Speed up the doublet finder #242
Closed
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
Introduce the inner loop parallelization in the doublet finder using the stride pattern already used in the "fishbone", and make use of a 2D grid instead of a hand-made stride.
|
I confirm that on volta hand made is definitively faster than the 2d grid https://github.com/VinInn/ctest/blob/master/cuda/combiXY.cu ) maybe we should investigate with Nvidia HM 2D |
|
is it possible the cause is the different number of blocks vs threads in the two cases ?
|
|
Updating the block size in ==411836== Profiling application: ./combiHM
==411836== Profiling result:
Type Time(%) Time Calls Avg Min Max Name
GPU activities: 47.85% 21.660ms 16 1.3537ms 1.3499ms 1.3588ms void nn<int=1>(unsigned int*, float const *, float const *, unsigned int*, int, float)
26.52% 12.007ms 16 750.41us 747.86us 753.27us void nn<int=2>(unsigned int*, float const *, float const *, unsigned int*, int, float)
13.71% 6.2061ms 16 387.88us 386.14us 389.82us void nn<int=4>(unsigned int*, float const *, float const *, unsigned int*, int, float)
7.33% 3.3163ms 16 207.27us 206.17us 208.54us void nn<int=8>(unsigned int*, float const *, float const *, unsigned int*, int, float)
4.21% 1.9058ms 16 119.11us 118.40us 121.15us void nn<int=16>(unsigned int*, float const *, float const *, unsigned int*, int, float)==411864== Profiling application: ./combiXY
==411864== Profiling result:
Type Time(%) Time Calls Avg Min Max Name
GPU activities: 46.86% 23.330ms 16 1.4581ms 1.4540ms 1.4618ms void nn<int=1>(unsigned int*, float const *, float const *, unsigned int*, int, float)
24.62% 12.255ms 16 765.95us 765.46us 766.81us void nn<int=2>(unsigned int*, float const *, float const *, unsigned int*, int, float)
14.22% 7.0786ms 16 442.41us 441.72us 443.16us void nn<int=4>(unsigned int*, float const *, float const *, unsigned int*, int, float)
8.77% 4.3651ms 16 272.82us 272.57us 273.69us void nn<int=8>(unsigned int*, float const *, float const *, unsigned int*, int, float)
5.19% 2.5830ms 16 161.44us 159.26us 163.13us void nn<int=16>(unsigned int*, float const *, float const *, unsigned int*, int, float)==411890== Profiling application: ./combiXY2
==411890== Profiling result:
Type Time(%) Time Calls Avg Min Max Name
GPU activities: 49.75% 23.314ms 16 1.4571ms 1.4538ms 1.4621ms void nn<int=1>(unsigned int*, float const *, float const *, unsigned int*, int, float)
25.47% 11.936ms 16 745.98us 743.38us 748.34us void nn<int=2>(unsigned int*, float const *, float const *, unsigned int*, int, float)
13.26% 6.2147ms 16 388.42us 386.97us 389.69us void nn<int=4>(unsigned int*, float const *, float const *, unsigned int*, int, float)
7.08% 3.3200ms 16 207.50us 206.56us 208.51us void nn<int=8>(unsigned int*, float const *, float const *, unsigned int*, int, float)
4.07% 1.9070ms 16 119.19us 118.62us 120.99us void nn<int=16>(unsigned int*, float const *, float const *, unsigned int*, int, float) |
|
ok, i think you are fully right. |
|
I updated the simple test and added more nt,nb combinations (of course the test is small, so there is no register pressure) |
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.
Second part of @VinInn 's #238.
Introduce the inner loop parallelization in the doublet finder using the stride pattern already used in the "fishbone", and make use of a 2D grid instead of a hand-made stride.