Sivagnanam Namasivayamurthy

Results 23 comments of Sivagnanam Namasivayamurthy

Getting seg fault during tuning. ``` | 307 | 370 | 0 64 64 32 8 8 8 8 2 1 1 0 0 0 0 1 | OK 447...

Thank you for your time @CNugteren . I did try removing `MWG==64` from `src/tunining/kernels/xgemm.hpp`, still tuning failed and caused device reboot. I've removed any option > 32 and started tuning...

I use these functions very frequently for Deep learning related stuff.Not sure about other users though. Please feel free to close this if you feel this isn't needed at the...

@kidapu Does this issue still exists? Were you able to get it working in Linux?

@moskewcz @roserg What is the optimum global & local size did you use to reproduce the resuts for FP 32? I would like to check the performance on Adreno 330....

> But it's very confusing when I simply changes some lines, the performance reduces much. Like below: @Thu-Chris Yes, I too faced similar problems when modifying certain lines of the...

@Thu-Chris Its not only Qualcomm's compiler, even ARM Mali T764 shows similar degradation.

@Thu-Chris Any possible suggestions (other than using L1 cache - with image as input) to improve the performance?

@Thu-Chris From your previous comments, I already tried 1, 2, 3 & 4. > For different set of (m,n,k), select the most efficient tiling method for(m,n,k). But it's hard to...

``` from matplotlib.ticker import EngFormatter formatter = EngFormatter(unit='Hz', places=0) img = axes.imshow(Z, extent=[0, duration, f_min, fs/2], aspect="auto") ``` I made these changes in plot.py and it worked for me.