nihuini
cd7559c639
more fix for fp16p, still disabled by default
7 years ago
nihuini
4b6bffa560
Mat row should be elemsize-aware
7 years ago
harhar539
5e317b98c5
fix illegal memory access at conv layer of vulkan ( #1011 )
* 1.fix pad tail bug in commit d1ea2a3 at pooling layer
* fix illegal memory access at conv layer of vulkan
fix illegal memory access at conv layer of vulkan when bias term is 0
7 years ago
nihui
25b9736f82
shader fp16 packed
7 years ago
nihuini
4b50a97e31
implement vulkan winograd23
7 years ago
nihuini
37e150162a
do not retrieve timestamp availabitliy bits
7 years ago
nihuini
738fb6bb14
print gpu per layer benchmark
7 years ago
nihuini
8e2fb2e710
expose timestamp_period and timestamp_valid_bits
7 years ago
nihuini
c9a9486307
merge command submit and wait, expose queue_count, concurrent queue submission shall work
7 years ago
nihuini
2b21cf9e02
move mutex class family to platform.h
7 years ago
nihuini
aa94e77e68
fix pipeline object leak
7 years ago
kalcohol
a6aab42f95
add himix200 toolchain for Hi3516CV500, Hi3516DV300, Hi3519AV100. ( #989 )
7 years ago
nihui
07260527fc
fix activation params
7 years ago
nihui
3e003ffd98
fuse sigmoid
7 years ago
nihui
5adfa290a5
1x1s1d1_lds_4_4_4 is non-optimal, delete it
7 years ago
nihuini
8ac300c3a2
mat4 type in shared memory makes some driver unhappy ..
7 years ago
nihuini
f5ba97e7c6
lds optimize for conv3x3s1, conv1x1s1 and fc
7 years ago
nihuini
8322a14964
set fixed local size
7 years ago
nihuini
e46a3e428a
cmake warning--
7 years ago
nihuini
7a8f68aca6
move vulkan code to subdir, new layer interface create_pipeline and destroy_pipeline for post-loading works
7 years ago
nihui
d46c5989e1
mention practical-pytorch-to-onnx-to-ncnn
7 years ago
nihui
8a3955dde7
delete tensorflow2ncnn/pytorch2ncnn, it never works
7 years ago
nihuini
7c5fd53855
parse tensor info
7 years ago
nihuini
dbef56df47
write graph
7 years ago
nihuini
ed2e441a1d
fix write fp16 weight with tag
7 years ago
nihuini
c6e075cef7
fuse deconv/innerproduct relu arm
7 years ago
nihuini
81fda4d6d7
ncnnoptimize save fp16 storage weight
7 years ago
nihuini
01807482f2
parse pt file
7 years ago
nihuini
60e8261812
input param whc order
7 years ago
nihuini
48a1cb4052
more optimize routines ...
7 years ago
ShuangLiu1992
98bb0bb243
Update onnx2ncnn.cpp ( #948 )
* Update onnx2ncnn.cpp
num_filter for ConvTranspose is wrong (line 1075), should be the same sa Conv, which reads the number of filters from dim(0) (line 998)
int num_filter = W.dims(0);
* Update onnx2ncnn.cpp
7 years ago
Cocoa Oikawa
102a68d193
adding detailed how to build for jetson ( #953 )
* adding how to build for jetson
* update build instruction for NVIDIA Jetson
7 years ago
nihui
be81ecf1f6
fix build on msvc
7 years ago
nihui
a75d45fa9a
chmod -x
7 years ago
nihui
2fe769f314
update fused param files, enable ncnnoptimize tool build
7 years ago
nihui
99b81b2ee9
eliminate dropout
7 years ago
nihuini
528fe8e9e3
gpu convolution/deconvolution/innerproduct fuse activation
7 years ago
nihuini
3f85cafc08
fuse relu leakyrelu clip into convolution/deconvolution/innerproduct
7 years ago
Sungmann Cho
59c476a6b5
Suppress a CMake warning: implicitly converting 'string' to 'STRING' type ( #942 )
The latest CMake generates lots of the same warning message as below,
when building ToT ncnn for iOS:
CMake Warning (dev) at toolchains/ios.toolchain.cmake:451 (set):
implicitly converting 'string' to 'STRING' type.
This PR fixes 'string' to 'STRING' to suppress the warning messages above.
7 years ago
nihuini
5e98a8799c
fuse deconvolution batchnorm
7 years ago
nihuini
7984ffcb4d
ncnnoptimize tool
7 years ago
nihuini
b81e1f3906
get rid of the old workaround :)
7 years ago
791136190
e2e8e1b9d7
mxnet2ncnn tool support symbol.softmax op ( #938 )
* [CHG]when use HybridBlock to call F.softmax,the softmax op name is "softmax"(mxnet_version:10301)
* [CHG]Remove type mismatch error when using static code detection tool
7 years ago
nihuini
5d86014d9c
add missing barrier for transfer dst, fix softmax pack4, fix #932
7 years ago
nihuini
4729ea3505
bottom blob memory never alias, reuse blob memory more elegantly relying on refcount
7 years ago
nihuini
1dea9fe353
more elementwise binaryop alias, convert priorbox steps param
7 years ago
nihui
274392eb80
convolution padding same on gpu
7 years ago
nihui
469e274150
require cmake 3.0 for list append
7 years ago
gfjiangly
9ffe2b84e9
Optimization of the softmax layer ( #914 )
* Optimize the loop structure to improve the speed of the softmax layer & Reduce memory consumption
* use 4 space instead of tab
7 years ago
nihui
8724440c59
bind wait barrier count member to memory, fix #932
7 years ago