I'm just curious if there is any particular reason for not including the ATen native cuda headers?
|
set(INSTALL_HEADERS ${base_h} ${ATen_CORE_HEADERS} ${native_h} ${native_cpu_h} ${native_quantized_h} ${cuda_h} ${cudnn_h} ${hip_h} ${miopen_h}) |
I was toying around with fusing some element-wise operations in the TorchScript+libtorch tandem. So, I stumbled over CUDALoops.cuh etc. for some quick fusing tests.
But unfortunately, they didn't get "installed".
Thanks!
cc @ezyang @seemethere @malfet @ngimel
I'm just curious if there is any particular reason for not including the ATen native cuda headers?
pytorch/aten/src/ATen/CMakeLists.txt
Line 371 in a6a31bc
I was toying around with fusing some element-wise operations in the TorchScript+libtorch tandem. So, I stumbled over CUDALoops.cuh etc. for some quick fusing tests.
But unfortunately, they didn't get "installed".
Thanks!
cc @ezyang @seemethere @malfet @ngimel