[pull] master from tensorflow:master#1676
Merged
Merged
Conversation
…nate dimension PiperOrigin-RevId: 891685545
The XLA FFI registration explicitly requires an `execute` handler to be present in the `XLA_FFI_Handler_Bundle` for all registered FFI targets, even if the target is exclusively used for its `instantiate` stage (like `xla.gpu.ext.cub_scan_unassigned_scratch_size` which computes the scratch buffer size). If `execute` is `nullptr`, `xla::ffi::RegisterHandler` fails validation and returns an `InvalidArgument` error. However, the `XLA_FFI_REGISTER_HANDLER` macro evaluates this registration inside a static initializer and assigns the returned `XLA_FFI_Error*` to an unused static variable. This causes the error to be silently swallowed without crashing or logging. As a result, without the dummy execute handler, the FFI handler is silently omitted from the registry. This leads to a `NOT_FOUND` error later during compilation (e.g., in the `EstimateCubScratchSize` pass) when XLA tries to resolve the handler to determine the necessary scratch size. PiperOrigin-RevId: 891689079
…PING_FOR_DEBUG`. That name is misleading. - It may look like it's needed on Windows because there's no `mmap` call. That's not the case. On Windows that flags is extremely harmful and slows things down to a halt when building the cache. - It makes it look like it can be used in production (apparently `*_FOR_TEST` is not explicit enough). PiperOrigin-RevId: 891701455
PiperOrigin-RevId: 891701526
PiperOrigin-RevId: 891703723
Partial migration of IndexingMap::GetAffineMap. Still some users are there to be migrated, but I do not want to make a giant CL. The changes include: - Updating call sites in XLA GPU backend emitters (reduction, scatter, transpose) - Updating HWIR's hlo_expansion.cc to use SymbolicMap. - Updating various files in /emitters... PiperOrigin-RevId: 891709384
- Create module from HLO text instead of Builder. - Check output with FileCheck. PiperOrigin-RevId: 891727583
…pile. PiperOrigin-RevId: 891729474
PiperOrigin-RevId: 891735440
I'm about to refactor the test and need a way to validate that the configs are not changed by the refactor. PiperOrigin-RevId: 891758501
Add the tag optonly, it does not pass without -c opt on some internal backends. PiperOrigin-RevId: 891761700
…A:TPU:SC compiler. PiperOrigin-RevId: 891763488
Also set shardings on shardy outliner also when func has no result shardings but the outsharding of the namedcomputation has. PiperOrigin-RevId: 891783867
Imported from GitHub PR openxla/xla#40017 The previous progress tracking design pre-allocated one `se::Event` per thunk and overwrote its timestamp on each execution. This meant that for thunks inside while loops, only the *last* iteration's state was visible -- earlier iterations were silently lost. When diagnosing deadlocks inside loops, the tracker could not tell which iteration hung or show the full execution history. This PR redesigns the progress tracker to append a new event record for every thunk execution, preserving the complete chronological history including loop iteration state. **Test coverage**: Extended `thunk_executor_test` to verify double-execution accumulation (2x event count, correct timestamps) and while loop nest tracking (trip count, loop depth, iteration number recorded per body thunk execution). Copybara import of the project: -- e2abfa8281c0a8f05f2d45caff3f9525bef4638a by Eugene Zhulenev <ezhulenev@openxla.org>: [xla:gpu] Correctly track thunk progress in presence of loops Merging this change closes #40017 PiperOrigin-RevId: 891790703
PiperOrigin-RevId: 891793109
PiperOrigin-RevId: 891803849
- Pass `Config` structs instead of a bunch of function arguments. - Make helper function for launching clients. - Use gmatchers to avoid boilerplate. PiperOrigin-RevId: 891806662
PiperOrigin-RevId: 891814918
PiperOrigin-RevId: 891819181
PiperOrigin-RevId: 891829849
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.
13 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 : )