Add head dimension 80 to compiled kernels

[njhill]

No description provided.

[WoosukKwon]

@njhill Can you please execute the generate_kernels.py and add the generated files for hdim80? I tried to add them by myself, but it seems your repo doesn't allow it.

[WoosukKwon]

@njhill Oh actually the compiled head size should be a multiple of 32 :(

flash-attention/csrc/flash_attn/src/kernel_traits.h

Line 69 in 013f0c4

static_assert(kHeadDim % 32 == 0);

[njhill]

ah, thanks @WoosukKwon :( .. I thought there might be something like that, sorry should have checked before I opened this