nihui
d476191ff1
pnnx export_onnx function ( #3784 )
4 years ago
nihui
0aba8af1d3
pnnx swin transformer ( #3783 )
4 years ago
村长大人
c697e988b0
fixbug: linux-arm-hisiv500 load mem "Bus error" ( #3779 )
4 years ago
Zhiqiang Wang
a7aa0fe70d
Update README.md ( #3778 )
4 years ago
Zhuo Zhang
3dfc10647c
docs: update QQ group info ( #3777 )
4 years ago
NaLan ZeYu
5388f9f312
test: fix printf arguments mismatch ( #3774 )
4 years ago
jasonZhang
663b42e0d2
add tanh avx512 optimize ( #3770 )
4 years ago
nihui
f9c1787de9
implement einsum layer and pnnx conversion ( #3768 )
4 years ago
jasonZhang
c34e305902
Add SSE&AVX optimized for tan ( #3765 )
4 years ago
nihui
ee6402553c
layernorm for vector and mat along w, pnnx convnext end2end test ( #3764 )
4 years ago
nihui
318dc4f5f5
convert Tensor.select to ncnn, fuse multiheadattention unpack, add vit_b_32 end2end test ( #3763 )
4 years ago
nihui
b516485528
update ci qemu ( #3756 )
* update ci qemu
* update rvv qemu args
* fix bus error on armv7
* apply downstream qemu patches
4 years ago
dependabot[bot]
2f2e678cce
Bump docker/setup-qemu-action from 1 to 2 ( #3761 )
Bumps [docker/setup-qemu-action](https://github.com/docker/setup-qemu-action ) from 1 to 2.
- [Release notes](https://github.com/docker/setup-qemu-action/releases )
- [Commits](https://github.com/docker/setup-qemu-action/compare/v1...v2 )
---
updated-dependencies:
- dependency-name: docker/setup-qemu-action
dependency-type: direct:production
update-type: version-update:semver-major
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
4 years ago
nihui
a2b25c88b0
pnnx convert torch.einsum ( #3744 )
* normalize einsum equation
* replace dot pair with alphabet word
* shorten permute equation
4 years ago
841973620
a2e25d3fd3
fix Multiplication result may overflow 'int' before it is converted to 'size_t'. ( #3757 )
4 years ago
jasonZhang
e62d674e5d
Add unittest and SSE&AVX optimized for BNLL ( #3759 )
4 years ago
nathannaveen
d7cc08b54a
chore: Set permissions for GitHub actions ( #3758 )
Restrict the GitHub token permissions only to the required ones; this way, even if the attackers will succeed in compromising your workflow, they won’t be able to do much.
- Included permissions for the action. https://github.com/ossf/scorecard/blob/main/docs/checks.md#token-permissions
https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#permissions
https://docs.github.com/en/actions/using-jobs/assigning-permissions-to-jobs
[Keeping your GitHub Actions and workflows secure Part 1: Preventing pwn requests](https://securitylab.github.com/research/github-actions-preventing-pwn-requests/ )
Signed-off-by: nathannaveen <42319948+nathannaveen@users.noreply.github.com>
4 years ago
dependabot[bot]
1634de33df
Bump pypa/cibuildwheel from 2.4.0 to 2.5.0 ( #3755 )
Bumps [pypa/cibuildwheel](https://github.com/pypa/cibuildwheel ) from 2.4.0 to 2.5.0.
- [Release notes](https://github.com/pypa/cibuildwheel/releases )
- [Changelog](https://github.com/pypa/cibuildwheel/blob/main/docs/changelog.md )
- [Commits](https://github.com/pypa/cibuildwheel/compare/v2.4.0...2.5.0 )
---
updated-dependencies:
- dependency-name: pypa/cibuildwheel
dependency-type: direct:production
update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
4 years ago
nihui
615d63e6a1
use single thread in ci swiftshader ( #3754 )
4 years ago
nihui
40c50f198b
move linux gpu ci to cvm2 ( #3750 )
4 years ago
nihui
5cdb9a8f4a
move pnnx ci to cvm hosted ( #3749 )
4 years ago
nihui
fff16a025d
pnnx fuse static conv ( #3743 )
4 years ago
nihui
f9c4e90962
catch more foldable constant in static shape inference mode ( #3738 )
4 years ago
Lry89757
ca9abd1c4a
Update the add-custom-layer.zh.md ( #3741 )
1. 🐛 Fix Bug of float and int. 修复了std::max()中参数int和float参数不符合的Bug
2. 👀 The structure of ncnn changes. ncnn文件结构变动,所有testlayer.cpp改成在tests/文件夹中
4 years ago
dependabot[bot]
4e6b5d9779
Bump github/codeql-action from 1 to 2 ( #3737 )
Bumps [github/codeql-action](https://github.com/github/codeql-action ) from 1 to 2.
- [Release notes](https://github.com/github/codeql-action/releases )
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md )
- [Commits](https://github.com/github/codeql-action/compare/v1...v2 )
---
updated-dependencies:
- dependency-name: github/codeql-action
dependency-type: direct:production
update-type: version-update:semver-major
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
4 years ago
BUG1989
30dec7ae28
Update the benchmark of jetson agx xavier ( #3735 )
4 years ago
nihui
57f3017afd
pnnx torch bmm max min ( #3736 )
* convert torch.bmm
* convert torch.min torch.max
4 years ago
BUG1989
f14388a150
Add the benchmark of Jetson AGX Orin ( #3733 )
4 years ago
FeiGeChuanShu
3f6fbcc7a2
add auto num_class for yolov5_pnnx example ( #3729 )
4 years ago
Yoh
bf1109c10b
fix elempack16 bug ( #3728 )
4 years ago
Xavier Hsinyuan
cea2aff3bb
RVV Optimized erfc and GELU ( #3726 )
* RVV optimized GELU(fp32) and erfc(fp32)
4 years ago
jasonZhang
98e35ded36
arm neon optimization for arcsin/arccos pack4 ( #3725 )
4 years ago
nihui
6125c9f47c
add tips for disabling android ui and setting cpu/gpu performance mode
4 years ago
nihui
b4a31638f9
update rk3399 rk3288 gpu benchmark data
4 years ago
jasonZhang
8e2e4db3f4
arm neon optimization for tan pack4 ( #3722 )
Co-authored-by: jasonZhang892 <jasonZhang892@users.noreply.github.com>
4 years ago
nihui
308965b7e9
sanitize cooperative matrix option in tests
4 years ago
nihui
50fa6d39c0
enable fp16a for mali t760 v2
4 years ago
nihui
7600270430
create uop in spirv-1 mode for vulkan 1.0 compatibility ( #3721 )
4 years ago
nihui
a061871c1c
use getauxval since android api 18 ( #3718 )
4 years ago
UNeedCryDear
f3129be69a
Reduce yolov5 post-processing time. ( #3649 )
* Reduce yolov5 post time.
sigmoid(X)<1,
IF:box_confidence<prob_threshold
THEN: confidence=box_confidence*sigmoid(class_score)<box_confidence<prob_threshold,
So there is no need to calculate class_score when box_confidence<prob_threshold,and the confidence of most anchors is lower than the prob_threshold.
Co-authored-by: UNeedCryDear <UNeedCryDear@users.noreply.github.com>
4 years ago
nihui
ae75a093fa
update print 4d mat, remove deprecated content
4 years ago
nihui
88da92f9bf
fix pnnx macos install rpath
4 years ago
nihui
a14acbc6dd
pnnx save ncnn bin with fp16 storage ( #3715 )
4 years ago
nihui
0623f94eca
arm neon optimization for softmax ( #3713 )
4 years ago
nihui
0ea327b557
x86 sse/avx/avx512 optimization for softmax ( #3712 )
4 years ago
nihui
9826f3dbf8
shader include vulkan activation, workaround for moltenvk tanh half4 issue ( #3711 )
4 years ago
jasonZhang
2b0064c3db
fix armv7 platform build tests error ( #3710 )
4 years ago
nihui
49e70c81a6
update linking glslang libraries
4 years ago
nihui
376f14bba4
windows ci matrix ( #3709 )
4 years ago
nihui
eadb2381dd
force fp32 accumulator for global pooling ( #3708 )
4 years ago