Our current implementation of Reverse Cache Read/Write hasn't considered itervars in outer blocks when there are nested blocks in the program.
This is problematic when loops in outer blocks are bound to physical threads, workloads in different threads might re-use the same region in generated cache read/write buffer.
Our current implementation of Reverse Cache Read/Write hasn't considered itervars in outer blocks when there are nested blocks in the program. This is problematic when loops in outer blocks are bound to physical threads, workloads in different threads might re-use the same region in generated cache read/write buffer.