Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Sep 19, 2018

• fix the dequantize arm bug (Tencent#580)

  Show description for 3489d02
  

  BUG1989

  authored and

  nihui

  committed
  3489d02

  Copy full SHA for 3489d02

• innerproduce layer with int8 impl,the type of top_blob shoud be integer. (Tencent#578)

  

  BUG1989

  authored and

  nihui

  committed
  7d2d18d

  Copy full SHA for 7d2d18d

Commits on Sep 15, 2018

• aarch64 assembly for conv3x3s1, about 18% speed gain

  nihui

  committed
  a862a34

  Copy full SHA for a862a34

Commits on Sep 12, 2018

• delete out.fill operate in the armv7 int8 convdw

  

  BUG1989

  authored and

  nihui

  committed
  5f1e678

  Copy full SHA for 5f1e678

• add armv7 int8 convdw3x3 and conv3x3s2 left4 case

  

  BUG1989

  authored and

  nihui

  committed
  9b8e968

  Copy full SHA for 9b8e968

Commits on Aug 30, 2018

• implement the missing dequantize image on armv7, prefer neon-optimized 3-dim dequantize, fix Tencent#547

  nihui

  committed
  ef36d79

  Copy full SHA for ef36d79

• fix build without neon

  nihui

  committed
  19ad4cf

  Copy full SHA for 19ad4cf

• disable the wrong mnasnet benchmark :|

  nihui

  committed
  c4e3a9c

  Copy full SHA for c4e3a9c

• Update qcom820,rk3399,rk3288,hisi3519 benchmark (Tencent#551)

  Show description for 09b0c72
  

  BUG1989

  authored and

  nihui

  committed
  09b0c72

  Copy full SHA for 09b0c72

Commits on Aug 27, 2018

• update qcom410 i.mx7 dual benchmark

  nihui

  authored
  49c956d

  Copy full SHA for 49c956d

• update qcom810 iphone5s benchmark

  nihui

  authored
  a1ce093

  Copy full SHA for a1ce093

• add some comment

  nihui

  committed
  6e02923

  Copy full SHA for 6e02923

• add mnasnet benchmark

  nihui

  committed
  9a19bd9

  Copy full SHA for 9a19bd9

Commits on Aug 24, 2018

• as we already have the int8_scale_term switch, do not have to rely on the actual scale value

  nihui

  committed
  23de61f

  Copy full SHA for 23de61f

• convolution depthwise 3x3 stride 1 with unroll 8 on aarch64

  nihui

  committed
  dda21b8

  Copy full SHA for dda21b8

Commits on Aug 23, 2018

• fix the caffe2ncnn that "DepthwiseConvolution" loss group num param (Tencent#540)

  Show description for d5df230
  

  BUG1989

  authored and

  nihui

  committed
  d5df230

  Copy full SHA for d5df230

• NCNN_XADD returns int, though rarely used, fix Tencent#538

  nihui

  committed
  9e49144

  Copy full SHA for 9e49144

• Mat fill int optimization

  nihui

  committed
  144f818

  Copy full SHA for 144f818

• packed int8 convolution 3x3 stride 1 for armv7, 7%~25% faster than vanilla one, but God knows how hard I try :|

  nihui

  committed
  8526a69

  Copy full SHA for 8526a69

• fix null sub group elemsize/allocator in depthwise deconvolution, fix Tencent#539

  nihui

  committed
  a76a07e

  Copy full SHA for a76a07e

Commits on Aug 20, 2018

• quantized padding in convolution, use range sweets

  nihui

  committed
  6f1b0b0

  Copy full SHA for 6f1b0b0

• padding is elemsize aware, copy_make_border is now a padding wrapper

  nihui

  committed
  bf1c58b

  Copy full SHA for bf1c58b

• Mat range for zero-copy slicing reference

  nihui

  committed
  d321d65

  Copy full SHA for d321d65

Commits on Aug 16, 2018

• use max threads as cpu count on openmp capable runtime

  nihui

  committed
  f78e8fd

  Copy full SHA for f78e8fd

Commits on Aug 15, 2018

• fix bus error for int8 storage

  nihui

  committed
  2790f3c

  Copy full SHA for 2790f3c

• ShuffleChannel = Reshape + SwapAxis + Reshape

  nihui

  committed
  28b35b8

  Copy full SHA for 28b35b8

Commits on Aug 14, 2018

• Rasp bench (Tencent#531)

  Show description for 837e6b0
  

  diegodgs

  authored and

  nihui

  committed
  837e6b0

  Copy full SHA for 837e6b0

Commits on Aug 13, 2018

• add armv7 int8 conv3x3s1,using vaddw to replace vadd and vmovl

  

  BUG1989

  authored and

  nihui

  committed
  1b0e334

  Copy full SHA for 1b0e334

Commits on Aug 12, 2018

• restore the old conv3x3s2 as reference, fast dilation convolution fails on striding

  nihui

  committed
  72411b7

  Copy full SHA for 72411b7

• pack weight and more unroll makes improvement, ~20% faster for conv3x3s2

  nihui

  committed
  1f20eb4

  Copy full SHA for 1f20eb4

Commits on Aug 10, 2018

• fix asm "invalid operand" error for target iOS armv7 on file dequantize_arm.cpp

  

  chensy

  authored and

  nihui

  committed
  30cc738

  Copy full SHA for 30cc738

• fix gettid call for glibc

  

  Diego Gomes

  authored and

  nihui

  committed
  4d73407

  Copy full SHA for 4d73407

• fix auxv read for elf64

  

  Diego Gomes

  authored and

  nihui

  committed
  534f38e

  Copy full SHA for 534f38e

• store int8 scale in binary

  nihui

  committed
  2dbaf6f

  Copy full SHA for 2dbaf6f

Commits on Aug 5, 2018

• enable pool allocator in sample project, display unscaled image

  nihui

  committed
  cebded1

  Copy full SHA for cebded1

Pagination

Previous

Next