masahi commented on code in PR #12785: URL: https://github.com/apache/tvm/pull/12785#discussion_r974611391
########## include/tvm/tir/builtin.h: ########## @@ -720,6 +720,16 @@ TVM_DLL const Op& texture2d_load(); */ TVM_DLL const Op& mem_copy(); +/*! + * \brief Initiate a non-blocking DMA copy from source to destination + */ +TVM_DLL const Op& dma_copy(); + +/*! + * \brief Wait until the number of DMAs in flight is less than or equal to some maximum + */ +TVM_DLL const Op& dma_wait(); Review Comment: For CUDA async copy, I added an option that users need to enable manually: https://github.com/apache/tvm/blob/036aa722aefd7126c17063cddb260b9347f17598/src/driver/driver_api.cc#L546-L551 For Hexagon, we may just enable this pass in `driver_api.cc` whenever the target is Hexagon. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
