aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

+1

Browse Source
This commit is contained in:
Evghenii
2014-01-28 21:29:20 +01:00
parent 354e535b37
commit 651cc04ee1
1 changed files with 19 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
examples_ptx/radixSort/radixSort.ispc
View File

@@ -81,6 +81,7 @@ void sortPass(
localCounts[i] = 0; localCounts[i] = 0;
const int unitScan = exclusive_scan_add(1); const int unitScan = exclusive_scan_add(1);
const int unitSum = exclusive_scan_add((int)(programIndex < programCount/2));
foreach (i = 0 ... nloc) foreach (i = 0 ... nloc)
@@ -102,11 +103,26 @@ void sortPass(
else else
{ {
#ifdef __NVPTX__ #ifdef __NVPTX__
/* buf fix, somehow atomic w/o branching fails */
if (programIndex < 16) if (programIndex < 16)
rel = atomic_add_global(&localCounts[key],1); {
if (reduce_equal(key) == true)
{
rel = localCounts[key] + unitScan;
localCounts[key] = rel+1;
}
else
rel = atomic_add_global(&localCounts[key],1);
}
else else
rel = atomic_add_global(&localCounts[key],1); {
if (reduce_equal(key) == true)
{
rel = localCounts[key] + unitScan - unitSum;
localCounts[key] = rel+1;
}
else
rel = atomic_add_global(&localCounts[key],1);
}
#else #else
rel = atomic_add_local(&localCounts[key],1); rel = atomic_add_local(&localCounts[key],1);
#endif #endif