George Hotz
8137e0b400
typo in __imatmul__ -> cleanup
2023-02-23 20:13:44 -08:00
George Hotz
4c54adeb18
adding tuples is fine
2023-02-23 19:42:48 -08:00
George Hotz
661812ffef
don't ignore type
2023-02-23 19:38:52 -08:00
George Hotz
10c6ccf7e0
better order in Tensor class
2023-02-23 19:33:37 -08:00
George Hotz
f2ca81c66d
reduce cleanup
2023-02-23 19:19:49 -08:00
George Hotz
8446f77fe9
image_conv2d_decorator
2023-02-23 18:49:23 -08:00
George Hotz
82caa2d5b7
MULACC -> FusedOp
2023-02-23 18:17:57 -08:00
George Hotz
758515dcc0
conv2d is an hlop ( #589 )
...
* conv2d is an hlop
* shorter conv
* KOPT=-1
* alt imp
* MULACC
* smarter mulacc
* pop conv
* 7x7 -> 5x5
* didn't fix, that's not going to work
* this is faster and matches old behavior
* oh, non lazy just won't work with mulacc
* mulacc in torch
* bool types were creeping in
* optimizer is actually better with hlop conv
* fix pushing permutes issue
* refactor einsum_mulacc
* fix up readme
* update readme
* _image_conv2d
* fix bias addition location
* pushing permutes gets back to 200 kernels
* conv cleanup
* disable hlop conv
* don't hide that in helpers
2023-02-23 17:52:31 -08:00
Sohaib
8835df7a5c
upgrade onnx to 1.13.0 ( #588 )
...
- remove protobuf from direct dependencies
- replace deprecated mapping.TENSOR_TYPE_TO_NP_TYPE
Co-authored-by: Sohaib Errabii <sohaib.errabii@ipops.io >
2023-02-23 13:59:23 -08:00
calledit
81f7c6800a
Added info on simdgroup availability ( #586 )
...
* Add info on simdgroup availability
* "osx" not "os x"
* Update metal_matmul.py
* Update metal_matmul.py
2023-02-23 13:59:02 -08:00
George Hotz
d22e19536b
onnx: support low quality Resize. stuck on ConvTranspose will have to wait for convless
2023-02-23 09:05:23 -08:00
George Hotz
ab3a2ae9a2
fix test_resnet in onnx now that maxpool works
2023-02-23 08:41:47 -08:00
George Hotz
fd6082dcef
support all _pool2d. conv will eventually be an hlop
2023-02-23 08:19:47 -08:00
George Hotz
94ccab941e
compile_tensorflow: no cast required
2023-02-22 21:14:21 -08:00
George Hotz
135d0ddb78
compile_tensorflow: read weights from disk
2023-02-22 21:12:35 -08:00
George Hotz
0615dcffe7
compile_tensorflow: save the weights
2023-02-22 21:05:45 -08:00
George Hotz
c537fd0614
compile_tensorflow: add initialize and tests
2023-02-22 20:50:53 -08:00
George Hotz
dc914cde50
compile_tensorflow
2023-02-22 20:08:58 -08:00
George Hotz
10ba96029c
device: ImportError -> Exception
2023-02-22 19:48:59 -08:00
George Hotz
76b4d0577d
yolov8 works up to the MaxPool
2023-02-22 19:32:13 -08:00
Mischa Untaga
5190784cbb
Fix Tensor random functions determinism with same seed ( #580 )
...
* fix Tensor random functions determinism with same seed
* long lived rng
* TIL ClassVar typing
2023-02-22 19:08:43 -08:00
George Hotz
c8d89eb20e
avg/max pool strides
2023-02-22 18:00:48 -08:00
George Hotz
ac8daaeea5
refactor device, credit martinloretzzz
2023-02-22 17:29:40 -08:00
George Hotz
a3ddc1d484
CLANG doesn't support float4
2023-02-22 10:41:15 -08:00
George Hotz
c4c2c28738
a sustainable approach to float4 ( #582 )
...
* a sustainable approach to float4
* can_float4
* fix tests
* fix float4
* delete dead code
* types and minor cleanup
2023-02-22 09:45:08 -08:00
George Hotz
50e541174c
fix image copyout
2023-02-22 07:27:13 -08:00
George Hotz
628ce067a1
add tests to mypy
2023-02-22 07:07:38 -08:00
George Hotz
104c3c5e73
oops, forgot that debug
2023-02-22 06:58:27 -08:00
George Hotz
c5e2126d49
move DEBUG to helpers
2023-02-22 06:52:11 -08:00
Mischa Untaga
14bb2c40a2
Fix yolov3 example ( #577 )
2023-02-21 09:24:00 -08:00
Martin Loretz
8550b3e168
Fix kernel cache key ( #570 )
2023-02-21 03:53:07 -08:00
George Hotz
66b4b3bdd3
clang: linux needs fPIC
2023-02-20 21:09:33 -08:00
George Hotz
4f4b7d05a7
clang needs -lm and is very slow
2023-02-20 21:02:04 -08:00
George Hotz
4126bf2982
remove six (hopefully not needed)
2023-02-20 20:44:23 -08:00
George Hotz
efcb3f0cdd
fix metal dep
2023-02-20 20:43:32 -08:00
Connor Henderson
9670bf1fd1
Add unsqueeze ( #574 )
...
* Add unsqueeze
* remove UNSQUEEZE from llops part of readme
* make it an hlop
2023-02-20 20:14:59 -08:00
George Hotz
cfad2902d5
clang: that might have been getting GCed
2023-02-20 20:13:05 -08:00
Diogo
506970414a
added metal packages to setup and release metal buffers after del ( #571 )
2023-02-20 20:01:07 -08:00
George Hotz
d9fa47ecc9
use the TinyJit in the efficientnet runner, 200ms -> 20ms
2023-02-20 19:58:16 -08:00
George Hotz
714bf4b108
clang backend ( #572 )
...
* start clang backend
* mostly working
* no group for reduce w clang
* it compiles
* compiles
* a11y
* minor fixups
* formatting
* add a test
* rename test
2023-02-20 18:18:18 -08:00
George Hotz
353e81fa8f
need hand opt with IMAGE==2
2023-02-20 10:18:19 -08:00
George Hotz
b925d05e22
no networkx is okay
2023-02-20 09:57:12 -08:00
George Hotz
4d232c7c95
optional networkx + DEBUGCL=2
2023-02-20 09:50:46 -08:00
George Hotz
8b0082540b
openpilot compile cleanups
2023-02-20 09:16:03 -08:00
George Hotz
ea13504f35
fix METAL_XCODE
2023-02-19 20:02:12 -08:00
George Hotz
bbfec2fde7
8.46 TFLOPS
2023-02-19 13:21:25 -08:00
George Hotz
1ba847963d
reshape and retain metal_matmul
2023-02-19 13:07:23 -08:00
George Hotz
60008e55cd
sick of that failing
2023-02-19 13:05:37 -08:00
George Hotz
0b3f686530
Good fast triton ( #567 )
...
* runtime fixups
* uints and printbufs
* uints don't work
2023-02-19 12:21:55 -08:00
Jacky Lee
cb679cd051
Fix weight initialization ( #566 )
...
* Fix weight initialization
* Use scaled_uniform in serious_mnist
2023-02-19 11:25:29 -08:00
