Use 2D grid for inner loop parallelization #260
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
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.
|
Performances HEAD #260 - this PR |
|
If we look to nvprof instead #260 - this PR what is difficult to understand is
|
|
The total time spent in the kernels decreases (both with #242 and #260), so something else must be taking longer:
For example, I do not know if the time spent in creating and launching the blocks is accounted, nor what happens to the overall GPU occupancy. |
|
Here is summary of the comparison of the time spent in the kernels, with the modified ones in bold:
It seems that |
|
we need to run regression on MC to verify that MTV is identical. |
|
Here are the improvements I see if I apply only the changes to on a pair of P100
on a pair of V100
|
|
MTV http://innocent.home.cern.ch/innocent/RelVal/pixOnlyPU50_gpuPR260/plots_summary.html
A: HEAD |
|
ok with I get from this PR and so it seems that , at least for this workflow on V100, 2D parallelization is not worth for Doublet Building with "stride=1" we reach "1420.1 ± 5.3 ev/s" with stride=8" in fishbone is "1426.3 ± 4.8 ev/s" |
This was referenced Jan 23, 2019
|
OK... does #261 look good ? |
|
#261 "looks" good , maybe I should test it.... |
|
#261 gives me "1419.6 ± 3.7 ev/s" equivalent to this PR and stride=1 in the doublet finder. |
|
This version should be equivalent to #261 with the advantage of having the ability to modify the 2D grid params for fishbone and doublet finder as well |
|
for completeness the nvprof report for this version |
|
@fwyzard , will you have the time to update your curves with the latest version of this PR? |
|
sure |
|
Updated plots on V100: Updated plots on P100: |
Validation summaryReference release CMSSW_10_4_0 at b8365c6
|
|
|
|
I'm running the validation on the P100 to avoid |
/RelValTTbar_13/CMSSW_10_4_0_pre3-PU25ns_103X_upgrade2018_realistic_v8-v1/GEN-SIM-DIGI-RAW summary
/RelValZMM_13/CMSSW_10_4_0_pre3-103X_upgrade2018_realistic_v8-v1/GEN-SIM-DIGI-RAW summary
|
fwyzard
pushed a commit
that referenced
this pull request
Oct 8, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Oct 20, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Oct 20, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Oct 23, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Nov 6, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Nov 6, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Nov 16, 2020
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.
fwyzard
added a commit
that referenced
this pull request
Nov 27, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Dec 26, 2020
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.
fwyzard
pushed a commit
that referenced
this pull request
Jan 15, 2021
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.
fwyzard
pushed a commit
that referenced
this pull request
Mar 23, 2021
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.
fwyzard
pushed a commit
that referenced
this pull request
Apr 1, 2021
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.
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.
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.
w/tr/ #242 here we keep the same number of threads per block as in the baseline
(this was already the case for the doublet finder)