{"author":"George Hotz","author_email":"72895+geohot@users.noreply.github.com","author_time":1715994018,"commit_time":1715994018,"committer":"GitHub","committer_email":"noreply@github.com","hash":"07b350a8f40cc0450c403b7c31b2db4f5802e24d","message":"new uops is an actual graph (#4560)\n\n* new uops is an actual graph\r\n\r\n* it's way slower\r\n\r\n* simpler\r\n\r\n* fix define acc\r\n\r\n* render_loop unique\r\n\r\n* ops test pass\r\n\r\n* add pattern matcher back, there's bugs\r\n\r\n* rewrite\r\n\r\n* use priority queue\r\n\r\n* recursive children\r\n\r\n* fix tests\r\n\r\n* fix tests with SINK\r\n\r\n* fix abstractions\r\n\r\n* fix assembly\r\n\r\n* simpler\r\n\r\n* link define_acc\r\n\r\n* fix DEFINE_ACC placement\r\n\r\n* type verify\r\n\r\n* full cmp\r\n\r\n* fix cmp\r\n\r\n* ACCESS_ACC\r\n\r\n* insert DEFINE_ACC\r\n\r\n* fix PHI\r\n\r\n* recursive rewrite\r\n\r\n* fix many tests\r\n\r\n* sum collapse\r\n\r\n* more patterns\r\n\r\n* correct change\r\n\r\n* fold arange\r\n\r\n* fix that lin test\r\n\r\n* space\r\n\r\n* big folding rule works\r\n\r\n* close\r\n\r\n* has more maxes, meh\r\n\r\n* cached node replace\r\n\r\n* set changed\r\n\r\n* simplest folding yet\r\n\r\n* works\r\n\r\n* works\r\n\r\n* DIV\r\n\r\n* all tests pass\r\n\r\n* del\r\n\r\n* fuzz linearizer fails\r\n\r\n* sum_collapse\r\n\r\n* test depth 2 cf\r\n\r\n* fix lin test 14\r\n\r\n* fix clang depth\r\n\r\n* disable that\r\n\r\n* failure 14 is fixed\r\n\r\n* fix ptx\r\n\r\n* failure 27 is fixed\r\n\r\n* fix llama\r\n\r\n* run_cnt\r\n\r\n* Revert \"Optimize PTX gated loads index calculation (#4304)\"\r\n\r\nThis reverts commit d97d5a76899a3caf448b093f69a2e21bdd76be97.\r\n\r\n* fix uops loop\r\n\r\n* fix ptx bugs\r\n\r\n* add barrier\r\n\r\n* print\r\n\r\n* mem_type in ptx direct\r\n\r\n* bypass tests that fail in CI but pass locally\r\n\r\n* ptx remove ptr_ar\r\n\r\n* more ptx passing\r\n\r\n* fix ptx tests\r\n\r\n* assert compile support\r\n\r\n* remove  model inference benchmark from red","parents":["daf57af3eb1f02003bef53cf953426b961e99c69"],"tree_hash":"e161d7a893aa5a0ab3208288d2a8244fe4ab9679"}