Steven Robertson a439bf671d Fix occupancy issues (1 block/SM when shuffle was on). ...

There are 16 bar.sync() registers available per *chip*, not per block, and I
was using number 8 in the shuffle code. Evidently the driver rewrites them per
SM, but does not compact their range. Good to know.

2010-09-12 11:09:47 -04:00

cuburn

Fix occupancy issues (1 block/SM when shuffle was on).

2010-09-12 11:09:47 -04:00

helpers

Experiments with larger CTAs for IterThread

2010-09-12 02:01:03 -04:00

bench.py

Fixed bench.py, with the help of Device Assertions™!*

2010-09-11 00:16:43 -04:00

main.py

A fake log filter stage while I work on other stuff

2010-09-12 02:32:03 -04:00

TODO

Lots-o-stuff.

2010-09-09 11:36:14 -04:00

Description

PyCUDA implementation of a GPU-accelerated fractal flame renderer.

2.3 MiB

Languages

Python 92.8%

Cuda 6%

Shell 0.6%

C 0.6%