diff --git a/mace/kernels/transpose.h b/mace/kernels/transpose.h index 5f42b8ad211a9841c5e225e15013f658b34cd5b7..3e52de1a105ff96c2fc93b6f0ce12f70078f4aa1 100644 --- a/mace/kernels/transpose.h +++ b/mace/kernels/transpose.h @@ -73,7 +73,8 @@ static void TransposeNCHWToNHWCC2(const float *input, for (w = 0; w + 3 < width; w += 4) { float32x4_t vi0 = vld1q_f32(input + in_offset); float32x4_t vi1 = vld1q_f32(input + in_offset + image_size); - vst2q_f32(output + out_offset, {vi0, vi1}); + float32x4x2_t vi = {vi0, vi1}; + vst2q_f32(output + out_offset, vi); in_offset += 4; out_offset += 8; diff --git a/mace/ops/conv_2d_benchmark.cc b/mace/ops/conv_2d_benchmark.cc index d935c5030e680c7990054a7a0f08836467e061e1..4a5d80e4cfbfabcd7d948d874fdc8c2f144fbfd4 100644 --- a/mace/ops/conv_2d_benchmark.cc +++ b/mace/ops/conv_2d_benchmark.cc @@ -168,7 +168,6 @@ BM_CONV_2D(1, 1024, 7, 7, 1, 1, 1, 1, SAME, 1024); BM_CONV_2D(64, 32, 34, 34, 3, 3, 1, 1, VALID, 32); BM_CONV_2D(1, 32, 34, 34, 3, 3, 1, 1, VALID, 32); -// bokeh BM_CONV_2D(1, 32, 256, 256, 1, 15, 1, 1, SAME, 2); BM_CONV_2D(1, 32, 256, 256, 15, 1, 1, 1, SAME, 2); BM_CONV_2D(1, 64, 64, 64, 15, 1, 1, 1, SAME, 2);