commit | f11ca65fb23d37d50bc214e149f4bdd9722cb3eb | [log] [tgz] |
---|---|---|
author | Hua Jiang <huaj@xilinx.com> | Thu Mar 18 13:44:32 2021 -0700 |
committer | GitHub <noreply@github.com> | Thu Mar 18 16:44:32 2021 -0400 |
tree | 4aae689c8eb0d3cb66494cd5f7586561769d8a32 | |
parent | f187a39f273961b1c2acc52088cc0a2305e5b247 [diff] |
Use TVMArrayCopyFromBytes API to replace VTA internal function VTABufferCPUPtr (#23) Issue: Current vta deploy example directly use VTABufferCPUPtr function to access VTA memory, this method may have issue once the VTA memory need cache flush/invalide and such logic not leverage VTA runtime memory movement logic which can default handle cache related issue. Solution: VTA runtime already implemented memory copy logic which integrate with TVMArrayCopy* API and here we use TVMArrayCopyFromBytes function replace this VTABufferCPUPtr function.
VTA (versatile tensor accelerator) is an open-source deep learning accelerator complemented with an end-to-end TVM-based compiler stack.
The key features of VTA include: