mirror-github-bspeice/cuburn
1
0
Fork 0
mirror of https://github.com/stevenrobertson/cuburn.git synced 2025-02-05 03:30:05 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

Work around an overflow condition for now.

Browse Source 
I'm not sure what's going wrong; the math still holds up at higher
densities, but when you crank up the samples-per-pixel count the
accumulators start overflowing stochastically, and when they do
they dump nonsense into the output. Until I have time, take a small
perf hit by flushing much more often.
This commit is contained in:
Steven Robertson 2015-02-14 17:48:22 -08:00
parent 8da1821616
commit e08444f74b
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
cuburn/code/iter.py
View File

@ -346,17 +346,17 @@ iter(uint64_t out_ptr, uint64_t atom_ptr,
cvt.u64.u32 ptr, off;
add.u64 ptr, ptr, %4;
// 97% of the time, do an atomic add, then jump to the end without
// 80% of the time, do an atomic add, then jump to the end without
// stalling the thread waiting for the data value
setp.le.f32 p, %5, 0.97;
setp.le.f32 p, %5, 0.80;
@p red.global.add.u64 [ptr], val;
@p bra oflow_end;
// 3% of the time, do the atomic add, and wait for the results
// 20% of the time, do the atomic add, and wait for the results
atom.global.add.u64 val, [ptr], val;
mov.b64 {lo, hi}, val;
// If the density is less than 256, jump to the end
// If the density is less than 32, jump to the end
setp.lo.u32 p, hi, (256 << 22);
@p bra oflow_end;