This repository has been archived by the owner on Aug 11, 2020. It is now read-only.
Accelerate AddTakeGrad + Support Sorting #153
Merged
Conversation
This file contains 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
| @@ -6,6 +6,11 @@ | |||
| */ | |||
| #ifndef MSHADOW_CUDA_TENSOR_GPU_INL_CUH_ | |||
| #define MSHADOW_CUDA_TENSOR_GPU_INL_CUH_ | |||
| #include <thrust/device_ptr.h> | |||
There was a problem hiding this comment.
Is thrust installed with cuda by default? Does this complicate compiling and linking? Do we need to update config and cmake?
There was a problem hiding this comment.
Yes, thrust is included in cuda toolkit. Also, since it's a template library, there is no need to install. http://docs.nvidia.com/cuda/thrust/#axzz4EPrET26A . (I find that thrust has been added to CUDA toolkit long ago, maybe since CUDA 5 -- http://developer.download.nvidia.com/compute/cuda/5_0/rel/docs/CUDA_Toolkit_Release_Notes_And_Errata.txt).
Revise type of RangeExp Add repeat option to range Add new backward kernel for embedding + Add sort functions. Fix format Fix header Fix Lint + Doc Turn off the sorting support for CUDA < 7.0
sxjscience
force-pushed
the
acc_addgrad_add_sort
branch
from
9b12a5c to
32ffa76
Compare
|
@piiswrong I choose to use |
|
@piiswrong Is it OK to merge now? |
6 tasks
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.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.
rangeoperation to MShadow, which is similar to python's rangeAddTakeGradusing Torch's kernel, the backward pass of embedding layer will be much faster. Slow speed of the AddTakeGradKernel in embedding layer apache/mxnet#2612. I will make a PR to the MXNet side once it's merged.SortByKeyandVectorizedSort. The GPU side is implemented using Thrust. Also, sorting forhalf_tis currently not implemented. May add later.