Search results

TensorDomain::flatten should squeeze broadcast IDs as done by the usual reshape transform

Status: Open.

Bug

#3691 In NVIDIA/Fuser;

· naoyam opened on Jan 9, 2025

Optimize matrix multiplication performance using tma multicast

Matmuls

Status: Open.

#3689 In NVIDIA/Fuser;

· rdspring1 opened on Jan 9, 2025

Inlining error in Hopper matmul with AxisMapping and grid swizzling

Matmuls

Status: Open.

Bug

#3671 In NVIDIA/Fuser;

· jacobhinkle opened on Jan 6, 2025

wrong result in matmul kernel

Status: Open.

#3665 In NVIDIA/Fuser;

· jjsjann123 opened on Jan 2, 2025

pad on broadcast dimensions hitting assert during transform replay

Status: Open.

#3660 In NVIDIA/Fuser;

· jjsjann123 opened on Dec 31, 2024

Feature request: Extend the uop forwarding in the fusion segmenter to include other single-input trivial ops

rope

Status: Open.

#3647 In NVIDIA/Fuser;

· naoyam opened on Dec 25, 2024

Feature request: Schedule a small matmul op as a reduction (or pointwise) op

rope

Status: Open.

#3646 In NVIDIA/Fuser;

· naoyam opened on Dec 25, 2024

Vectorization analysis returns wrong Vectorization Factor

Status: Open.

#3640 In NVIDIA/Fuser;

· jjsjann123 opened on Dec 24, 2024

Split by warp tile in Hopper matmul scheduler

Matmuls

Status: Open.

Enhancement

#3636 In NVIDIA/Fuser;

· jacobhinkle opened on Dec 23, 2024

Feature request: Extend the remove broadcast + squeeze pass

rope

Status: Open.

#3635 In NVIDIA/Fuser;

· naoyam opened on Dec 23, 2024

Performance gap between manual nvfuser definition and thunder.jit

Status: Open.

#3629 In NVIDIA/Fuser;

· Priya2698 opened on Dec 20, 2024

INTERNAL ASSERT FAILED at "nvfuser/csrc/device_lower/analysis/sync_information.cpp":812 Inconsistent parallelization found between TV48 and TV2

Thunder

Status: Open.

Bug

#3609 In NVIDIA/Fuser;

· crcrpar opened on Dec 18, 2024