Revert "Tensor construction codemod(raw_mutable_data) (#16373)" (#18680)
authorJunjie Bai <bai@in.tum.de>
Mon, 1 Apr 2019 21:30:09 +0000 (14:30 -0700)
committerFacebook Github Bot <facebook-github-bot@users.noreply.github.com>
Mon, 1 Apr 2019 21:39:13 +0000 (14:39 -0700)
commit246f5c412eea453d42245ff496316934c01bda73
tree32fa9df00a521533d0eee7d119944b7c24dde5fc
parentbdfdf6c2b936bc5ae34c6fe52dfbb92847ae4205
Revert "Tensor construction codemod(raw_mutable_data) (#16373)" (#18680)

Summary:
This reverts commit d73c830e236f5b980e5c91914b818d150b60278c.

We have observed significant perf drop when training ResNext101 with multiple amd GPUs:

Before:
https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-bench/1636/console
2 GPUs ResNext training got 150\~160 imgs/sec
4 GPUs ResNext training got 270\~280 imgs/sec

After:
https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-bench/1637/console
Both 2 and 4 GPUs ResNext training drop to 110\~120 imgs/sec

Similar perf drop are seen on ResNet50 training jobs as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18680

Differential Revision: D14702941

Pulled By: bddppq

fbshipit-source-id: 828141805afc23f25c08d4a2eb6d4b99f817c128
27 files changed:
caffe2/operators/boolean_mask_ops.cc
caffe2/operators/boolean_mask_ops.cu
caffe2/operators/boolean_unmask_ops.cc
caffe2/operators/boolean_unmask_ops.cu
caffe2/operators/concat_split_op.h
caffe2/operators/conditional_op.cc
caffe2/operators/copy_op.cu
caffe2/operators/copy_op.h
caffe2/operators/crf_viterbi_op.cc
caffe2/operators/dataset_ops.cc
caffe2/operators/ensure_cpu_output_op.h
caffe2/operators/flatten_op.h
caffe2/operators/gather_ranges_to_dense_op.h
caffe2/operators/lengths_tile_op.cc
caffe2/operators/pack_segments.cc
caffe2/operators/pack_segments.cu
caffe2/operators/partition_ops.h
caffe2/operators/prepend_dim_op.h
caffe2/operators/remove_data_blocks_op.h
caffe2/operators/reservoir_sampling.cc
caffe2/operators/reshape_op.h
caffe2/operators/sequence_ops.cc
caffe2/operators/text_file_reader.cc
caffe2/operators/tile_op.h
caffe2/operators/utility_ops.h
caffe2/quantization/server/fully_connected_dnnlowp_op.cc
caffe2/queue/rebatching_queue.cc