-
Gabriel Wu authored
* [fix] NVRTC execution backend * [fmt] run pre-commit * [fix] coderabbit reviews * [test] add cuda-python to test dep * [fix] coderabbit reviews * [fix] CUDA 13 compatibility * [fix] sm90 * [fix] CUDA 13 compatibility * [fix] pre-commit * [fix] always use cuda::std::__atomic_ref_impl * [fix] restore to external API * Revert "[fix] restore to external API" This reverts commit 49bd875638fb631d270015f408991d38fd1e9a5d. * [fmt] use space instead tabs for py codegen * [fix] im2col API * [fix] revert atomic.h * [fix] dynamic shape * [refactor] extract common utils * [feat] support L2 persistent map * [fix] l2 persistent map * [fix] pre-commit * [fix] restore _TYPE_MAP * [fix] pre-commit * [fix] avoid duplicate TMA descs * [docs] add docstring * [fix] coderabbit * [fix] coderabbit * [fix] coderabbit * [fix] coderabbit
eb415744