Optimize the rowwise add function. #7047
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
chengduoZH
reviewed
Dec 27, 2017
| template <typename T> | ||
| __global__ void RowwiseAddKernel(const T* a, const T* b, T* c, int64_t height, | ||
| int64_t width) { | ||
| int64_t num = height * width; |
There was a problem hiding this comment.
num can be passed in as a parameter.
There was a problem hiding this comment.
Done. Thank you!
| for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < num; | ||
| i += blockDim.x * gridDim.x) { | ||
| int h = i / width; | ||
| int w = i % width; |
There was a problem hiding this comment.
The integer modulo (%) and division operations are expensive in GPU hardware.
The division seems can be replaced by the multiplication. And modulo (%) can be replaced by subtraction and multiplication.
There was a problem hiding this comment.
Done. Thank you!
qingqing01
force-pushed
the
rowwise_add
branch
from
01bc012 to
5c94725
Compare
chengduoZH
reviewed
Dec 27, 2017
| __global__ void RowwiseAddKernel(const T* a, const T* b, T* c, int64_t width, | ||
| int64_t num) { | ||
| T tmp = 1.0 / width; | ||
| for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < num; |
There was a problem hiding this comment.
It would be better to change the type of num to int. Otherwise, there is a comparison of int data and int64_t data.
qingqing01
force-pushed
the
rowwise_add
branch
from
58efe9b to
1936738
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Fix #6842
Mainly for the broadcast in Eigen. The time changes after optimization are as follows:
Experiments Env:
Total time of 2 epoc: