mirrors/tinygrad
2
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-24 02:14:17 +00:00
Code Issues Projects Releases Packages Wiki Activity
13,471 commits 280 branches 15 tags 325 MiB
Commit graph

13,471 commits

This branch
This branch
All branches
Author SHA1 Message Date
b1tg
0fbc551622
 		train bert with fp8 (#13874) 
* fp8 train

* clean

* lint

* test fix from #13439

* skip first/last layer

* rm __init__, restore unroll <=32 check

* tests

* clean test, remove unused

* multi-gpu test, clean quantize_to_fp8

* remove bert contiguous

* run script

* test: better check

* run script search

* add seed in bert data shuffle

* move script to mi350x folder

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2026-01-09 09:21:59 -05:00
nimlgen
ba209d6305
 		am: utc_l1_enable on all sdma inst (#14083) 2026-01-09 17:17:05 +03:00
nimlgen
6b308b89b7
 		viz: timeline time (#14080) 
* viz: timeline time

* less lines

* cut
 2026-01-09 16:43:45 +03:00
nimlgen
40f9fa2db4
 		autogen: new kfd (#14082) 2026-01-09 16:08:17 +03:00
qazal
2917ed1616
 		roc: propagate decoder errors to main thread (#14081) 
* roc: propagate decoder errors to main thread

* types

* add cause
 2026-01-09 21:10:45 +09:00
qazal
f3f4d9b387
 		viz: fix disasm node width (#14079) 2026-01-09 16:37:37 +09:00
anu
c70c112254
 		fix CUDA=1 disassembly (VIZ=1) by stripping null terminator (#14046) 
* fix ptxas disassembly bug

* single '

* move fix to get_bytes

* move rstrip

---------

Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com>
 2026-01-09 15:19:59 +09:00
qazal
13e5d00d0e
 		viz: exclude comma in register highlight (#14078) 
* viz: exclude comma in register highlight

* simplify
 2026-01-09 15:10:30 +09:00
qazal
a071adffc0
 		viz: amdgpu disassembly register highlighting UI (#14059) 
* viz: amdgpu disassembly register highlighting

* minor details

* details from IDA

* more details from IDA

* refactor token colors

* move tokenizer to python

* simplify

* minimal tokenizer for registers

* all the operand types
 2026-01-09 11:27:09 +09:00
chenyu
b878f9d5a4
 		reuse Tensor init with const path [pr] (#14076) 2026-01-08 17:49:37 -05:00
chenyu
efcb32f6a9
 		unique const when requires_grad is set to True (#14075) 
* unique const when requires_grad is set to True

* fix pyrender
 2026-01-08 16:30:45 -05:00
chenyu
b34c637767
 		support bfloat16 for CL (#14073) 2026-01-08 14:14:29 -05:00
Garret Castro
16b652302e
 		skip bf16 test if not supported by device (#14070) 2026-01-08 13:37:24 -05:00
nimlgen
3f61a96d79
 		am: SetSoftMaxByFreq on gfx10+ (#14068) 2026-01-08 17:00:03 +03:00
George Hotz
e7b5d8a434
 		assembly/amd: more RDNA4 asm (#14062) 
* rdna4 more

* asm

* fixes

* assembly/amd: handwritten wmma failing test

* passes

* wmma default hacks

* space

* 0 skips in rdna3/rdna4 disasm

* more RDNA4 tests

---------

Co-authored-by: qazal <qazal.software@gmail.com>
 2026-01-08 05:09:37 -08:00
nimlgen
e372c841ba
 		hevc: beam in decode (#14067) 
* hevc: beam in decode

* fine

* g
 2026-01-08 15:47:16 +03:00
nimlgen
1732a4ec4b
 		am: rework set_clocks (#14065) 2026-01-08 15:33:32 +03:00
nimlgen
f3aceaa08b
 		hevc: fast decoder (#14057) 2026-01-08 15:20:37 +03:00
qazal
309197bca5
 		assembly/amd: test_roundtrip for cdna/rdna4 (#14066) 2026-01-08 21:03:13 +09:00
qazal
15a056715d
 		fix amd assembly IDE tests on macbook (#14063) 2026-01-08 17:27:52 +09:00
wozeparrot
027b935269
 		tk: fix grouped load store (#14035) 2026-01-07 22:38:02 -08:00
George Hotz
2db04d0696
 		assembly/amd: start adding RDNA4 support (#14060) 
* assembly/amd: start adding RDNA4 support

* rdna4 asm
 2026-01-07 21:19:30 -08:00
George Hotz
cb500466c2
 		assembly/amd: amd_asm_matmul (#13989) 
* amd_asm_matmul

* dsl transform

* asm roundtrip

* fixed

* less

* better

* more

* simpler

* simplify

* lil

* simpler

* compact

* work

* cleanups

* simplify

* simpler

* cleanup

* name the regs

* simp

* big simp

* big simp

* simp

* acc grid

* fast

* stuff

* fast

* simpler

* owrks

* save vgprs

* save vgprs

* Compact

* less VGPRs

* after

* SQTT support

* fastest

* faster

* lil faster

* tile regs

* faster

* readable

* one more

* simpler

* lil simpler

* NO_GLOBAL skips early globals

* stock kernel

* cleanups

* cleanups

* one b reg

* safe reg changes

* acc is compact now

* remove confusing stuff

* sregs

* lds cleanups

* vopd
 2026-01-07 20:11:05 -08:00
chenyu
3caa1e2c98
 		fix cast HALF with PYTHON backend (#14058) 2026-01-07 16:52:05 -05:00
chenyu
5f1ede7f7e
 		clean up test_dtype (#14055) 
use less lambda
 2026-01-07 15:45:42 -05:00
nimlgen
5bd4593eda
 		hevc: cleaner decoder (#14056) 
* hevc: cleaner decoder

* nn
 2026-01-07 18:29:30 +03:00
b1tg
241f0402b4
 		add seed in bert data shuffle (#14054) 2026-01-07 10:02:05 -05:00
nimlgen
25c82dd242
 		nv: profile nvdec (#14053) 2026-01-07 15:56:54 +03:00
qazal
35900290b2
 		viz: configure text height for cfg (#14052) 2026-01-07 18:58:56 +09:00
chenyu
87f4bc5446
 		update variable names around jit [pr] (#14049) 
lbs, st_vars_dtype_device and rawbuffers no more
 2026-01-06 22:32:41 -05:00
chenyu
2833c5a54b
 		few more jit tests with multi tensor inputs (#14047) 2026-01-06 22:05:22 -05:00
chenyu
72a3f78d19
 		jit includes tensor inputs in containers (#14043) 
* jit includes tensor inputs in containers

* cleanup
 2026-01-06 19:42:06 -05:00
chenyu
c714881832
 		don't allow jit input to be const (#14045) 
* don't allow jit input to be unbuffered like const

* just const to fix multi

* fix rnnt
 2026-01-06 18:15:22 -05:00
chenyu
a8896f28e1
 		test_unrealized_const_input_frozen (#14044) 
unrealized const is not replaced in jit
 2026-01-06 14:17:43 -05:00
nimlgen
325f4006ff
 		amd: copies w/o sdma (#14036) 
* amd: copies w/o sdma

* as_args

* fixes

* f
 2026-01-06 21:15:58 +03:00
chenyu
7fb18f7e47
 		raise when jit fxn returns non-Tensor output (#14042) 2026-01-06 12:59:20 -05:00
chenyu
4491ec0c9e
 		JitError (#14041) 
* JitError

* test_symbolic_jit
 2026-01-06 12:19:50 -05:00
chenyu
6ddddc68af
 		test jit tolist failure (#14040) 
also moved tests to test_jit_footguns
 2026-01-06 11:16:57 -05:00
chenyu
b699b9f763
 		test case for jit a function with item call (#14039) 
* test case for jit a function with item call

output is silently wrong now

* no dtype
 2026-01-06 10:40:43 -05:00
nimlgen
02084f5376
 		mockdsp: use dsp allocator (#14037) 
* mockdsp: use dsp allocator

* fix

* ?
 2026-01-06 16:04:47 +03:00
wozeparrot
2b3e01e79c
 		tk: support sliced local -> reg load (#14034) 2026-01-06 05:33:24 -05:00
George Hotz
45f7fd073d
 		assembly/amd: pcode bug fixes (#14032) 
* bring over pcode parser

* fixes

* pdf test

* delay alu
 2026-01-06 00:15:48 -08:00
wozeparrot
21d0f6bb76
 		tk: flat global -> local load (#14033) 2026-01-05 23:35:53 -08:00
qazal
3170365a5b
 		visualize SQTT with the same cfg infrastructure (#13870) 
* start

* rough sketch

* post render dag

* art

* intro g key

* work

* custom color scale

* colors

* more blue

* better

* smaller

* use for loop in test
 2026-01-06 14:53:20 +09:00
Christopher Milan
0120d69caa
 		autogen: avcodec (and simplify workflow) (#14031) 
* simplify autogen workflow and add avcodec verification

- Consolidate all regeneration into single steps (delete + import)
- Remove continue-on-error and individual diff checks
- Use git diff at end to catch all differences
- Show artifact URL in failure message
- Add avcodec.py verification

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* patch avcodec

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-05 23:30:25 -05:00
George Hotz
20653d2996
 		assembly/amd: make pdf.py code shine (#14029) 
* assembly/amd: make pdf.py code shine

* no merge

* pdf2 is the future

* something

* regen enums

* test

* work

* remove junk

* write

* pcode extraction

* pdf2 passes all tests

* simplify

* simpler pdf

* late filter

* remove hacks

* simplify pdf2.py

* field type

* remove defaults

* don't export srcenum

* simple pdf.py

* simpler

* cleaner

* less hack in PDF
 2026-01-05 18:49:40 -08:00
qazal
ea7b149ca5
 		viz command line tool (#14030) 2026-01-06 10:19:47 +09:00
Christopher Milan
f86c728440
 		load libclang as 'libclang.so' too (#14028) 2026-01-05 16:56:16 -05:00
chenyu
eda6a73897
 		clean up canonicalize_device (#14027) 
centralize the type check
 2026-01-05 10:29:55 -05:00
chenyu
ce464b147a
 		clean up comments that mentioned outdated terms (#14026) 
no MultiLazyBuffer and no ShapeTracker in comments
 2026-01-05 09:42:58 -05:00