github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-14 17:38:06 -05:00
Code Issues Packages Projects Releases Wiki Activity
11,106 Commits 203 Branches 14 Tags
python_speed
Commit Graph

11106 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
qazal
c7bb561ef9 remu: add v_rsq_f32_e32 instruction (#11947) 
https://github.com/tinygrad/tinygrad/pull/11936 introduces a change to
the AMD LLVM renderer that outputs this instruction. Adding both 32 and
64 bit variants.
 2025-09-01 11:29:31 +03:00
Sieds Lykles
d9560a631c remove cast between ints if safe (#11946) 2025-09-01 05:56:49 +02:00
Sieds Lykles
a19d689481 fix vec dtype _min_max (#11944) 2025-09-01 03:24:07 +02:00
Sieds Lykles
f32f3464d6 Can safe cast from certain ints to floats (#11941) 
* add rule

* add some tests

* prevent infinite loop with bfloat16

* add some ints to double and float can_safe_cast

* add tests
 2025-09-01 00:51:24 +02:00
Sieds Lykles
1c6e43c203 Double cast is one cast if intermediate cast is safe (#11939) 
* add rule

* add some tests

* prevent infinite loop with bfloat16

* prevent more infinite rewrite
 2025-09-01 00:36:29 +02:00
wozeparrot
7e68045fb2 feat: small llama3 training (#11829) 2025-08-31 13:41:47 -07:00
nimlgen
020abe0556 hcq: finalize without synchronization when in error state (#11872) 
* hcq: finalize without synchronization when in error state

* ooops

* fix

* fix

* fix
 2025-08-31 18:39:13 +03:00
qazal
2004c9757d tracing: add default clock (#11935) 2025-08-31 18:24:44 +03:00
b1tg
c1eeb3b99c only skip AMD_LLVM (#11934) 
Co-authored-by: b1tg <b1tg@users.noreply.github.com>
 2025-08-31 18:15:47 +03:00
b1tg
75d380a77c fix transcendentals in python renderer (#11932) 
* fix transcendentals in python renderer

* add test

---------

Co-authored-by: b1tg <b1tg@users.noreply.github.com>
 2025-08-31 09:37:17 -04:00
Sieds Lykles
61e4dc6ad5 render special arg in cstyle if arg is UOp (#11931) 2025-08-31 07:01:29 +02:00
Sieds Lykles
d3252ccd85 fix special vmax when arg is UOp (#11930) 2025-08-31 06:54:39 +02:00
qazal
0bacd9fc9b viz: give disassembly its own node (#11927) 2025-08-31 00:28:52 +03:00
chenyu
af89be317e relax rtol for bfloat16 test_dtype_alu (#11926) 2025-08-30 17:16:08 -04:00
George Hotz
632c2fb119 lowerer works on rangeifed + print exception (#11925) 2025-08-30 12:05:44 -07:00
qazal
c27b99d68f viz: refactor to indexed rewrite traces (#11923) 2025-08-30 20:01:47 +03:00
qazal
9aff00a6ea switch viz command line args to pathlib (#11922) 2025-08-30 18:13:47 +03:00
qazal
c86ee5bfaf viz: canonicalize device name colors (#11921) 2025-08-30 18:12:30 +03:00
nimlgen
a4f05ebd1a ci: rebuild gpuocelot with boost libs (#11920) 2025-08-30 17:24:19 +03:00
qazal
bf0d055b39 viz: color by name (#11919) 2025-08-30 16:04:58 +03:00
Sieds Lykles
0bc34c000f simplify range mod its own upper bound (#11917) 
* add rules

* add tests
 2025-08-30 08:37:35 +02:00
chenyu
561318fea7 Tensor.cos in test_stype_alu (#11916) 
* Tensor.cos in test_stype_alu

* need this fix anyway
 2025-08-29 20:26:36 -04:00
NoahKusaba
0838021753 remove np from beautiful_cifar (#10988) 
* remove np from beautiful_cifar

* remove np from cifar

* rename variable and rename tensor.arrange to just tensor.randperm

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2025-08-29 19:34:16 -04:00
nimlgen
cf9d8c8142 ci: pin boost for macos runners (#11910) 2025-08-30 01:38:06 +03:00
nimlgen
c6e342cdac mockgpu: no hang if gpuocelot failed (#11915) 2025-08-30 00:44:49 +03:00
chenyu
26d03a86a1 test_symbolic_ops.py cleanup (#11895) 2025-08-29 17:11:59 -04:00
b1tg
b2cc06218a python bfloat16 (#11912) 
* python bf16

* _to_torch_storage_type

---------

Co-authored-by: b1tg <b1tg@users.noreply.github.com>
 2025-08-29 15:18:02 -04:00
George Hotz
afad7d0cd1 remove dtype from range, it will be dtypes.index soon [pr] (#11914) 
* remove dtype from range, it will be dtypes.index soon [pr]

* a few more
 2025-08-29 09:52:07 -07:00
qazal
30e72d5820 multi device and copy tracing for NULL device (#11913) 
* add device name to NULL programs

* trace transfers
 2025-08-29 15:31:00 +03:00
qazal
d8e1e4dc61 tracing: show NULL programs (#11911) 2025-08-29 14:09:33 +03:00
nimlgen
75678b2cbe amd: retire pm4 xcc sync (#11835) 
* amd: aql default when several xccs

* amd: retire om4 xcc sync

* remove more

* more

* more
 2025-08-29 09:56:27 +03:00
George Hotz
394c2d1db1 update Kernel API in tests + move optimize_local_size (#11907) 2025-08-28 15:12:47 -07:00
nimlgen
fa695ac1ce ci: mac gpuocelot (#11906) 
* gm

* fix?

* ops

* imp

* xx

* add file
 2025-08-28 23:29:43 +03:00
George Hotz
b9b438c516 small updates from postopt (#11903) 
* tests from postopt

* modernize

* skip lin tests

* that's fixed?

* skip, not failure
 2025-08-28 12:34:52 -07:00
nimlgen
bb55a3001f nv: flush reset message (#11897) 2025-08-28 22:17:20 +03:00
nimlgen
e8289c75b1 ci: do not reinstall existing pkgs in macos (#11900) 2025-08-28 21:20:15 +03:00
chenyu
134cf56904 update cache name for gpuocelot (#11896) 2025-08-28 13:11:10 -04:00
Ben Waldron
ea1be2e4cd [bounty] Remove using reshape to register symbolic shape (#11771) 
* Modify tests and start work towards removing symbolic reshape

* Refactor symbolic reshape

* fix small error

* much cleaner + fix more tests

* Can remove this now

* Update test_symbolic_ops and test_tiny

* Couple more tests

* Unused import

* More tests and add EXPAND to Tensor.empty

* Fix test beam search

* all int

* Fix rangeify by adding shrink

* Remove OOB check and so fix test_symbolic_jit

* test_symbolic_jit doesn't need OOB Context anymore either

* Should remove that test now

* Cleanups part 1

* fix linters

* Final cleanups

* Don't reassign inside for loop

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2025-08-28 12:30:49 -04:00
qazal
53853ae49b viz: switch to Path2D (#11892) 2025-08-28 18:58:16 +03:00
nimlgen
874c1db4af am: init support for aql (#11888) 2025-08-28 18:41:46 +03:00
Ben Waldron
17ecaf4682 Add test_variable_empty (#11889) 
* Add test_variable_empty

* Move test and add TODO

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2025-08-28 11:38:27 -04:00
Nino Risteski
54be477152 rope cache optim for jit prune in llm.py (#11678) 
* rope cache optim for jit prune

* rope test

* tests in test attention

* Revert "rope test"

This reverts commit 69ede543d0.

* lint
 2025-08-28 08:31:29 -07:00
quortus
5f8fe9a331 Replace ASSIGN with STORE in test_linearizer (#11821) 2025-08-28 07:33:20 -07:00
geohotstan
4e8370309c Support onnx If OP (#11648) 
* start

* tiny clean up

* whoops, didn't mean to accidentally fix this

* fix .to(device), kinda hacky and this fix makes it slower?

* merge properly

* FINALLY figured out slowness, also hack pylint for now

* add DEBUGONNX print for subgraph

* oops

* WOOOOOOOO SHAPE CACHE 50% SPEED INCREASE

* small fix, but maybe all deterministic Tensor creation in fp should be cached

* cache condition

* sliiiightly cleaner

* better abstraction?

* remove sam from model_benchmark

* remove shape cache speed up for now

* less lines

* isinstance fix

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2025-08-28 10:17:35 -04:00
George Hotz
6d6f0dada7 support for tuple ranges (#11890) 
* support for tuple ranges

* breaks it
 2025-08-28 07:02:31 -07:00
nimlgen
60dd9a162c memory: tiny tlsf cleanup (#11887) 2025-08-28 14:07:18 +03:00
chenyu
beb5982165 FUSE_ATTENTION (#11884) 2025-08-27 19:59:17 -04:00
George Hotz
cb5295168d postrange boilerplate work (#11881) 2025-08-27 15:22:59 -07:00
George Hotz
fd579433bc pre expander shouldn't go in gpudims (#11880) 2025-08-27 14:52:24 -07:00
nimlgen
44816218b5 memplan: fix large buffers planning (#11878) 
* memplan: fix large buffers planning

* fix

* fix dsp
 2025-08-27 23:54:27 +03:00