- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I am developing an n linear sorter using n work-items. Therefore, I am using two barrier(CLK_LOCAL_MEM_FENCE) to ensure that compares and shifts are done properly. My code works on FPGAs and GPUs.
I am a bit confused by this AOCL warning: "Compiler Warning: Threads may reach barrier out of order - allowing at most 2 concurrent workgroups" I thought threads (work-items) reaching a barrier out of order is OpenCL default behaviour or can we assume that on FPGA the work-items within a work-group are always implemented in a lockstep manner? Further, is the compiler trying to implement concurrent work-groups? ThanksLink Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have met the warning. The work-items within a work-group are not implemented in a lockstep manner, but pipeline (one by one) manner. So, the out-of-order is mainly because of "if-else" statement, when the first work-item(e.g. if) has complex logic and the second work-item(e.g. else) is easy to finish.
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page