Open jeromeku opened 5 months ago
We haven't yet done the plumbing to emit the correct EVT arguments structures for creating a PyTorch extension for a kernel that uses EVT. Apologies that this hasn't been better documented and lacks a clear error indicating the lack of support.
@jackkosaian Thanks for the response.
Are there any examples or documentation on how to properly construct arguments for an EVT
, other than the streamk example?
Moreover, I'm having trouble with the different epilogue interfaces, #1459, for a relatively simple example. Would appreciate any help!
This issue has been labeled inactive-30d
due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d
if there is no activity in the next 60 days.
This issue has been labeled inactive-90d
due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.
Describe the bug The Python pytorch emitter does not output functioning code when compiling
Gemm
with anEVT
.Steps/Code to reproduce bug The script below reproduces the bug.
Switch
jit
toTrue
when callingcutlass.emit.pytorch
to see the generated code (see additional context, as well).Expected behavior Expect the
jitted
pytorch module to work per the non-pytorch version (usingplan.run
, which compiles and runs the kernel directly throughpycuda
/C
interface).Environment details (please complete the following information):
A6000
nvidia-cutlass
:3.5.0
Additional Context Below is the generated extension module (with
jit
set toFalse
).Issues:
DeviceKernel
but none is generatedEVT
is declared, none of the interface functions provide args for the visitor func