[Pallas] Fix ZeroDivisionError in block spec for int64 1D tensors#1950
Draft
[Pallas] Fix ZeroDivisionError in block spec for int64 1D tensors#1950
Conversation
meta-cla
bot
added
the
CLA Signed
The tiling factor formula `128 * (32 // bitwidth)` evaluates to 0 for dtypes wider than 32 bits (e.g. int64 where bitwidth=64) because `32 // 64 = 0` in integer division. This causes `bs % tiling_1d` to divide by zero in `_compute_block_spec_info`. Fix by changing the parenthesization to `128 * 32 // bitwidth` so the product is divided as a whole: 4096 // 64 = 64. Affected examples: cross_entropy, jagged_dense_add, jagged_mean, jagged_softmax, jagged_sum (all use int64 index/offset tensors).
norx1991
force-pushed
the
yifeixu/fix-int64-tiling-zerodiv
branch
from
1d68b6a to
a33f104
Compare
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
ZeroDivisionError: integer modulo by zeroin_compute_block_spec_infowhen processing 1D tensors with dtypes wider than 32 bits (e.g.int64)tiling_1d = 128 * (32 // bitwidth)evaluates to 0 forbitwidth=64because32 // 64 = 0in Python integer division. The modulo checkbs % tiling_1dthen divides by zero.128 * 32 // bitwidthso the product4096is divided as a whole (4096 // 64 = 64)cross_entropy,jagged_dense_add,jagged_mean,jagged_softmax,jagged_sum(all use int64 index/offset tensors)