mirrors/tinygrad
2
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-24 02:14:17 +00:00
Code Issues Projects Releases Packages Wiki Activity
13,371 commits 280 branches 15 tags 325 MiB
Commit graph

5,652 commits

Author SHA1 Message Date
George Hotz
f597ec8d5f
 		Merge branch 'master' into minigen 2026-05-20 11:22:40 -07:00
chenyu
4dbe6a2ee7
 		remove _force_unique from Tensor init (#16277) 2026-05-20 14:13:05 -04:00
qazal
1e0fffe256
 		fused ce llama kernel in UOps (#16263) 
* work

* using uops

* delete things

* work

* work

* higher level uops

* cleanups
 2026-05-20 19:45:28 +09:00
chenyu
e1715b3b92
 		extent jit const error to deviceless inputs (#16276) 2026-05-20 02:02:45 -04:00
chenyu
170b857da9
 		clean up deviceless const _buffer (#16274) 
process on CPU similar to multi
 2026-05-19 22:47:45 -04:00
chenyu
188d7ec15e
 		clone can take device (#16271) 
useful to materialize const on a specific device
 2026-05-19 21:29:27 -04:00
George Hotz
55515747b7
 		Remove Ops.VCONST (#16267) 
* start removing vconst

* remove a lot of vconst

* const folding + strict ordering

* update tests

* spec from minigen

* move that
 2026-05-19 16:35:24 -07:00
Christopher Milan
7cdd9cbdeb
 		PYTHONREMU: V_CVT_PK_BF8_F32 saturation (#16268) 2026-05-19 19:29:59 -04:00
George Hotz
651031e3da
 		Merge branch 'master' into minigen 2026-05-19 14:12:51 -07:00
Christopher Milan
bb2a51f1ea
 		fix mypy mockgpu and add tinygrad.renderer.isa to packages (#16265) 2026-05-19 16:45:03 -04:00
chenyu
890b731b1e
 		more prerequisuite test changed for deviceless const (#16264) 2026-05-19 15:43:45 -04:00
ttomsa
aa1e59ab97
 		X86 with Ops.INS (#14873) 
* draft

* cleanup test_encodings

* cleanup test_isel

* model flag state and support rematerialization

* woops

* add vbroadcastss instruction

* don't fuse load if used multiple times in src

* add movabs instruction and fix idiv

* fixes

* add x86 backend to tests

* float16 fix

* rm TwoAddress2nd

* add BARRIER

* test windows ci

* yup isel fixes the mask stuff too and its beautiful

* add cmoves to the spec

* support storing imms

* no TUPLE_ORDER, breaks tests

* fix remaining seg faults

* add float max

* always fuse index

* minor

* fix DEFINE_VAR/SPECIAL and enable multithreading

* linter

* more linter

* more

* more

* more

* let's try this

* perhaps

* start new scheduler

* more scheduling info

* cleaner shuffle functions

* fixup isel tests

* skip bounds check when NOOPs exist

* skip inf rewrite tests

* fix const tag hack and add x86ops to _shape

* fix

* skip a few tests

* func arg order independent from op value

* x86 goes in own linearize

* switch to PARAM

* more

* add min x86op and neg in decomps

* do mulacc in isel

* use def_reg in test_encodings

* enable emulated int64 tests

* how much does this fix

* Ops becomes OpType

* fix

* rm noqa

* rm machine scheduler stuff

* and this

* allow for extending enums and move X86Ops out of uop

* fix imports

* rm X86GroupOp from ops.py

* spacing

* tell mypy to shut up

* more linter

* add x86op test

* allow set[X86Ops] in upat

* move NOOPs to pre_isel_matcher and rm NOOP from spec

* more asserts

* also this

* cleanup encode

* simplify live range

* fix idiv

* add Ops.INS to x86

* more changes

* more changes

* more changes

* fix

* fix

* fix

* fix

* print formatted assembly

* fix 8bit idiv?

* oops

* enable float16  and unaligned vector load/store

* actually no

* move x86 tests

* no more bool cast

* fix

* linter

* linter

* move X86Ops to x86.py

* fix vpbroadcast

* cleanups

* linter

* print correct reg names

* canonical max

* move max/min and add test

* support float16 vector load/store

* rm bad rewrite

* vpsrldq can't access memory

* regalloc takes renderer

* enable vector load/store on all dtypes

* more isel tests

* rm this for now

* a lot better

* fix

* fix

* fix

* deal with flags correctly

* fix

* enable gep noop rule

* fix

* fix

* fix

* add callee saved registers

* use Ops.CONST instead of X86Ops.IMM

* fix

* enable TUPLE_ORDER

* fix

* rm x86 code in linearizer

* fix

* fix

* fix

* move isa rewrites to codegen

* fix

* fix

* skip test_linearizer.py

* skip more tests

* fix

* fix for idiv/mod changes

* fix

* don't use fmadd if it duplicates fused op

* hacky

* fix

* cleanups

* cleanups

* fix

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2026-05-19 12:42:54 -07:00
Sachith Shetty
74567c1958
 		fix: pass input device to ONNX helper internal tensors (#16242) 
* fix: pass input device to onnx methods internal tensors

* test: onnx helper internal tensors use input device
 2026-05-19 11:16:33 -07:00
Christopher Milan
a178301dbe
 		PYTHONREMU: fix CDNA VOP3 conditional writes (#16258) 2026-05-19 13:31:31 -04:00
George Hotz
79ad8f3ef4
 		Merge branch 'master' into minigen 2026-05-18 22:14:50 -07:00
George Hotz
a120709671
 		tighten shape spec for broadcasting (#16206) 
* tighten shape spec for broadcasting

* use IndexError, not ValueError

* needs size
 2026-05-18 22:12:04 -07:00
George Hotz
8a25d18135
 		Merge branch 'master' into minigen 2026-05-18 20:40:14 -07:00
George Hotz
3f2d401464
 		all tests pass with NOOPT=1 (#16257) 
* all tests pass with NOOPT=1

* fix a few more

* noopt 100% pass

* noopt 100% pass
 2026-05-18 20:39:51 -07:00
chenyu
e694d7f222
 		more deviceless const prerequisites [pr] (#16256) 
* more deviceless const prerequisites [pr]

* remove that

* arange.contiguous -> arange.clone in tests

arange will become deviceless const soon, update tests where it needs to be a buffer
 2026-05-18 23:14:12 -04:00
chenyu
c1076ed56c
 		Tensor.device and UOp.device can be None (#16255) 2026-05-18 22:08:10 -04:00
George Hotz
474ec6e44d more amd fixes 2026-05-18 18:34:16 -07:00
George Hotz
7274239315 emu fixes 2026-05-18 18:23:10 -07:00
chenyu
d532b4f533
 		multi alu with deviceless const (#16251) 2026-05-18 19:31:53 -04:00
Christopher Milan
7515824a6d
 		ci: actually use clang-20, enable bfloat16 (#16249) 2026-05-18 19:06:43 -04:00
chenyu
754344087a
 		assign for deviceless const source (#16248) 2026-05-18 17:39:53 -04:00
chenyu
73e6b4963b
 		to and shard is noop for deviceless uop (#16247) 2026-05-18 16:11:10 -04:00
chenyu
db639ebe3e
 		deviceless const from UOp (#16243) 2026-05-18 14:14:12 -04:00
chenyu
5ae4dbd599
 		make slow tests faster (#16244) 2026-05-18 11:42:02 -04:00
chenyu
dcee90aa3f
 		remove requires_grad use in extra/examples (#16238) 
except the ones fed into optimizer
 2026-05-16 18:40:26 -04:00
chenyu
8631b6f17d
 		remove use of requires_grad in test/ (#16237) 2026-05-16 17:21:07 -04:00
chenyu
0ddc50d050
 		do not gate backward on requires_grad (#16230) 
DETACH is filtered in _deepwalk. instead of None, it gets 0 grad now
 2026-05-16 12:29:49 -04:00
qazal
ebcb7b7cc0
 		fp8 gemm tests with scale args (#16231) 
* update atol

* update fp8 path

* more work

* update profile.sh
 2026-05-16 20:47:58 +09:00
wozeparrot
2d48d7ab09
 		remove more invalid (#16227) 2026-05-16 02:52:27 -07:00
Christopher Milan
79c0ae5b89
 		metal: arch is GPU family (#16223) 2026-05-15 21:22:48 -04:00
chenyu
d62c1d83c0
 		remove Tensor.eye override (#16219) 
* remove Tensor.eye override

was only needed for requires_grad arg

* README
 2026-05-15 15:40:34 -04:00
chenyu
07a172dbbb
 		remove noop requires_grad_ calls (#16213) 2026-05-15 13:31:10 -04:00
chenyu
c6cf9e8f0c
 		remove test_svd_nonfull_5_5 (#16217) 
flaky, kinda overlap with test_svd_general
 2026-05-15 13:10:02 -04:00
chenyu
409bb0c9ad
 		requires_grad cannot be None (#16212) 
final goal is to remove requires_grad, first change the default to True, and don't allow None
 2026-05-15 02:01:04 -04:00
chenyu
a612b88abb
 		better assert when setitem a refed tensor (#16210) 
also decouple from requires_grad
 2026-05-14 23:40:29 -04:00
chenyu
a75c14f010
 		some setitem tests (#16209) 2026-05-14 22:36:25 -04:00
Christopher Milan
891a1ae7c2
 		onnx: remove dtype_fallback (#15717) 2026-05-14 22:06:57 -04:00
chenyu
ffa1aac7b1
 		gradient for STORE/AFTER ala clone (#16205) 2026-05-14 20:17:27 -04:00
chenyu
09096ea565
 		test_gradient_through_clone (#16203) 
backward through clone crashes now
 2026-05-14 19:26:47 -04:00
George Hotz
83ec66da34
 		fix a fastdiv edge case (#16199) 2026-05-14 13:12:18 -07:00
George Hotz
3b8cc31759
 		disable fast idiv by default, it's broken (#16197) 
* disable fast idiv by default, it's broken

* fix fast idiv tests
 2026-05-14 11:48:27 -07:00
C T
1b779a9058
 		add gelu approximate="none" (match pytorch) (#16162) 
* add gelu approximate="none" (match pytorch)

* lint

* pass through onnx Gelu approximate

* type annotate

* explicit math.sqrt

* keep tinygrad's gelu approximate="tanh" default
 2026-05-13 18:53:24 -07:00
b1tg
3c806ff406
 		clean up gguf (#16160) 2026-05-12 21:16:10 -07:00
chenyu
38d407fd58
 		simplify svd more (#16181) 
all the slowness is scheduling
 2026-05-12 23:48:22 -04:00
chenyu
32138c2418
 		svd to mixin (#16175) 2026-05-12 22:29:01 -04:00
chenyu
2172363be5
 		don't use Tensor indexing in svd (#16174) 
prepare mixin, also about 4X faster for 8x8 input
 2026-05-12 21:56:19 -04:00