Matthew Bentham
Matthew Bentham
Thanks @rhughesgeomerics. I think we don't need ViewOriginCoords any more, and the ConcatAxis is way simpler to use. The work needed would be to rewrite the subtensor handling and reference...
I think for non-Android, using a direct delegate is quite a bit simpler to set up than getting NNNAPI and the NN HAL driver service to work. Using the direct...
While I agree that NNAPI _could_ be made to work on (non-Android) Linux, it doesn't seem to be a supported or promoted option within the NNAPI project, for example I've...
I've just checked the internal performance tracking tests within Arm, we did see a small regression in some test cases from 21.05 to 21.08, and those were all fixed in...
6010440 clock cycles / 0.038s indicates 158169473 clocks/sec ie. roughly 158Mhz, is that the correct clock speed of the NPU in your system?
Hmm, from the kernel messages, 1829.964914-1829.957455 = 0.007459 ie. 7.5ms, is the actual length of inference on the NPU. And 6010440 clocks / 0.0075s = 801,392,000 or roughly 800Mhz, which...
At a guess I'd say that OpenCL isn't working on the device. Does clinfo show that the Mali-G31 is providing OpenCL?
That's very interesting, I've not seen that before. Maybe ClQLstmEndToEndTest needs OpenCL 3.0. Ah, now I look at the test output, I think it's just this one test which is...
Thanks Anton - would you be able to attach the Arm NN event profiles from the two runs? Perhaps for some reason sub-optimal kernels are being selected, and we should...