Update changelog for 1.8.0, fix error introduced in block_reduce_raking

NVlabs · Feb 15, 2018 · 17fbfba · 17fbfba
1 parent 96d6daa
commit 17fbfba
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 1 deletion.
diff --git a/CHANGE_LOG.TXT b/CHANGE_LOG.TXT
@@ -1,3 +1,14 @@
+1.8.0    02/15/2018
+    - API change: change to the interfaces of cub::ShuffleIndex, cub::ShuffleUp, and 
+      cub::ShuffleDown to better compute the PTX shfl control constant for 
+      logical warps smaller than 32 threads 
+    - Bug fixes: 
+        - Issue #112: Bug in WarpScan's broadcast of warp-wide aggregate for 
+          logical warps < 32 threads 
+
+
+//-----------------------------------------------------------------------------
+
 1.7.5    02/08/2018
     - Added radix-sorting support for __half keys
     - Updated sorting policies for improved 8b-key performance 

diff --git a/cub/block/specializations/block_reduce_raking.cuh b/cub/block/specializations/block_reduce_raking.cuh
@@ -192,7 +192,7 @@ struct BlockReduceRaking
 
                 int valid_raking_threads = (IS_FULL_TILE) ?
                     RAKING_THREADS :
-                    (num_valid + RAKING_THREADS - 1) / RAKING_THREADS;
+                    (num_valid + SEGMENT_LENGTH - 1) / SEGMENT_LENGTH;
 
                 partial = WarpReduce(temp_storage.warp_storage).template Reduce<IS_FULL_TILE && RAKING_UNGUARDED>(
                     partial,