the supported dtypes are not align with cuda #2253

Activity

1. daisyden

   added

   skippedUsed for temp UT failure to parallel fixUsed for temp UT failure to parallel fix

   on Oct 31, 2025
2. daisyden

   self-assigned this

   on Oct 31, 2025
3. daisyden

   added

   test: utUnit-test failure surfaceUnit-test failure surface

   and removed

   test: utUnit-test failure surfaceUnit-test failure surface

   on Nov 1, 2025
4. chuanqi129

   added this to the PT2.10 milestone on Nov 12, 2025
5. daisyden

   added

   duplicateThis issue or pull request already existsThis issue or pull request already exists

   on Nov 19, 2025
6. daisyden commented on Nov 19, 2025

   

   daisyden

   on Nov 19, 2025

   ContributorAuthor

   More actions

   duplicated with #2289
7. daisyden

   added

   ut_upstreamUpstream unit-test dependencyUpstream unit-test dependency

   on Nov 19, 2025
8. daisyden commented on Nov 24, 2025

   

   daisyden

   on Nov 24, 2025

   ContributorAuthor

   More actions

   Aligned with CUDA in commit pytorch/pytorch@6874f0a, but need a review of @CuiYifeng and @EikanWang
9. riverliuintel

   modified the milestones: PT2.10, PT2.11 on Nov 27, 2025
10. mengfei25 commented on Dec 4, 2025

    

    mengfei25

    on Dec 4, 2025

    Contributor

    More actions

    ✅ op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv_transpose2d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv_transpose3d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_softsign_xpu Passed in nightly testing
11. mengfei25 commented on Dec 4, 2025

    

    mengfei25

    on Dec 4, 2025

    Contributor

    More actions

    ✅ op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv_transpose2d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv_transpose3d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_softsign_xpu Passed in nightly testing
12. chuanqi129

    modified the milestones: PT2.11, UT Backlog on Mar 2, 2026
13. 2 remaining items

    Load more

14. Stonepia commented on May 14, 2026

    

    Stonepia

    on May 14, 2026

    Contributor

    More actions

    ❌ Agent error in code_fix step:

    OpenCode failed (rc=-9)
    

    Traceback

        output, log_path, session_id, token_usage = backend.run(
                                                    ^^^^^^^^^^^^
      File "~/torch-xpu-ops/.github/issue-handler/issue_handler/utils/agent_backend.py", line 325, in run
        raise RuntimeError(
    RuntimeError: OpenCode failed (rc=-9), log: ~/torch-xpu-ops/.github/issue-handler/logs/agent-issue-2253-implementing-20260514-014329.log
    dd `dtypesIfXPU` matching `dtypesIfCUDA` for the affected ops.No previous fix attempt in the tree. Let me find the OpInfo definitions for the affected ops.Now I need to identify which ops from the issue need `dtypesIfXPU`. Let me check if `dtypesIfXPU` already exists anywhere:Good, so `dtypesIfXPU` is already a supported parameter. Now I need to find each affected op and add `dtypesIfXPU` matching `dtypesIfCUDA`. Let me look at each op mentioned in the issue. I'll search for them systematically.
    
15. Stonepia commented on May 14, 2026

    

    Stonepia

    on May 14, 2026

    Contributor

    More actions

    ❌ Agent error in code_fix step:

    Command '['gh', 'api', '/repos/chuanqi129/pytorch/pulls', '--method', 'POST', '-f', 'title=the supported dtypes are not align with cuda', '-f', "body=## the supported dtypes are not align with cuda\n\
    

    Traceback

      File "~/torch-xpu-ops/.github/issue-handler/issue_handler/utils/git.py", line 77, in _gh_api
        raw = _gh(cmd, token=token)
              ^^^^^^^^^^^^^^^^^^^^^
      File "~/torch-xpu-ops/.github/issue-handler/issue_handler/utils/git.py", line 60, in _gh
        raise subprocess.CalledProcessError(
    subprocess.CalledProcessError: Command '['gh', 'api', '/repos/chuanqi129/pytorch/pulls', '--method', 'POST', '-f', 'title=the supported dtypes are not align with cuda', '-f', "body=## the supported dtypes are not align with cuda\n\nFixes https://github.com/intel/torch-xpu-ops/issues/2253\n\n**Root Cause:** The OpInfo definitions for ops like addmm, addmv, baddbmm, mm, mv, histogram, histogramdd, conv1d/2d/3d, conv_transpose*, einsum, embedding_bag, linear, addbmm, bmm, linalg_multi_dot use 'dtypesIfCUDA' to restrict supported dtypes (e.g. floating_and_complex_types only), but lack a corresponding 'dtypesIfXPU'. As a result, XPU falls back to the general 'dtypes' field (which includes int64, complex32, and other integer types), causing test_dtypes to fail because XPU does not actually support those dtypes for these ops. The complex32 failure in true_divide is a separate manifestation of the same missing 'dtypesIfXPU' restriction.\n\n**Failed Tests:**\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmm_decomposed_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmv_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_baddbmm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_histogram_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_histogramdd_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_mv_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv1d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv2d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv3d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv_transpose1d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_einsum_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_python_ref__refs_true_divide_xpu_complex32`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_mm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_python_ref_executor__refs_true_divide_executor_aten_xpu_complex32`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_embedding_bag_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_linear_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_matmul_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_mm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_mv_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_bilinear_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv1d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv2d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv3d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose1d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose2d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose3d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_linear_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_tensordot_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_numpy_ref_addbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmm_decomposed_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmv_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_baddbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_bmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_linalg_multi_dot_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_mm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_mv_xpu_int64`\n\n\n---\n\n**Diff stat:**\n```\n.../_internal/common_methods_invocations.py        | 113 +++++++++++++--------\n .../testing/_internal/opinfo/definitions/linalg.py |   1 +\n 2 files changed, 70 insertions(+), 44 deletions(-)\n```\n\n\n", '-f', 'head=agent/issue-2253', '-f', 'base=main', '-F', 'draft=true']' returned non-zero exit status 1.
    
16. Stonepia commented on May 14, 2026

    

    Stonepia

    on May 14, 2026

    Contributor

    More actions

    ❌ Agent error in code_fix step:

    Command '['gh', 'api', '/repos/chuanqi129/pytorch/pulls', '--method', 'POST', '-f', 'title=the supported dtypes are not align with cuda', '-f', "body=## the supported dtypes are not align with cuda\n\
    

    Traceback

      File "~/torch-xpu-ops/.github/issue-handler/issue_handler/utils/git.py", line 77, in _gh_api
        raw = _gh(cmd, token=token)
              ^^^^^^^^^^^^^^^^^^^^^
      File "~/torch-xpu-ops/.github/issue-handler/issue_handler/utils/git.py", line 60, in _gh
        raise subprocess.CalledProcessError(
    subprocess.CalledProcessError: Command '['gh', 'api', '/repos/chuanqi129/pytorch/pulls', '--method', 'POST', '-f', 'title=the supported dtypes are not align with cuda', '-f', "body=## the supported dtypes are not align with cuda\n\nFixes https://github.com/intel/torch-xpu-ops/issues/2253\n\n**Root Cause:** The OpInfo definitions for ops like addmm, addmv, baddbmm, mm, mv, histogram, histogramdd, conv1d/2d/3d, conv_transpose*, einsum, embedding_bag, linear, addbmm, bmm, linalg_multi_dot use 'dtypesIfCUDA' to restrict supported dtypes (e.g. floating_and_complex_types only), but lack a corresponding 'dtypesIfXPU'. As a result, XPU falls back to the general 'dtypes' field (which includes int64, complex32, and other integer types), causing test_dtypes to fail because XPU does not actually support those dtypes for these ops. The complex32 failure in true_divide is a separate manifestation of the same missing 'dtypesIfXPU' restriction.\n\n**Failed Tests:**\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmm_decomposed_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_addmv_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_baddbmm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_histogram_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_histogramdd_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_mv_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv1d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv2d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv3d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_conv_transpose1d_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_einsum_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_python_ref__refs_true_divide_xpu_complex32`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_mm_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_python_ref_executor__refs_true_divide_executor_aten_xpu_complex32`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_embedding_bag_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_dtypes_nn_functional_linear_xpu`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_matmul_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_mm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_mv_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_bilinear_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv1d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv2d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv3d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose1d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose2d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_conv_transpose3d_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_nn_functional_linear_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_noncontiguous_samples_tensordot_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_numpy_ref_addbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmm_decomposed_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_addmv_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_baddbmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_bmm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_linalg_multi_dot_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_mm_xpu_int64`\n- `test_ops_xpu.py::TestCommonXPU::test_multiple_devices_mv_xpu_int64`\n\n\n---\n\n**Diff stat:**\n```\naten/src/ATen/native/mps/operations/Im2Col.mm      | 12 -----\n aten/src/ATen/native/mps/operations/ReduceOps.mm   | 24 ++++++++-\n test/inductor/test_torchinductor_strided_blocks.py | 11 ----\n test/test_mps.py                                   |  8 ---\n test/test_nn.py                                    | 61 ++++++++++------------\n torch/_inductor/codegen/triton.py                  | 20 ++-----\n torch/_inductor/runtime/triton_heuristics.py       |  4 +-\n 7 files changed, 54 insertions(+), 86 deletions(-)\n```\n\n\n", '-f', 'head=agent/issue-2253', '-f', 'base=main', '-F', 'draft=true']' returned non-zero exit status 1.
    
17. Stonepia

    added

    agent:needs-humanAgent lifecycle: real bug the agent could not fix on its ownAgent lifecycle: real bug the agent could not fix on its own

    and removed

    agent:activeAgent lifecycle: activeAgent lifecycle: active

    on May 14, 2026
18. daisyden

    added

    dependency component: communityMatching open pytorch/pytorch issueMatching open pytorch/pytorch issue

    on Jun 10, 2026
19. daisyden commented on Jun 10, 2026

    

    daisyden

    on Jun 10, 2026

    ContributorAuthor

    More actions

    The upstream PyTorch PR (pytorch/pytorch#176688) is open and expected to resolve this dependency. This issue is labeled as a community component dependency.
20. github-actions commented on Jun 23, 2026

    

    github-actionsBot

    on Jun 23, 2026 – with GitHub Actions

    More actions

    ✅ op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_python_ref_executor__refs_true_divide_executor_aten_xpu_complex32 Passed in nightly testing
21. github-actions commented on Jul 22, 2026

    

    github-actionsBot

    on Jul 22, 2026 – with GitHub Actions

    More actions

    ✅ op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_addmm_decomposed_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_addmm_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_addmv_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_baddbmm_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_einsum_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_mm_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_mv_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv1d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv2d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv3d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_conv_transpose1d_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_embedding_bag_xpu op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_dtypes_nn_functional_linear_xpu Passed in nightly testing
22. libohao1201 commented on Aug 20, 2026

    

    libohao1201

    on Aug 20, 2026

    Contributor

    More actions

    The following XPU UT case(s) tracked by this issue have been skipped in #5043 until a fix is available:

    • test_dtypes_addmm_decomposed_xpu
    • test_dtypes_addmm_xpu
    • test_dtypes_addmv_xpu
    • test_dtypes_baddbmm_xpu
    • test_dtypes_histogram_xpu
    • test_dtypes_histogramdd_xpu
    • test_dtypes_mv_xpu
    • test_dtypes_nn_functional_conv1d_xpu
    • test_dtypes_nn_functional_conv2d_xpu
    • test_dtypes_nn_functional_conv3d_xpu
23. github-actions commented on Aug 27, 2026

    

    github-actionsBot

    on Aug 27, 2026 – with GitHub Actions

    More actions

    ✅ op_ut,third_party.torch-xpu-ops.test.xpu.test_ops_xpu.TestCommonXPU,test_python_ref__refs_true_divide_xpu_complex32 Passed in nightly testing on BMG
24. daisyden

    added

    dtype: int64int64 / long dtypeint64 / long dtype

    test: utUnit-test failure surfaceUnit-test failure surface

    on Nov 11, 2026
25. daisyden

    added the

    Bug

    issue type on Nov 11, 2026
26. daisyden commented on Nov 11, 2026

    

    daisyden

    on Nov 11, 2026

    ContributorAuthor

    More actions

    label-issue: #2253

    Root cause: XPU does not register the same dtype set as CUDA for a family of linear-algebra / histogram / embedding ops (addmm, addmv, baddbmm, histogram, histogramdd, mv, conv*, einsum, mm, embedding_bag, linear, matmul, tensordot, bmm, addbmm, bilinear, linalg.multi_dot, true_divide). test_dtypes fails with "The supported dtypes for on device type xpu are incorrect!"; test_noncontiguous_samples / test_numpy_ref / test_multiple_devices fail because int64 (and complex32) paths are missing. This is a dtype-coverage feature gap, not a numeric bug.

    Umbrella issue spanning many ops — strong need_split candidate (one tracking issue per op family, or at least separate histogram vs matmul-family).

    Analyzed representative case: test/xpu/test_ops_xpu.py::TestCommonXPU::test_dtypes_histogram_xpu (first non-struck case; one group).

    axis	value	reason
    type	Bug	Native Type unset; reported as UT failures. Underlying cause is a dtype-coverage gap, but surfaced as failing op_ut.
    test	test: ut	op_ut cases in test_ops_xpu.py (TestCommonXPU). Existing skipped + ut_upstream labels kept.
    module	(none confident)	Spans addmm/mm/bmm/histogram/embedding_bag/conv/einsum — no single XPU module; left unset rather than mislabel.
    priority	High	Many distinct failing op_ut cases across the op family (>6).
    dtype	dtype: int64	Most non-struck failures are int64 non-contiguous/numpy-ref; histogram cases are dtype-support.
    symptom	Functionality	"supported dtypes ... are incorrect" — op rejects/accepts wrong dtype set vs CUDA.
    dependency	dependency component: community (existing, kept)	Upstream OpInfo dtype claims; ut_upstream.
    triage	need_split, agent:needs-human (existing)	Umbrella across many ops; existing human duplicate label has no linked target.

    Existing human labels preserved: duplicate, skipped, dependency component: community, ut_upstream, agent:needs-human.

    Representative cases (subset; see extract.json):

    • test_dtypes_histogram_xpu
    • test_dtypes_histogramdd_xpu
    • test_noncontiguous_samples_matmul_xpu_int64
    • test_noncontiguous_samples_mm_xpu_int64
    • test_noncontiguous_samples_mv_xpu_int64
    • test_multiple_devices_bmm_xpu_int64
    • test_multiple_devices_linalg_multi_dot_xpu_int64
    • test_numpy_ref_addbmm_xpu_int64
    • test_noncontiguous_samples_einsum_xpu_int64

    Warning

    apply_labels.py flagged items that need a human:

    • warning: dropping label(s) not defined in labels.json: ['(none confident)', 'Functionality', 'dependency component: community (existing, kept)', 'need_split, agent:needs-human (existing)']

    ---

    Applied to this issue. Generated by a local label-issue run.