fixed macro in reduce.cl to use preprocessors size for warp

lanl · Sep 14, 2015 · 84df834 · 84df834
1 parent e82817a
commit 84df834
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 5 deletions.
diff --git a/mesh/mesh_kern.cl b/mesh/mesh_kern.cl
@@ -1914,8 +1914,8 @@ __kernel void finish_reduction_scan2_cl(
 
 void reduction_minmax_within_tile4(__local  int4  *tile)
 {
-   const unsigned int tiX  = get_local_id(0);
-   const unsigned int ntX  = get_local_size(0);
+    const unsigned int tiX  = get_local_id(0);
+    const unsigned int ntX  = get_local_size(0);
 
     for (int offset=ntX>>1; offset > MIN_REDUCE_SYNC_SIZE; offset >>= 1){
       if (tiX < offset){
@@ -3260,8 +3260,8 @@ __kernel void set_boundary_refinement(
 
 void reduction_sum_int2_within_tile(__local  int2  *itile)
 {
-   const unsigned int tiX  = get_local_id(0);
-   const unsigned int ntX  = get_local_size(0);
+    const unsigned int tiX  = get_local_id(0);
+    const unsigned int ntX  = get_local_size(0);
 
     for (int offset = ntX >> 1; offset > MIN_REDUCE_SYNC_SIZE; offset >>= 1)
     {

diff --git a/mesh/reduce.cl b/mesh/reduce.cl
@@ -123,7 +123,7 @@ real MIN(real a, real b)
         }                                                                       \
         barrier(CLK_LOCAL_MEM_FENCE);                                           \
     }                                                                           \
-    if (tiX < 32)                                                               \
+    if (tiX < MIN_REDUCE_SYNC_SIZE)                                             \
     {                                                                           \
         for (int offset = MIN_REDUCE_SYNC_SIZE; offset > 1; offset >>= 1)       \
         {                                                                       \