George Hotz
d1f6701eb7
hotfix: lower amd threshold + improve block reorder test
2025-04-22 20:44:29 +01:00
nimlgen
db51133537
rename HWInterface -> FileIOInterface ( #9989 )
...
* rename HWInterface -> FileIOInterface
* ugh
2025-04-22 22:18:57 +03:00
George Hotz
c1539b0319
putting add first orders loads as expected ( #9991 )
2025-04-22 20:12:05 +01:00
nimlgen
bd580d8ea4
hcq: use mmio interface in nv ( #9986 )
...
* hcq: start mmio interface
* allow double cast
* revert
* faster?
* simpler, not needed more now
* dd
* types
* fix
2025-04-22 21:58:12 +03:00
George Hotz
feee6986c9
faster block reorder ( #9990 )
...
* faster block reorder [pr]
* that shouldn't change order
* key just in sorted
* ind
2025-04-22 19:18:57 +01:00
qazal
6cb2d18c03
refactor schedule linearize to defaultdict [pr] ( #9984 )
...
* refactor schedule linearize to defaultdict [pr]
* skip that
* don't need .get
2025-04-23 00:00:23 +08:00
chenyu
9e5e371999
make DISABLE_COMPILER_CACHE a ContextVar [pr] ( #9983 )
2025-04-22 10:32:54 -04:00
qazal
bbc324f5dc
remove CAST_AFTER_EXPAND ( #9980 )
2025-04-22 21:06:11 +08:00
George Hotz
c519b553db
non recursive toposort is 2x+ faster ( #9979 )
...
* non recursive toposort is 2x+ faster
* don't change the order
2025-04-22 13:59:38 +01:00
qazal
0d9014d021
place create_ast last, type_verify in the end (once) [pr] ( #9977 )
2025-04-22 20:15:23 +08:00
chenyu
fb89d9a584
retinanet eval combine output on GPUS[0] ( #9966 )
...
eval 35 sec -> 20 sec. it was spending 13 seconds assembling output tensor on CPU backend. GPUS[0] seems to have enough memory, otherwise we can lower EVAL_BS
2025-04-22 07:43:51 -04:00
qazal
7b55846e08
prep STORE UOp creation for multi output [pr] ( #9975 )
...
* prep STORE UOp creation for multi output [pr]
* test_multioutput_ast
2025-04-22 19:34:52 +08:00
George Hotz
e358e0a0c6
move metadata set to tensor [pr] ( #9976 )
...
* move metadata set to tensor [pr]
* only track that in tensor.py
2025-04-22 12:30:35 +01:00
qazal
f6271515fe
refactor UOp.st [pr] ( #9973 )
2025-04-22 18:46:56 +08:00
George Hotz
f5dc70c624
microbenchmarks + micro speed ups ( #9972 )
...
* microbenchmarks
* forgot the ubenchs
* clean up type verify
2025-04-22 11:30:46 +01:00
qazal
1cf4e24ca5
fix kernelize usage with pm_gradient ( #9953 )
...
* fix kernelize usage with pm_gradient
* remove that
2025-04-22 17:26:05 +08:00
deftdawg
32bbff942c
amd: add nbio 7.2.0 for some rdna2 ( #9964 )
...
* - Updated of #9700 which fixes #9665 but for the Steam Deck which was erroring on NBIO 7.2.0
* unrelated change
---------
Co-authored-by: nimlgen <138685161+nimlgen@users.noreply.github.com >
2025-04-22 12:10:48 +03:00
Ignacio Sica
0e79aee706
use_tensor_cores bugfix ( #9969 )
2025-04-21 22:58:17 -03:00
chenyu
5294c32279
dev scripts for retinanet ( #9968 )
...
also BASE_DIR -> BASEDIR for consistency, and move wandb up a bit for more accurate timing
2025-04-21 17:54:56 -04:00
nimlgen
4340197132
am: download fw from web ( #9956 )
...
* am: download fw from web
* tested
* link works
* default to web
* this is default
* not used
2025-04-21 23:26:33 +03:00
nimlgen
7244ca863c
am: fix double read of sdma fw ( #9965 )
2025-04-21 23:04:34 +03:00
uuuvn
b35f94b6ec
Don't hardcode default CLOUDDEV ( #9935 )
2025-04-21 18:46:55 +01:00
Francis Lata
defa1e77f6
get the proper dataset count ( #9962 )
2025-04-21 12:11:37 -04:00
qazal
36ed3c3253
fix kernelize with VIEW children ( #9961 )
2025-04-21 23:38:46 +08:00
uuuvn
757533cbe6
Less verbose cloud multiprocessing start ( #9960 )
...
The set name before starting part used to be required for #9935 when
CLOUDDEV was a global variable, now just readability improvement
2025-04-21 16:19:54 +01:00
Francis Lata
d7e247f329
RetinaNet INITMLPERF support ( #9950 )
...
* fixes to make fake data work
* fix eval beam
* fix merge issue
2025-04-21 10:32:05 -04:00
kamilisjon
014f870733
rm ( #9959 )
...
Co-authored-by: KamilisJonkus <kamilis.jonkus@agmis.com >
2025-04-21 15:23:45 +01:00
chenyu
f68c7041c4
doc fix is_floating_point dtype.float -> dtypes.float ( #9958 )
2025-04-21 09:23:59 -04:00
akhuntsaria
2d423e6737
fix assertion message for supported device in export_model ( #9957 )
2025-04-21 09:23:44 -04:00
ttomsa
783a191925
rm mul from _masked_setitem ( #9951 )
2025-04-21 06:41:50 -04:00
nimlgen
46469f00a2
am: tiny changes in psp load ( #9952 )
2025-04-21 11:52:02 +03:00
qazal
0bee225a58
Tensor.kernelize docs ( #9946 )
...
* Tensor.kernelize docs
* syntax
* test_kernelize_bw
* Tensor.kernelize docstring
* pruning
* tiny details
* details 2
* becomes_map terminology
* more changes to becomes
2025-04-21 16:34:03 +08:00
Francis Lata
ea4cb2c715
small cleanups ( #9947 )
2025-04-20 20:33:20 -04:00
qazal
e8910540f6
Kernelize can be called multiple times on a Tensor ( #9949 )
...
* Kernelize can be called multiple times on a Tensor
* add (failing) test_kernelize_bw
2025-04-21 06:28:47 +08:00
qazal
1d90be2cff
match kernelize API in process replay ( #9948 )
2025-04-21 05:23:41 +08:00
qazal
343a5eb588
dedup assigns in grouper VIZ name function [pr] ( #9942 )
2025-04-20 21:42:25 +08:00
qazal
e20ef7196a
Tensor.kernelize ( #9845 )
...
* add kernelize
* remove that
* kernelize returns self
* update abstractions2.py
* kernelize in test_schedule
* temp: assert BUFFER_VIEW's existence
* ASSIGN must have a buffer or subbuffer target
* assert and shrink
* fix
* padded setitem
* var
* toposort once
* extra
* base_buffer
* end with BUFFER_VIEW
* setitem for disk
* test_setitem_becomes_subbuffer
* mul slice test
* torch backend fix 1
* non-deterministic
* keep subbuffer
2025-04-20 20:53:49 +08:00
qazal
dd16087f62
fold double ASSIGN to same target ( #9941 )
2025-04-20 19:06:38 +08:00
qazal
9a9aba4cd5
setitem tests (some failing) from kernelize ( #9940 )
2025-04-20 18:47:55 +08:00
chenyu
6c30948df6
hand_coded_optimizations returns list[Opt] [pr] ( #9938 )
...
new api looks like `k.apply_opts(hand_coded_optimizations(k))`
2025-04-19 20:26:59 -04:00
chenyu
720f20865b
remove required_optimizations ( #9848 )
2025-04-19 16:51:16 -04:00
qazal
218e01833d
update scheduler section for abstractions2.py [pr] ( #9927 )
2025-04-19 12:09:14 +03:00
chenyu
3fdba48fc7
update bert green and README ( #9934 )
...
submission candidate
2025-04-18 21:21:28 -04:00
George Hotz
b359125ebf
rewrite the linearizer ( #9885 )
...
* random speedups [pr]
* speeding up linearizer
* test_gemm passes
* progress
* test_gemm passes
* working
* simpler
* blockstart unneeded
* simpler
* bugfix
* work
* don't compare
* faster
* progress
* cleanups
* work
* cleanups
* working
* reorder
* name is dumb
* fix tests
* lin2 works
* clean ctx
* mostly bottom up
* passes
* same speed now
* new lin is faster
* dedup
* lines and tuples
* track that
* lin
* revert that
* tests should pass
* merge siblings
* cleaner expression
* only lin2
* finally, some speed
* simpler
* fix unmergables with blockends
2025-04-18 22:35:40 +01:00
Ignacio Sica
023b1c28a2
test_tensor_cores_padded refactor (#9724 )
...
* set pad t 3 for amd padded tc test
* change pad for amd regardless CI
* test tc padded uops and correctness separately
* add test_tensor_cores_padded_uops test to ci
* remove redundant chack for amd device
* cleanup
2025-04-18 17:05:54 -03:00
Ignacio Sica
afff82ba0f
fix ptx linearizer bug [pr] ( #9926 )
...
* fix ptx bug
* align 16
* revert align because it breaks pr
* smallest diff that fixes ptx bug
2025-04-18 13:48:43 -03:00
chenyu
617b45748f
fuse embedding for bert on red ( #9925 )
...
also updated BEAM param and use AMD driver for actual run. 535ms step
2025-04-18 07:20:25 -04:00
qazal
b58decac0c
fix diamond assigns before mapping tensors UOps to assigns ( #9855 )
...
* keep tensor_map until diamond assign fixup
* ctx
2025-04-18 14:17:43 +03:00
qazal
a37d921917
get name from SINK in process replay ( #9924 )
...
* get name from SINK in process replay
* space
2025-04-18 13:51:11 +03:00
George Hotz
aa98aff4cd
don't use ops name, just keep sink ( #9922 )
...
* don't use ops name, just keep sink
* fix test
* endif sink
2025-04-18 08:59:18 +01:00
