Commit Graph

5417 Commits

Author SHA1 Message Date
chenyu
5c8de2d044 revert a mod pattern (#5864)
fixed UOP_IS_SYMBOLIC=1 linearizer failure 47
2024-08-01 17:24:26 -04:00
nimlgen
34168a64e3 optimize nv profiler (#5856)
* nv profiler fix

* cleanup hcq a bit

* fixes

* fix

* typo

* all signals put timestamp

* a bit cleaner

* merge fields

* type

* import

* tiny fix
2024-08-01 23:57:45 +03:00
George Hotz
2d3c7e4d4e some TestPickleJIT tests (#5860)
* some TestPickleJIT tests

* hotfix: print which opencl device we are using
2024-08-01 12:39:59 -07:00
George Hotz
e347f10d33 hotfix: print which opencl device we are using 2024-08-01 12:39:46 -07:00
chenyu
0c8d202348 revert some UOp IDIV bound (#5863)
* revert some UOp IDIV bound

breaks conv with UOP_IS_SYMBOLIC, added some conv tests in CI

* those are correct

* skip slow ones
2024-08-01 15:09:06 -04:00
George Hotz
53fcac9e80 hotfix: increase time on flaky NV test 2024-08-01 10:20:07 -07:00
qazal
cedf459843 infra for multi view reduce_info [run_process_replay] (#5861) 2024-08-01 19:46:55 +03:00
qazal
26d0265d66 test schedule of LazyBuffers [run_process_replay] (#5859) 2024-08-01 19:06:29 +03:00
George Hotz
0e34d83777 hotfix: don't include the old input_rawbuffers in all_resources 2024-08-01 09:00:11 -07:00
chenyu
d609206a4a move UOp patterns around [run_process_replay] (#5857)
group lt / div / mod together and minor cleanups
2024-08-01 11:32:08 -04:00
qazal
3e95e2bb0b mutate reduceop shapes pre ast creation [run_process_replay] (#5855) 2024-08-01 15:00:05 +03:00
qazal
ba0a0008aa early update the reduceop axis [run_process_replay] (#5854) 2024-08-01 14:08:40 +03:00
David Hou
eb91423cb4 MLB support reshape for uneven shards (#5804)
* cleaner uneven reshape

* update test
2024-08-01 02:36:03 -07:00
David González Martínez
0f09b94c43 add failing test for second order derivatives (#5772)
* add failing test

* fix lint

* fix bad merge

* fix again

* fix test

* more minimal
2024-08-01 02:34:47 -07:00
George Hotz
9d05dfb6f4 move JIT graphing into CapturedJit (#5852)
* move JIT graphing into CapturedJit

* better

* _jit_cache

* clear inputs cleanup

* test_pickle_jit with graph + cleanup

* 0 is fine to start

* support None in bufs

* alloc real buffers

* cleaner
2024-07-31 20:48:17 -07:00
chenyu
0ec732b494 test lin fail 47 for UOP_IS_SYMBOLIC (#5853)
failed arange example with UOP_IS_SYMBOLIC
2024-07-31 23:09:22 -04:00
George Hotz
c6a8395f1b CapturedJit is fun to pickle [run_process_replay] (#5851)
* CapturedJit is fun to pickle

* export input replace
2024-07-31 17:23:01 -07:00
George Hotz
5ff3e46718 diff symbolic with uops [run_process_replay] (#5841)
* diff symbolic with uops

* mergable symbolic diff
2024-07-31 15:15:01 -07:00
George Hotz
72621d9e7c count the specials in uops [run_process_replay] (#5848)
* count the specials in uops [run_process_replay]

* cleanups
2024-07-31 14:53:18 -07:00
chenyu
c2ffcf6887 remove the wrong mod UOp pattern (#5847)
don't think we are hitting it because the stride construction, and it's wrong and not needed
2024-07-31 16:24:25 -04:00
qazal
8174c438a3 pad test_failure_45 (#5846) 2024-07-31 23:08:48 +03:00
George Hotz
8672a9db3f add test to validate lazyops dims (#5845) 2024-07-31 12:59:38 -07:00
chenyu
4fe5b95568 fix UOp ALU bound (#5844)
* fix UOp ALU bound

root cause of resnet bug, the ALU bound is only correct for scalar, not vectorized

* it can be nan...
2024-07-31 15:19:31 -04:00
George Hotz
5eedd9e3ad raise the line ceiling to 8600. USE LINES CAREFULLY 2024-07-31 09:56:39 -07:00
nimlgen
f768935be8 add RING_ALLREDUCE_THRESHOLD (#5835)
* add RING_ALLREDUCE_THRESHOLD

* becnhmark

* fixes

* fix n_gpus

* unused import

* remove debug=2
2024-07-31 16:13:09 +03:00
nimlgen
431749dc21 hcq fix timestamp around kernel (#5837) 2024-07-31 16:12:27 +03:00
chenyu
2e087ca8e4 UOp bound for div negative number (#5808) 2024-07-31 02:10:23 -04:00
qazal
bcbd925001 hcopts failing test for fused arange kernel (#5815)
* add failure_43

* n 45
2024-07-31 09:02:44 +03:00
chenyu
93c5989c84 add UOp bound for BinaryOps.CMPLT (#5833)
and remove the redundant lt folding rule
2024-07-31 01:46:48 -04:00
chenyu
5560bda509 remove redundant mod 1 pattern [run_process_replay] (#5832)
it's folded because min==max
2024-07-31 01:12:05 -04:00
qazal
ed556c260e UOps.IF rules more tests (#5831)
* init tests

* split tests

* assert multiple gates simplicity
2024-07-31 00:11:02 -04:00
Vyacheslav Pachkov
610e454132 fix opencl_ioctl on comma (#5814)
- remove unused code
- add CP_REG_TO_MEM opcode
- fixed parse_cmd_buf for more than 1 command object by correcting
an offset
- fixed memory mappings for cases when memory was allocated with
KGSL_MEMFLAGS_USE_CPU_MAP.
KGSL_MEMFLAGS_USE_CPU_MAP: If set on call and return, the returned GPU
address will be 0. Calling mmap() will set the GPU address.
So there are no IOCTL_KGSL_GPUOBJ_INFO ioctls for that type of memory
and it resulted to crash right after get_mem.
2024-07-30 20:44:06 -07:00
David Hou
9a485f36e4 shard kvcache (#5830) 2024-07-30 20:29:54 -07:00
David Hou
492a696d14 allow specify splits in shard, handle multiple different splits in MLB.e (#5599)
* allow specify splits in shard, handle multiple different splits in MLB.e

* line width

* linter

* don't use Device in docstring

* specify size of shards instead of boundaries

* adjust docstring for specify size of shards instead of boundaries

* don't allow splits on symbolic axis?

* just allow sint in splits_to_bounds

* add message for assert

* bounds instead of splits to save lines

* fix types

* reduce diff

* fix

* tuple

* golf :(

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
2024-07-30 19:33:04 -07:00
George Hotz
3f6a43ba12 less lines [run_process_replay] (#5829) 2024-07-30 19:32:54 -07:00
chenyu
c3da458bc3 UOp if min==max folds to CONST (#5828)
* UOp if min==max folds to CONST

* fix test
2024-07-30 22:14:22 -04:00
George Hotz
4e89d45513 hotfix: put contiguous back in llama 2024-07-30 18:43:48 -07:00
George Hotz
21c5e8e1b7 extreme llama speed, 57.34 tok/s (#5827)
* extreme llama speed

* mergable
2024-07-30 18:32:09 -07:00
George Hotz
e6879035a0 work to make GEMV fast (#5824)
* work to make GEMV fast

* half8 cast

* align struct

* fix amd

* float8 is a later problem
2024-07-30 17:41:40 -07:00
chenyu
2d90b7a103 remove redundant max boolean pattern (#5826)
covered by generic max folding [run_process_replay]
2024-07-30 20:27:54 -04:00
chenyu
02f0be03f2 tests on UOp div negative number and arange opts (#5825) 2024-07-30 20:06:57 -04:00
George Hotz
4dd24dc439 use decimal for timestamps for more precision [run_process_replay] (#5823)
* use decimal for timestamps for more precision

* err, didn't get saved

* fix types + 38 -> 40
2024-07-30 15:06:14 -07:00
chenyu
d072e628da UOp bounds for max (#5820) 2024-07-30 17:54:44 -04:00
George Hotz
3630208a01 lil transcendental folding cleanup [run_process_replay] (#5822)
* lil transcendental folding cleanup [run_process_replay]

* idk why function isn't Callable
2024-07-30 14:10:17 -07:00
George Hotz
693990a346 swap src[2] and src[3] in load [run_process_replay] (#5821)
* swap src[2] and src[3] in load [run_process_replay]

* cleanups + bugfix

* fix ptx
2024-07-30 14:04:13 -07:00
George Hotz
17a2f74412 new style load/store folder (#5784)
* remove old index reorder

* new style folder

* works better

* dedup

* one failure

* this is fine now...

* expander_rewrite

* images broken, but all else should work

* cleanups

* make tests work with old

* fix images

* cleanups + bugfix

* minor fixes

* fix gated store folding

* flip gate_creator and expander

* fix gated store

* remove unneeded rules

* lines getting close

* line count good
2024-07-30 13:17:20 -07:00
chenyu
e8a42b945c simpler src variables in UOp._min_max [run_process_replay] (#5819)
s0,s1 instead of self.src[0] and self.src[1]
2024-07-30 15:18:42 -04:00
Francis Lata
a0baff7a3d update dataloader script example (#5818) 2024-07-30 15:18:29 -04:00
wozeparrot
eebb1b9922 feat: temperature 0 llama3 benchmark (#5806) 2024-07-30 12:05:36 -07:00
qazal
03d866b84f UOps.IF with rewrite rules (#5812)
* expand merge

* merge barriers

* gate_folder

* test_linearizer_failures

* this can be here

* bring the new repr back

* gate_folder2

* gate_creator is better

* gate_folder

* dedup conditions

* early gate folding

* dedup barrier

* fold noop conditions

* all consts can go away

* free lines
2024-07-30 20:50:56 +03:00