github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-30 01:08:26 -05:00
Code Issues Packages Projects Releases Wiki Activity
6,570 Commits 210 Branches 14 Tags
b086584d643c21d2993f6ff4dce3eee3b76df256
Commit Graph

6570 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
George Hotz
b086584d64 add vcount to PtrDtype (#7388) 2024-10-30 10:43:54 +08:00
uuuvn
06a8700bfa Replace sqrtl (long double) with sqrt (double) for double (#7366) 2024-10-30 10:20:41 +08:00
gonutz
e7cbc6dc23 Fix ValueError in Yolo 8 example (#7387) 
Calling

    python3 examples/yolov8.py ./test/models/efficientnet/Chicken.jpg

used to result in this error

    ValueError: Calling nonzero on 0d arrays is not allowed.

Using np.atleast_1d makes sure we avoid a zero-dimension array.

Co-authored-by: gonutz <gonutz@fake.mail>
 2024-10-30 10:18:39 +08:00
chenyu
f389e1a8a0 test more special values for sin/cos/tan [pr] (#7386) 2024-10-29 21:13:37 -04:00
chenyu
33acbaeb24 reuse polyN in trig_poly float64 (#7385) 
similar speed, less alu (151 v.s. 154 per sine) and simpler, the power of 2 thing should probably be done in polyN if needed
 2024-10-29 20:45:56 -04:00
chenyu
6bf38c35e5 clean up transcendental frexp [pr] (#7384) 
also added some unit tests for frexp
 2024-10-29 18:51:37 -04:00
chenyu
99b82f5708 minor cleanup payne_hanek_reduction [pr] (#7383) 2024-10-29 17:59:18 -04:00
chenyu
f6abde95fa clean up Tensor._reduce (#7382) 
use make_tuple and self.ndim
 2024-10-29 17:23:57 -04:00
nimlgen
4ed2c40d48 qcom a bit cleaner (#7380) 2024-10-29 23:50:28 +03:00
chenyu
07ad6d20ed simpler commutative flipping condition (#7377) 
`x.src[1].tuplize < x.src[0].tuplize` implies `x.src[0] is not x.src[1]`

also renamed cc -> op
 2024-10-29 13:51:24 -04:00
chenyu
d3c192b056 Device method cleanup [pr] (#7375) 2024-10-29 12:49:47 -04:00
chenyu
f8a623b386 fix typing in Conv2d (#7374) 
* fix typing in Conv2d

`self.padding: Union[int, List[int]]` was wrong

* fix that
 2024-10-29 11:27:46 -04:00
qazal
7bd79f4922 pass viz render errors (#7369) 
* pass viz render errors

* pcall
 2024-10-29 22:48:27 +08:00
qazal
51c0c8d27e cachable small graph rewrite (#7371) 2024-10-29 22:28:13 +08:00
chenyu
9b81931a36 make_pair -> make_tuple [pr] (#7372) 
it's used more often as generic tuple, also removed the default 2.
 2024-10-29 10:27:39 -04:00
qazal
d803a9c7c8 global metadata try 2 (#7367) 2024-10-29 20:21:00 +08:00
George Hotz
2cfc7b6695 Index everywhere 2 (#7363) 
* indexing everywhere [pr]

* fix tests
 2024-10-29 19:29:40 +08:00
qazal
7149eabb34 assert set equality in TestTensorMetadata [pr] (#7364) 2024-10-29 19:29:29 +08:00
qazal
0ebdb136e8 revert metadata with graph_rewrite (#7353) (#7362) 
This reverts commit 540e4179e7.
 2024-10-29 19:16:31 +08:00
qazal
f2044cfb22 hotfix: if getenv("RUN_PROCESS_REPLAY") (#7361) 2024-10-29 18:51:29 +08:00
George Hotz
0af1212164 use assertEqual with new style uops [pr] (#7360) 2024-10-29 18:43:21 +08:00
George Hotz
0beb2d8f84 ptx indexing (#7359) 
* ptx indexing

* shorter

* fix load/store
 2024-10-29 18:29:44 +08:00
George Hotz
572499c71a add indexing to ops_python (#7358) 
* add indexing to ops_python

* fix image
 2024-10-29 18:11:03 +08:00
qazal
540e4179e7 global UOp to Metadata mapping + inverse DEBUG=2 metadata order [pr] (#7353) 
* add ctx.buf_metadata [pr]

* revert metadata insertion order

* lint rename
 2024-10-29 17:12:00 +08:00
George Hotz
2fdfcffe4c improve ci speed [pr] (#7357) 2024-10-29 17:00:35 +08:00
qazal
8fab7b21df everything is load (#7355) 
* everything is load

* rename to ops
 2024-10-29 16:47:33 +08:00
George Hotz
b647fa7514 rename MathTraits to maximum [pr] (#7356) 2024-10-29 16:43:04 +08:00
George Hotz
2bf55d8eda make ops more like tensor [pr] (#7352) 
* make ops more like tensor [pr]

* tensor is simple math trait

* no shifts
 2024-10-29 16:23:41 +08:00
George Hotz
3989bd2682 idiv + reciprocal [pr] (#7354) 
* idiv + reciprocal

* remove upcast from div

* fix docs
 2024-10-29 15:54:19 +08:00
Bhavya Gada
3419ae282d VIZ UI improvement: generic function for vscode opener (#7338) 
* generic function for vscode opener

* eslint

* shorter

---------

Co-authored-by: qazal <qazal.software@gmail.com>
 2024-10-29 15:02:04 +08:00
qazal
c03e1693fc shorter gate folding [pr] (#7350) 2024-10-29 14:49:32 +08:00
George Hotz
3e8225299c ext gate indexing (#7349) 
* ext gate indexing

* copy paste better
 2024-10-29 14:46:10 +08:00
Bhavya Gada
13ea4979d5 VIZ UI improvement: autoscroll kernel list when using arrow buttons (#7344) 2024-10-29 14:40:42 +08:00
George Hotz
d9d4dd6756 faster ci [pr] (#7348) 2024-10-29 14:01:44 +08:00
George Hotz
a5e0f59e41 move autogen to different CI runner [pr] (#7346) 
* move autogen to different CI runner [pr]

* balance a bit

* readme back there

* compile enet in autogen
 2024-10-29 13:35:22 +08:00
George Hotz
4cb236a495 index in cstyle (#7328) 
* index only in cstyle

* fix prefix dtypes

* fix tests

* global indexing

* Revert "global indexing"

This reverts commit 4d507e8abb.

* fix image

* fix image

* ptx tests

* fix CUDA dtype rendering
 2024-10-29 13:06:26 +08:00
George Hotz
f55c3dcff8 hotfix: bump ocelot 2024-10-29 12:46:24 +08:00
George Hotz
4fe1945df6 llvm if load (#7345) 
* llvm if load

* unneeded line

* local llvm CI
 2024-10-29 11:33:22 +08:00
chenyu
8625dd4eea minor changes reading ops.py (#7343) 2024-10-28 19:04:12 -04:00
chenyu
6021bf87f4 unify T = TypeVar("T") (#7342) 2024-10-28 18:43:44 -04:00
chenyu
293adc141a clean up get_shape [pr] (#7341) 
* clean up get_shape [pr]

aapi is literal false

* more
 2024-10-28 18:25:37 -04:00
chenyu
c398f2467c test uop mul min/max do not have nan in 0*inf (#7340) 2024-10-28 17:52:01 -04:00
chenyu
0843734927 clean up nan handling in transcendental (#7332) 
* clean up nan handling in transcendental

* skip remu crash
 2024-10-28 16:21:49 -04:00
Sieds Lykles
75dcd98e79 Fix calculation of vmin and vmax in multiplication when one src is negative and the other src has negative min and positive max (#7333) 
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-10-28 16:01:46 -04:00
chenyu
603fcc96f2 limit UOps.ALU min/max to non-float only (#7336) 
does this impact anything? some inf is incorrect now
 2024-10-28 15:34:19 -04:00
ignaciosica
32fa297e6c cleaner nan rendering (#7337) 2024-10-28 14:36:36 -04:00
qazal
00362a117c scheduler bfs renames [pr] (#7335) 2024-10-29 00:24:23 +08:00
qazal
d8820644e0 split preschedule from ast rewrite [pr] (#7334) 2024-10-28 17:45:09 +02:00
chenyu
6b0e8cb04f remove float_to_bits in transcendental [pr] (#7331) 
it's just bitcast, and removed the weird bits_to_float indirection
 2024-10-28 10:20:19 -04:00
qazal
b9b28e6883 viz stuff [pr] (#7330) 
* viz stuff [pr]

* button
 2024-10-28 21:46:18 +08:00