4,663 commits

Author	SHA1	Message	Date
George Hotz	1afa3c0877	vmap on full model (#13340) ... * vmap on full model * vmap gemm * reduce sums on end * outer reduce * only if there's ranges * put those rules in symbolic * ranges * do opt later * add zero range	2025-11-18 16:06:06 -08:00
George Hotz	9c59b3d19e	vmap grad needs reduce_backward (#13336) ... * vmap grad needs reduce_backward * fuse and outer	2025-11-18 10:08:30 -08:00
George Hotz	06e39a88a9	outer vmap works (#13334) ... * outer vmap works * fuse works * vmap outer works * outer ranges work * grad work * should be good to merge	2025-11-18 09:27:48 -08:00
George Hotz	583560ab72	this is the right way to write vmap (#13328)	2025-11-17 20:20:52 -08:00
George Hotz	e4fead8a86	write scan in uops (#13321) ... * write scan in uops * ops range * no need for variable * meh, later * shorter	2025-11-17 16:58:08 -08:00
George Hotz	6d3385c284	print special ops in postrange (#13318) ... * print special ops in postrange * fix on OSX	2025-11-17 14:43:23 -08:00
wozeparrot	33773fda87	tk initial mi350 (#13289)	2025-11-17 11:46:32 -08:00
nimlgen	9bb17c53ea	amd: timer fix (#13267)	2025-11-17 13:59:03 +03:00
George Hotz	cabd4add48	more work parsing SQTT, separate VIZ/PROFILE (#13308) ... * more work parsing SQTT * more minimal runner * sep VIZ/PROFILE * parse print new * improve parser * more filter * that * split them * lil cleanup * skip flaky test * AQL in mmapeak	2025-11-16 10:40:39 -08:00
George Hotz	295600dc5a	saturday coffee shop work parsing the att format (#13295) ... * saturday coffee shop work parsing the att format * add examples * parser * classes of packets * fully vibe coded parser * vibing * empty * some vibe names * vibes * most of these are wrong * more vibes * better names * parsing * parse * cleanup parser * touchups	2025-11-16 08:25:51 -08:00
chenyu	8f0e747b3a	Tensor._tri with arange (#13297)	2025-11-16 10:21:16 -05:00
chenyu	e8844853ed	Tensor.eye with arange (#13287) ... with rangify we can write these with arange	2025-11-15 12:32:27 -05:00
George Hotz	22c08b470c	fold using outerworld range (#13286) ... * scan using outerworld range * almost * sched * simple range * mypy * woooo outer range * spec passes * print the numbers * lol it runs * real test	2025-11-14 20:43:41 -08:00
George Hotz	567066f51f	tests for cast there and back (#13195) ... * fix cast folding in llama * dtypes that work everywhere * Skip test_cast_there_and_back for backend casts Skip test due to backend casting issues.	2025-11-14 16:56:09 -08:00
George Hotz	6c5fa349e1	add (unused) outer range (#13285)	2025-11-14 16:47:52 -08:00
George Hotz	e5351699bd	openpilot warp (#13283) ... * openpilot image warp test * 0.4 ms on metal, 1 ms on CPU * new inputs each time * reshape	2025-11-14 13:55:32 -08:00
chenyu	888aaab151	test_tiny cleanup (#13276)	2025-11-14 11:11:32 -05:00
nimlgen	3e63831b98	nv: support 580+ drivers (#13269) ... * nv: 580+ support * start * f * fake * fix	2025-11-14 21:44:16 +08:00
nimlgen	c80d459d99	autogen: fix packed args structs (#13274) ... * autogen: fix packed args structs * and test this	2025-11-14 20:24:06 +08:00
nimlgen	14eb48b13a	autogen: rename nv_gpu to nv_570 (#13273) ... * autogen: rename nv_gpu to nv_570 * rename	2025-11-14 20:07:19 +08:00
nimlgen	f72b1fbca4	nv: read numClasses (#13271) ... * nv: read numClasses * fix * d	2025-11-14 19:43:25 +08:00
Christopher Milan	09f3aae169	In-tree autogen: all C libraries (#13220) ... * checkout files from autogen branch * ioctl with payload * fix am generations * properly fix generations This reverts commit b2a54f4f41. * revert discovery.h * support pragma pack(1) * typo * better getter * typo * NVCEC0_QMDV05_00_RELEASE[01]_ENABLE * align support * anon handling fix --------- Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>	2025-11-13 18:57:44 -08:00
wozeparrot	7eb0d8e744	feat: mixins on tiles (#13246)	2025-11-13 16:52:52 -08:00
Ayman Jabr	256f81bb02	Fix tracemeta 0 (#13049) ... * chore: tclesius branch resolved * fix: indentation --------- Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>	2025-11-13 09:07:11 -08:00
George Hotz	bcdfc109b5	hotfix: disable flaky test	2025-11-13 06:19:28 -08:00
wozeparrot	759557f633	feat: move tk tests to testextra (#13242)	2025-11-12 17:06:53 -08:00
chenyu	3f939f3d3c	update pm_simplify_valid (#13241) ... * update pm_simplify_valid fixed openpilot conv regression * IMAGE training is broken	2025-11-12 19:40:02 -05:00
George Hotz	ab9fa964d8	DISABLE_COMPILER_CACHE -> CCACHE (#13234) ... * DISABLE_COMPILER_CACHE -> CCACHE * Fix cachekey assignment in Compiler constructor	2025-11-12 15:07:09 -08:00
Jan Akhremchik	bc8e537423	Add NONZERO op to onnx backend (#13211)	2025-11-12 08:55:51 -08:00
qazal	7a6853fa40	viz: show python callstack in the first graph (#13218)	2025-11-12 20:52:28 +08:00
wozeparrot	371c1f2355	tk: move tiles to class (#13224)	2025-11-11 21:53:46 -08:00
Christopher Milan	41a098a82d	In-tree autogen: libc.py (#13217) ... * checkout changes from autogen branch * parents * pylint happy * move sys to system in helpers.py * typo * typo	2025-11-11 19:13:48 -08:00
wozeparrot	222bb12ddf	tk softmax (#13205)	2025-11-11 15:13:16 -08:00
qazal	bc55bc4849	cleanup test_viz profiler tests (#13221)	2025-11-12 03:46:48 +08:00
wozeparrot	73497af4c0	clean: use np for allclose (#13204)	2025-11-10 23:02:43 -08:00
chenyu	22b8579234	one last regressed dm kernel (#13201)	2025-11-10 23:30:52 -05:00
chenyu	829cdafccc	update openpilot slow conv uop ast (#13197) ... the two remaining slow ones	2025-11-10 17:03:20 -05:00
wozeparrot	6252831ceb	feat: initial tk library (#13160)	2025-11-09 22:54:29 -08:00
chenyu	2ba8b4946f	external_benchmark_op_cat.py (#13168) ... * external_benchmark_op_cat.py cat kernel that's 1ms on master and 50us with no GROUP and with NOLOCALS * fix	2025-11-08 01:54:10 -05:00
George Hotz	ffb9e8396f	fix indexing bug with convs ... * minimal difference for ONE_POOL=1 * fix indexing bug * improve indexing debugger * more debugger improvements * always for reshape	2025-11-07 16:45:19 -08:00
Ahmed Harmouche	3ecff3a8da	Fix dim splitting bug for len(dim) == len(limited) case (#13142) ... * Fix gpudims bug on webgpu * Fix split dim bug * Remove webgpu_bug from examples * Add test for shape correctness * Fix 3D indexing --------- Co-authored-by: chenyu <chenyu@fastmail.com>	2025-11-07 12:31:06 -05:00
nimlgen	b8e48effcb	device: no compilers message with reasons (#13146) ... * device: no compilers message with reasons * typings * mypy	2025-11-07 23:01:45 +08:00
chenyu	bb8cf948f2	variation of (x%c)+(x//c)*c = x (#13135) ... when x is in the form of y//b, the idiv term might have combined	2025-11-06 18:53:28 -05:00
George Hotz	42b34cf83d	bottom up linearizer (#13133) ... * bottom up linearizer * late stores * more complete * remove broken heuristic * upcast size * opt * more conservative * it needs that * disable opencl half on QCOM * fix * make that a real test * cpu test okay * ptx skip * end is after the range	2025-11-06 15:30:32 -08:00
chenyu	bfb0c0391f	test custom eye function (#13134) ... this version is also faster with NOOPT	2025-11-06 14:51:55 -05:00
nimlgen	dafdb4bfb1	test hcq open with pytest (#13124) ... * test hcq open with pytest * fi	2025-11-06 20:09:51 +08:00
nimlgen	05e2ff4d87	system: fix flock on pcidevs (#13123) ... * system: fix locking of hcq devices * rename and fullrun * force ok * fix * fix	2025-11-06 19:02:13 +08:00
chenyu	f33c182393	test custom qkv kernel (#13118) ... adding the online softmax hits infinite loop so starting with this	2025-11-05 23:32:13 -05:00
George Hotz	9b2b535fa4	fix issue with multi flip (#13115)	2025-11-05 15:28:50 -08:00
George Hotz	4027eef264	fix test warnings (#13114) ... * fix test warnings * precommit passes * ignore std_mean warning	2025-11-05 15:06:29 -08:00