[Torch] Disable building with MKL#5583
Conversation
0e79969 to
b5bd876
Compare
0cb084d to
860bfa1
Compare
|
What needs to be done here to get this merged? |
6e35c22 to
109d5cc
Compare
Fighting a bit with the build process to have it find LAPACK - and BLAS. ... and considering |
|
I suggest not worrying about 1.6 or even 1.7. These capabilities are fast moving, and we'll have a new LTS next year (I hope - not official statement). That will make it much easier to work with LBT. |
da0852b to
bef6194
Compare
|
I have confirmed that the generated Torch_jll can be imported on |
Also, * Changed recipe to use OpenBLAS (OpenBLAS32) and not BLIS. Building with BLIS cannot find LAPACK (in OpenBLAS) - and OpenBLAS takes precedence over BLIS. * Added Declarations.yaml FileProduct - needed for generating Torch wrapper.
bef6194 to
ffaf28c
Compare
bff3507 to
e06e3c0
Compare
|
This may need a @giordano review. |
|
Thanks a lot @stemann! |
| Dependency("XNNPACK_jll"; compat = "0.0.20210622"), | ||
| Dependency(PackageSpec("protoc_jll", Base.UUID("c7845625-083e-5bbe-8504-b32d602b7110")); compat="~3.13.0"), | ||
| HostBuildDependency(PackageSpec("protoc_jll", Base.UUID("c7845625-083e-5bbe-8504-b32d602b7110"), v"3.13.0")), | ||
| RuntimeDependency("CUDA_Runtime_jll"), |
There was a problem hiding this comment.
Was this supposed to be a dependency for all platforms? Or only the CUDA ones?
There was a problem hiding this comment.
It’s required, I believe, by the CUDA augment block that I added - I guess that is necessary in order to be able to get hold of the CUDA artifacts.
No description provided.