[STF] Improved cache mechanism for executable CUDA graphs #3768
+251
−63
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
caugonnet
commented
Feb 11, 2025
| fprintf(stderr, "Reclaimed %s in cache graph (asked %s remaining %s)\n", pretty_print_bytes(reclaimed).c_str(), pretty_print_bytes(to_reclaim).c_str(), pretty_print_bytes(total_cache_footprint).c_str()); | ||
| } | ||
|
|
||
| // TODO we should not have to redefine this one again |
There was a problem hiding this comment.
this should rely on existing ::std::hash<pair<...>> unless we have a cyclic dep
caugonnet
commented
Feb 11, 2025
| } | ||
| }; | ||
|
|
||
| // TODO per device ! |
There was a problem hiding this comment.
or have a cache per device in the async resource
caugonnet
commented
Feb 11, 2025
| last_use = cache->index; | ||
| } | ||
|
|
||
| void update() { |
caugonnet
commented
Feb 11, 2025
|
|
||
| executable_graph_cache *cache; | ||
| ::std::shared_ptr<cudaGraphExec_t> exec_g; | ||
| size_t use_cnt; |
There was a problem hiding this comment.
remove, and / or replace by cache stats
caugonnet
commented
Feb 11, 2025
| // Check if there is a matching entry (and update it if necessary) | ||
| ::std::pair<bool, ::std::shared_ptr<cudaGraphExec_t>> query(size_t nnodes, size_t nedges, ::std::shared_ptr<cudaGraph_t> g) | ||
| { | ||
| auto range = cached_graphs.equal_range({nnodes, nedges}); |
There was a problem hiding this comment.
TODO reorder by oldest ?
caugonnet
commented
Feb 12, 2025
| executable_graph_cache() | ||
| { | ||
| cache_size_limit = 512 * 1024 * 1024; | ||
| const char* str = getenv("CUDASTF_GRAPH_CACHE_SIZE_MB"); |
There was a problem hiding this comment.
Define what happens for MB = 0 (infinite ?) and perhaps we'd better make this the default ...
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.
Description
CUDA graph consumes around 10KB per kernel node, which quickly adds up when we have large CUDA graphs kept in cache. This implements a mechanism to cap resources due to cached CUDA graphs.
closes
Checklist