#18332: Fix BN hang for FPU Kernel #18634
Merged
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
4 tasks
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
2 times, most recently
from
1b87bcf to
f08a501
Compare
VirdhatchaniKN
requested review from
ayerofieiev-tt,
dmakoviichuk-tt,
rfurko-tt,
cfjchu,
TT-BrianLiu,
razorback3,
dongjin-na,
bbradelTT,
omilyutin-tt,
yugaoTT,
tt-aho,
vsureshTT and
edwinleeTT
as code owners
umadevimcw
approved these changes
Mar 5, 2025
bbradelTT
approved these changes
Mar 5, 2025
mcw-anasuya
approved these changes
Mar 5, 2025
VirdhatchaniKN
force-pushed
the
virdhatchani/BN_hang
branch
2 times, most recently
from
9ff289e to
57c18ab
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
from
f08a501 to
4bc2169
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/BN_hang
branch
from
57c18ab to
75ec194
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
from
4bc2169 to
adfebf2
Compare
tt-aho
approved these changes
Mar 5, 2025
| tile_regs_commit(); | ||
|
|
||
| tile_regs_wait(); | ||
| pack_tile_with_dt(dst0, cb_affine_or_out); | ||
| pack_tile_with_dt(0, cb_affine_or_out); |
There was a problem hiding this comment.
Minor: Why was this change needed?
There was a problem hiding this comment.
dst0 is also 0. the value is same. Its just that for binary_dest_reuse_tiles I passed 0 and I followed the same here.
VirdhatchaniKN
force-pushed
the
virdhatchani/BN_hang
branch
from
75ec194 to
736407b
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
from
adfebf2 to
dbb139b
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
from
dbb139b to
c4d5a9c
Compare
VirdhatchaniKN
force-pushed
the
virdhatchani/fpu_bn_hang
branch
from
c4d5a9c to
dc3aed0
Compare
|
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.
Ticket
#18332
Problem description
Batch norm kernel hangs for larger shapes. This was because the buffer size is set to 2. The kernel reserves
freqnumber of tiles. So when this gets more than 2, there is an hang as it tries to reserve more than the size. Hence unpack and math threads got hung at thecopy_tilecall while the pack thread was stuck atcb_reserve_back.What's changed
freqloops in the kernel to ensure that as each tile gets reserved, they also get popped before the next tile gets reservedcb_numbuffer as I have rearranged the computationChecklist