Enable compile-time optimisation for prebuilt wheels #19
Conversation
This file contains hidden or 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
02895e4 to
ab89607
Compare
achaikou
approved these changes
May 13, 2025
Compile-time optimisation is crucial for good performance in si4ti,
most likely because of Eigen. In tests on a x86_64 platform (Intel Xeon
Platinum 8462Y+ using 8 cores), careful choise of compile-time optimisation
could decrease the runtime by about 40%.
Number of runs per case was 5. We report the lowest, average, and median
runtime recorded in seconds. The data set is a sample data set from the
domain expert (2 files, one baseline and one monitor cube). The input
data set has the following dimensions 361 x 221 x 376 (number of rows x
number of columns x number of layers) as xtgeo.Cube.
| Target | Min | Avg | Median |
| --------- | ----- | ----- | ------ |
| generic | 63.97 | 64.06 | 64.03 |
| x86-64-v3 | 39.44 | 39.95 | 40.06 |
| native | 51.04 | 51.40 | 51.36 |
The table shows that the target `x86-64-v3` [1] showed a reduction of
runtime ~40% (from 60s to 38s wallclock time). Enabling newer
archtitectures or `-march=native` did not show improved performance
during our tests.
The `compute_impedance` function was called with the following parameters:
_a, _b = si4ti.compute_impedance(
cubes,
segments = 1,
max_iter = 100,
damping_3D = 0.001,
damping_4D = 0.001,
latsmooth_3D = 0.01,
latsmooth_4D = 0.05,
)
Enabling compile-time optimisations impacts portability of the wheels
as the wheels only work on devices with support for the used instruction
set. This should still be true for all reasonably new CPUs as as
`x86-64-v3` refers to common CPUs with AVX2 support such as Intel
Haswell (released 2013) and AMD Excavator (released 2015) and newer "big
cores" [2]. It is expected that people running the si4ti application are
on powerful, reasonably new machines, and appreciate the performance
gains. Mac users on arm64 Macs should be unaffected since we do not turn
on any additional vectorisation features besides the ones that are
activated by default.
[1]: https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html
[2]: https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels
Tagging with the new version allows us and users to specifically install the bindings with vectorisation enabled. This helps with testing. A new shall be published after merging this change.
cd0a410 to
0a614e1
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.
Vectorisation is crucial for good performance in si4ti as Eigen relies heavily on vectorisation. In tests on a x86_64 platform (Intel Xeon Platinum 8462Y+ using 8 cores), enabling target specific optimisation
x86-64-v31 showed a reduction of runtime ~40% (from 60s to 38s wallclock time). Enabling newer archtitectures or-march=nativedid not show improved performance during our tests.This impacts usability of the wheels as they only work on devices with support for the used instruction set. This should still be true for all reasonably new CPUs as as
x86-64-v3refers to common CPUs with AVX2 such as Intel Haswell (released 2013) and AMD Excavator (released 2015) and newer "big cores" 2. It is expected that people running the si4ti application are on powerful, reasonably new machines, and appreciate the performance gains. Mac users on arm64 Macs should be unaffected since do not turn any vectorisation features on besides the ones that are activated by default.