Issues

#3560

· HaoYuan-Gao opened

on Aug 27, 2026

Conv device wrappers: update() leaves stale geometry and fused-output pointers, two wrappers miss the 2 GiB gate, negative output extents not rejected

#3545

· VaggelisGian opened

on Aug 25, 2026

Small profiler defects: device-ID validation off by one, kTensorNCHW mapped to TensorNHWC, sparse-meta host/device type mismatch

#3544

· VaggelisGian opened

on Aug 25, 2026

Three host reference utility defects: wrong MRE formula, operator precedence in padded-diagonal fill, unseeded per-element RNG in gaussian fill

#3543

· VaggelisGian opened

on Aug 25, 2026

Validation gaps: grouped/split-K can_implement accept crashing configs; batch grid wraps at 65536; Sm80 EVT mode gates are assert-only

#3542

· VaggelisGian opened

on Aug 25, 2026

Several 2.x device wrappers have update()/operator() paths that never compile or silently apply stale state

#3541

· VaggelisGian opened

on Aug 25, 2026

GemmUniversalWithBroadcast ColumnMajor-output transpose leaves the bias vector indexed in the wrong coordinate system

#3540

· VaggelisGian opened

on Aug 25, 2026

EllGemm ColumnMajor-output workspace sizing uses the unswapped problem shape and undersizes split-K semaphores

#3539

· VaggelisGian opened

on Aug 25, 2026

GemmWithKReduction parallel split-K: workspace initialization is skipped, partials land in the user output buffer

#3538

· VaggelisGian opened

on Aug 25, 2026

Blockwise-scaled split-K GEMM applies slice-0 scale factors to every K slice

#3537

· VaggelisGian opened

on Aug 25, 2026

Blockscaled GEMV drops the K tail when floor(K/tile) is a multiple of the stage count: example 91 fails at --k=1152

#3536

· VaggelisGian opened

on Aug 25, 2026

smem_atom_layoutSFB branches on majorSFA instead of majorSFB

#3535

· VaggelisGian opened

on Aug 25, 2026

Navigation Menu

Uh oh!

All issues

Question about `recast()` with negative strides

Conv device wrappers: update() leaves stale geometry and fused-output pointers, two wrappers miss the 2 GiB gate, negative output extents not rejected

Small profiler defects: device-ID validation off by one, kTensorNCHW mapped to TensorNHWC, sparse-meta host/device type mismatch

Three host reference utility defects: wrong MRE formula, operator precedence in padded-diagonal fill, unseeded per-element RNG in gaussian fill

Validation gaps: grouped/split-K can_implement accept crashing configs; batch grid wraps at 65536; Sm80 EVT mode gates are assert-only

Several 2.x device wrappers have update()/operator() paths that never compile or silently apply stale state

GemmUniversalWithBroadcast ColumnMajor-output transpose leaves the bias vector indexed in the wrong coordinate system

EllGemm ColumnMajor-output workspace sizing uses the unswapped problem shape and undersizes split-K semaphores

GemmWithKReduction parallel split-K: workspace initialization is skipped, partials land in the user output buffer

Blockwise-scaled split-K GEMM applies slice-0 scale factors to every K slice

Blockscaled GEMV drops the K tail when floor(K/tile) is a multiple of the stage count: example 91 fails at --k=1152

smem_atom_layoutSFB branches on majorSFA instead of majorSFB

Uh oh!

Search results