Fixed a bug that caused a crash on Fermi.

42cbcb89 · Peter Eastman · 9f4bd981 · 42cbcb89
Commit 42cbcb89 authored Sep 02, 2010 by Peter Eastman
Show whitespace changes
Inline Side-by-side

Showing with 4 additions and 5 deletions

platforms/opencl/src/kernels/sort.cl platforms/opencl/src/kernels/sort.cl +4 -5

No files found.
--- a/platforms/opencl/src/kernels/sort.cl
+++ b/platforms/opencl/src/kernels/sort.cl
@@ -76,9 +76,8 @@ __kernel void computeBucketPositions(int numBuckets, __global int* bucketOffset,
        // Perform a parallel prefix sum.
        for (int step = 1; step < get_local_size(0); step *= 2) {
-            int add = buffer[get_local_id(0)-step];
+            int add = (get_local_id(0) >= step ? buffer[get_local_id(0)-step] : 0);
            barrier(CLK_LOCAL_MEM_FENCE);
-            if (get_local_id(0) >= step)
            buffer[get_local_id(0)] += add;
            barrier(CLK_LOCAL_MEM_FENCE);
        }