FAIL in matrixMul example

[vvolkl]

The matrixMul example fails for several CPU backends, for example -DALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLE=ON:

...
Error! Matrix[51197]=0.00000000, ref=1.59999990 error term is > 1.000000E-06
Error! Matrix[51198]=0.00000000, ref=1.59999990 error term is > 1.000000E-06
Error! Matrix[51199]=0.00000000, ref=1.59999990 error term is > 1.000000E-06
Result = FAIL

Note: For peak performance, please refer to the matrixMulCUBLAS example.

Looking at the CI setup, it seems like this is expected, is that correct? If that is the case, can I update the install instructions to use the vectorAdd example instead?

[sbastrakov]

Good catch, thanks for reporting. I am a bit confused that this comment implies that ALPAKA_ACC_CPU_B_OMP2_T_SEQ should work, but the following condition is different. I will look into that

[sbastrakov]

Ah, but the documentation you linked even contradicts this comment. So some things among those are wrong for sure

[psychocoderHPC]

Unfortionally I never added the element layer to the example. The required changes are in the GTC video linked in the README.md.
We should update the example.

[sbastrakov]

So does changing CUPLA_KERNEL to CUPLA_KERNEL_OPTI (as suggested by the docs) not help here?

[psychocoderHPC]

No it is not enough because the kernel does not use the element layer. The element layer must be added then you can use OPTI kernel call. Am 20. August 2021 11:59:42 MESZ schrieb Sergei Bastrakov ***@***.***>:

…

So does changing `CUPLA_KERNEL` to `CUPLA_KERNEL_OPTI` (as suggested by the docs) not help here? -- You are receiving this because you commented. Reply to this email directly or view it on GitHub: #212 (comment)