Fix for Issue #128: 7-bit sorting passes fail for sm61 w/ large-values

https://github.com/NVlabs/cub/issues/128
NVIDIA · Feb 8, 2018 · a921e86 · a921e86
1 parent 5b065ee
commit a921e86
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 2 deletions.
diff --git a/CHANGE_LOG.TXT b/CHANGE_LOG.TXT
@@ -4,6 +4,7 @@
     - Bug fixes: 
         - Syntax tweaks to mollify Clang
         - Issue #127: DeviceRunLengthEncode::Encode returns wrong results 
+        - Issue #128: 7-bit sorting passes fail for sm61 w/ large-values
 
 
 //-----------------------------------------------------------------------------

diff --git a/cub/block/block_radix_rank.cuh b/cub/block/block_radix_rank.cuh
@@ -140,7 +140,7 @@ public:
     enum
     {
         /// Number of bin-starting offsets tracked per thread
-        BINS_TRACKED_PER_THREAD = CUB_MAX(1, RADIX_DIGITS / BLOCK_THREADS),
+        BINS_TRACKED_PER_THREAD = CUB_MAX(1, (RADIX_DIGITS + BLOCK_THREADS - 1) / BLOCK_THREADS),
     };
 
 private:
@@ -495,7 +495,7 @@ public:
     enum
     {
         /// Number of bin-starting offsets tracked per thread
-        BINS_TRACKED_PER_THREAD = CUB_MAX(1, RADIX_DIGITS / BLOCK_THREADS),
+        BINS_TRACKED_PER_THREAD = CUB_MAX(1, (RADIX_DIGITS + BLOCK_THREADS - 1) / BLOCK_THREADS),
     };
 
 private: