commaai
/
tinygrad
mirror of https://github.com/commaai/tinygrad.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
274 Commits 10 Branches 0 Tags 52 MiB
Commit Graph

274 Commits

Author SHA1 Message Date
George Hotz df64658a2c weee, opencl tests in CI 2020-11-10 10:04:45 -08:00
George Hotz d47a128812 pocl 2020-11-10 10:02:13 -08:00
George Hotz c05401a9ca sudo maybe 2020-11-10 09:53:49 -08:00
George Hotz 09bc8eddfe clinfo 2020-11-10 09:51:38 -08:00
George Hotz 58e703d099 fix tests 2020-11-10 09:49:19 -08:00
George Hotz 23405cec43 intel opencl 2020-11-10 09:41:40 -08:00
George Hotz 33090c4b0d install more 2020-11-10 09:34:56 -08:00
George Hotz a52590e76c cpu opencl maybe 2020-11-10 09:32:54 -08:00
George Hotz f513302955 refactor profiler 2020-11-10 07:31:16 -08:00
adamritter f27628b21c
No separate pad2d kernel needed (#99) 
Co-authored-by: holonomicjl <58403584+holonomicjl@users.noreply.github.com>
 2020-11-10 06:47:53 -08:00
George Hotz 2d4a5d5950 readme 2020-11-10 01:27:04 -08:00
George Hotz 6e6bcbe5f2 shapes on backward 2020-11-10 01:23:31 -08:00
Ryan Neph 56f71ae8e5
Cleanup (#96) 
* init GPU supsample retbuf to 0

* reduce GPU kernel source lines

ref: #94
 2020-11-10 01:20:04 -08:00
George Hotz 55012d21bb debug in backward pass too 2020-11-10 01:19:52 -08:00
George Hotz 5d1985312c miniprofiler is real 2020-11-10 01:05:29 -08:00
George Hotz c76a20b4be 4s and 7s work 2020-11-10 00:54:17 -08:00
George Hotz ae0cd17c2d debug is env var, and simpler faster pad2d 2020-11-10 00:42:23 -08:00
George Hotz f7d10d5639 DEBUG flag 2020-11-10 00:36:59 -08:00
George Hotz 6a56d5d030 remove pyopencl 2020-11-09 23:58:35 -08:00
George Hotz 943ff6490d logic error okay too 2020-11-09 23:55:21 -08:00
George Hotz 29b22d117f will gpu tests work? 2020-11-09 23:51:29 -08:00
George Hotz abbf0d1328 cleanup logsoftmax 2020-11-09 23:47:04 -08:00
George Hotz aeb90226a8 reduce lines with reduce_op 2020-11-09 23:36:44 -08:00
George Hotz 55c914912d minor cleanup 2020-11-09 23:14:59 -08:00
George Hotz d41ad2bf37 uint2 saves lines 2020-11-09 23:11:20 -08:00
George Hotz 8b23033fa9 support all the enet sizes 2020-11-09 18:04:16 -08:00
George Hotz 9db95ab942 fix enet padding 2020-11-09 17:56:57 -08:00
George Hotz 866b759d3b match torch api for pad2d 2020-11-09 17:48:56 -08:00
George Hotz daf073535f new -> zeros 2020-11-09 17:31:52 -08:00
Ryan Neph 16d564a53c
finish unsupporting strided pool, add global avg pool test (#92) 2020-11-09 17:31:22 -08:00
George Hotz 7ac1b163a5 add backward to enet train 2020-11-09 16:05:52 -08:00
George Hotz 8ca9c0205f train_efficientnet is broken still 2020-11-09 16:01:16 -08:00
George Hotz 870b84a893 test pad2d backward on GPU 2020-11-09 15:50:43 -08:00
adamritter b541c05d88
Pad2d backward pass on GPU (#89) 
* Pad2d backward pass on GPU

* Faster Pad2D GPU backward pass (no zeroing needed)

* Fix out of bounds error

* Don't save prg

Co-authored-by: holonomicjl <58403584+holonomicjl@users.noreply.github.com>
 2020-11-09 15:49:37 -08:00
George Hotz e46d122f65 not supporting stride 2020-11-09 15:06:58 -08:00
Ryan Neph c21c2a0b62
revert b0c0c5d: Strided Pool funcs (#74) (#87) 
Strided CPU Pooling was introduced but assumes small kernel size
(<=(10,10)), but efficientnet.py feeds kernel_size=(112,112).

This causes a huge array buffer allocation in stack_for_pool() that
hangs inference for a long time or until system OOM.

Revert CPU Pooling for now, and re-introduce #74 later with a new
global-average-pooling op that can be used instead of avgpool2d with
large kernel size for efficientnet inference.

Co-authored-by: Ryan Neph <ryanneph@google.com>
 2020-11-09 14:58:18 -08:00
George Hotz 53157fb876 add back scale 2020-11-09 10:20:56 -08:00
George Hotz 3ffbd47335 Revert "Revert "pygame is fine, cv2 can also do the trick (#79)" (#85)" 
This reverts commit 6b982621f8.
 2020-11-09 10:18:48 -08:00
George Hotz 6b982621f8
Revert "pygame is fine, cv2 can also do the trick (#79)" (#85) 
This reverts commit e7f2f43331.
 2020-11-09 10:03:38 -08:00
dustcollector12 e7f2f43331
pygame is fine, cv2 can also do the trick (#79) 
* pygame is fine, cv2 can also do the trick

* retimg and copy constructor not needed

* shape is missing without copy constructor

* retimg put back

* addressing capture buffering
 2020-11-09 10:02:06 -08:00
Ryan Neph 7e515308a5
label op subtests by params (#83) 2020-11-09 06:25:06 -08:00
Ryan Neph 5bedf566d1
tests should use rtol unless special case (#82) 2020-11-08 17:25:11 -08:00
Ryan Neph 04b9312a34
Fix GPU Pooling bug at boundary + better Pooling test coverage (#81) 
* fixed Pooling bug

* Clarify Pooling tests
 2020-11-08 17:25:01 -08:00
niclaswue c57b1b9e7d
deleted unnecessary import in utils (#78) 2020-11-08 15:55:16 -08:00
Ryan Neph b0c0c5d0d6
strided Pool funcs (#74) 
* *Pool2D GPU forward supports stride

* kernel_size from ctx instead of saved_tensors

* *Pool2D CPU forward supports stride

* update ctx.stride properly
 2020-11-08 11:45:55 -08:00
George Hotz 06504a5824 bump version 2020-11-08 09:34:07 -08:00
ziofil db3eccc16b
implemented backward for Pad2D & test (#73) 2020-11-07 21:58:42 -08:00
George Hotz 75d69e956f readme more 2020-11-07 21:58:20 -08:00
Dimitar Vagalinski 35a5c82a2a
done as he said (#71) 2020-11-07 18:28:39 -08:00
Ryan Neph 5265f6c578
add AvgPool2D backward pass on GPU (#68) 2020-11-07 12:27:29 -08:00