perf: Remove x87 elliptic integral in solenoid B-field #1469
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
As it stands, the solenoid B-field code uses the `boost::math::ellint_1` and `boost::math::ellint_2` methods. For reasons which I do not understand, these functions use 80-bit x87 floating point numbers, which leads to a lot of conversion between x87 and SSE floating points. It also has the downside that x87 operations are generally slower then their SSE counterparts on modern CPUs. The interesting thing is that Boost's method for deciding on the floating point precision seems to be quite arbitrary; even when the functions are explicitly templated to use 64-bit `double` types, the internals of these functions still use 80-bit floating points. The best solution I have found is to use a compiler preprocessor flag, `BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS`, which tells Boost not to use x87 floating points at all. Enabling this flag only in the solenoid B-field translation unit should be safe and improve performance. Very unscientifically benchmarked on my Intel i5-7300U CPU, the solenoid B-field benchmark with the old code takes 615.7±7.3µs per iteration, while the new code takes 353.7±7.0µs, a speedup of about 75%.
stephenswat
added
Component - Core
stephenswat
changed the title
Codecov Report
@@ Coverage Diff @@
## main #1469 +/- ##
=======================================
Coverage 48.59% 48.59%
=======================================
Files 380 380
Lines 20589 20589
Branches 9431 9431
=======================================
Hits 10005 10005
Misses 4097 4097
Partials 6487 6487
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
|
Seeing this only now, I'm guessing you checked that the solenoid field output is approximately compatible @stephenswat? (not identical of course) |
paulgessinger
added
the
backport develop/v19.x
|
The backport to To backport manually, run these commands in your terminal: # Fetch latest updates from GitHub
git fetch
# Create a new working tree
git worktree add .worktrees/backport-develop/v19.x develop/v19.x
# Navigate to the new working tree
cd .worktrees/backport-develop/v19.x
# Create a new branch
git switch --create backport-1469-to-develop/v19.x
# Cherry-pick the merged commit of this pull request and resolve the conflicts
git cherry-pick -x --mainline 1 3795148974ec2682f84218b7f8c14d1f3be121c1
# Push it to GitHub
git push --set-upstream origin backport-1469-to-develop/v19.x
# Go back to the original working tree
cd ../..
# Delete the working tree
git worktree remove .worktrees/backport-develop/v19.xThen, create a pull request where the |
paulgessinger
pushed a commit
to paulgessinger/acts
that referenced
this pull request
Sep 2, 2022
…1469) As it stands, the solenoid B-field code uses the `boost::math::ellint_1` and `boost::math::ellint_2` methods. For reasons which I do not understand, these functions use 80-bit x87 floating point numbers, which leads to a lot of conversion between x87 and SSE floating points. It also has the downside that x87 operations are generally slower then their SSE counterparts on modern CPUs. The interesting thing is that Boost's method for deciding on the floating point precision seems to be quite arbitrary; even when the functions are explicitly templated to use 64-bit `double` types, the internals of these functions still use 80-bit floating points. The best solution I have found is to use a compiler preprocessor flag, `BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS`, which tells Boost not to use x87 floating points at all. Enabling this flag only in the solenoid B-field translation unit should be safe and improve performance. Very unscientifically benchmarked on my Intel i5-7300U CPU, the solenoid B-field benchmark with the old code takes 615.7±7.3µs per iteration, while the new code takes 353.7±7.0µs, a speedup of about 75%.
kodiakhq bot
pushed a commit
that referenced
this pull request
Sep 22, 2022
…to develop/v19.x] (#1487) Backport 3795148 from #1469 --- As it stands, the solenoid B-field code uses the `boost::math::ellint_1` and `boost::math::ellint_2` methods. For reasons which I do not understand, these functions use 80-bit x87 floating point numbers, which leads to a lot of conversion between x87 and SSE floating points. It also has the downside that x87 operations are generally slower then their SSE counterparts on modern CPUs. The interesting thing is that Boost's method for deciding on the floating point precision seems to be quite arbitrary; even when the functions are explicitly templated to use 64-bit `double` types, the internals of these functions still use 80-bit floating points. The best solution I have found is to use a compiler preprocessor flag, `BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS`, which tells Boost not to use x87 floating points at all. Enabling this flag only in the solenoid B-field translation unit should be safe and improve performance. Very unscientifically benchmarked on my Intel i5-7300U CPU, the solenoid B-field benchmark with the old code takes 615.7±7.3µs per iteration, while the new code takes 353.7±7.0µs, a speedup of about 75%.
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.
As it stands, the solenoid B-field code uses the
boost::math::ellint_1andboost::math::ellint_2methods. For reasons which I do not understand, these functions use 80-bit x87 floating point numbers, which leads to a lot of conversion between x87 and SSE floating points. It also has the downside that x87 operations are generally slower then their SSE counterparts on modern CPUs.The interesting thing is that Boost's method for deciding on the floating point precision seems to be quite arbitrary; even when the
functions are explicitly templated to use 64-bit
doubletypes, the internals of these functions still use 80-bit floating points.The best solution I have found is to use a compiler preprocessor flag,
BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS, which tells Boost not to use x87 floating points at all. Enabling this flag only in the solenoid B-field translation unit should be safe and improve performance. Very unscientifically benchmarked on my Intel i5-7300U CPU, the solenoid B-field benchmark with the old code takes 615.7±7.3µs per iteration, while the new code takes 353.7±7.0µs, a speedup of about 75%.