nihui
9858d203df
padding handle autopad same lower, bottom_shapes is hint only, respect the actual one_blob_only property, fix #1828
6 years ago
Tijmen Verhulsdonck
e3b31511ad
Added AVX implementation to cast to/from bfloat and float32 ( #1836 )
6 years ago
nihuini
16aa0bff7b
fix cpu affinity mask unsigned overflow
6 years ago
nihuini
f72b394588
skip crop for identical shape
6 years ago
nihuini
554890cda8
fp16p and fp16s cannot be both enabled in shader source
6 years ago
nihuini
1a3a99d7c9
old qcom driver cannot handle binding id alias
6 years ago
Tijmen Verhulsdonck
da09e5e7f1
Adding channel padding support for blazeface model. ( #1826 )
* Add channel padding and blazeface model support.
* remove python binding
* remove std::min usage
* fix reference blob usage
* Increased padding test coverage
* implement requested changes
6 years ago
tpoisonooo
8e1c3ac4d1
Add crop para check ( #1825 )
* add copy_cut_border check; fix compile warnings
6 years ago
JackieWu
ce2251db05
Improve ROIAlign (accelerate ROIAlign, support sampling ratio and aligned ROIAlign) ( #1820 )
* update roialign
* update tool for roialign
* fix ceil for roialign
* fix ios build
* fix
* fix
* make it compatibile with the old version
* fix ios build
* trigger CI
* fix test
* order
* build
* remove code
* merge roialign
* accelerate ROIAlign
* update note
* rename func
* roialign version
* trigger CI
* fix roialign
* use ref for pre-calc in roialign
* retrigger CI
* pre-alloc
6 years ago
JackieWu
c61a60bfc6
fix cmake ( #1823 )
6 years ago
zhiliu6
63d7e2c88d
Add support for darknet EfficientNetB0-Yolov3 conversion. ( #1821 )
* fuse mish and depthwise convolution
* darknet2ncnn: support enet-coco conversion.
6 years ago
nihui
5178724d79
static openmp with ndk r21+
6 years ago
nihui
13c59d9ad2
parse float by ourself, drop locale hack
6 years ago
nihui
1b27ae4e1c
fix msvc build
6 years ago
nihui
509a14e4a6
enable per thread locale unconditionally
6 years ago
nihui
23f79a661d
force c locale for paramdict parsing
6 years ago
nihuini
64985809a3
fix crash in load_model when gpu is not used
6 years ago
nihui
478e15d7dc
fix memorydata test on nvidia gpu
6 years ago
nihuini
f87f21779f
resolve cast from type properly, no more fp16p to/from fp16s conversion
6 years ago
nihuini
bb56b5439f
fix vkmat download on integrated gpu, workaround priorbox fp16s with online spirv, fix #1700 fix #1805
6 years ago
nihuini
989b0f70cc
convert shader source to hex data at build time
6 years ago
nihui
8fec0038ba
fix ci test
6 years ago
nihuini
116869594c
fix cpu-only build
6 years ago
nihuini
aeba24b371
enable implicit fp16a on arm mali variants, add bug tag for layout binding id alias
6 years ago
nihuini
054ec09195
adreno device blacklist
6 years ago
nihuini
c16105bcbd
use android log printf for easier error reporting
6 years ago
nihuini
765003a615
fix build with old vulkan sdk
6 years ago
tpoisonooo
c1cf258e21
Approximately preproc ( #1788 )
* add nv21 to rgb with half resize
* remove const decoration for int
6 years ago
zhiliu6
d23cef320c
Add Swish layer ( #1799 )
* add swish layer
* add swish test
6 years ago
nihui
2133121e6d
destroy descriptorsets properly
6 years ago
nihui
85fb76a6b9
clear all internal state on command reset
6 years ago
nihuini
6788384595
query gpu heap budget api
6 years ago
nihuini
cbdaf4b3e1
wrap cmake variables with ncnn prefix
6 years ago
nihui
17c445480f
runtime spir-v compilation with libglslang ( #1779 )
6 years ago
nihuini
d232272db0
lower end gpu friendly
6 years ago
xfan1024
f9a66465c2
fix compile error ( #1775 )
6 years ago
nihuini
b71f22d074
report adreno info, benchncnn enable image storage on adreno
6 years ago
nihuini
ed7bca6de0
deepcopy image shader
6 years ago
nihuini
c94d1b39ad
force diable image storage on macos and ios, fix #1738
6 years ago
Charles.Wang
72ac36ea60
add statisticspooling layer ( #1768 )
* add statisticspooling layer
* update my license, delete "#include <algorithm>, #include <functional>" in src/layer/statisticspooling.cpp, add ncnn_add_layer entry in src/CMakeLists.txt.
* Move ncnn_add_layer(StatisticsPooling) to tail of the layer list. Because the layer enum index is generated from this order. Inserting would break binary compatibility.
6 years ago
nihui
ce6abe24b8
memory type required by buffer and image can be different
6 years ago
zhiliu6
3bfabf1d6a
Add fused convolution and mish layer support. ( #1761 )
6 years ago
nihui
c35bf2f091
more stable tanh_ps, fix #1769
6 years ago
nihui
b85ba38e5b
scale image shader
6 years ago
nihui
4a4fa0889a
instancenorm image shader
6 years ago
nihuini
36cbf3e3af
batchnorm image shader
6 years ago
nihuini
5b4f648834
prelu image shader
6 years ago
nihuini
b5f85eee13
fix image1d_xx8 macro, normalize image shader
6 years ago
nihuini
1c045620a6
silence interp resize_type warning during ncnnoptimize
6 years ago
nihuini
a394f198e4
lrn image shader
6 years ago