Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

use Intel OpenMP to speedup seq2batch when WITH_MKL #6622

Merged
merged 2 commits into from
Dec 16, 2017
Merged

use Intel OpenMP to speedup seq2batch when WITH_MKL #6622

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
84cb542
use intel openmp to speedup seq2batch when WITH_MKL
tensor-tang Dec 15, 2017
9c27c13
follow comments using macro to separate the original implements
tensor-tang Dec 16, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 22 additions & 3 deletions paddle/gserver/layers/SequenceToBatch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,12 +171,31 @@ void SequenceToBatch::sequence2BatchCopy(Matrix &batch,
hl_sequence2batch_copy(
batchData, seqData, idxData, seqWidth, batchCount, seq2batch);
} else {
for (int i = 0; i < batchCount; ++i) {
if (seq2batch) {
if (seq2batch) {
#ifdef PADDLE_USE_MKLML
const int blockMemSize = 8 * 1024;
const int blockSize = blockMemSize / sizeof(real);
#pragma omp parallel for collapse(2)
for (int i = 0; i < batchCount; ++i) {
for (int j = 0; j < seqWidth; j += blockSize) {
memcpy(batch.rowBuf(i) + j,
sequence.rowBuf(idxData[i]) + j,
(j + blockSize > seqWidth) ? (seqWidth - j) * sizeof(real)
: blockMemSize);
}
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里能保留原来代码在么？
openblas的情况下，是不是原来的memcpy更快速。现在多了一层循环了。

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

哦好的，那我保留原来的区分下好。thx

#else
for (int i = 0; i < batchCount; ++i) {
memcpy(batch.rowBuf(i),
sequence.rowBuf(idxData[i]),
seqWidth * sizeof(real));
} else {
}
#endif
} else {
#ifdef PADDLE_USE_MKLML
#pragma omp parallel for
#endif
for (int i = 0; i < batchCount; ++i) {
memcpy(sequence.rowBuf(idxData[i]),
batch.rowBuf(i),
seqWidth * sizeof(real));
Expand Down