nihuini
3631c1933d
non-inlined addref and release slows down overall speed, move them to header
2021-04-07 18:59:19 +08:00
nihui
e9cc637573
arm neon optimization for int8 packing kernels ( #2809 )
2021-04-07 14:06:29 +08:00
nihui
d7cbc055f3
fix illegal instruction on pi4 when NCNN_ARM82 enabled
...
compiler may compile inline member functions as noinline blocks for different architectures, and linker may pick the newer arch, that results illegal instructions on old hardware
2021-04-07 11:08:55 +08:00
teng
cc8e7a13fa
Update operation-param-weight-table ( #2811 )
2021-04-02 22:27:05 +08:00
DC Technology
86e53b35db
add Apple M1 benchmark ( #2810 )
...
* add Apple M1 benchmark
* Update README.md
2021-04-02 16:08:37 +08:00
nihuini
256754bff9
fix build with old gcc, fix #2805
2021-04-01 10:51:42 +08:00
teng
a90a31d340
doc spelling mistakes ( #2804 )
2021-04-01 00:33:45 +08:00
zhiliu6
61cd9da55b
optimize x86 3x3 pack8 leftover ( #2797 )
2021-03-30 15:14:28 +08:00
nihuini
912e81d086
fix tanh neon, fix #2751
2021-03-29 19:51:18 +08:00
nihuini
6c41822d35
do not remove optional hidden output for lstm/gru/rnn
2021-03-29 19:10:22 +08:00
nihuini
5e85f447e6
convert mxnet channel padding, fix some lgtm warnings
2021-03-29 15:07:08 +08:00
Cai Shanli
8011aea8c8
Update release-python.yml ( #2794 )
...
remove build wheel for ppc64le
2021-03-29 10:38:37 +08:00
nihui
32b48f0157
fix int8 auto pack layout
2021-03-28 22:58:44 +08:00
nihui
1ea8bfbd2e
x86 avx2 conv3x3s1 pack8 direct optimization, fix #2789
2021-03-28 11:30:51 +08:00
ncnnnnn
6e6cb9f4f3
simple sort ncnn_add_layer_test ( #2790 )
...
for obsessive
2021-03-27 22:46:28 +08:00
nihui
a48bf43ef7
test conv/fc int8 with activation
2021-03-27 20:06:18 +08:00
nihui
5fe75f19ef
architecture changes for int8 packing ( #2771 )
...
* quantize and dequantize tests
* unify activation and usability function
* drop NCNN_REQUANT cmake option, test dequantize requantize pack8, fix webassembly build
* benchmark use requantize int8 model
2021-03-27 18:49:13 +08:00
nihui
d4a7abc218
fix onnx2ncnn clip without max blob, fix #2788
2021-03-26 23:58:35 +08:00
nihui
eaee64c782
export vkcommand and vkcompute
2021-03-25 00:00:34 +08:00
nihuini
e86799e95f
fix get_big_cpu_count return zero on smp cpu
2021-03-22 19:13:50 +08:00
nihui
7c079b853e
default to big cpu count
2021-03-22 18:50:51 +08:00
5f00ba89d2
feat(ncnnoptimize): replace denormals to zero on layers with weights ( #2690 )
...
* feat(ncnnoptimize): replace denormals to zero on layers with weights
Co-authored-by: youngsoo.lee <youngsoo15.lee@gmail.com >
Co-authored-by: Restyled.io <commits@restyled.io >
2021-03-22 10:26:57 +08:00
nihui
19c183cacc
fix ncnnoptimize default cut range
2021-03-21 22:47:35 +08:00
5e565cfa8a
Restyle add net Cut Function ( #2763 )
...
* add ncnnoptimise cut net function
* add ncnnoptimise cut net function
* Restyled by clang-format
* Restyled by astyle
* Restyled by clang-format
* Restyled by astyle
Co-authored-by: chenty <admin@chenty.com >
Co-authored-by: Restyled.io <commits@restyled.io >
2021-03-19 22:14:37 +08:00
Cai Shanli
e9a61b3e64
fix python return value policy ( #2756 )
2021-03-16 21:46:53 +08:00
nihui
67e24e0703
use local pool allocator ( #2736 )
...
* use local pool allocator
* detach extract feat from local allocator
* fix test
2021-03-16 16:23:28 +08:00
Cai Shanli
64ef781a9c
add test python vulkan ( #2754 )
...
* add test python vulkan
* add python vulkan ci on linux with 3.6 and 3.8
* fix ci vulkan icd
2021-03-15 23:00:33 +08:00
nihuini
15d63ec0f5
fuse onnx multiheadattention with same qkv blob
2021-03-15 19:08:44 +08:00
Cai Shanli
f5b307689b
fix net and extractor destroy order when use vulkan ( #2732 )
2021-03-13 20:54:06 +08:00
RBelogorodtsevFBase
1212ed6e94
implements gelu activation ( #2749 )
2021-03-12 22:25:25 +08:00
PENGUINLIONG
2d31868389
Trouble shooting guide for Vulkan on NVIDIA GPUs ( #2731 )
2021-03-06 20:27:13 +08:00
nihui
b58cd14678
fix non arm-neon build
2021-03-06 15:46:26 +08:00
nihui
0870bf45b1
optimize warpaffine family
2021-03-06 15:38:06 +08:00
nihuini
e449435dbe
fix mlir2ncnn warning, prettier alignment :)
2021-03-05 16:33:29 +08:00
nihuini
c8ccccf045
adapt mlir changes
2021-03-05 16:21:25 +08:00
nihuini
c17eb4e208
multiheadattention layer
2021-03-05 11:45:02 +08:00
Zhiqiang Wang
2c370b2c58
Fix typo in docs ( #2727 )
2021-03-04 19:56:15 +08:00
nihuini
b51959802c
fix buffer2host copy, fix #2725
2021-03-04 16:49:25 +08:00
nihuini
b0d16325b1
fuse onnx binaryop with scalar
2021-03-04 11:40:29 +08:00
Cai Shanli
2edb3ed7a4
nanodet python demo ( #2723 )
...
* nanodet python demo
* add clip
* fix clip wh
* remove nms package requirement
2021-03-03 22:17:17 +08:00
nihuini
f7cbcaa72b
fix onnx normalize expand ghost shape
2021-03-03 17:21:45 +08:00
nihuini
c910574b5b
fuse onnx multiheadattention
2021-03-03 16:25:33 +08:00
teng
c3466a7798
fix array index out of bounds in examples/yolact.cpp ( #2722 )
2021-03-02 20:22:23 +08:00
RangiLyu
ecf1f413b4
fix duplicate variable name in examples/nanodet.cpp ( #2719 )
2021-03-02 18:34:33 +08:00
nihuini
f2a5ea7678
fix layernorm ghost input without affine
2021-03-02 17:32:36 +08:00
nihuini
7ac23ab34d
fuse onnx layernorm, fix 2-dim layernorm implementation, add test
2021-03-02 15:42:03 +08:00
zhiliu6
57397c418d
Optimize general AVX2 convolution. ( #2714 )
2021-03-01 08:30:20 +08:00
Xu Yang
fd634e9a58
remove unnecessary mat clone when NCNN_BENCHMARK enabled ( #2708 )
2021-02-25 19:05:17 +08:00
Dahan Gong
cbd410c237
fix broken inplace forward ( #2709 )
2021-02-25 19:04:15 +08:00
8c9bea2322
Restyle faster bbox calculation by background score ( #2693 )
...
* faster bbox calculation by background score
* Restyled by clang-format
* Restyled by astyle
* Restyled by clang-format
* Restyled by astyle
Co-authored-by: Qoo <r97922153@gmail.com >
Co-authored-by: Restyled.io <commits@restyled.io >
2021-02-24 23:39:12 +08:00
nihuini