trunk/3ee461fd8f0bc4b7b10c578a1677ee70c587fab0: [ROCm][CK] Remove host-side FMHA sequence padding workaround (#195498)
- PyTorch: 407 events in the last 90 days
- PyTorch: 404th Release in the last 90 days
- Previous: earlier the same day · ciflow/trunk/195663: Update
What happened
Summary: Context #187152 added a gfx950 host-side pad/copy stopgap after unpadded CK FMHA TRLOAD instances faulted on tile-unaligned Q/K sequence lengths. ROCm/rocm-libraries#11519 guards those generated dispatch arms so unaligned requests select CK's existing tail-safe kPadSeqLenQ/K instances. What Remove the external Q/K/V sequence padding, O/LSE scratch allocation, and output copies from the PyTorch CK FMHA wrapp…
Summary assembled by rule from the sources below