[xpu] Fix conv2d incorrect results and alignment errors for non-64-byte-aligned tensors on XPU.#2

Draft

BBBela wants to merge 92 commits intomainfrom

bbela/issue-2238-part-6

Commits on Mar 19, 2026

Revert "`torch.bcomplex32` shell dtype. (#173783)"
pytorchmergebot
committed
Revert "Support negative index slicing with backed symints (#177308)"
pytorchmergebot
committed
Revert "Slicing with backed should produce backed output when possible (#175819)"
pytorchmergebot
committed
[Bugfix] Fix onehot runtime error (#177160)

Lucaskabela
authored and
pytorchmergebot
committed
[BE] Remove libtorch Linux docker images (#177802)

atalman
authored and
pytorchmergebot
committed
Fix type annotation for _sym_get_coordinate (#177446)

aorenste
authored and
pytorchmergebot
committed
[macOS wheel] Add wheel platform tag vs dylib minos validation (#177609)

yangw-dev
authored and
pytorchmergebot
committed
[Profiler] Add a test to validate kernel metadata is present in output (#177745)

ryanzhang22
authored and
pytorchmergebot
committed
[Profiler] Add external id handling to events() (#177662)

ryanzhang22
authored and
pytorchmergebot
committed
[DTensor] Fix _to_copy to reduce Partial before non-linear dtype conversions (#172696)

stmcgovern
authored and
pytorchmergebot
committed
Revert "Fix type annotation for _sym_get_coordinate (#177446)"
pytorchmergebot
committed
test: big gpu required to run test_original_aten_preserved_split_addmm (#175806)

ddelgadovargas-cyber
authored and
pytorchmergebot
committed
Fix dispatch macro strings for `F.interpolate` (#177835)

NicolasHug
authored and
pytorchmergebot
committed
Revert "Fix rocm embedding backward dynamic (#176567)"
pytorchmergebot
committed
[DTensor] index single-dim strategy (#176038)

wconstab
authored and
pytorchmergebot
committed
[dynamo] Support tracing _enter_autocast and _exit_autocast (#173877)

rtimpe
authored and
pytorchmergebot
committed
[dynamo] Support some more context manager functions (#176521)

rtimpe
authored and
pytorchmergebot
committed
Bring back the executable bit on build_host_protoc.sh (#165991)

blawrence-ont
authored and
pytorchmergebot
committed
[DTensor] Fix test_comm_mode_with_dtensor for Dijkstra sharding propagation (#177798)

wconstab
authored and
pytorchmergebot
committed
Revert "Fix dispatch macro strings for `F.interpolate` (#177835)"
pytorchmergebot
committed
Revert "[BE] Rename separable and channels-last paths in UpSampleKernel.cpp (#176714)"
pytorchmergebot
committed
Revert "[BE] Rename non-separable interpolation path in UpSampleKernel.cpp (#176713)"
pytorchmergebot
committed
Revert "Fixes assertEqual and assert_close for DTensor (#176895)"
pytorchmergebot
committed
Make Dtensor have consistent cache key in compile (#173526)

tugsbayasgalan
authored and
pytorchmergebot
committed
Revert "[ROCm] Enable expandable segments (#173330)"
pytorchmergebot
committed
[export] Share param/buffer dicts between GraphModule and root module (#177841)

tugsbayasgalan
authored and
pytorchmergebot
committed
[pytorch] address violations of warning unreachable-code-return in aten (#177795)

yfeldblum
authored and
pytorchmergebot
committed
Handle div by 0 in optimization hint when fallback is 0 (#177709) (#177709)

laithsakka
authored and
pytorchmergebot
committed
[dynamo] Add synchronize_event custom op for Event.synchronize() (#177613)

mlazos
authored and
pytorchmergebot
committed
[inductor] Add lowering and control_deps support for synchronize_event (#177614)

mlazos
authored and
pytorchmergebot
committed
[dynamo] Add end-to-end test for Event.synchronize() with D2H copies (#177615)

mlazos
authored and
pytorchmergebot
committed
[xpu][fix] Fix xpu CI failures. (#177724)

etaf
authored and
pytorchmergebot
committed
Revert "Where kernel with opaque types (#174204)"
pytorchmergebot
committed
[fully_shard][DTensor] Support fully_shard with DTensor on full SPMD mesh (#176334)

fegin
authored and
pytorchmergebot
committed
[CI] Add ARC runner label mapping config (#177803)

huydhn
authored and
pytorchmergebot
committed
[CI] Add ARC experiment support to runner determinator (#177804)

huydhn
authored and
pytorchmergebot
committed
Fix stale recursive dict-tag tensor guards (#177525)

bobrenjc93
authored and
pytorchmergebot
committed
[MPS] Remove some usages of double dtype in MPS tests (#177766)

kurtamohler
authored and
pytorchmergebot
committed
Revert "Fix behavor of CTC loss cuDNN backend when `zero_infinity=True` (#176911)"
pytorchmergebot
committed
[CI] Forward fix lint from #177804 (#177899)

huydhn
authored and
pytorchmergebot
committed
[torchTLX] Consolidate TLX knobs + TLXInductorChoices subclass (#177748) (#177748)

dshi7
authored and
pytorchmergebot
committed
[BE]: Propogate typing from functions via TypeGuard and TypeIs (#177879)

Skylion007
authored and
pytorchmergebot
committed
[decomp] Add max_pool2d_with_indices_backward decomposition using gather+reduction (#167318)

stmcgovern
authored and
pytorchmergebot
committed
[inductor] Batch cubin-to-obj conversion using .incbin assembly (#177864)

mergennachin
authored and
pytorchmergebot
committed
Use fp8 conversion intrinsics on Hopper+ to work around ptxas codegen bug (#177870)

ngimel
authored and
pytorchmergebot
committed

Commits on Mar 20, 2026

Commits on Mar 24, 2026

Comment out xpu SHA checkout in CMakeLists.txt.

BBBela
authored and
pytorchmergebot
committed
Enforce the zero-offset tensors before launching oneDNN conv2d kernel.

BBBela
authored and
pytorchmergebot
committed
Revert "Comment out xpu SHA checkout in CMakeLists.txt."

BBBela
authored and
pytorchmergebot
committed
Change approach to more general.

BBBela
authored and
pytorchmergebot
committed
Adjust functions names.

BBBela
authored and
pytorchmergebot
committed