George Hotz
47d39a6b8b
add sqtt support to the emulator ( #14791 )
...
* add sqtt support to the emulator
* more sqtt
* cleanup
* cleanups
* simpler tests
* some decent tests
* test branch
2026-02-16 16:48:26 +08:00
wozeparrot
45aebe1572
hipkittens fa backward ( #14723 )
2026-02-16 00:38:44 -08:00
Nicolas Pinto
20b658b786
fuse MULACC after MUL->SHL ( #14788 )
...
* decompositions: fuse (x << n) + c to MULACC
MUL→SHL converts x*(2^n) to x<<n before MULACC can fuse (x*c)+y.
Add pattern to also fuse (x<<n)+c → MULACC(x, 2^n, c) for backends
that support both MULACC and SHL.
* test: add test_mulacc_shl for SHL->MULACC fusion
* test: relax test_mulacc_unrolled to >= 4
SHL->MULACC fusion now also catches power-of-2 address calculations,
increasing MULACC count from 4 to 6 on PTX. the test's intent is that
each unrolled multiply is individually fused (not grouped), so >= 4
is the correct assertion.
---------
Co-authored-by: Prithvish <deformercoding@gmail.com >
Co-authored-by: Nicolas Pinto <41171+npinto@users.noreply.github.com >
Co-authored-by: Nicolas Pinto <npinto@mbp23.local >
2026-02-16 16:26:44 +08:00
qazal
ac62d28ddc
viz: amdgpu arch cleanup ( #14790 )
...
* viz: amdgpu arch cleanup
* don't do that
* simpler sqttmap
* work
* self.arch
2026-02-16 16:48:12 +09:00
George Hotz
401095e3e7
emulator barrier tests ( #14789 )
2026-02-16 15:31:01 +08:00
Bautista Garcia
0f1ca8eb43
torch_load: fix shared storage slicing ( #14771 )
...
* faster zip_extract + usage in torch load
* clean zip in torch load
* working zipextract in torchload
* tar_extract in tar path
* faster tar path
* tests passing, cleanup needed
* faster tar with 1MB buffer
* comments
* unify storage_source with all paths
* use bufferedreader in zip path
* fix ruff
* clean
* removed unnecessary string conversion
* fix for tensors that share storage
* less hacky
* shared storage test
* test comment
* linter
---------
Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com >
2026-02-16 14:30:13 +08:00
George Hotz
dff9cf35c2
amd asm emulator fixes + run it in CI ( #14786 )
...
* amd asm fix, try 2
* fix tests
2026-02-16 13:24:21 +08:00
qazal
55a4dfa2e0
cdna4 asm_gemm tests in CI on the null backend ( #14785 )
...
* cdna4 asm_gemm tests in CI on the null backend
* no .numpy() in null
* better
* gemm/asm: device comes from renderer
2026-02-16 14:06:23 +09:00
qazal
c2be31e75b
move Estimates to rewrite rules [pr] ( #14782 )
...
* move Estimates to rewrite rules [pr]
* don't need this cached_property
* tuple
* return
2026-02-16 12:59:42 +09:00
George Hotz
0abcb9aac2
move more to mixins ( #14780 )
...
* move more to mixins
* revert
* move some
* do not change
* more
* fix tests
* Revert "more"
This reverts commit d942d59fa4
2026-02-16 11:35:00 +08:00
qazal
8e7c5f5b09
remove Tensor.training = True in test_arange ( #14781 )
2026-02-16 11:19:42 +09:00
kevvz
33b2ade8cd
Rdna4 emulator test_ops, dtypes pass ( #14773 )
...
* test_ops, test_dtypes pass
* merge cdna4
* ruff + more tests
* reorganize
* /backend
* again
* again...
* add rdna4
2026-02-16 10:13:39 +08:00
qazal
156b6cb7e4
native bf16 cast in cdna4 ( #14574 )
...
* native bf16 cast in cdna4
* don't need contig backward
* simpler
* contig bw still wins in those cases
2026-02-16 10:51:32 +09:00
qazal
33b31d9cd6
tinykittens flash attention dtype fix, add CI ( #14770 )
...
* don't hardcdoe amd device
* add failing tests, ci too
* fix: fix for dtype mixin
* bump to rocm 7.1
---------
Co-authored-by: Woze Parrot <wozeparrot@gmail.com >
2026-02-16 01:15:11 +09:00
chenyu
352845d8cc
update cast to uint tests ( #14768 )
...
result in valid range should work, add intermediate cast to NIRRenderer since it's UB for [128, 256)
2026-02-15 10:55:13 -05:00
qazal
ceccc8eb86
unskip now passing multi tests [pr] ( #14759 )
2026-02-15 20:30:00 +09:00
qazal
9da7f5e733
disable process replay for AMD emulator renderer [pr] ( #14766 )
...
* disable process replay for AMD emulator renderer [pr]
* line
* skip
2026-02-15 18:52:37 +09:00
George Hotz
9759fd6193
dtype mixin ( #14763 )
...
* dtype mixin
* dtype mixin methods
2026-02-15 16:03:48 +08:00
qazal
42b6bf0b7a
fix sdpa causal failing test on multi ( #14762 )
...
* simple failing test
* device is from xq
2026-02-15 16:54:33 +09:00
George Hotz
0e215c433d
remove hack from cast ( #14760 )
...
* remove hack from cast
* skip tests
* linters to 3.12, another skip
* fix rand
* m_
2026-02-15 13:56:38 +08:00
George Hotz
d176af6269
start outerworld call test, fix gate ( #14758 )
2026-02-15 12:35:01 +08:00
chenyu
ca68037f26
lazy basic setitem to unrealized Tensor ( #14756 )
...
undo the view and make it a mask, this fuses the setitem with any pending compute too.
one behavior change is that for target not backed by a buffer (const and arange), rangeify makes output contiguous under the hood.
this is stricter better than raise and ask user to call contiguous, as that would no longer be fuse-able.
2026-02-14 20:27:03 -05:00
George Hotz
32980c74d1
hotfix: skip flaky tests, looped many times on tinymac3
2026-02-15 07:46:29 +08:00
chenyu
902dc7c09c
fix test_numpy_parity_and_backward_2d ( #14755 )
...
test setup issue, test failed locally with `RUN_SLOW=1`
2026-02-14 17:59:00 -05:00
chenyu
043f5dbfa0
fix write-after-read tracking ( #14754 )
...
AFTER-AFTER was silently dropped, which breaks write-after-read
2026-02-14 17:23:05 -05:00
chenyu
d79c63a0ff
test_multi_step_assign_read_write_same_buffer ( #14752 )
...
pattern in LAMB that can be off subtly
2026-02-14 16:39:08 -05:00
chenyu
95f4c7e90a
fix limit_bufs to not limit index ( #14751 )
...
index is not real buffer. also made MAX_KERNEL_BUFFERS a ContextVar
2026-02-14 16:00:03 -05:00
chenyu
0ce4a55dad
clean up test_setitem_slice ( #14750 )
...
moved to test_setitem_schedule, and use contiguous zeros as scheduler handles empty differently now
2026-02-14 14:29:16 -05:00
chenyu
8f6772fd8c
more setitem kernel mem tests ( #14749 )
...
* more setitem kernel mem tests
test only the slice is accessed
* update
2026-02-14 11:01:03 -05:00
chenyu
446909fb7a
more setitem kernel tests ( #14748 )
...
check where realize happened
2026-02-14 09:57:46 -05:00
nimlgen
e1a18dadae
fix devices for copies ( #14747 )
...
* fix devices for copies
* add test
2026-02-14 17:39:41 +03:00
Christopher Milan
eaa9506a00
disallow subnormals in emulated test_dtype ( #14744 )
2026-02-14 00:11:57 -05:00
qazal
c88bb075f0
hotfix: correct way to get renderer arch ( #14743 )
2026-02-14 12:38:20 +08:00
qazal
6dc7ea58fd
make flash attention tests run on DEV=NULL EMULATE=AMD_CDNA4 ( #14742 )
...
* make flash attention tests run on DEV=NULL EMULATE=AMD_CDNA4
* no if CI, this is just the arch
2026-02-14 12:24:37 +09:00
George Hotz
e8bd432bf6
move amd emulator out of tree ( #14740 )
...
* move amd emulator out of tree
* move the readme too
2026-02-14 10:32:00 +08:00
chenyu
dca7819f76
more setitem into unrealized tests ( #14737 )
...
* more setitem into unrealized tests
into empty, const with alu, and arange
* typo
2026-02-13 20:28:51 -05:00
chenyu
8b205a007e
lazy setitem for realized target ( #14735 )
2026-02-13 12:20:14 -05:00
nimlgen
3bee6638e3
external_test_hive_reset ( #14729 )
...
* external_test_hive_reset
* add fault
2026-02-13 19:08:36 +03:00
George Hotz
c0fe78f73b
BUG: metadata is lost with partial assign ( #14732 )
2026-02-13 21:35:21 +08:00
George Hotz
5289b4e882
renderer/amd: add cdna emulator ( #14721 )
...
* renderer/amd: add cdna emulator
* fixes
* no predecode
* no early
* REMU_PATH
* delete that
* round
* Fix cache invalidation check in _compile_smem
2026-02-13 16:06:58 +08:00
Christopher Milan
08a555c875
skip test_expand_buffer_before_cast on WEBGPU metal ( #14724 )
2026-02-13 00:01:05 -05:00
chenyu
50cb40be88
clean up test/null/test_indexing.py ( #14720 )
2026-02-12 22:36:53 -05:00
qazal
5b624b5e93
viz: better error message for out of range timestamps ( #14722 )
...
* test_timestamp_out_of_range
* rel_ts helper
* linter
2026-02-13 12:13:40 +09:00
George Hotz
4088d686b2
remove llvm requirement from amd ( #14717 )
...
* remove llvm requirement from amd
* tests pass
* test
* sink kernarg_size
* move stuff
* amd_asm_matmul to new style
* default type
* fix tests, simpler
* cu mode is faster and simpler
* darken
2026-02-13 10:50:12 +08:00
George Hotz
d3adb8428e
Revert "hotfix: skip test/amd in macpytest" ( #14704 )
...
* Revert "hotfix: skip test/amd in macpytest"
This reverts commit b7dade2adf
2026-02-13 08:00:24 +08:00
Christopher Milan
c30bb0f006
fix WEBGPU isnan check ( #14711 )
2026-02-12 17:01:18 -05:00
chenyu
787998fac3
fix getitem tensor indexing detection ( #14712 )
...
issue with sint
2026-02-12 16:04:37 -05:00
chenyu
86352988d8
update test_uops_stats for setitem ( #14710 )
...
realize both full tensor and the slice should not add to global_mem
2026-02-12 12:26:13 -05:00
chenyu
56caf6a3a2
fix Estimate.from_uops for sliced access ( #14695 )
...
"assume all DEFINE_GLOBAL memory is accessed" is wrong for partial load. get accessed accumulated from INDEX, then cap at full size. now mem_est never exceeds lds_est
2026-02-12 11:18:07 -05:00
chenyu
8551fa50d3
support bitcast in sym_infer ( #14708 )
...
fixed `DEBUG=2 DEV=WEBGPU python -m pytest test/backend/test_tensor_variable.py::TestTensorVariable::test_symbolic_pad`
2026-02-12 10:21:05 -05:00
