KEMBAR78
Accuracy issues with Jitterated complex kernels for acos, acosh, asin, asinh, tan and tanh · Issue #100842 · pytorch/pytorch · GitHub
Skip to content

Accuracy issues with Jitterated complex kernels for acos, acosh, asin, asinh, tan and tanh #100842

@peterbell10

Description

@peterbell10

Issue description

The following complex kernels cause CI failures when implemented with jitted_gpu_kernel.

Some may simply require a small tolerance increase to the tests but for larger differences we need to figure out which implementation is correct, thrust or jitterator.

Note that gpu_kernel implemenations use thrust while jitted_gpu_kernel calls the complex math implementations we vendor from libc++. These can be found in llvm_complex.cpp.

cc @ngimel @mruberry @ezyang @soumith @msaroufim @wconstab @bdhirsh @anijain2305

Metadata

Metadata

Assignees

No one assigned

    Labels

    good first issuemodule: cudaRelated to torch.cuda, and CUDA support in generalmodule: jiteratortriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions