Implementation of SVE in gemv routines.

51 views
Skip to first unread message

Jai

unread,
Dec 11, 2023, 11:11:11 PM12/11/23
to OpenBLAS-users
Hello All,
I am rying to implement sve in gemv(levle2) routines, I see that for implementing sve in level 3 and level1(dot product) there is not any changes made in the gemv.c programs under interface, only sve code is added under kernel/arm64. Function gevm.c under interface that is calling the gemv_n.S is a function pointer(correct me if I am wrong) which points/calls according the type of the function (like dgemv, sgemv etc.). I wrote sve code under /kernel/arm64 . After that I tried rebuilding the library but that is not working, still gemv_n.S is being called. Please help/guide me how can I solve the problem.
I don't have the idea of writing generic code which can be called by the gemv function of interface and if there are any reference materials please share that also.
Thank You.

Zhang Xianyi

unread,
Dec 11, 2023, 11:13:34 PM12/11/23
to OpenBLAS-users
Hi,

Please check KERNEL files in kernel/arm64.  It will compile different gemv_*.S into gemv_n.o

Regards,

Xianyi

Jai

unread,
Dec 14, 2023, 2:14:55 AM12/14/23
to OpenBLAS-users
Thank Xianyi for quick reply,
I can see the different gemv_n.o under kernel, but I am not able to figure out how to integrate my code for gemv using sve. I am following the way it is done for dot product. They wrote the dot.c and dot_kernel_sve.c under kernel/arm64 (no other change i could find). The same gemv program under interface calls this dot.c which was earlier calling dot.S. 
Thanks.

Jai

unread,
Dec 14, 2023, 2:14:55 AM12/14/23
to OpenBLAS-users
Thanks Xianyi for replying.
I can see the gemv_n.o files under  kernel, but I am not able to get how to integrate my code so that gemv function under interface calls my function( goes to my sve code for matrix vector multiplication added under kernel/arm64). I am following the implementation they have done for dot product.
Regards.
Jai.

On Tuesday 12 December 2023 at 09:43:34 UTC+5:30 Zhang Xianyi wrote:
Reply all
Reply to author
Forward
0 new messages