Loop Blocking for fn GPU Backend #1787
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
|
Hi there, this is jenkins continuous integration... |
|
launch jenkins |
|
launch perftests |
|
launch jenkins |
1 similar comment
|
launch jenkins |
|
launch jenkins |
|
launch perftest |
|
launch perftest |
|
launch perftest |
havogt
requested changes
Sep 25, 2024
| return index_at_dim<I>(blockIdx) * (ThreadBlockSize::value * LoopBlockSize::value) + | ||
| index_at_dim<I>(threadIdx) * LoopBlockSize::value; | ||
| } else { | ||
| return integral_constant<int, 0>(); |
There was a problem hiding this comment.
I am a bit lost here, maybe you can add a few comments.
There was a problem hiding this comment.
Added some at various places, let me know if that’s enough.
|
launch perftest |
|
launch perftest |
|
launch jenkins |
|
launch perftest |
|
launch perftest |
|
launch jenkins |
|
launch jenkins |
|
launch perftest |
havogt
pushed a commit
that referenced
this pull request
Oct 30, 2024
Implements loop blocking for the GPU fn backend. Thread block size (that is, CUDA/HIP threads per block) and loop block size (that is, loop iterations per CUDA/HIP thread) can now be specified as template parameters. Further changes: - Set `__launch_bounds__` in the fn GPU kernel based on the thread block size. - Activate vertical loop blocking in the fn nabla kernels on newer CUDA versions that support `GT_PROMISE`. Performance changes: - `__launch_bounds__` affects performance of the `fn_cartesian_vertical_advection` benchmark significantly (positively or negatively, depending on domain size). - Performance of fn nabla benchmarks improves significantly on newer CUDA versions. - Performance on Daint is currently reduced due to too old CUDA version.
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.
Implements loop blocking for the GPU fn backend. Thread block size (that is, CUDA/HIP threads per block) and loop block size (that is, loop iterations per CUDA/HIP thread) can now be specified as template parameters.
Further changes:
__launch_bounds__in the fn GPU kernel based on the thread block size.GT_PROMISE.Performance changes:
__launch_bounds__affects performance of thefn_cartesian_vertical_advectionbenchmark significantly (positively or negatively, depending on domain size).