Skip to content

[FIX] Gradlib OOM on Navi and sometimes on MI #124

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

Merged
merged 3 commits into from
Aug 8, 2024
Merged

[FIX] Gradlib OOM on Navi and sometimes on MI #124

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9f17ee2
add memory clean up after every shape and parameter to reduce cache i…
Aug 7, 2024
5112161
small typo
Aug 7, 2024
00f0c1c
syntax change
Aug 7, 2024
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
8 changes: 7 additions & 1 deletion gradlib/gradlib/GemmTuner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import os
import random
from pathlib import Path

Expand All @@ -13,6 +14,8 @@
rtol = 1e-5
atol = 1

CACHE_INVALIDATE_BUFFERS = int(os.getenv("CACHE_INVALIDATE_BUFFERS", "37"))


class Gemm:

Expand All @@ -24,7 +27,7 @@ def __init__(self, m, n, k, indtype, outdtype, rocblas_decode=False):
self.outdtype = outdtype
self.use_rocblas = (indtype == outdtype
and indtype is not torch.float8_e4m3fnuz)
self.nb = 37
self.nb = CACHE_INVALIDATE_BUFFERS
self.inp = torch.randn((self.n, self.k),
device='cuda').to(self.indtype)
self.weights = torch.randn((self.m, self.k),
Expand Down Expand Up @@ -283,6 +286,9 @@ def find_best_sols(self):
soldf.loc[i, 'libtype'] = gemmobj.best_libtype
soldf.loc[i, 'solidx'] = gemmobj.best_solidx
soldf.loc[i, 'soltimems'] = gemmobj.best_soltime
del gemmobj
torch.cuda.empty_cache()

soldf['indtype'] = self.indtype
soldf['outdtype'] = self.outdtype
finaldf = pd.concat([self.gemm_problems, soldf], axis=1)
Expand Down