{"author":"nimlgen","author_email":"138685161+nimlgen@users.noreply.github.com","author_time":1713801020,"commit_time":1713801020,"committer":"GitHub","committer_email":"noreply@github.com","hash":"e6227bdb15a15481c974b6be9d099c14e6259a67","message":"nv driver (#4044)\n\n* start\r\n\r\n* fix err 93\r\n\r\n* gpu\r\n\r\n* ioctl mappings\r\n\r\n* alloc like cuda\r\n\r\n* semaphores\r\n\r\n* wait for semaphores value\r\n\r\n* start ops_nv\r\n\r\n* very simple kernels work\r\n\r\n* init several gpus\r\n\r\n* qmd dumper\r\n\r\n* dirty, but most of kernels work\r\n\r\n* always all test_ops\r\n\r\n* progress, more tests, stable\r\n\r\n* test_ops passes, gpt2 works\r\n\r\nbut wth big fifo, wrap of fifo doesn't work, i think it's something coherency releated\r\n\r\n* need better sync\r\n\r\n* fix sync\r\n\r\n* alloc2\r\n\r\n* all tests pass!\r\n\r\n* cleanup 1\r\n\r\n* cleanup\r\n\r\n* multigpu, simple transfer\r\n\r\n* fix sync\r\n\r\n* correct init\r\n\r\n* nv_gpu autogen + sync bug fix\r\n\r\n* clean extra/nv_gpu_driver\r\n\r\n* p2p\r\n\r\n* clean up\r\n\r\n* remove old gen\r\n\r\n* small fixes\r\n\r\n* cleanup\r\n\r\n* cleanup 2\r\n\r\n* small fixes\r\n\r\n* bigger queue size\r\n\r\n* cleanups\r\n\r\n* wait\r\n\r\n* fixed signals for devs\r\n\r\n* fix hang + parallel beam\r\n\r\n* small fixes\r\n\r\n* detect when local memory is big in kernel\r\n\r\n* correct assert\r\n\r\n* small fixes\r\n\r\n* correct tls size est\r\n\r\n* one va space\r\n\r\n* less lines\r\n\r\n* shorter\r\n\r\n* save 2 lines\r\n\r\n* save some lines\r\n\r\n* remove type ignores\r\n\r\n---------\r\n\r\nCo-authored-by: George Hotz <72895+geohot@users.noreply.github.com>","parents":["77a378000527e921bfd6a6d3cf55cc84789359c9"],"tree_hash":"15cb1150ecbeb1232bfe00ac31fe377f4313d86c"}