Getting the generated kernels

[blueberry]

Is there a way to see how a generated kernel for a certain blas operation looks like? I am looking at how to use clBLAS kernels in another library, but can not follow the generation process in clBLAS. Is there something like "dump the generated kernels to a string"?

[CNugteren]

One option would be to simply printf the kernels yourself by modifying the source. I did that myself a while back, and, if I recall it correctly, it can be done like this (example for GEMM):

• Open src/library/blas/gens/gemm.c
• Add printf(“%s”, buf); just before the return of the blockGen function

[gregsomers]

Compile clBLAS from source. You need the cmake variable BLAS_DUMP_CLBLAS_KERNELS set to ON. You can do the interactive config cmake -i, it will prompt to enter a value.

After linking with the newly compiled lib, it will dump all of the kernels to the directory the program was executed from. It should dump the kernel for every single clBLAS call.

[blueberry]

Thank you, Greg.

On Tue, Jun 9, 2015 at 8:33 PM, Greg Somers notifications@github.com wrote:

Compile clBLAS from source. You need the cmake variable BLAS_DUMP_CLBLAS_KERNELS set to ON. You can do the interactive config cmake -i, it will prompt to enter a value.

After linking with the newly compiled lib, it will dump all of the kernels to the directory the program was executed from. It should dump the kernel for every single clBLAS call.

— Reply to this email directly or view it on GitHub https://github.com/clMathLibraries/clBLAS/issues/104#issuecomment-110457342 .

[TimmyLiu]

Can we close this issue?