Commit Graph

5148 Commits

Author SHA1 Message Date
George Hotz
2611907afb start ripping out old scheduler -- no maps (#14909)
* start ripping out old scheduler -- no maps

* no more metadata
2026-02-20 21:05:04 +08:00
nimlgen
1b3b94a72a fix mockam mypy (#14908) 2026-02-20 15:15:05 +03:00
George Hotz
55d3a5def9 preallocate all realized buffers (#14823)
* preallocate all realized buffers

* contiguous

* work

* comment that out

* move to schedule

* better

* correct fix

* just buffer

* disk bufs

* fixes disk tensor stuff

* fix symbolic stuff

* fix multi

* 162 failures

* bugfixes

* don't check that anymore

* fix schedule tests

* mnist should be contiguious

* type and buffer

* fix tests

* shrink axis correction

* mypy fixes

* tests skips

* same 37 failures

* dedup

* no shrink in the graph

* 29 failures

* skips

* fix custom kernel

* fix training

* those optimizations aren't supported currently

* simpler

* more correct

* tests

* 14 failures

* works

* fix that test

* broken

* 11 failures

* only kernel counts left

* fixes

* all tests pass

* remove tensor_map

* op test

* 200 -> 230

* test fixes

* fixes

* revert test_tiny thing

* guard

* revert that

* test tiny passes

* no contigs there

* base realize back

* Revert "no contigs there"

This reverts commit c45bb9fcfd.

* revert that

* chop many assigns

* 12 failures

* fix tests

* tests

* apply after

* pre-commit

* remove old code

* delete that

* fix types

* remove extra contig

* fix dataloader

* torch fix

* disk fix

* update kernel fusion numbres

* runs on amd

* restore kernel count

* add that rule back

* that

* disable that

* wrong

* add the correct rule for that folding

* more tests

* guard c1.arg

* no newlines

* realize those

* split into a different file

* remove detach/contig back

* skip 2

* update that
2026-02-20 20:05:54 +08:00
nimlgen
dbf894215a init mockam (#14889)
* mockam

* more tests

* linter

* x
2026-02-20 14:09:11 +03:00
George Hotz
6610255654 add the correct rule for gcd div/mod folding (#14905)
* add the correct rule for that folding

* more tests

* guard c1.arg
2026-02-20 18:11:54 +08:00
George Hotz
a28fc2fba7 hotfix: remove wrong symbolic rule 2026-02-20 17:09:18 +08:00
qazal
16ae96fa58 finish rdna4 sqtt (#14903)
* unskip

* it's a wave pair in rdna4

* work

* that

* hidden archive

* generic s_delay, mystery InstOpRDNA4.UNK_60

* branch failing test

* UNK_60 is OTHER_VMEM_STORE

* rdna4 has both s_delay_alu and s_wait_alu

* real branch failing test

* rdna4 doesn't have JUMP_NO, it's NEXT with a flag for no jump

* make inst_delay skips recursive

* all rdna4 tests pass

* simm16 unwraps

* that has a name
2026-02-20 16:06:13 +09:00
qazal
52b51a0324 test fixes from rdna4 sqtt (#14902) 2026-02-20 14:42:33 +09:00
qazal
32f569b573 viz/sqtt: decoder fixes pre rdna4/cdna4 work (#14900)
* viz/sqtt: decoder fixes pre rdna4/cdna4 work

* fix

* branch_inst + more tests

* smaller
2026-02-20 12:10:15 +09:00
qazal
e9ae3da711 viz: click on CALL node goes to codegen (#14609)
* viz: click on CALL node goes to codegen

* colored name
2026-02-20 11:13:11 +09:00
George Hotz
fc5677c28b resnet dataloader + more test cleanups (#14899)
* resnet dataloader

* tests
2026-02-20 10:05:47 +08:00
chenyu
b9744ab62b one more test_gpudims test (#14898)
failure from the bad simplification attempt
2026-02-19 18:18:44 -05:00
chenyu
9d6cf00be2 fix gpudim bug and test_split_2d_to_3d (#14896) 2026-02-19 16:46:24 -05:00
chenyu
2b31823ef9 update test_gpudims to prove bijectivity (#14895)
* update test_gpudims to prove bijectivity

* one more
2026-02-19 16:18:59 -05:00
chenyu
19ce7a3f7f use z3 to verify gpudims output index (#14894)
found a bug with z3
2026-02-19 15:24:38 -05:00
chenyu
52f727738b move test_grouped_dims to test/null (#14893)
it's a pure helper
2026-02-19 14:50:53 -05:00
chenyu
7400362a86 remove UOp.vars [pr] (#14891) 2026-02-19 12:09:39 -05:00
chenyu
06ef8a26b7 add a test case that triggers CALL passthrough_multi (#14887) 2026-02-19 10:45:40 -05:00
Kartik Vashishta
9a9c7648e9 system: fix pci_scan_bus vendor filter (#14885)
* system: fix pci_scan_bus vendor filter

* fix: formatting
2026-02-19 17:23:32 +03:00
George Hotz
f6c1cf343c new symbolic rule from prealloc_bufs (#14883)
* new symbolic rule from prealloc_bufs

* optim
2026-02-19 20:57:30 +08:00
qazal
911399bee5 assembly/amd: move the kernel capture stuff out of helpers (#14881) 2026-02-19 16:28:48 +09:00
George Hotz
2f0f8b5776 more test relaxations from prealloc_bufs (#14880) 2026-02-19 14:23:28 +08:00
George Hotz
ab61c16730 fixes and test relaxations from prealloc_bufs (#14875)
* fixes and test relaxations from prealloc_bufs

* fix error type and guard _mop

* revert that

* contiguous makes extra/torch_backend/test_kernel_fusion.py fail
2026-02-19 11:37:25 +08:00
chenyu
0c85b93938 support shink sharded and non-sharded axes (#14874)
simpler to just support it
2026-02-18 20:54:10 -05:00
chenyu
e8252e6e4f use offical gguf in test (#14872)
also deleted bad test_load_sample_mxfp4, added some hard coded simple tests
2026-02-18 19:55:09 -05:00
chenyu
8c830c5b44 test_full_like_shrink_on_shard_axis (#14870)
* test_full_like_shrink_on_shard_axis

add a test case that triggers non-copy branch in mstack_early_shrink

* 0
2026-02-18 19:23:44 -05:00
Ananta Ranganathan
4005e9db6d Mxfp4 fix (#14866)
* double e2m1 values for mxfp4

* check if assert equal works in ci

* Revert "check if assert equal works in ci"

This reverts commit 8cf902ce0d.

* remove unnecessary whitespace change

* add test case that fails for old implementation but passes for new

* add note that the previous test is bad

* clarification on the methodology for the test

* fix the indent problem that happened to skip this test

* for now update mxfp4 block test to similarly use allclose (bad)

* add gist link and clearer explanation of process for computing test data
2026-02-18 18:50:59 -05:00
chenyu
f771de6738 gc.collect() to get the correct GlobalCounters.mem_used in tests (#14868)
test can be flaky if gc happens in between
2026-02-18 15:01:23 -05:00
chenyu
f84a11bb9f delete uneven shard tests and mentions (#14867) 2026-02-18 14:10:33 -05:00
chenyu
5746a605ce UOp.axis raises for invalid reshape (#14863)
reshape is lazy now, so better to raise from the .axis call and not have caller to handle invalid case
2026-02-18 11:28:56 -05:00
George Hotz
af839b2bd1 remove all the outerworld stuff, it was too complex (#14852) 2026-02-18 17:44:11 +08:00
George Hotz
d5636fba90 assign after copy shouldn't contig (#14847)
* assign after copy shouldn't contig

* fix assign copy
2026-02-18 12:23:49 +08:00
George Hotz
ab55e8c6b9 assign should be used as output buffer (#14845)
* assign should be used as buffer

* late removed

* the fix

* better fix

* backward slice
2026-02-18 09:37:46 +08:00
chenyu
e3c120c8e1 exclude 100 in test_assign_add (#14846)
this can crash, not sure why. skip 100 to see if it's better
2026-02-17 19:12:47 -05:00
chenyu
72cf603805 removed if self.buffer.is_allocated() in realized (#14836)
automatically fixes is_realized issue for empty
2026-02-17 15:35:56 -05:00
chenyu
aec8a6c85b Revert "one run_schedule for assign realize (#14835)" (#14837)
This reverts commit df7c37f611.
2026-02-17 14:34:26 -05:00
chenyu
df7c37f611 one run_schedule for assign realize (#14835)
concat schedules. separate out the execution part
2026-02-17 14:01:55 -05:00
chenyu
61867c2f35 TestRealizeIsRealized (#14834)
test after calling .realize(), uop.is_realized is True. currently not working for empty (thus disk tensor), and const
2026-02-17 13:30:35 -05:00
chenyu
f147791105 update test to reset and test kernel_count directly (#14832) 2026-02-17 11:48:46 -05:00
chenyu
9d4937ab5e remove assign test @unittest.skip("this test is crashing!") (#14831) 2026-02-17 10:30:58 -05:00
nimlgen
dda5ccf63b hcq: fix usb<->cpu mappings (#14827)
* hcq: fix usb<->cpu mappings

* non cpu

* um
2026-02-17 18:04:18 +03:00
chenyu
f2f039cc0f fix chained full-buffer assign (#14828)
this shows issue that pm_remove_bufferize drops tags, will fix in bufferize next. this also fixed rand being different in jit vs no-jit
2026-02-17 09:11:04 -05:00
chenyu
58fa82eef5 stronger test_assign_add (#14826)
also test self add 10 and 100 times
2026-02-17 08:36:09 -05:00
George Hotz
ff60dab622 Revert "big sink is on base (#14819)" (#14825)
This reverts commit 5fc3d8109f.
2026-02-17 19:18:06 +08:00
George Hotz
5fc3d8109f big sink is on base (#14819)
* big sink is on base

* contiguous fixes tests
2026-02-17 18:32:56 +08:00
qazal
f590564bf7 gemm multiple is only for cdna4 asm (#14814)
* gemm multiple is only for cdna4 asm

* move to backend

* and arch

* path
2026-02-17 14:00:02 +09:00
George Hotz
f081f154ae parameterize the CDNA asm gemm (#14813)
* parameterize the CDNA asm gemm

* fix llama test

* fix

* add more gemmt ests

* confirm all match

* test these asm gemms
2026-02-17 11:35:18 +08:00
George Hotz
bc3487d607 VIZ display cleanups (#14811)
* exclude reshape/expand broadcasts from viz

* limit src lines
2026-02-17 10:03:08 +08:00
chenyu
5bca5be2d2 test slice assign twice retains the buffer (#14807) 2026-02-16 20:01:47 -05:00
chenyu
9b44fbe0b8 update test_assign_add_twice (#14806)
failed test case to show that `+=1` twice returns a different buffer
2026-02-16 17:52:11 -05:00