How does one build pytorch with int64_t indexing version of openBLAS to support LAPACK operations on inputs of sizes > 2^32
Once you have your OpenBLAS with int64_t indexing built, you can either put it in a known OpenBLAS install folder like
/opt/OpenBLAS/lib or specify its path in
LD_LIBRARY_PATH. Then you can run
setup.py script with
BLAS=OpenBLAS to ensure that it uses OpenBLAS and not another BLAS library.
How do one built openBLAS with int64_t support ?
I don’t know But I’m sure you can find instructions in their repo or online.