This is a better name since we have multiple 64-bit things that could
be referred to.
LP64 : integer=32, long=64, pointer=64
ILP64 : integer=64, long=64, pointer=64
This makes packages use lapack and blas, which can wrap different
BLAS/LAPACK implementations.
treewide: cleanup from blas/lapack changes
A few issues in the original treewide:
- can’t assume blas64 is a bool
- unused commented code
This is based on previous work for switching between BLAS and LAPACK
implementation in Debian[1] and Gentoo[2]. The goal is to have one way
to depend on the BLAS/LAPACK libraries that all packages must use. The
attrs “blas” and “lapack” are used to represent a wrapped BLAS/LAPACK
provider. Derivations that don’t care how BLAS and LAPACK are
implemented can just use blas and lapack directly. If you do care what
you get (perhaps for some CPP), you should verify that blas and lapack
match what you expect with an assertion.
The “blas” package collides with the old “blas” reference
implementation. This has been renamed to “blas-reference”. In
addition, “lapack-reference” is also included, corresponding to
“liblapack” from Netlib.org.
Currently, there are 3 providers of the BLAS and LAPACK interfaces:
- lapack-reference: the BLAS/LAPACK implementation maintained by netlib.org
- OpenBLAS: an optimized version of BLAS and LAPACK
- MKL: Intel’s unfree but highly optimized BLAS/LAPACK implementation
By default, the above implementations all use the “LP64” BLAS and
LAPACK ABI. This corresponds to “openblasCompat” and is the safest way
to use BLAS/LAPACK. You may received some benefits from “ILP64” or
8-byte integer BLAS at the expense of breaking compatibility with some
packages.
This can be switched at build time with an override like:
import <nixpkgs> {
config.allowUnfree = true;
overlays = [(self: super: {
lapack = super.lapack.override {
lapackProvider = super.lapack-reference;
};
blas = super.blas.override {
blasProvider = super.lapack-reference;
};
})];
}
or, switched at runtime via LD_LIBRARY_PATH like:
$ LD_LIBRARY_PATH=$(nix-build -E '(with import <nixpkgs> {}).lapack.override { lapackProvider = pkgs.mkl; is64bit = true; })')/lib:$(nix-build -E '(with import <nixpkgs> {}).blas.override { blasProvider = pkgs.mkl; is64bit = true; })')/lib ./your-blas-linked-binary
By default, we use OpenBLAS LP64 also known in Nixpkgs as
openblasCompat.
[1]: https://wiki.debian.org/DebianScience/LinearAlgebraLibraries
[2]: https://wiki.gentoo.org/wiki/Blas-lapack-switch
* Switch to default buildPhase & installPhase
* In preConfigure
* Do not add -DNPARTITION to CHOLMOD_CONFIG. That would disable the use of Metis but we already have that.
* Do not remove -lrt on Darwin, Darwin compiler can handle that and the code no longer exists anyway.
* With CUDA enabled
* Do not replace CUDA_ROOT. It does not exist any more. Instead we are setting CUDA_PATH in makeFlags.
* Do not replace GPU_BLAS_PATH, it defaults to CUDA_PATH so it will end up with the same value.
* Do not add -DCHOLMOD_OMP_NUM_THREADS to GPU_CONFIG. Why would be having the library use the same number of threads as the builder a good idea?
* Do not replace CUDA_PATH, we are setting it in makeFlags now.
* Do not replace CUDART_LIB and CUBLAS_LIB. They were being replaced incorrectly (cuda libs are located in lib directory, not lib64). Instead set the correct paths in makeFlags.
* Do not replace CUDA_INC_PATH. Its default looks like it will end up with the same value.
* Do not replace NV20, NV30, NV35 – not used any more.
* Do not replace NVCC, defaults to the same.
* Do not replace NVCCFLAGS, we just used the default from SourceSparse 4.4.7 with -gencode=arch=compute_60,code=compute_60 tacked on top. Current upstream default looks much better.
* Stop adding -DNTIMER to CFLAGS on Darwin – clock_gettime is supported by macOS 10.12 SDK.
* In buildPhase
* Move the make arguments to makeFlags and library to buildFlags, allowing us to drop the manual make call. I did not verify all of these are still needed.
* Remove the creation of libsuitesparse.so. As far as I could tell it is some kind of remnant of our old expression – perhaps due to past deficiencies of the build scripts, we created the individual libraries as symlinks to libsuitesparse.so: e36b3ec0a5 But since the build script can now build individual .so libraries, there should be no need for this abomination. No other distros do this either.
* In installPhase
* No need to copy things manually, there is an install target. We just need to pass INSTALL=$out flag to make to let it know where to install the files.
* I do not have means of verifying the darwin dylib name fix but it looks like it might be fixed in an upcoming release.
* I dropped the rpath fixup as it does not seem to be needed any more (ldd does not report any unresolved libraries).
* Split to multiple outputs
* Fetch source from GitLab (the only source for future releases)
* Re-order attrset to be more idiomatic
* Format with nixpkgs-fmt
Naive concatenation of $LD_LIBRARY_PATH can result in an empty
colon-delimited segment; this tells glibc to load libraries from the
current directory, which is definitely wrong, and may be a security
vulnerability if the current directory is untrusted. (See #67234, for
example.) Fix this throughout the tree.
Signed-off-by: Anders Kaseorg <andersk@mit.edu>
