Lines Matching refs:batch
96 const int batch = thread_id / out_depth / out_width / out_height;
110 const int input_offset_temp = in_height * batch;
178 const int num_batches = args.batch;
247 const int batch = b / batch_blocks;
248 const int block = b - batch * batch_blocks;
252 const int inout_offset = batch * in_size + filter_offset;
338 const int batch = thread_id / out_width / out_height / out_depth;
354 // for each sample in the batch.
364 // pixels for a given batch and input depth. The following
371 (batch * in_depth + in_channel) * (in_height * in_width);
458 const int num_batches = args.batch;
601 args.batch * DivUp(args.out_depth, kBlockDepth) * kBlockDepth;
610 DivUp(args.batch * args.out_depth, kBlockDepth) * kBlockDepth;
698 args.batch * args.out_rows * args.out_cols * args.out_depth;
786 const int batch = thread_id / in_depth / in_width / in_height;
803 out_depth * out_width * (out_row + out_height * batch);
823 in_depth * (in_col + in_width * (in_row + in_height * batch));
858 const int batch = thread_id / in_depth / in_width / in_height;
889 (batch * out_depth * out_height * out_width) +
898 const int in_backprop_offset = (batch * in_height * in_width * in_depth) +
927 args.batch * args.in_rows * args.in_cols * args.in_depth;
1012 const int batch = thread_id / out_depth / out_width / out_height;
1026 out_depth * (out_col + out_width * (out_row + out_height * batch));
1034 const int input_offset_temp = in_width * (in_row + in_height * batch);
1055 const int input_offset_temp = in_width * (in_row + in_height * batch);
1074 // example in a batch, for example). This can reduce the
1124 const int num_batches = args.batch;
1196 const int batch = b / batch_blocks;
1197 const int block = b - batch * batch_blocks;
1201 const int inout_offset = batch * in_size + filter_offset;
1293 const int batch = thread_id / out_depth / out_width / out_height;
1306 (batch * out_depth * out_height * out_width) +
1318 (batch * in_depth * in_height * in_width) +
1340 (batch * in_depth * in_height * in_width) +
1359 // example in a batch, for example). This can reduce the
1394 const int num_batches = args.batch;
1551 args.batch * DivUp(args.out_depth, kBlockDepth) * kBlockDepth;
1558 DivUp(args.batch * args.out_depth, kBlockDepth) * kBlockDepth;
1670 args.batch * args.out_rows * args.out_cols * args.out_depth;
