[pull] master from tensorflow:master#1688

Merged

pull[bot] merged 39 commits into

GesuBackups:masterfrom

tensorflow:master

Apr 3, 2026

Commits on Mar 21, 2026

Commits on Apr 2, 2026

Stabilize SymbolicExpr and SymbolicMap string representation
tensorflower-gardener
committed
[XLA:GPU] Use XLA_VLOG_DEVICE where device id is useful in logs.

PatriosTheGreat
authored and
tensorflower-gardener
committed
[XLA:GPU] Do not run large HLO fusions concurrently in command buffers to avoid instruction cache thrashing.

thomasjoerg
authored and
tensorflower-gardener
committed
Add TargetMachineOptions::Native() to create options for the host machine.

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change

derekmauro
authored and
tensorflower-gardener
committed
Bump rules_python 1.8.4 -> 1.8.5 to restore correct URL-based req processing.

belitskiy
authored and
tensorflower-gardener
committed
[mpmd] Use reserved_hbm_bytes attribute to reserve per-fragment memory

ICGog
authored and
tensorflower-gardener
committed
Make XLA:GPU propagate all CPU HW features to FFI and XLA:CPU

beckerhe
authored and
tensorflower-gardener
committed
Merge pull request #112909 from codeXsidd:fix-typo-metrics
tensorflower-gardener
committed
PR #40311: [xla:gpu] Delete vestigial WaitForStreamsThunk

ezhulenev
authored and
tensorflower-gardener
committed
Consolidate sanitizer build checks for tfrt_session_python_test.
tensorflower-gardener
committed
Reverts cd712c81dacec1d737af7e27dc01c8fc0a7f5020

beckerhe
authored and
tensorflower-gardener
committed
HLO builder tuple_test is hardware-independent.

nvgrw
authored and
tensorflower-gardener
committed
Original change is causing presubmit tests failure. Rolling back for further investigation.
tensorflower-gardener
committed
Rollback of PR #39854

yuriivcs
authored and
tensorflower-gardener
committed
Make `HloProgram::name()` return a unique and stable name even if the module does not have a name

junwhanahn
authored and
tensorflower-gardener
committed
Remove unneeded condition variable from coordination service.

mwhittaker
authored and
tensorflower-gardener
committed
Reverts e6f36bd7a12f87aac6f21a5068719dfc91bed3eb

Varcho
authored and
tensorflower-gardener
committed
Populate device_memory_bytes_limit attribute while reconstructing TopologyDescription from proto

krishnaharidasan
authored and
tensorflower-gardener
committed
PR #40117: Update XLA oneAPI presubmit build targets

mraunak
authored and
tensorflower-gardener
committed
Drop deduplicating any calls on outliner.

ekayaaslan
authored and
tensorflower-gardener
committed
Centralize PjRt Transpose Caching in CommonPjRtClient.

junwhanahn
authored and
tensorflower-gardener
committed
Implement non-descending layout support in `CpuRawBuffer::CopyToLiteralAsync`

junwhanahn
authored and
tensorflower-gardener
committed
[XLA:CPU] Add Pi approximation to fusion microbenchmark.

penpornk
authored and
tensorflower-gardener
committed
[IFRT IR] Add compile option to set strict memory reservations

ICGog
authored and
tensorflower-gardener
committed
Run `MakeArraysFromHostBuffer` and `CopyToHostBuffer` tests with strides on all memory kinds

junwhanahn
authored and
tensorflower-gardener
committed
[XLA:CPU] Add Opt Preset FAST_COMPILE for interactive and test users.

seantalts
authored and
tensorflower-gardener
committed
Use salted hashing for key_hash so that we have fewer collisions in bucket.table.

ezbr
authored and
tensorflower-gardener
committed
[XLA] Wrap loops containing only DCHECKs with ifndef NDEBUG.

zvikinoza
authored and
tensorflower-gardener
committed
[tsl] Clean up #includes in errors.h and status.h

d0k
authored and
tensorflower-gardener
committed
Migrate multithreaded_compilation_test to PjRt runtime.

nvgrw
authored and
tensorflower-gardener
committed
[XLA:CPU] Rename is_fusion_emitters to use_fusion_emitters for consistency.

seantalts
authored and
tensorflower-gardener
committed
Remove unused reset logic from coordination agent.

mwhittaker
authored and
tensorflower-gardener
committed
[IFRT IR] Do not apply identity IfrtArraType conversion in lowering to IFRT.

ICGog
authored and
tensorflower-gardener
committed
[IFRT] Add `Value::ByteSize` method

hyeontaek
authored and
tensorflower-gardener
committed

Commits on Apr 3, 2026

Erase async collectives in XLA CPU.

mwhittaker
authored and
tensorflower-gardener
committed
Integrate LLVM at llvm/llvm-project@7ccd92e5e6e5

alinas
authored and
tensorflower-gardener
committed