Add support for reductions to TensorIterator #11908
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
colesbury
requested review from
apaszke,
ezyang,
gchanan,
soumith and
zdevito
as code owners
|
Awesome stuff! I can't wait to switch einsum and tensordot to this. |
fmassa
reviewed
Sep 21, 2018
aten/src/ATen/native/cpu/Reduce.h
Outdated
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
ngimel
reviewed
Sep 22, 2018
fmassa
reviewed
Sep 22, 2018
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
ping @colesbury |
This adds support for reductions like sum() and mul() to TensorIterator. Performance is similar to existing optimized code for CPU, and generally better than existing code for CUDA kernels. The templatized CUDA kernel requires fewer instantiations than the existing THCReduce/THCReduceAll code. For example, sum() previously generated 43 CUDA kernels, while it now requires only one (larger) CUDA kernel. I suspect this should reduce code-size and compilation time, but I haven't measured it.
colesbury
force-pushed
the
tensor_iterator_sum
branch
from
aeb0b9d to
2156e2b
Compare
colesbury
force-pushed
the
tensor_iterator_sum
branch
2 times, most recently
from
59a787e to
18b9b19
Compare
colesbury
force-pushed
the
tensor_iterator_sum
branch
from
18b9b19 to
0d5e19b
Compare
There was a problem hiding this comment.
colesbury has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
gchanan
reviewed
Oct 16, 2018
aten/src/ATen/native/cuda/Reduce.cuh
Outdated
| } | ||
|
|
||
| template <typename scalar_t, typename func_t> | ||
| struct Reduction { |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
|
||
| #include <sstream> | ||
|
|
||
| namespace at { namespace native { namespace { |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| static ScalarType get_dtype(Tensor& result, const Tensor& self, optional<ScalarType> dtype, | ||
| bool promote_integers=false) { | ||
| if (dtype.has_value()) { | ||
| return dtype.value(); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
There was a problem hiding this comment.
colesbury has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Oct 25, 2018
Summary: This adds support for reductions like sum() and mul() to TensorIterator. Performance is similar to existing optimized code for CPU, and generally better than existing code for CUDA kernels. The templatized CUDA kernel requires fewer instantiations than the existing THCReduce/THCReduceAll code. For example, sum() previously generated 43 CUDA kernels, while it now requires only one (larger) CUDA kernel. I suspect this should reduce code-size and compilation time, but I haven't measured it. Below are timings for sum() on [CPU](https://ark.intel.com/products/81908/Intel-Xeon-Processor-E5-2680-v3-30M-Cache-2_50-GHz) (12 threads and 1 thread) and CUDA with various tensor sizes. CPU | Reduction (dim) | Master | PR | Master (1 thread) | PR (1 thread) | |----------------------|---------|---------|-------------------|---------------| | 1024x1024 (all) | 22 us | 34 us | 136 us | 147 us | | 1024x1024 (0) | 30 us | 28 us | 160 us | 160 us | | 1024x1024 (1) | 25 us | 25 us | 171 us | 146 us | | 1024x10x1024 (all) | 542 us | 550 us | 4.14 ms | 3.11 ms | | 1024x10x1024 (0) | 658 us | 690 us | 6.80 ms | 5.93 ms | | 1024x10x1024 (1) | 761 us | 757 us | 3.34 ms | 3.52 ms | | 1024x10x1024 (2) | 538 us | 545 us | 3.73 ms | 3.04 ms | | 1024x1024x1024 (all) | 72 ms | 71 ms | 364 ms | 357 ms | | 1024x1024x1024 (0) | 94 ms | 90 ms | 935 ms | 927 ms | | 1024x1024x1024 (1) | 80 ms | 86 ms | 881 ms | 688 ms | | 1024x1024x1024 (2) | 71 ms | 71 ms | 456 ms | 354 ms | CUDA | Reduction (dim) | M40 base | M40 PR | P100 base | P100 PR | |----------------------|----------|---------|-----------|-----------| | 1024x10x1024 (all) | 238 us | 182 us | 136 us | 97 us | | 1024x10x1024 (0) | 166 us | 179 us | 105 us | 84 us | | 1024x10x1024 (1) | 181 us | 182 us | 89 us | 91 us | | 1024x10x1024 (2) | 180 us | 168 us | 88 us | 79 us | | 1024x1024x1024 (all) | 17.5 ms | 16.4 ms | 8.23 ms | 7.48 ms | | 1024x1024x1024 (0) | 27.2 ms | 28.6 ms | 7.63 ms | 7.38 ms | | 1024x1024x1024 (1) | 16.5 ms | 16.3 ms | 7.66 ms | 7.40 ms | | 1024x1024x1024 (2) | 17.8 ms | 16.4 ms | 8.37 ms | 7.31 ms | Timings were generated with this script: https://gist.github.com/colesbury/d3238b266d8a9872fe6f68f77619b379 Pull Request resolved: pytorch/pytorch#11908 Differential Revision: D10071760 Pulled By: colesbury fbshipit-source-id: 40e37a0e6803f1628b94cc5a52a10dfbb601f3d6
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds support for reductions like sum() and mul() to TensorIterator.
Performance is similar to existing optimized code for CPU, and generally
better than existing code for CUDA kernels.
The templatized CUDA kernel requires fewer instantiations than the
existing THCReduce/THCReduceAll code. For example, sum() previously
generated 43 CUDA kernels, while it now requires only one (larger)
CUDA kernel. I suspect this should reduce code-size and
compilation time, but I haven't measured it.
Below are timings for sum() on CPU (12 threads and 1 thread) and CUDA with various tensor sizes.
CPU
CUDA
Timings were generated with this script:
https://gist.github.com/colesbury/d3238b266d8a9872fe6f68f77619b379