PR #23619: Fix the EmitSort checking after enabling NVLS and user buffer #23935
+55
−3
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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
copybara-service
bot
force-pushed
the
test_738419324
branch
2 times, most recently
from
d24947d to
f5741c4
Compare
Imported from GitHub PR #23619 There is a reported bug from NVIDIA that running Midjourney model triggers XLA error after enabling NVLS and user buffer by setting NCCL_NVLS_ENABLE=1 and --xla_gpu_enable_nccl_user_buffers=true: ``` jaxlib.xla_extension.XlaRuntimeError: INTERNAL: RET_CHECK failure (external/xla/xla/service/gpu/ir_emitter_unnested.cc:1676) LayoutUtil::LayoutsInShapesEqual(keys_shape, sort->operand(i)->shape()). ``` This is because after enabling NVLS and user buffer, one of the operand of `sort` operation is from a different memory space (user buffer), and the previous `LayoutsInShapesEqual` check is too strong to pass as it also checks if operands are from the same memory space. This MR makes the sort layout check weaker as operands do not have to be in the same memory space as long as they all on the device. Copybara import of the project: -- 8265352 by Chenhao Jiang <[email protected]>: Making the sort layout check weaker as operands do not have to be in the same memory space as long as they all on the device. Merging this change closes #23619 FUTURE_COPYBARA_INTEGRATE_REVIEW=#23619 from serach24:chenhao/fix_nvsl_check_failed 8265352 PiperOrigin-RevId: 738419324
copybara-service
bot
force-pushed
the
test_738419324
branch
from
f5741c4 to
6d4f1d5
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Outdated suggestions cannot be applied.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.
PR #23619: Fix the EmitSort checking after enabling NVLS and user buffer
Imported from GitHub PR #23619
There is a reported bug from NVIDIA that running Midjourney model triggers XLA error after enabling NVLS and user buffer by setting NCCL_NVLS_ENABLE=1 and --xla_gpu_enable_nccl_user_buffers=true:
This is because after enabling NVLS and user buffer, one of the operand of
sortoperation is from a different memory space (user buffer), and the previousLayoutsInShapesEqualcheck is too strong to pass as it also checks if operands are from the same memory space.This MR makes the sort layout check weaker as operands do not have to be in the same memory space as long as they all on the device.
Copybara import of the project:
--
8265352 by Chenhao Jiang [email protected]:
Making the sort layout check weaker as operands do not have to be in the same memory space as long as they all on the device.
Merging this change closes #23619
FUTURE_COPYBARA_INTEGRATE_REVIEW=#23619 from serach24:chenhao/fix_nvsl_check_failed 8265352