qazal
bf2d9d138f
viz: simplify amdgpu cfg ( #14326 )
...
* viz: replace llvm disasm with our disasm
* it starts with more code
* then it becomes less
* simpler, cdna disassembles with decimal simm16
* s_branch is upper case, add test
* simm16s and others
2026-01-25 15:21:45 +09:00
qazal
647e527a7e
viz: replace llvm disasm with our disasm ( #14325 )
2026-01-25 13:56:56 +09:00
nimlgen
4280a8eef2
am: update fw ( #14323 )
2026-01-25 01:08:47 +03:00
chenyu
7e41da1ae8
fix generate_dataset.sh ( #14324 )
...
added `set -e` so wrong pathes would fail the script, then fixed the path
2026-01-24 16:47:10 -05:00
chenyu
311bfd91d6
clean up where_on_load [pr] ( #14322 )
...
no repeated split_uop and general cleanup
2026-01-24 14:43:43 -05:00
nimlgen
8b282ba6d2
memory: reserved vram ( #14318 )
2026-01-24 19:39:24 +03:00
chenyu
00e9ba0b82
update type for split_uop and where_on_load [pr] ( #14319 )
...
also variable names in where_on_load, before logic update
2026-01-24 11:17:41 -05:00
chenyu
cb69b7b2b2
comment out fold_where_closure ( #14316 )
2026-01-24 10:15:42 -05:00
wozeparrot
d74587f16d
fa multi fix 2 ( #14314 )
2026-01-23 23:35:02 -08:00
chenyu
d9f0ad1d87
update return type for Tensor.tolist ( #14313 )
...
since sequence is incorrect since it can be list of list, use Any to avoid recursive type
2026-01-23 23:21:49 -05:00
qazal
807bc40931
assembly/amd: dsl and disasm cleanup ( #14311 )
...
* rdna4 inst helper
* remove dsl aliases
2026-01-24 11:36:12 +09:00
Christopher Milan
e782d44918
WEBGPU/NIR truncates ints ( #14307 )
...
* WEBGPU truncates ints
* nir has this bug too
2026-01-23 19:28:06 -05:00
nimlgen
26220a472e
no core_id ( #14265 )
...
* no core_id
* kwargs
* est
* linters
* ugh
* revert this
* deps
* glb
* should work?
* nn
* line
* fx
* ym
* z
* d
* um?
* revert
* this one?
* first half
* um p2
* all?
* um
* cleaner
* um
2026-01-23 21:30:12 +03:00
chenyu
e65bc7a7c5
where closure folding ( #14304 )
2026-01-23 10:55:13 -05:00
chenyu
d5a3b02a9c
clean up xpow ( #14295 )
...
mostly for `ret * (base < 0).where(adj, ret.const_like(1))` -> `(base < 0).where(neg_base, ret)`, since it's good for NAN neg_base but not generic
2026-01-23 10:19:47 -05:00
qazal
b913c910c5
assembly/amd: rdna4 passing test_roundtrip ( #14300 )
...
* test_roundtrip on different archs
* failing tests
* take RDNA4 xml changes from the emu branch
* work
* min diff to disasm flat
* test_add passes, rdna4 first
* correct vgpr field for the multi dword store stuff
* amdllvm
* recompile in roundtrip, get sources from emulator
* amdllvm, 2
* clean clean
* note, don't rely on that os.environ
---------
Co-authored-by: George Hotz <geohot@gmail.com >
2026-01-23 21:33:53 +09:00
qazal
f3b0e42863
remove extra sqtt pickles in gfx1200 ( #14302 )
2026-01-23 20:13:48 +09:00
George Hotz
d116312b1a
get cdna sqtt working ( #14301 )
...
* get cdna sqtt working
* cnd aprser
* wavestart/waveend
* names
* cdna
* test that
2026-01-23 18:46:15 +08:00
George Hotz
a5c4fa39d1
RDNA4 support in SQTT ( #14299 )
...
* table test
* cleanups
* dead file
* delta short
* tests
* delta test
* work
* l4 tests pass
* l0
* cnda
* print
* reverT
* wave failure
* wave failure
* test
* encs
* no l0 crap
* L4
* rdna4 sqtt
* notes
* linter
2026-01-23 16:16:45 +08:00
wozeparrot
963c59ebdb
fix: pull fixes from gradacc branch ( #14296 )
2026-01-22 23:07:54 -08:00
Christopher Milan
68668b8f28
fix WEBGPU NEG ( #14298 )
...
* fix WEBGPU NEG
* add test
* parenthesize
2026-01-23 01:44:52 -05:00
qazal
3b8a7bb8c9
use existing roc.py infra for sqtt tests ( #14297 )
...
* add pc, per kernel tracing
* work
* remove those imports
* min diff
2026-01-23 14:07:11 +09:00
chenyu
5f32f7a06b
fix winograd padding order ( #14294 )
2026-01-22 23:00:14 -05:00
George Hotz
52b989c6c8
don't place consts early + fixes from anthropic challenge ( #14286 )
...
* don't place consts early
* add anthropic challenge
* with ref
* do we still have to devectorize bools?
* tests pass
* just WHERE
* fine, revert that
* fine, revert
* only index
* z3 validator doesn't support vectorized
* Revert "z3 validator doesn't support vectorized"
This reverts commit 1b7930ecb3
2026-01-23 10:48:39 +09:00
chenyu
0903782bc0
remove few dead or unneeded codes [pr] ( #14275 )
2026-01-22 20:05:43 -05:00
chenyu
3eb5cd7d32
stronger test_rand_is_lazy ( #14293 )
2026-01-22 18:58:53 -05:00
chenyu
c15b6e6709
update test_randn_finite skipped device ( #14292 )
2026-01-22 18:26:02 -05:00
chenyu
073c6a81b5
raise if Tensor._buffer is called during jit ( #14114 )
...
* raise if Tensor._buffer is called during jit
* cleaner
2026-01-22 17:30:18 -05:00
nimlgen
8cd22df2dd
amd: alive wgps ( #14149 )
...
* amd: disabled wgps
* l
* wgp
* uoops
* mockgpu
* drm
* ad this
* fi
* reg
2026-01-23 00:08:45 +03:00
chenyu
a738c4bb22
test symbolic view broken with jit ( #14290 )
2026-01-22 13:44:47 -05:00
chenyu
f22fa6a5be
test rand is lazy ( #14289 )
2026-01-22 13:07:55 -05:00
chenyu
1726b884f2
update test_jit_v_nojit_random_regen ( #14288 )
...
current behavior is that jit and non-jit consume random seed differently, still the random values are different
2026-01-22 12:21:47 -05:00
chenyu
fbed36fa15
jit graph handle input==output aliasing ( #14287 )
...
a position that wasn't an input during capture should never become an input during execution, but graph cannot tell this by jit_cache and input_buffers only
2026-01-22 11:37:41 -05:00
chenyu
8bb61c2490
stronger test_graph_input_output_aliasing ( #14282 )
...
* stronger test_graph_input_output_aliasing
* comfirmed failure
2026-01-22 09:59:34 -05:00
qazal
d7afa02085
clean up the extra/sqtt directory ( #14284 )
...
* remove legacy test_timing stuff
* remove legacy test_pmc, update active_sqtt_parse
2026-01-22 19:10:59 +09:00
qazal
dff5f361b0
support rendering assembly kernels on the NULL backend ( #14283 )
...
* assembly custom kernels in DEV=NULL, use renderer arch
* update mmapeak
* llvm
2026-01-22 15:49:07 +09:00
qazal
dfefeddeed
add tflops to cdna gemm custom kernel ( #14281 )
2026-01-22 12:48:28 +09:00
qazal
18f408a35a
custom assembly kernel with variable tests ( #14280 )
...
* custom assembly kernel with variable tests
* different threads
* sink
* zeros like / flatten
2026-01-22 11:34:17 +09:00
chenyu
4de107b764
jit graph bug when input is output ( #14278 )
...
* jit graph bug when input is output
wrong result in llm
* not just metal
2026-01-21 18:49:52 -05:00
wozeparrot
76a9242a66
fa: merge kv bwd into one kernel ( #14277 )
2026-01-21 15:24:41 -08:00
chenyu
6279ae4a94
remove llm generate always reset start_pos ( #14276 )
...
* remove llm generate always reset start_pos
by itself seems like a bug, also added a test to repro forward_jit.reset() issue
* issue is jit graph, so revert that test
2026-01-21 16:54:30 -05:00
nimlgen
da1fedc3c8
working ioctls ( #14272 )
2026-01-21 20:29:04 +03:00
chenyu
574d171fa6
fix onnx Pad constant_value=None ( #14271 )
...
also removed a dead branch in _resolve_pool_pads
2026-01-21 11:51:34 -05:00
chenyu
a18d34be1e
simpler split_store outer range check [pr] ( #14273 )
...
also fixed comment
2026-01-21 11:51:14 -05:00
chenyu
e64111ad08
update all_same [pr] ( #14270 )
...
add type annotation and unit test
2026-01-21 11:26:15 -05:00
chenyu
9ad3c865ac
fix bug in logsumexp keepdim=True ( #14268 )
2026-01-21 09:49:55 -05:00
George Hotz
41d00a046d
add device to local, fix PCONTIG=2 ( #14266 )
...
* add device to local, fix PCONTIG=2
* regression test
* remove the device when we render
* viz slowness
* no long
2026-01-21 22:12:18 +09:00
wozeparrot
c1d14ea832
llama8b train fixes ( #14264 )
2026-01-20 20:34:47 -08:00
qazal
549dbabfcb
move ALLOW_DEVICE_USAGE=0 to get_program [pr] ( #14263 )
2026-01-21 12:56:05 +09:00
qazal
78a28227c6
assembly/amd: cdna4 mfma support ( #14206 )
2026-01-21 09:12:05 +09:00
