deepcrayon
/
tinygrab
1
0
Fork 0
Code
2,980 Commits
1 Branch
4 Tags
68 MiB
Commit Graph

20 Commits (bbeba8ec856b1cae519eba6a44bd1b799a7eccbb)

Author SHA1 Message Date
George Hotz bbeba8ec85
use default dict for external_model_benchmark (#2592) 
* device default

* Device.DEFAULT

* half max for cuda

* CUDA_INCLUDE_PATH

* closer to working

* cuda fixups

* Update ops_cuda.py
 2023-12-03 15:25:43 -08:00
chenyu a58736fdf1
print DEBUG=2 stats after stats update (#2590) 2023-12-03 17:13:37 -05:00
George Hotz d6b404ac11
No dtype alloc (#2570) 
* fix all allocs

* improve docs

* ugh fix fake alloc
 2023-12-02 13:29:40 -08:00
George Hotz 5068e99d18
refactor to remove extra kernel params (#2563) 
* refactor to have compiled kernel

* bugfixes

* docs/beautiful.py

* revert that

* fix tests
 2023-12-02 00:32:25 -08:00
George Hotz 6733425095
lower schedule (#2559) 
* lower schedule

* remove RAND, and don't put load in the JIT yet

* better fix for that test
 2023-12-01 19:17:46 -08:00
George Hotz 9c306be282 better name for fast path 2023-12-01 15:32:47 -08:00
chenyu 67f4e03724
rewrite 0 size loadop into a CONST (#2556) 
* rewrite 0 size loadop into a CONST

* check alloc size

* EMPTY is better

* Revert "EMPTY is better"

This reverts commit 574fe0f9ed28f1b97da5a81afdfd2cd5d9a94ff9.

* no ast is created

* fix test
 2023-12-01 18:29:06 -05:00
George Hotz 4447188051 gate METAL_FAST_LOAD 2023-12-01 15:28:40 -08:00
George Hotz eb595588bb device.py cleanups for -1 line 2023-12-01 14:59:33 -08:00
George Hotz f5de21e753
fast path for copy (#2548) 
* fast copy

* ruff first

* flat_mv on malloc

* order + webgpu test
 2023-12-01 11:34:47 -08:00
George Hotz 12fa846122
zero copy (#2531) 
* zero copy

* zero copy test

* loads coder in milliseconds

* zero copy for cpu and torch

* src_from_buffer is None

* SLOW_METAL_COPY there
 2023-11-30 18:38:41 -08:00
George Hotz 2c363b5f0b
new style device (#2530) 
* cpu tests pass

* torch works

* works

* metal works

* fix ops_disk

* metal jit works

* fix openpilot

* llvm and clang work

* fix webgpu

* docs are rly broken

* LRU works on metal

* delete comment

* revert name to ._buf. LRU only on Compiled

* changes

* allocator

* allocator, getting closer

* lru alloc

* LRUAllocator

* all pass

* metal

* cuda

* test examples

* linearizer

* test fixes

* fix custom + clean realize

* fix hip

* skip tests

* fix tests

* fix size=0

* fix MOCKHIP

* fix thneed

* copy better

* simple

* old style metal copy

* fix thneed

* np reshape

* give cuda a device
 2023-11-30 17:07:16 -08:00
chenyu d210f6a786
minor device.py cleanups (#2510) 2023-11-29 18:16:25 -05:00
George Hotz 889acefe85
Support weird loads in Image (#2498) 
* image support weird loads

* umm, that was always wrong

* openpilot compile fails with a weird error

* image test passes

* we have valids now

* clean that up

* no more required opts

* add fastvits test, fix bug

* minor cleanups
 2023-11-29 08:30:46 -08:00
George Hotz 5629fc368c
Use Buffer.STORE at the end of ASTs (#2494) 
* work

* store broken

* interpreteds work

* this passes

* symbolic cpu

* fix tests

* fix opt tests

* images fail

* fix InterpretedFlopCounter

* stupid hack for images
 2023-11-28 20:11:37 -08:00
George Hotz 756b01f46f
why were these ever called buffer (#2483) 2023-11-27 21:02:07 -08:00
George Hotz 9e07824542
move device to device.py (#2466) 
* move device to device.py

* pylint test --disable R,C,W,E --enable E0611

* fix tests
 2023-11-27 11:34:37 -08:00
George Hotz b9eae94ae9 move Device back into lazy 2023-02-11 11:26:53 -08:00
George Hotz 78795e3507 reduce line count by simplifying DeviceBuffer 2023-02-09 12:52:14 -06:00
George Hotz 0ac3286af0 factor out Device 2023-02-07 16:08:20 -06:00