-
For example, Can the CBLAS functions be linked with the optimised kernels under |
Beta Was this translation helpful? Give feedback.
Replies: 5 comments 1 reply
-
Your assumptions are misinformed. |
Beta Was this translation helpful? Give feedback.
-
Sorry I'm just a new person. To confirm, as you said the CBLAS entry points are emitted from the same source files, does that mean |
Beta Was this translation helpful? Give feedback.
-
Compile it once maybe? |
Beta Was this translation helpful? Give feedback.
-
The lapack-netlib subtree is a full copy of the Reference-LAPACK implementation from https://github.com/Reference-LAPACK/lapack (also sometimes called "netlib" LAPACK) but the BLAS and CBLAS parts of it are never built. Instead, the files under "interface" get compiled twice, once with "CBLAS" defined to generate the CBLAS interface, and once without it to generate the BLAS one. The interfaces either call the respective machine-specific BLAS kernels directly (typical for level 1 BLAS functions) or through the corresponding lecel2/level3 driver code that manages splitting the workload into cache-sized tiles (Goto's block matrix algorithm) |
Beta Was this translation helpful? Give feedback.
-
Take simple function and compare fortran and c calling conventions, double/single float/complex fortran/c exports are compiled from same C source. https://github.com/OpenMathLib/OpenBLAS/blob/develop/interface/dot.c |
Beta Was this translation helpful? Give feedback.
Take simple function and compare fortran and c calling conventions, double/single float/complex fortran/c exports are compiled from same C source.
https://github.com/OpenMathLib/OpenBLAS/blob/develop/interface/dot.c