oneMKL getri offload to GPU

[TestDR]

Hi,

I ran "getri.cpp" (oneAPI base toolkit examples @ /opt/intel/oneapi/mkl/2023.0.0/examples/examples_dpcpp.tgz.) through the Intel Advisor to compare  CPU to GPU offload analysis. It seems my old Xeon processor can do it as fast as a data center GPU Ponte Vecchio XT 512 XVE. Any idea?

Selected accelerator to analyze: Intel(R) Ponte Vecchio XT 512 XVE.
Baseline Host: Intel(R) Xeon(R) W-3225 CPU @ 3.70GHz.

Measured CPU Time: 4.520s Accelerated CPU+GPU Time: 4.520s
Speedup for Accelerated Code: 1.0x Number of Offloads: 0 Fraction of Accelerated Code: 0%

Compile and link settings

icpx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 -I$MKLROOT/include -L$MKLROOT/lib/intel64 -DSYCL_DEVICES_cpu -lmkl_sycl -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core -lsycl -lOpenCL -lpthread -lm -ldl -O2 -g getri.cpp -o getri

Advisor CLIs

advisor --collect=survey --auto-finalize --static-instruction-mix --project-dir=./getri_run512 -- ./getri
advisor --collect=tripcounts --flop --stacks --auto-finalize --cache-simulation=single --data-transfer=light --target-device=pvc_xt_512xve --project-dir=./getri_run512 -- ./getri
advisor --collect=projection --no-assume-dependencies --config=pvc_xt_512xve --project-dir=./getri_run512 -- ./getri

[Ruslan_M_Intel]

Hello,

That's right - Offload Modeling shows the same performance on PVC for this sample. Do you have PVC results as well? Could you please share them?

Best regards,

Ruslan Murtazin

[TestDR]

Hi Ruslan,

Do you mean the results on actual PVC GPU or any other PVC results generated by the intel advisor? Can you please be specific?

Thanks.

Rajitha

[Ruslan_M_Intel]

Hi Rajitha,

Yes, exactly! I meant Advisor GPU profiling result for the same app (but gathered on real PVC) to compare against Offload Modeling result.

Best regards,

Ruslan 

[RemyaP_Intel]

Hi,

We haven't received any response from your end. Is your issue resolved?

Regards,

Remya Premdas

[RemyaP_Intel]

Hi,

We assume that your issue is resolved. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Regards,

Remya Premdas