Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Improve mrpc efficiency by leveraging write combining
Previously, fill 1k mrpc input buffer took 1024 memwr tlps, with each payload 1 dwords, while only 1 byte is valid(enabled). In this case, too many of tlps within a timer windows introduce tlp throttling. By use of the write combining buffer, 1k data fillingtake 16 memwr tlps with each payload 16 dwords.
- Loading branch information