* gemm x86 multiply alpha beta in post gemm stage, enable one_blob_only
* relax mnk multiple restrictions
* make square tiles in each thread
* sanitize num_threads changes

* add some c_api interfaces related to allocator setup.

* fix errors in allocator parameters in c_api.

* test c api allocator

Co-authored-by: zhangtongshe <yuyuyezi@vip.qq.com>

Co-authored-by: LRY89757 <LRY89757@users.noreply.github.com>
Co-authored-by: nihuini <nihuini@tencent.com>
Co-authored-by: nihui <shuizhuyuanluo@126.com>

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

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>

* Add the test samples for elempack==16

* Add the AVX512 Support for batchnorm

* 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

* 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

* discard weight memory for x86 and vulkan
* drop arm innerproduct weight
* drop arm convolution weight
* drop arm convolutiondepthwise weight
* drop x86 vulkan deconvolution deconvolutiondepthwise weight
* drop arm deconvolution deconvolutiondepthwise weight
* arm neon assembly optimization for innerproduct pack4

* feat(src/layer): add vision_transformer benchmark and relative layer

* refactor(testutil.h): add para for RandomMat

* convolution winograd pack16to1

* x86 deconvolution and deconvolutiondepthwise

* simpleomp allow 32 arguments

* drop shadow variable workaround

* less winograd test error

* unified relu avx512

* unifed clip avx512

* unaryop avx512

* sigmoid avx512

* binaryop avx512

* padding convolution avx512

* convolutiondepthwise avx512

* innerproduct avx512

* reshape avx512

* slice avx512

* hardsigmoid hardswish avx512

* swish avx512

* pooling avx512

* crop avx512

* convolution sgemm pack16

* convolution 3x3 winograd pack16

* interp avx512

* convolution sgemm pack1to16

* convolution sgemm pack16to8

* convolution sgemm pack8to16

* convolution sgemm pack16to4

* fix vulkan permute pack8

* fix vulkan convolution gemm pack8to1

* query and enable cooperative matrix

* fix build with old vulkan sdk

* implement cooperative matrix optimization

* add nvidia-t4 coverage

* adjust test option for more coverage

* mix vulkan winograd f23 and f43

* larget epsilon for winograd optimization test

* vulkan deconvolution sgemm col2im

* vulkan convolution winograd43

* improve fp16s numeric stablity

* vulkan convolution im2col sgemm

* check squeezenet top2, as top3 vs top4 score too close..

* fix sigmoid returns nan with very large input