chenyu and GitHub
a187dfd3df
bert BEAM_UOPS_MAX 3000->4000 ( #9603 )
...
more stable for the final step time
green 410ms (master) -> 397ms (BEAM=4) -> 392ms (this)
red 561ms (master) -> 550ms (this)
2025-03-27 11:58:47 -04:00
qazal and GitHub
088a677e25
rescale to fit viz graph [pr] ( #9599 )
...
* zoom to fit the graph in viz [pr]
* always on screen fit graph
* space key recenters
2025-03-27 23:33:51 +08:00
nimlgen and GitHub
3737821b9e
prepare for clang graph ( #9600 )
...
* prepare for clang graph
* emu
* ops
* ops2
* better type
* fix
2025-03-27 20:09:37 +07:00
qazal and GitHub
bf94924d5a
fix viz with nested graph_rewrite ( #9595 )
2025-03-27 13:14:28 +08:00
qazal and GitHub
c011751b41
statically define viz arrow heads ( #9594 )
2025-03-27 12:22:04 +08:00
qazal and GitHub
0877497bad
hotfix: use captured uops in viz render [pr] ( #9593 )
...
* hotfix: use captured uops in viz render [pr]
* better error
2025-03-27 11:52:12 +08:00
qazal and GitHub
e5ff7b23d7
refactor to @track_matches + add failing test_nested_rewrite ( #9592 )
...
* test_nested_rewrite
* refactor to track_matches
* positional arg
2025-03-27 11:11:56 +08:00
chenyu and GitHub
62888614f6
lower bert eval bs to 24 ( #9590 )
...
oom during eval
2025-03-26 21:25:23 -04:00
nimlgen and GitHub
dc9da1d917
memplan into one buffer ( #9526 )
...
* new memplanner
* new should works
* fix
* VALIDATE_MEMORY_PLANNER
* hm?
* ugh
* fix alignment
* fix2
* rm
* tiny fixes
* test
* comments and fixes
* fix2
* liiiinetr
* t
* fix
2025-03-27 01:46:50 +07:00
qazal and GitHub
8b717c345c
cache viz worker at launch ( #9589 )
2025-03-27 01:10:02 +08:00
George Hotz and GitHub
d62ced8981
symbolic -> symbolic_flat ( #9588 )
2025-03-26 23:34:43 +08:00
George Hotz and GitHub
8aaa5e1ec5
generate the individual indexes ( #9587 )
2025-03-26 22:32:06 +08:00
George Hotz and GitHub
5c6cd884e3
multiple simplifies is faster [pr] ( #9586 )
...
* multiple simplifies is faster [pr]
* cleanup
* cleanup
2025-03-26 21:42:52 +08:00
George Hotz and GitHub
1e6e75e39a
little changes from dsp branch ( #9582 )
...
* little changes from dsp branch
* not that one
* need the where
* Revert "need the where"
This reverts commit 140f89c878 .
2025-03-26 20:01:21 +08:00
nimlgen and GitHub
e88a640ca5
fix _access_resources for offset buffers ( #9580 )
...
* fix _access_resources for offset buffers
* test
2025-03-26 18:42:43 +07:00
Andrey and GitHub
7b865ed03d
use tuple in isinstance for type checking ( #9583 )
2025-03-26 19:36:48 +08:00
George Hotz and GitHub
9115ce8860
linearizer fixups from DSP branch ( #9581 )
2025-03-26 18:28:15 +08:00
qazal and GitHub
e799df537e
prep viz UI cleanup for grid scales ( #9579 )
...
* less ways to make a button
* move collapse out
* work
* do not create extra resizers
* better
* ul
* safari
2025-03-26 17:48:15 +08:00
nimlgen and GitHub
ccbcdca473
add memplanner tests ( #9577 )
2025-03-26 10:59:39 +07:00
qazal and GitHub
c03dadfcb9
add TORCHVIZ=1 to beautiful_mnist_torch ( #9576 )
2025-03-26 11:17:08 +08:00
qazal and GitHub
93bcb974c5
select torch device in examples/beautiful_mnist_torch.py ( #9575 )
2025-03-26 11:01:25 +08:00
uuuvn and GitHub
2c32126fc8
am: AMRegister refactor ( #9572 )
2025-03-26 00:52:40 +07:00
chenyu and GitHub
cddd750d68
add a failed test case for jit/nojit rand [pr] ( #9574 )
...
currently adding jit produced different rand values
2025-03-25 13:32:44 -04:00
nimlgen and GitHub
4cf2b68ca8
am_smi: fix init for newer versions ( #9559 )
2025-03-25 23:48:05 +07:00
qazal and GitHub
a6a5c0aec5
add NULL=1 backend ( #9573 )
...
* add NULL=1 backend
* NullAllocator
* line
* metadata should still work
* it shouldn't have memory usage
* Revert "it shouldn't have memory usage"
This reverts commit a9080fdd43 .
* back
* null flops
2025-03-25 22:20:52 +08:00
qazal and GitHub
b60d9976b4
better yaxis formatting in viz memory graph ( #9570 )
...
* better bytes format
* pluralize
* 1 less line
2025-03-25 16:50:22 +08:00
qazal and GitHub
faf3b5b245
display kernel metadata in memory viz ( #9569 )
...
* display kernel metadata in memory viz
* fix that
2025-03-25 13:14:54 +08:00
qazal and GitHub
52301fe68e
move Buffer refcount increment out of schedule.py ( #9564 )
...
* move Buffer refcount increment out of schedule.py
* add TestGC.test_assign_refcount
* refcount refers to Ops.BUFFER UOps
2025-03-25 12:08:27 +08:00
qazal and GitHub
262f5a2bd3
hotfix: replace link in viz/readme ( #9568 )
2025-03-25 10:24:49 +08:00
chenyu and GitHub
6427272bf6
minor update to rand [pr] ( #9566 )
2025-03-24 18:49:50 -04:00
chenyu and GitHub
b0e070e737
remove MOCKGPU workaround in rand ( #9565 )
...
also `requires_grad_` to save a line
2025-03-24 17:49:45 -04:00
qazal and GitHub
d7c754ce49
failing test for UOp buffer ref count ( #9563 )
...
* failing test for UOp buffer ref count
* lint
2025-03-25 00:10:48 +08:00
f90001e1a6
amd llvm render (no_comgr prereq) ( #9543 )
...
* amd llvm render
* skip test_div_rounding_mode
---------
Co-authored-by: b1tg <[email protected] >
2025-03-24 22:50:51 +08:00
Priyank Patel and GitHub
4f5e03bd60
better fix inplace detach ( #9557 )
2025-03-24 22:50:28 +08:00
qazal and GitHub
1c40873962
show buffer info in memory viz ( #9562 )
2025-03-24 22:12:30 +08:00
qazal and GitHub
efaee75656
start viz of memory usage ( #9561 )
...
* start viz of memory usage
* polygons/bars + use d3
2025-03-24 19:05:35 +08:00
qazal and GitHub
1cfe6d02fe
refactor uop_to_json to return a dict [pr] ( #9560 )
2025-03-24 16:38:17 +08:00
nimlgen and GitHub
edf9e1bf8d
am: move out soc21 to a sep module ( #9551 )
...
* am: soc module is not part of am
* am: soc module is not part of am
2025-03-24 14:17:42 +07:00
George Hotz and GitHub
74d98eafb8
add onnx frontend stub [pr] ( #9558 )
2025-03-24 12:24:34 +08:00
geohot
de7d6cec3a
hotfix: DEBUG 5 prints the ast
2025-03-24 11:43:11 +08:00
chenyu and GitHub
ba41076e94
update embedding test to not use dtypes.long [pr] ( #9556 )
2025-03-23 21:33:38 -04:00
chenyu and GitHub
c965f4c20b
update bert config ( #9555 )
...
BEAM 4->5 for green, 2% faster
use AMD driver instead of AM for red, 5% faster
2025-03-23 16:14:41 -04:00
chenyu and GitHub
d734e24c01
minor WEBGPU_PATH cleanup [pr] ( #9552 )
...
also mypy recognizes `sys.platform == 'win32'` but does not recognizes it if wrapped inside a helper...
2025-03-23 09:10:02 -04:00
Ahmed Harmouche and GitHub
7ce7fe0574
Refactor webgpu_dawn lib finding ( #9547 )
...
* Refactor webgpu_dawn lib finding
* Fix ruff
2025-03-23 08:23:29 -04:00
uuuvn and GitHub
c631c72f22
HCQ: Increment timeline signal before submitting ( #9550 )
...
`AMDComputeQueue.__del__` frees `hw_page` which is safe because
`AMDAllocator._free` does `self.dev.synchronize()` which is supposed
to wait for execution of IB to finish, however that doesn't happen if
AMDComputeQueue is dropped right after submit before timeline signal is
incremented, which it is in most places leading to a race if .bind() is
also used (required for multi-xcc because bug in mec fw treats all
PACKET3_PRED_EXECs outside IBs as if they had EXEC_COUNT of zero).
2025-03-23 18:30:38 +07:00
nimlgen and GitHub
d5667419af
am: move out pte creation logic ( #9548 )
...
* am: move out pte creation logic
* emu
* ops
2025-03-23 18:29:10 +07:00
geohotstan and GitHub
309afa20b7
add Tensor.max_unpool2d ( #9518 )
...
* why does max_unpool2d feel slower than out.gradient ...
* slightly cleaner
* what happened to ruff
* need to think about this some more
* slightly faster now?
* clean up, 1 more failing edge case
* ok good
* working TINY_BACKEND
* nit doc wording
* retry CI
2025-03-22 12:11:33 -04:00
quortus and GitHub
bdd44d4255
Fix DSP transcendentals ( #9542 )
2025-03-22 11:08:18 +08:00
Ignacio Sica and GitHub
eddafb84e5
Bugfix for TC=3 ( #9464 )
...
* wrong but uses less shared
* for size 8 tc1 with devectorize in 0 loads into local before wmma and works
* improvements over tc1 devectorize
* fix tc=3
* works for handcoded tc opts
* clean bugfix tc=3
* fix
* revert changes
2025-03-21 16:43:42 -07:00
chenyu and GitHub
6da78164f9
assert Kernel ast.op to be Ops.SINK [pr] ( #9539 )
...
rest of the code assumes self.ast is defined anyway
2025-03-21 18:09:44 -04:00