Segmentation Fault in DGEMM with AMD BLIS Library using Flang Compiler

[AkhilAkkapelli]

I am encountering a segmentation fault while running a DGEMM program with AMD BLIS. I am using the Flang compiler with AOCC and linking against the BLIS library. The program fails during the DGEMM operation with the following output:
Run: flang DGEMMExp.f90 -L/opt/AMD/aocl/aocl-linux-aocc-5.0.0/aocc/lib -lblis
./a.out

Output:

 transpose done
starting dgemm
libblis: Selecting sub-configuration 'generic'.
Segmentation fault (core dumped)

Program:

program DGEMMExp
implicit none

integer, parameter :: n = 2**5 + 16
double precision, dimension(n,n) :: A, At, AAt
integer(8) :: i, starttime, endtime, countrate

CALL random_number(A)
At = transpose(A)
print*, "transpose done"
call system_clock(starttime, countrate)

print*, "starting dgemm"
do i = 1, 3
   call DGEMM('N', 'N', n, n, n, 1.0d0, A, n, At, n, 0.0d0, AAt, n)
end do

call system_clock(endtime, countrate)
print*, "end DGEMM"
print*, "CPU Time:", real(endtime-starttime,8) / countrate

end program DGEMMExp

[vkirangoud]

Use i8 or -fdefault-integer-8 while compiling. Thanks Harish for suggesting this fix.

[AkhilAkkapelli]

@vkirangoud How to get verbose output like :
== Matrix multiplication using AMD BLIS DGEMM ==
and how to enable multithreading and specify threads for BLIS routines. and also verify the multithreading information in verbose output.

[vkirangoud]

Refer to https://docs.amd.com/r/en-US/57404-AOCL-user-guide/AOCL-User-Guide-57404Thanks,Kiran VOn 16 Nov 2024, at 2:24 AM, Akhil Akkapelli ***@***.***> wrote: How to get verbose output like : == Matrix multiplication using AMD BLIS DGEMM == and how to enable multithreading and specify threads for BLIS routines. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>