local
740           const size_t nr_block_size = min(nc - nr_block_start, nr);  in xnn_pack_f32_deconv_goki_w()
743               packed_w[nr_block_offset] = b[nr_block_start + nr_block_offset];  in xnn_pack_f32_deconv_goki_w()
753 …k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + round_down_po2(kr_block_start,…  in xnn_pack_f32_deconv_goki_w()
764 …k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + (kr_block_start + kr_block_off…  in xnn_pack_f32_deconv_goki_w()