On-the-fly opacities: "too many resources requested for launch"

[iancrossfield]

When trying to run HELIOS3 with "on-the-fly" opacities on a computer system that's a few years old, i received the error pycuda._driver.LaunchError: cuLaunchKernel failed: too many resources requested for launch with a reference to line ~1361 of helios/source/computation.py.

[iancrossfield]

I was able to get things running again by dialing down the grid/block size used by the GPU code. In two places in computation.py I changed lines
from:

                     block=(32, 32, 1),
                     grid=((int(quant.nbin) + 31) // 32, (int(quant.nlayer) + 31) // 32, 1)

to:

                     block=(16, 16, 1)
                     grid=((int(quant.nbin) + 15) // 16, (int(quant.nlayer) + 15) // 16, 1)

After this edit, running with on-the-fly opacities and the lodders_m0 Solar abundances file produces a T-P profile very close (though not quite indistinguishable) from that produced using the 'premixed' solar-abundance mixture.