* Finish the gelu x86 intrinsics
* Finish the fast tanh x86 simd impl

Co-authored-by: LRY89757 <LRY89757@users.noreply.github.com>
Co-authored-by: nihuini <nihuini@tencent.com>

Co-authored-by: LinHeLurking <LinHeLurking@users.noreply.github.com>
Co-authored-by: LRY89757 <LRY89757@users.noreply.github.com>
Co-authored-by: nihuini <nihuini@tencent.com>

* Add DeformableConv2D
* add unittest and docs
* pnnx torchvision deformconv2d conversion

Co-authored-by: miemie2013 <miemie2013@users.noreply.github.com>
Co-authored-by: nihui <shuizhuyuanluo@126.com>

* implement 4d memorydata

* fix ncnnoptimize memorydata 4d

* Add the test samples for elempack==16

* Add the AVX512 Support for batchnorm

* clear bom marker, avoid vector data function

* asimdfhm in fc
* move neon bf16 conversion function to arm_usability header
* fix cmake option
* fix build with newer gcc
* arm84 coverage
* arm asimdfhm optimization for innerproduct gemm fp16s

Signed-off-by: sodo <djdisodo@gmail.com>

* fix mat::fill gcc4 avx sse conflict bug
* fix build and crash with gcc 4.4.0

Co-authored-by: nihui <shuizhuyuanluo@126.com>

* armv7 vfpv4 infrastructure
* optional fp16 format ieee
* arm neon assembly optimization for cast fp16/bf16

* riscv convolution dot function

* move rvv ci to centos

* apply code-format changes

* fix gcov path

* move newlib riscv ci to centos

* fine grained winograd options

* x86 optimization for convolution winograd f23 pack4/pack8/pack16

* fix avx512 and t4 ci

* fix fast direct conv path

* winograd63 is actually slower than winograd43 on very large channel

src/stb_image.h
src/layer/detectionoutput.cpp
tools/quantize/imreadwrite.cpp

* split arm82 source, wip

* check compiler arm82 only for arm 64bit target

* drop arm82 registery

* strict check compiler support arm82