fixes run-length decode oob access

elstehle · elstehle · commit f41027b60339 · 2023-10-27T09:38:21.000-07:00
diff --git a/cub/cub/block/block_run_length_decode.cuh b/cub/cub/block/block_run_length_decode.cuh
@@ -382,7 +382,10 @@ public:
     {
       decoded_items[i] = val;
       item_offsets[i]  = thread_decoded_offset - assigned_run_begin;
-      if (thread_decoded_offset == assigned_run_end - 1)
+
+      // A thread only needs to fetch the next run if this was not the last loop iteration
+      const bool is_final_loop_iteration = (i + 1 >= DECODED_ITEMS_PER_THREAD);
+      if (!is_final_loop_iteration && (thread_decoded_offset == assigned_run_end - 1))
       {
         // We make sure that a thread is not re-entering this conditional when being assigned to the last run already by
         // extending the last run's length to all the thread's item

Original file line number	Diff line number	Diff line change
`@@ -382,7 +382,10 @@ public:`
`382`	`382`	`{`
`383`	`383`	`decoded_items[i] = val;`
`384`	`384`	`item_offsets[i] = thread_decoded_offset - assigned_run_begin;`
`385`		`- if (thread_decoded_offset == assigned_run_end - 1)`
	`385`	`+`
	`386`	`+ // A thread only needs to fetch the next run if this was not the last loop iteration`
	`387`	`+ const bool is_final_loop_iteration = (i + 1 >= DECODED_ITEMS_PER_THREAD);`
	`388`	`+ if (!is_final_loop_iteration && (thread_decoded_offset == assigned_run_end - 1))`
`386`	`389`	`{`
`387`	`390`	`// We make sure that a thread is not re-entering this conditional when being assigned to the last run already by`
`388`	`391`	`// extending the last run's length to all the thread's item`