CUTLASS 4.6.2

[Junkai-Wu]

CuTe DSL

• Bug fixing and improvements
  • Reverted the TMA bulk copy elect_one change from 4.6.0 to reset the behavior to align with 4.5.x releases
  • Fixed a vectorized fp32->f8 conversion issue (!3382)
  • Fixed a CuTe DSL fp8 grouped_gemm_dglu kernel compilation issue (!397)
  • Fixed opt-level setting issue for ptxas (!3389)
  • set_name_prefix of kernel now supports full customization on compiled kernel name (!3389)
  • Fixed issue in (!3349)
  • Fixed issue in (!3351)
  • Reduced JIT compile overhead by ~50 ms per cute.compile
  • Speed up importing cutlass.cute by 3.8x with Torch installed, 1.25x without.

This release has been tested against the following packages:

• FlashAttention: main (00756db)
• Quack: main (79517ae)
• FlashInfer: main (766f94b)

---

This discussion was created from the release CUTLASS 4.6.2.