{"author":"qazal","author_email":"77887910+Qazalin@users.noreply.github.com","author_time":1702060160,"commit_time":1702060160,"committer":"GitHub","committer_email":"noreply@github.com","hash":"a29538a094b4c87ae390c33b0c366f2471ff35bd","message":"green more dtypes tests (#2656)\n\n* universal test cast\r\n\r\n* disable div\r\n\r\n* midcast fixup\r\n\r\n* add 64-bit types\r\n\r\n* hack maximum\r\n\r\n* use Metal precise::sin instead of default\r\n\r\nThis is because the default sin function defaults to single-percision math: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf#page=164\r\n\r\n* LLVM code_for_op support for var_dtype\r\n\r\n* comment out maximum for now with a TODO explaining it\r\n\r\n* Revert \"hack maximum\"\r\n\r\nThis reverts commit d170048c5fc029eab41f8472dd53f44c448370a1.\r\n\r\n* make the comment more specific\r\n\r\n* slightly more forgiving\r\n\r\n* ok does this fail in all backends?\r\n\r\n* weird its only Metal CI\r\n\r\n* add graph\r\n\r\n* skip sin of nan for CUDACPU\r\n\r\nThis is only happening in the CUDACPU runtime and not CUDA itself. https://github.com/tinygrad/tinygrad/actions/runs/7128973726/job/19412000385#step:16:36\r\n\r\n* METAL and CUDACPU behave differently in overflows with numpy running on CI\r\n\r\n* that skip is wrong\r\n\r\n* skip fp16 tests on LLVM similar to test_dtype\r\n\r\noriginal commit that skipped LLVM in CI 1826ff6b89b5c13d6ae6b4e4cfff35d3cae463e2\r\n\r\n* remove all of sin from CUDACPU\r\n\r\n* limit range of values in CUDACPU and METAL CI\r\n\r\n* Revert \"use Metal precise::sin instead of default\"\r\n\r\nThis reverts commit d960094d4a22fe69a9b6cb23ff7cd88e86a3c675.\r\n\r\n* change atol and rtol for Metal sin\r\n\r\n* METAL CI is more imprecise\r\n\r\n* cleanup\r\n\r\n---------\r\n\r\nCo-authored-by: George Hotz <geohot@gmail.com>","parents":["4164d0ebbdd1e723382bb216f4fef2c58e2534e4"],"tree_hash":"6f7a4829017e26c6622c898396dfdb8f9df865fd"}