1bc52c60df
fix: minor tweaks to view ( #1842 )
...
Co-authored-by: Roelof van Dijk <[email protected]
2023-09-10 15:55:57 -07:00
nimlgen and GitHub
31fca43706
kopt works with local+grouped reduce and tests ( #1824 )
2023-09-09 13:22:09 -07:00
George Hotz and GitHub
6100d7425f
add 2 to locals, uops debug 5 ( #1782 )
2023-09-05 19:44:43 -07:00
Adrian Kretz and GitHub
3473c9e88d
Metal conv tensor cores ( #1696 )
...
* Benchmark 5x5 conv kernel which is optimized
* Use Metal tensor cores in 2d convs
2023-09-04 15:14:46 -07:00
David Hou and GitHub
3151d91f6e
3x3 winograd convs ( #1675 )
...
* winograd
* simplify local groups code
* comment
* respects self.opts.has_local
* always simplify ones
* make mypy happy
* move reshape, WINO flag
* wino flag, simple forward backward test for wino
* extra wino test
* merge oops
* comments
* axis_needs_valid -> axis_is_masked
* don't delete needs_valid (it's unused though)
* make linter happy
* make linter happy
* smaller test
* change number
* make wino tests very small
2023-09-03 07:29:43 -07:00
nimlgen and GitHub
a96e54d8bb
search for grouped reduces ( #1732 )
2023-09-01 14:21:10 -07:00
George Hotz and GitHub
5c403d43b9
New >3 indexing ( #1729 )
...
* move reindexing into linearizer
* get_grouped_dims
* don't limit for clang
2023-08-31 21:24:15 -07:00
George Hotz and GitHub
c18a497dde
minor global dim cleanup ( #1724 )
2023-08-31 12:23:39 -07:00
62536d6000
perf: use enumerate where possible ( #1692 )
...
Co-authored-by: Roelof van Dijk <[email protected]
2023-08-30 10:41:51 -07:00
George Hotz and GitHub
fdd7f282cb
Reenable tensor cores for self-hosted Mac CI ( #1717 )
...
* debug 5 matmul
* allow tensor cores in CI
* tensor cores on arm64
* put debug back
2023-08-30 07:53:04 -07:00
George Hotz and GitHub
d37d092c14
split linearizer into 3 files ( #1654 )
2023-08-23 14:58:47 -07:00
George Hotz and GitHub
643cbdfd50
make embedding and GPT-2 fast ( #1631 )
...
* make embedding fast
* jit more, variable shape support
* print mem bw
2023-08-22 15:14:38 -07:00
geohot
696e4d20a1
fix KOPT=2 with variable shape
2023-08-22 11:34:34 -07:00
109100656f
refactor: no len if it is not needed ( #1598 )
...
Co-authored-by: Roelof van Dijk <[email protected]
2023-08-21 14:06:32 -07:00
George Hotz and GitHub
739f327d2d
Shorter ( #1582 )
...
* deleting lines
* remove insert dims
* if statement is never hit
* bug fixes
2023-08-20 08:12:16 -07:00
George Hotz and GitHub
e464442adf
WMMA for 7900XTX ( #1563 )
...
* go
* hip no LRU
* work
* works
* 16 TFLOPS
* 29 TFLOPS
* 30 TFLOPS
* never mind, it's 60 TFLOPS
* fix metal WMMA
* put hip alloc back
2023-08-19 09:07:23 -07:00
chenyu and GitHub
11dd9b1741
symbolic codegen and exec ( #1552 )
...
* symbolic codegen and exec
* fix and add test
* no sketchy
* merge_dicts type
* dtypes._arg_int32
2023-08-16 14:43:41 -07:00
George Hotz and GitHub
84c430355e
fix backends for new style ( #1443 )
...
* fix backends for new style
* fix method cache
* fix fakeless
* llvm blacklist
* fix kernel optimizer
2023-08-05 11:07:04 -07:00
wozeparrot and GitHub
a367f71fea
fix: don't put kernels into cache when optimizing ( #1409 )
2023-08-02 18:17:16 -04:00
George Hotz and GitHub
bfbb8d3d0f
fix ones, BS=2 stable diffusion, caching optimizer ( #1312 )
...
* fix ones, BS=2 stable diffusion
* caching optimizer
* print search time
* minor bug fix
2023-07-21 09:55:49 -07:00
George Hotz and GitHub
9746f6d094
move hand coded optimizer ( #1310 )
...
* move hand coded optimizer
* llvm can optimize
* fix llvm
* save linearizer
2023-07-21 07:53:12 -07:00
geohot
d963024a13
optimizer small fix: return if there's nothing to optimize
2023-07-20 16:57:30 -07:00
George Hotz and GitHub
9dffc9ba23
Use nevergrad to optimize kernels (try 2) ( #1301 )
...
* nevergrad try 2
* touchups
* no ones
* opt fixup
* cleanups
* touchup
* make new optimizer file
2023-07-20 16:46:45 -07:00
geohot