24 commits

Author	SHA1	Message	Date
George Hotz	962d980919	fuse hasn't worked since rangeify, remove it (#13057)	2025-11-02 14:01:52 +08:00
chenyu	285534ce64	delete DONT_REALIZE_EXPAND and DONT_GROUP_REDUCES (#12744) ... does nothing now	2025-10-16 14:11:33 -04:00
chenyu	cf8232ec6a	clean up more RANGEIFY flag (#12556)	2025-10-09 03:06:48 -04:00
George Hotz	44558a37f7	fix some rangeify tests (#12370) ... * fix bad range merges * fix rng * fix uop gc * fix some rangeify tests * now that needs rangeify 2 also	2025-09-30 20:12:08 +08:00
George Hotz	fd2e4f2353	failing rng test (#12328) ... * tighten spec: fixup devectorizer types / rangeify * tighten assign * failing rangeify test * simpler * otherwise contig * more tolerance cause rng seed changed	2025-09-29 16:06:45 +08:00
Sieds Lykles	29f0886395	skip test_softmax_fusion tests if RANGEIFY==1 (#12310)	2025-09-27 05:57:40 +02:00
ttomsa	220a2a88d7	a*(1/b) -> a/b on LLVM, CPU (#11743) ... * add fdiv rewrite * :) * use float_lop * use reciprocal() * revert * move to decompositions	2025-08-20 09:35:10 -04:00
chenyu	4666df71c1	fix test_fuse_and_tc_opt (#11699)	2025-08-16 21:10:53 -04:00
geohotstan	3d7c35d615	add fuse and tc opt bug repro (#11695) ... * FINALLY HAVE A SMALL REPRO OH BOY * show failure in CI * cleaner? * 1 possible fix * Revert "1 possible fix" This reverts commit 9e0fd215dd.	2025-08-16 18:24:49 -04:00
George Hotz	9764c6cdee	fix mismatch reduce, try 2 (#11560) ... * fix mismatch reduce, try 2 * fix heuristic * delete that test * don't start allowing ones	2025-08-07 07:57:58 -07:00
George Hotz	a1aa5670aa	Revert "fix mismatch reduce (#11547)" (#11549) ... This reverts commit 49d21a9055.	2025-08-06 22:43:15 -07:00
George Hotz	49d21a9055	fix mismatch reduce (#11547) ... * fix mismatch reduce * cleanups * fix shape * fix mypy * resolve	2025-08-06 21:12:51 -07:00
George Hotz	7c5e115747	test_mismatch_reduce (#11538)	2025-08-06 10:02:14 -07:00
chenyu	f02720ca2d	fix fuse gate_contiguous unique (#11504)	2025-08-04 23:43:31 -04:00
chenyu	e0106b6b25	1/(x*c) -> (1/c)*(1/x) (#11491) ... example: 2*(2*a).reciprocal() -> a.reciprocal() # TODO: bounds for reciprocal # TODO: should z3 work?	2025-08-03 23:35:46 -04:00
George Hotz	49a2583584	real new lowerer (#11419) ... * real new lowerer * fix group for reduce * skip missing ranges * fix wmma and unroll/contract * real fix for wmma * disable that test * fix if gate * simpler * flash attention fusion works * no end barriers * still broken * flash attention finally works	2025-07-29 15:35:51 -07:00
George Hotz	81ef879da3	non recursive top_down_rewrite (#10729) ... * non recursive top_down_rewrite * nicer algorithm * rewrite bottom up also * only top down is broken? * simpler iterative algo * no recursion errors * top down and bottom up * unified rewrite * simpler rewrite * clean up comments * move that comment	2025-06-09 16:33:04 -07:00
George Hotz	eaceafecae	do fusion locally (#10095) ... * do fusion locally * oops, that's the right way * explicit delete closure	2025-04-28 20:45:37 -04:00
George Hotz	dd52951dd0	fix single kernel softmax with cast (#9842) ... * fix single kernel softmax with cast * tolerate none * 3e-4 * skip on dtype	2025-04-11 12:12:02 +08:00
chenyu	7fa5f29582	add test_embedding to test_softmax_fusion (#9832)	2025-04-10 08:25:34 -04:00
George Hotz	53f0b2aad7	fix infinite loop in flash attention (#9827) ... * fix infinite loop in flash attention * get_contraction_with_reduce * skip that test * SINGLE_KERNEL_SOFTMAX + fix multi * default IGNORE_OOB * print change	2025-04-10 20:06:44 +08:00
George Hotz	fce432d2e3	Ops.FUSE makes softmax a single kernel (#9808) ... * KERNELIZE makes softmax a single kernel * single kernel works * softmax works * broken * correct * skip that test * kernelize tests * rename to fuse * better reduce_push_add_ones code * correct now * cleanups * oops * return None if we can't push ones * rename + docs * atol fixes group * flash attention broken test	2025-04-09 22:56:28 +08:00
chenyu	7a28133b37	failed test for single softmax backward (#9778) ... getting RecursionError with DONT_GROUP_REDUCES=1	2025-04-08 02:36:32 -04:00
George Hotz	fefee5d3ab	single kernel softmax (#9776) ... * real single kernel softmax * cleanup * fix blockend insertion * add to bert test	2025-04-08 12:35:48 +08:00