-
Notifications
You must be signed in to change notification settings - Fork 1.1k
[backport][rls-v3.8] cpu: x64: matmul: fix blocking heuristics for l2 set issues #3436
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
5f58eb0
to
2f34b95
Compare
make test |
2f34b95
to
15d3fc2
Compare
// The following consts are correct for all platforms supporting AMX | ||
constexpr size_t l2_ways = 16; | ||
constexpr size_t l2_ways_threshold = size_t(l2_ways * 0.75); | ||
constexpr size_t l2_sets = 2048; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Isn't number_of_sets = cache_size / number_ways
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fixed implemented in new function in platform.cpp:
uint32_t get_num_ways_in_cache(int level)
uint32_t get_num_sets_in_cache(int level)
and used them in heuristics.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
My bad I missed it was a backport. I guess you could just backport the original changes from main in this PR, and put these extra changes (query cache ways in xbyak_utils) to main in another PR.
15d3fc2
to
ad81169
Compare
ad81169
to
5169c5b
Compare
make test |
fix performance degradation related to issue:
https://jira.devtools.intel.com/browse/MFDNN-13324
backport original pull request:
#3403