4,081 commits

Author	SHA1	Message	Date
George Hotz	b2017dec02	move that	2025-08-05 16:23:41 -07:00
George Hotz	3b777a9e05	optimize in rewrite (#11516) ... * changes * fix test uops * dim shouldn't be 0 * huh, why did that one not save	2025-08-05 15:33:26 -07:00
nimlgen	fc4e713d1c	jit graph split tests (#11507) ... * jit graph split tests * fix * one more test * more tests * fix * xm * rmeote	2025-08-05 21:32:37 +03:00
chenyu	ace8e9a706	fix test_conv2d_winograd (#11511)	2025-08-05 12:15:46 -04:00
chenyu	223aaa0492	clean up more conv tests (#11510)	2025-08-05 12:15:30 -04:00
Garret Castro	76e62a1c23	extract conv layer test logic (#11488) ... * refactor: extract conv layer test logic * tuple is unnecessary * integrate _test_conv logic into all conv tests * fix linter, forgot dilation * undo winograd extraction adds too many if statements for a single case	2025-08-05 11:15:54 -04:00
uuuvn	011ef8fa9d	Fix incorrect jit current batch devs reset (#11505) ... `current_batch_devs = []` (in `flush_batch()`) happens between `new_batched_devs = ...` and `current_batch_devs = new_batched_devs` => doesn't actually reset anything leading to things not jitting properly which 2xs remote bert step time (should have similar effects on any non-hcq backend)	2025-08-05 08:16:16 +03:00
chenyu	f02720ca2d	fix fuse gate_contiguous unique (#11504)	2025-08-04 23:43:31 -04:00
qazal	846a2826ab	viz: remove TracingKey.fmt (#11482) ... * viz: remove TracingKey.fmt * remove from test too	2025-08-05 00:00:03 +03:00
leopf	4f0ee4e982	BPE tokenizer (#11415) ... * BPE works * refactor tok * oops * basic tests * fix eval * smaller diff * fix error * proper vocab decoding * use regex for splitting * escape ucatrange * full compat --------- Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>	2025-08-04 09:52:38 -07:00
b1tg	06af9f9236	fix double exception + add name,loc in error msg (#11487) ... Co-authored-by: b1tg <b1tg@users.noreply.github.com>	2025-08-04 13:41:23 +03:00
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
chenyu	dbc7807c61	enable WEBGPU tests with buffer limit (#11489) ... TestSample still fails?	2025-08-03 13:02:44 -07:00
chenyu	66be747908	few more dtype cast convinience methods (#11480)	2025-08-02 15:47:09 -04:00
chenyu	e22e5da9a5	move some test_dtype tests to unit (#11479)	2025-08-02 15:25:00 -04:00
nimlgen	da0b955be4	hcq: cpu can be graphed (#11474) ... * hcq: cpu can be graphed * ops * new jit decisions * fix test * fix remote * cleaner * fix	2025-08-02 21:01:19 +03:00
kevvz	ef7e01cadf	Fix SVD shape bug + Fix batched SVD bug (#11477) ... * failing test case * fix * better test * space	2025-08-02 09:47:41 -07:00
qazal	fa66d9772d	viz: show const node when it's root (#11456)	2025-08-01 01:01:58 +03:00
Eitan Turok	cba3655de5	Add Test for Setitem (#10559) ... * init * update * better * failing test * works * Delete test file * clean * lint * simplify variable name * rm contigious, rm int dtype, and add assertEqual --------- Co-authored-by: chenyu <chenyu@fastmail.com>	2025-07-30 22:03:41 -04:00
chenyu	4ca430e5bf	fix search dedup (#11439) ... it should check against pre real_axis axis in actions, not real_axis.	2025-07-30 17:24:16 -04:00
chenyu	d5fc6af4a2	remove unused ShapeTracker.consecutive [pr] (#11426)	2025-07-29 18:36:19 -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
chenyu	0e5d8d5c3c	remove tests that used .to_uop() (#11425) ... * remove tests that used .to_uop() * import	2025-07-29 15:52:16 -04:00
nimlgen	d38d285489	ci: add h machines (#11416) ... * ci: add h machines * more * fix names * names not collide * 20 * 10	2025-07-29 19:21:51 +03:00
George Hotz	8c10085459	assert shape on lowerer store [pr] (#11395) ... * assert shape on lowerer store [pr] * fix ptx	2025-07-27 10:41:57 -07:00
George Hotz	dfeee63d30	uop matmul work (#11388) ... * uop matmul work * works with locals	2025-07-26 21:23:55 -07:00
George Hotz	3923e78061	no_vectorized_acc keeps single DEFINE_REG (#11387) ... * no_vectorized_acc keeps single DEFINE_REG * fix ptx, skip flaky test	2025-07-26 11:44:09 -07:00
George Hotz	466ab5a3f2	store/load not pass through index (#11381) ... * noop * fix noop * store cat is NOOP * store dtype is void * stores aren't passed through anymore * meh, skip those for ptx * correct ptx skip * hl runs	2025-07-25 21:01:47 -07:00
chenyu	88c338bfcc	add kernelize to keccak for each data block (#11370) ... * add kernelize to keccak for each data block test_long works now. this prevents internal uops from growing propotional to data length and eventually too deep * this? * hash stuff * gate test * mv	2025-07-25 16:07:20 -04:00
nimlgen	3b3de8df61	hcq: graphed copies (#11302) ... * fast copies p2 * upd and fix * graph supports * fixes * fixes * fixes * fix * fix * fix mockgpu * fix alignment * smaller in ci	2025-07-24 17:36:19 +03:00
nimlgen	bf12041910	hcq: mapping of cpu to all hcq devices (#11354) ... * hcq: mapping of cpu to all hcq devices * fix kfd * nv * simpler * cleaner * correct skip * fix ifaces * system fixes * mypy	2025-07-24 12:52:38 +03:00
chenyu	82e6de7fc6	more keccak reference tests (#11329)	2025-07-23 22:06:39 -04:00
chenyu	5b570196e4	support DEV= to specify device (#11351)	2025-07-23 17:40:55 -04:00
George Hotz	7f0a41df4d	move optional out of devectorize [pr] (#11350) ... * move optional out of devectorize [pr] * fast idiv	2025-07-23 11:26:05 -07:00
chenyu	960da9319d	Remove StrEnum in onnx for python 3.10 (#11345) ... some training tests failed looks like parsing error?	2025-07-23 11:52:25 -04:00
nimlgen	304eb9cecb	allocate less memory in am tests (#11342)	2025-07-23 11:11:26 +03:00
George Hotz	e14b4fefa5	ranges on store (#11334) ... * ranges on store * fix store spec * fix that * fix gates * fix tests * fix ptx	2025-07-22 21:00:50 -07:00
George Hotz	53339e62f7	no gate store anymore (#11338) ... * no gate store anymore * fix up spec	2025-07-22 18:41:15 -07:00
George Hotz	09431d4ad1	make DEFINE_REG behave like the others (#11273) ... * simpler define reg * cast * PTRCAT define_acc * cleanups * fix uops stats * fix linearizer tests * llvm * define reg sets const * define reg sets const * no assign * collapse that * fix test_max_pool2d_bigger_stride_dilation * use index, fix webgpu * devec * fix tests * fix webgpu * fix llvm * threads for python * fix ops_python * only for reg * acc_half is real now in the emulator * fix llvm * fix webgpu init * fix wgpu test * fix some tests * fix ptx * fix ptx bool acc * cleanups * broken, meh. will fix with ENDRANGE * line count	2025-07-22 13:53:56 -07:00
chenyu	4535908679	update keccak test_long (#11331) ... it should compare with arg "shake_128"	2025-07-22 16:08:01 -04:00
George Hotz	affd83961c	small changes from define_reg (#11327) ... * small changes from define_reg * fix webgpu	2025-07-22 11:11:48 -07:00
chenyu	2d7c28de6a	clean up dup lambdas in helper_test_exception (#11325)	2025-07-22 12:21:57 -04:00
chenyu	c6aa8e58ca	fix TestDropoutProbabilityEdgeCases (#11322)	2025-07-22 11:13:56 -04:00
chenyu	fb42c84365	merge TestRollEdgeCases into test_ops (#11321)	2025-07-22 10:55:57 -04:00
chenyu	1d8b3e9d1c	movementop only Tensor.roll (#11317) ... * movementop only Tensor.roll * fixed	2025-07-22 10:34:15 -04:00
chenyu	a41140241b	truncate unsigned const in cstyle (#11318) ... it can be a warning or a hard error in clang PTX and PYTHON also need fix, skipping for now	2025-07-22 08:02:12 -04:00
qazal	6668d6d241	fix word_wrap with newlines in input string [pr] (#11319)	2025-07-22 12:03:13 +03:00
George Hotz	3b674df34b	generic changes from define_reg_2 (#11315) ... * generic changes from define_reg_2 * fix for ptx * ugh, that one	2025-07-21 15:14:06 -07:00
chenyu	6e9506e6fd	Tensor.roll supports dims=None (#11313)	2025-07-21 17:29:23 -04:00
George Hotz	108aac8af4	use AddrSpace instead of local (#11314) ... * use AddrSpace instead of local * addrspace in test	2025-07-21 14:00:06 -07:00