07:32 <Orum> I don't suppose there's an equivalent to RGA (https://github.com/GPUOpen-Tools/RGA) for nvidia, is there?

07:57 <Orum> jesus, the gpu binary is > 20x the size of the cpu binary

08:12 <angavrilov> the gpu compiler may be inlining everything

08:13 <angavrilov> I think there's no on-chip call stack support, so doing function calls requires spill to slow main memory

08:24 <Orum> well, actually the AMD APP binary is an actual binary, while the nVidia one is in ptx

08:25 <Orum> the problem is it's >23k lines of code that I don't want to look through... and more importantly I don't know why it fails some times and works flawlessly other times

08:26 <Orum> I suspect some shared resource is being grabbed by something else on my system and that's screwing it up, because the code itself works fine everywhere else and even on the gpu when it will run

08:30 <Orum> suppose at this point it's moot though, as now it runs even faster on the CPU than the GPU, and that's with copying buffers which I haven't optimized to just USE_HOST_PTR...