[pull] master from tensorflow:master - #8788
Merged
Merged
Conversation
PiperOrigin-RevId: 972507321
This pass removes unnecessary size-1 from the module. Degenerate dimensions are generally no-op, but it add unnecessary reshapes/bitcasts to the graph that can sometime prevent better fusion and tiling decision or cause problem with emitter pipelines, like Triton. Reverts 327ef3d PiperOrigin-RevId: 972515964
… kernel thunks. Create a symmetric memory for scratch buffers and use it to get a peer address of memory chunk instead of running separate rendzevous. PiperOrigin-RevId: 972518088
Imported from GitHub PR openxla/xla#47701 **NFC:** splitting functionality into separate library Extract `collective_rendezvous` from `collective_ops_utils` as it's really a runtime support and node used for collective ops analysis/optimization. Copybara import of the project: -- 6b91ae42a6453e1adcac7e7835a55179db1b9dd5 by Eugene Zhulenev <ezhulenev@openxla.org>: [xla] Extract a separate collective_rendezvous library Merging this change closes #47701 PiperOrigin-RevId: 972518994
… threshold is exceeded. Iterate by reference in the earlier collectives loop so reassigning earlier_coll updates the vector entry instead of being a local no-op. PiperOrigin-RevId: 972519816
Msan support is currently incomplete. For example, we don't add
the appropriate msan pass to LLVM's pipeline; instead, we manually
call __msan_unpoison here and there. In AOT (tfcompile) the situation
is even worse, since we only add these annotations to some of the
emitted code.
This change adds complete support for msan (and msan-track-origins)
to XLA:CPU in both JIT and AOT modes. In particular:
- Adds a single boolean config flag to determine whether msan should
be enabled in compiled code. This is then wired up through the
compiler, removing several #ifdef MEMORY_SANITIZER (or equivalent)
checks that we had scattered around the code base. (OK, this is
two knobs because we also support track-origins.)
- Adds the msan pass to LLVM as documented in
https://github.com/google/sanitizers/wiki/MemorySanitizerJIT
This requires a small TLS emulation library, which (1) in JIT
gives access to host symbols from JIT'ed code, and (2) in AOT
(tfcompile) must be linked in the final binary so that the
compiled code can reliably access msan symbols.
What's missing from this CL:
1. Removal of now obsolete annotations
2. Setting of the appropriate flags in tfcompile to enable the
new path in AOT.
These two changes will land once users have had time to recompile
their compilers (in case they embed tfcompile into them), about
1-2 weeks.
Note: the msan support implemented in this CL supersedes cl/959445928, i.e.
openxla/xla@70f2607611. That means the
xla_backend_extra option added there for msan on AOT will be removed when
landing the two missing pieces from this CL.
Reverts ef7547c
PiperOrigin-RevId: 972614006
PiperOrigin-RevId: 972616295
…s for which "keep_trivial_sync_annotation" returns true. Delete the redundant code block which only checks groups with a single op. PiperOrigin-RevId: 972643528
…eduction sizes. Using isolated log bounds (such as [-85, 85] for F32) led to downstream +inf failures in reduction fusions and expressions like x * exp(x). This change: - Precomputes downstream addition-reduction element counts per kExp in reverse post-order. - Scales kExp bounds using base thresholds per data type (e.g. 4.0 for F32, 2.5 for BF16/F16), dynamically tightening when N > 1. - Adds a small positive floor (0.1) to avoid negative intervals and preserve positive test coverage around zero. PiperOrigin-RevId: 972671223
Keep both legacy and new autotune cache formats in tests. PiperOrigin-RevId: 972671341
CommonPjRtClient using InferDispatchInfo. PiperOrigin-RevId: 972683293
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
11 participants
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.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )