No, there is no way to direct operations at this level. Greg’s SM overview gives a good explanation of operation - warps can change on a cycle by cycle basis.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Per-Thread Repeated Access into Small Shared Float Array | 8 | 1207 | March 26, 2019 | |
| How do CUDA cores on a SM execute warps concurrently? | 8 | 28631 | July 4, 2019 | |
| Scheduling Warps of different kernels in the same cycle on the same SM | 6 | 69 | December 6, 2024 | |
| Can warps from different CTAs be coscheduled? | 5 | 201 | July 6, 2024 | |
| Scheduling individual threads | 4 | 4579 | June 1, 2009 | |
| CUDA execution mapping onto GPUs | 0 | 2817 | March 2, 2009 | |
| Simple summary of CUDA execution model An attempt to simplify and summarize various sources on execu | 7 | 5554 | July 28, 2009 | |
| register allocation behaviour | 2 | 419 | January 9, 2019 | |
| Blocks/Warps/Threads Allocation I have some doubts about the allocation of blocks/warps/thread in CU | 5 | 2570 | November 1, 2012 | |
| About warp scheduller in one SM | 1 | 420 | September 20, 2023 |