geohot
964df8ec0e
move devectorizer remains into coalese
2026-07-04 15:27:12 -07:00
George Hotz and GitHub
a145fdce4b
remove the .gep method ( #16867 )
2026-07-04 14:07:09 -07:00
wozeparrot and GitHub
971800b46a
support none block divisible embed size ( #16865 )
2026-07-04 13:46:47 -07:00
George Hotz and GitHub
3fd6f3d28c
no more GEP ( #16866 )
2026-07-04 13:42:28 -07:00
George Hotz and GitHub
d76bcbaa02
deletions after new codegen ( #16864 )
...
* deletions after new codegen
* delete the expander
2026-07-04 13:12:11 -07:00
George Hotz and GitHub
c7e7687bd3
new codegen, try 3 ( #16781 )
...
* new codegen, try 3 [pr]
* reduce
* diffs from cg2
* minor fixes
* mergable
* new local buffers
* fixes
* correct fix
* add barrier, and image is after that
* fix get
* wmma work
* devec wmma
* fixes for wmma
* slightly more flexible
* fix tensor cores
* fix linter
* tests
* remove invalid?
* remove extra
* wmma merge
* flip reshape/permute on wmma
* broadcast_binary None
* update symbolic for vecless
* x86 fix
* fix custom llama kernel
* is_ptr hack
* fix pre-commit
* fix wmma
* simpler
* fixes
* better hreduce
* move reduce axes
* all wmma tests pass
* allow stacked wmma
* permute cleanups
* late permute
* correct permute flip
* fix group for reduce
* fix wmma permuted
* all pass
* mop cleanup
* new test
* push permute/reshape
* revert that
* half
2026-07-04 12:55:44 -07:00
qazal and GitHub
07f7383d29
llama: remove unused bf16 assembly gemm ( #16859 )
...
* only hk bf16 gemm
* rm asm gemm
* more cleanup
* half isn't supported in asm gemm anymore
* more test edits
* unused
* remove TestMagicGu
* uop gemm is still tested
* minimal diff
2026-07-04 18:41:54 +09:00
sirhcm and GitHub
353d8f1e13
use rusticl in ci ( #16852 )
2026-07-03 23:58:58 -04:00
chenyu and GitHub
709251e22d
more generic fold_add_divmod_recombine [pr] ( #16854 )
2026-07-03 22:00:59 -04:00
George Hotz and GitHub
41d6731bfd
reduce removes ones ( #16847 )
...
* reduce removes ones
* test changes
* lil clean
2026-07-03 13:10:08 -07:00
chenyu and GitHub
5fb3cfb9bc
remove early THREEFRY const folding [PR] ( #16850 )
...
it folds automatically once decomposed
2026-07-03 15:00:35 -04:00
chenyu and GitHub
ed7981f8f2
move divandmod next to symbolic [PR] ( #16848 )
2026-07-03 14:21:59 -04:00
chenyu and GitHub
6345c2883d
simpler get_grouped_dims ( #16842 )
...
works for more than 3D, also deleted get_contraction
2026-07-03 10:29:11 -04:00
geohot
0240998b06
hotfix: that test_ellipsis_with_int test was important
2026-07-02 19:19:38 -07:00
George Hotz and GitHub
e6fbede157
changes from new codegen ( #16836 )
2026-07-02 17:01:37 -07:00
chenyu and GitHub
682b098542
fix multi write after read in create_schedule ( #16831 )
2026-07-02 15:12:48 -04:00
chenyu and GitHub
3e73a2542b
_apply_map_to_tensors always use walk rewrite ( #16827 )
...
with assign it's wrong to recursively apply, fixed a random bug that counter does not update after round 2
2026-07-02 10:43:25 -04:00
chenyu and GitHub
2e62dd308d
move fs_load and fs_store to nn ( #16821 )
...
similar to safe_load and safe_save
2026-07-01 22:01:15 -04:00
chenyu and GitHub
c8aed121cf
fix WAR in create_schedule ( #16823 )
...
* fix WAR in create_schedule
combined dep building, fixed a case that realize(x, y) != realize(y, x)
* fix
2026-07-01 22:00:51 -04:00
chenyu and GitHub
efd256b2a3
fix precompiled STORE+AFTER ( #16822 )
2026-07-01 17:07:37 -04:00
chenyu and GitHub
4b3de041e2
delete Ops.DEVICE ( #16820 )
...
85 ops
2026-07-01 12:20:05 -04:00
chenyu and GitHub
59c7874724
fix isclose with scalar other ( #16813 )
2026-06-30 21:56:33 -04:00
chenyu and GitHub
372774fc41
empty to CreationMixin [PR] ( #16811 )
2026-06-30 20:42:42 -04:00
sirhcm and GitHub
8472d374ab
ci: parallelize fuzz tests ( #16805 )
2026-06-30 14:14:32 -04:00
qazal and GitHub
795559b3d4
viz/cli: support passing only a start marker ( #16802 )
...
* failing test
* support end interval
2026-06-30 17:06:32 +09:00
George Hotz and GitHub
b683ff9836
test updates + addrspace work from new codegen ( #16800 )
2026-06-29 19:12:10 -07:00
geohot
af0e339114
test fix
2026-06-29 18:37:49 -07:00
chenyu and GitHub
3a9b87e48f
remove Ops.DEVICE from Ops.PROGRAM [PR] ( #16797 )
2026-06-29 17:41:31 -04:00
nimlgen and GitHub
2ea8439d4d
viz: fix enter_calls ( #16793 )
2026-06-29 17:10:26 +03:00
chenyu and GitHub
e836899336
remove Tensor conv2d and dot override [PR] ( #16792 )
2026-06-29 09:58:21 -04:00
nimlgen and GitHub
d9c65bd843
viz: fix walk rewrite ( #16791 )
2026-06-29 16:32:32 +03:00
qazal and GitHub
905b405820
fix contiguous_view_offset for scalar index ( #16789 )
...
* simple failing test
* scalar passes
2026-06-29 12:50:21 +09:00
chenyu and GitHub
36c0e81979
deprecate Tensor.training [PR] ( #16774 )
2026-06-27 11:02:28 -04:00
George Hotz and GitHub
afe9fcaec8
remove dtype.vec from transcendental (glm) [pr] ( #16773 )
...
* remove dtype.vec from transcendental (glm) [pr]
* more cleanups
* fix fuzzer
2026-06-26 21:41:31 -07:00
George Hotz and GitHub
59aec35a7d
more devec cleanups (glm) ( #16772 )
2026-06-26 19:27:04 -07:00
George Hotz and GitHub
a6ed4237e5
remove Ops.VCAT and Ops.PTRCAT (glm) ( #16771 )
...
* remove Ops.VCAT and Ops.PTRCAT (glm)
* fix stack, and remove useless casts
2026-06-26 18:35:38 -07:00
chenyu and GitHub
5d1867a6c3
more Tensor.training -> TRAINING [PR] ( #16770 )
...
code and test and doc
2026-06-26 20:11:59 -04:00
chenyu and GitHub
8c9e8d38a9
fix permute copy ( #16766 )
...
flip and permute copy was wrong
2026-06-26 11:42:50 -04:00
chenyu and GitHub
434e233640
LUNIQUE -> ParamArg [PR] ( #16756 )
2026-06-26 09:04:49 -04:00
wozeparrot and GitHub
5857297e64
llama: benchmark offload copy script ( #16762 )
2026-06-26 00:26:32 -07:00
George Hotz and GitHub
5fcb21c1b3
remove vectorize/const symbolic rule [pr] ( #16759 )
...
* remove vectorize/const symbolic rule
* post index symbolic is late
* simpler
* less gates
* that's fine
* this
* work
* don't rerun
* this order
* late removal of index dtype
* skip that test
* it's fine with extra symbolic
2026-06-25 17:05:14 -07:00
sirhcm and GitHub
d5a852b9ee
ci: cleanup macos tests ( #16760 )
2026-06-25 19:47:29 -04:00
chenyu and GitHub
d953eb95d5
don't use Ops.UNIQUE in Buffer [PR] ( #16754 )
...
build new_buffer with ParamArg
2026-06-25 16:00:59 -04:00
George Hotz and GitHub
e72ecd101f
split decomp into files ( #16745 )
...
* split decomp into files
* that's in op
2026-06-25 08:40:02 -07:00
George Hotz and GitHub
54887cbbdf
move decomp files into codegen ( #16744 )
2026-06-25 08:19:02 -07:00
qazal and GitHub
daf8c8f509
viz: simpler test_link_sched_codegen, merge with the other codegen test ( #16739 )
...
* simpler test_link_sched_codegen
* merge test_codegen_tracing in the other test
* include BEAM.value
2026-06-25 15:02:00 +09:00
qazal and GitHub
695ffb3a08
viz: fix rendering custom Ops.PROGRAM source ( #16738 )
...
* failing test
* fix
* simplify test
2026-06-25 14:21:31 +09:00
George Hotz and GitHub
0a8e61d0c5
switch to the new memory coaleser [pr] ( #16716 )
...
* switch to the new memory coalese
* move that stuff
* copy in allowed length logic
* mulitple buffers
* new coalese is better
* fine
* earlier
* fixes
* work
* work
* valid
* stack on index const
2026-06-23 18:03:48 -07:00
chenyu and GitHub
ce87d80911
better _drop_valid_stmts [pr] ( #16719 )
...
also dropped the unused is_increasing
2026-06-23 19:35:01 -04:00
sirhcm and GitHub
116045cc8e
ci: remove tensorflow from testoptim ( #16717 )
2026-06-23 18:11:48 -04:00