Fixed race condition

de666e30 · peastman · b849c856 · de666e30 · de666e30
Commit de666e30 authored Feb 07, 2018 by peastman
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

platforms/cuda/src/kernels/rmsd.cu platforms/cuda/src/kernels/rmsd.cu +1 -0

platforms/opencl/src/kernels/rmsd.cl platforms/opencl/src/kernels/rmsd.cl +1 -0

No files found.
--- a/platforms/cuda/src/kernels/rmsd.cu
+++ b/platforms/cuda/src/kernels/rmsd.cu
@@ -6,6 +6,7 @@
 */
 __device__ real reduceValue(real value, volatile real* temp) {
    const int thread = threadIdx.x;
+    __syncthreads();
    temp[thread] = value;
    __syncthreads();
    for (uint step = 1; step < 32; step *= 2) {

--- a/platforms/opencl/src/kernels/rmsd.cl
+++ b/platforms/opencl/src/kernels/rmsd.cl
@@ -6,6 +6,7 @@
 */
 real reduceValue(real value, __local volatile real* temp) {
    const int thread = get_local_id(0);
+    barrier(CLK_LOCAL_MEM_FENCE);
    temp[thread] = value;
    barrier(CLK_LOCAL_MEM_FENCE);
    for (uint step = 1; step < 32; step *= 2) {