MasterJH5574 commented on code in PR #14182: URL: https://github.com/apache/tvm/pull/14182#discussion_r1125349558
########## include/tvm/tir/transform.h: ########## @@ -709,6 +709,12 @@ TVM_DLL Pass ManifestSharedMemoryLocalStage(); */ TVM_DLL Pass InstrumentProfileIntrinsics(); +/*! + * \brief Create default schedule for PrimFuncs to run on cuda device. + * \return The Pass. + */ +TVM_DLL Pass DefaultSchedule(); Review Comment: Though this pass is intended only for CUDA backend at this moment, I suppose it will actually work for some other GPU backends, as the tasks tuned by MetaSchedule on these backends are all following the same schedule rule. https://github.com/apache/tvm/blob/main/src/meta_schedule/utils.h#L512 So I personally would prefer the name of “DefaultGPUSchedule” and list the GPU backends this pass supports in the doc string. In the pass we have already checked if the target contains attribute “maximum # of threads per block.” We can fall back with no change if the target doesn’t have this attribute. -- 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]
