llvm/{cuda,}: Reduce and optimize data copies in model execution #2311
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
…e's no modulation The compiler is not able to eliminate these. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
…omposition functions Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
This has not been used since GPU accelerated evaluations were switched to range_evaluate. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
A 4B aligned version performs better on GPUs. A llvm.memcopy fallback is available via PNL_LLVM_DEBUG=unaligned_copy. Both variants much more register efficient (and usually faster) than store(load(...)). Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Move all shared params for GPU grid evaluate. Use PNL_LLVM_DEBUG='cuda_no_shared' to opt out. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Prefer large blocks if the kernel uses shared memory. Prefer warp sized blocks otherwise. Print kenel statistics and selected block size with PNL_LLVM_DEBUG=stat Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Print upload/download statistics when 'stat' switch is present. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
jvesely
added
compiler
|
This PR causes the following changes to the html docs (ubuntu-latest-3.7-x64): See CI logs for the full diff. |
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Outdated suggestions cannot be applied.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.
Do not create a copy of parameters if there are no parameter states.
Add memcopy helper to W/A heavy register usage by store(load(src), dst) idiom.
Upload base arguments to on-chip 'shared' memory for GPU evaluate (optional, default ON).
Add human-readable names to alloca instructions.
Drop unused 'evaluate' kernel wrapper code.
Drop 'cuda_data' debug switch.
Improves performance ~2.5x for both CPU and GPU execution (measured using predator-prey model with 101 levels of attention)