Skip to content
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.

Sign up for GitHub

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

Jump to bottom

Expanding the scope of 2D thread distribution to improve multi-threaded DGEMM performance #4655

Merged
merged 1 commit into from
May 14, 2024
Merged

Expanding the scope of 2D thread distribution to improve multi-threaded DGEMM performance #4655

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions driver/level3/level3_thread.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -826,6 +826,16 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IF
if (nthreads_m * nthreads_n > args -> nthreads) {
nthreads_n = blas_quickdivide(args -> nthreads, nthreads_m);
}
/* The nthreads_m and nthreads_n are adjusted so that the submatrix */
/* to be handled by each thread preferably becomes a square matrix */
/* by minimizing an objective function 'n * nthreads_m + m * nthreads_n'. */
/* Objective function come from sum of partitions in m and n. */
/* (n / nthreads_n) + (m / nthreads_m) */
/* = (n * nthreads_m + m * nthreads_n) / (nthreads_n * nthreads_m) */
while (nthreads_m % 2 == 0 && n * nthreads_m + m * nthreads_n > n * (nthreads_m / 2) + m * (nthreads_n * 2)) {
nthreads_m /= 2;
nthreads_n *= 2;
}
}

/* Execute serial or parallel computation */
Expand Down
Loading