[pull] master from tensorflow:master#1986
Merged
Merged
Conversation
It can happen that device assignment is just not very useful because of dummy data population by functional_hlo_runner. In such a case we should not rely on device assignments. PiperOrigin-RevId: 940591089
…ctions follow up to recent revert to hopefully prevent similar issues in the future + document multi output fusion bug PiperOrigin-RevId: 940597590
Remove TpuPlatform and TpuExecutor. PiperOrigin-RevId: 940600001
PiperOrigin-RevId: 940603017
…l-decode PiperOrigin-RevId: 940604157
PiperOrigin-RevId: 940606748
…dding checks for empty pattern lists and validating the index upon checkpoint restoration. PiperOrigin-RevId: 940617062
PiperOrigin-RevId: 940623404
…ag mappings Imported from GitHub PR openxla/xla#44823 This PR makes legacy non platforms apis not be forced down to downstream users. XLA now has toolchains, platforms and constraints counterparts for all of its --cpu, --crosstool_top, --compiler usage. This means that, using `platform_mapping`, we can now make non legacy platforms config settings and selects the default, and let XLA continue to use legacy flags while still being compatible (not sure if internal google usage forces keeping those but they could eventually go away too). This is mostly an attempt to see potential breakage downstream and will likely require some work on tensorflow and JAX before even knowing if internal checks accept this. This would be a good quality of life improvements for downstream users using legacy flags in selects for instance can create subtle bugs quite hard to diagnose, especially with selects and `//conditions:defaults` being evaluated if downstream users don't have platform_mappings themselves. Copybara import of the project: -- dd761c0d0783d19ed31b41cfd4db3901c862f91b by Corentin Kerisit <corentin.kerisit@gmail.com>: Make XLA CPU selects platform-based with legacy flag mappings Merging this change closes #44823 PiperOrigin-RevId: 940623405
In HandleAsyncStart, the output value set for the async-start instruction should be the union of the existing value set and the value set of the operand. The previous logic was overwriting the output value set instead of taking a union. PiperOrigin-RevId: 940666442
…lver. PiperOrigin-RevId: 940712587
PiperOrigin-RevId: 940717946
PiperOrigin-RevId: 940739760
Updates LLVM usage to match [f5faf2902744](llvm/llvm-project@f5faf2902744) PiperOrigin-RevId: 940766490
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.
10 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 : )