New visibility function #34
Open
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
…iguration file are requested for LOFAR and MWA
…ue in the case BIPP is linked to OpenBLAS built with openmp support
AdhocMan
requested changes
Nov 7, 2024
| @@ -50,8 +50,7 @@ auto gram(Queue& q, std::size_t n, const T* x, const T* y, const T* z, T wl, api | |||
| const dim3 block(blockSizeX, blockSizeY, 1); | |||
| const auto grid = kernel_launch_grid(q.device_prop(), {n, n, 1}, block); | |||
|
|
|||
| api::launch_kernel(gram_kernel<T>, grid, block, 0, q.stream(), n, x, y, z, | |||
| T(2 * 3.14159265358979323846 / wl), g, ldg); | |||
| api::launch_kernel(gram_kernel<T>, grid, block, 0, q.stream(), n, x, y, z, wl, g, ldg); | |||
There was a problem hiding this comment.
Nice catch. Maybe make a small PR with this fix, to get this in right away?
There was a problem hiding this comment.
Done through Bugfix gram gpu #36
…ance issue in the case BIPP is linked to OpenBLAS built with openmp support" This reverts commit 334cb2d.
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Outdated suggestions cannot be applied.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.
This PR to align the 8000 main branch of BIPP with that one used for validating and benchmarking BIPP for the Astro & Computing paper. Apart from the new visibility function introduced in
measurement_set.pyother commits are the results of investigating inconsistencies between the original Python implementation of BIPP (i.e. Bluebild) and BIPP {SS, NUFF} x {CPU, GPU}, hence it's a bit disparate.On top of the slides I presented on the 31th May 2024 (ppt, New Visibility Function), I reran a series of tests, among which:
MWA
Results in DP fully equivalent (processing the full dataset).
#@#IFPE_TVIS (old) 4.202 sec down to (new) 0.893 sec => speedup 4.7
LEAP42rx
Results in DP fully equivalent (processing the first 200 epochs and all 20 channels available).
#@#IFPE_TVIS (old) 31.578 sec down to (new) 2.372 sec => speedup 13.3
OSKAR simulation 9 point sources, 1000 epochs, 4096 channels, 16 TB
Results in DP fully equivalent (processing the first 10 epochs and 64 channels (1000..1063) among the 4096 available).
#@#IFPE_TVIS (old) 589.281 sec down to (new) 45.842 sec => speedup 12.9
To enable the new visibility function switch from:
to
(calls to
filter_dataare not necessary anymore)The old and new visibility functions takes two additional arguments:
sort_timeboolean, set to True by default, to request sorting the MS entries over timelog_levelinteger, set to 0 (no logging) by default, to print some information.Note: processing the MWA dataset revealed a bug in my original implementation of considering the weights from WEIGHT_SPECTRUM when available in the MS. Mainly the sum of weights was overestimated if diagonal values are non zero. Observed differences are small, at the level of the second / third digit.
Example when processing channel 4 of the above mentioned MWA dataset:
New visibility function:
Old visibility function:
Would be good if @shreyamkrishna and @hiboubbouter could check the new visibility function on other datasets (ideally containing a WEIGHT_SPECTRUM column).