support gpu send/recv thunk #1
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
wbmc
pushed a commit
that referenced
this pull request
Dec 5, 2023
Imported from GitHub PR openxla#6599 FP8 cublasLt matmul uses fast accumulation when both operands' precision are DEFAULT. Otherwise fall back to high precision acuumulation. Issue#openxla#6168 This PR is closely related to Flax PR-. Copybara import of the project: -- a4140da by shuw <[email protected]>: Add FP8 fast accumulation support for cublasLt. -- 9684568 by shuw <[email protected]>: Improve based on review #1 -- e906d76 by shuw <[email protected]>: Improve based on review #2 Merging this change closes openxla#6599 COPYBARA_INTEGRATE_REVIEW=openxla#6599 from wenscarl:fp8_fast_accumulation e906d76 PiperOrigin-RevId: 578948593
mars1248
pushed a commit
to mars1248/xla
that referenced
this pull request
Dec 22, 2023
Imported from GitHub PR openxla#7751 Due to fast accumulation being turned on in the forward mode, the cublasLt fp8 gemm with gelu epilogue can efficiently operate with a fused kernel. Compared against the XLA-generated gelu kernel on H100, the performance demonstrates some improvement for size of [8192, 4096] x [4096, 16384] + gelu: Execution time for matmul using cublasLt and gelu (XLA): 1.28ms Execution time for matmul_gelu using cublasLt: 1.25ms Copybara import of the project: -- e8abce3 by Shu Wang <[email protected]>: Support cublasLt Fp8 Approx Gelu epilogue fusion. -- 818127c by shuw <[email protected]>: Remove F32 check -- 5ce3108 by shuw <[email protected]>: Improve based on review intelligent-machine-learning#1 Merging this change closes openxla#7751 COPYBARA_INTEGRATE_REVIEW=openxla#7751 from wenscarl:cublaslt_fp8_gelu 5ce3108 PiperOrigin-RevId: 591236441
ApsarasX
pushed a commit
that referenced
this pull request
Mar 28, 2024
…execution scope Instead of always constructing while operation conditional in the default scope use the scope of a while operation itself. This generates correct CUDA graph: https://gist.github.com/ezhulenev/a84192fe8b46a4bf1a934a8baa08ea60 Memeset operation launched in a scope #1 is not synchronized with initial condition handle update PiperOrigin-RevId: 609742672
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.
No description provided.