* 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

* basic mips msa optimization for convolution int8

* mips msa optimization for convolution int8 gemm

* mips msa optimization for convolution int8 winograd pack8to4/pack8to1

* mention msa maddv/msubv intrinsics bug

* add:multiheadattention_arm

* pass the test in local

* add omp

* return naive

Co-authored-by: EdVince <EdVince@users.noreply.github.com>

* 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

* implicit gemm

* unroll direct conv by 2x2x2

* vulkan local memory optimization for conv1x1 pack4 and winograd on dgpu

* unified innerproduct pipeline creation

* reorder deconvolution weight layout

* flexible local memory data type

* more local memory optimization for conv/deconv gemm

Co-authored-by: MegEngine <megengine@megvii.com>
Co-authored-by: tpoisonooo <tpoisonooo@users.noreply.github.com>

* 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..

* Make some typos for SSE2 floor.

* Improve the implementation of SSE2 abs.

* Improve the implementation of SSE2 ceil.

* Add SSE2 implementation of floor and ceil in x86 targets.

* apply code-format changes

* Update the SSE2 floor implementation.

Co-authored-by: MouriNaruto <MouriNaruto@users.noreply.github.com>

* fix sigmoid returns nan with very large input

Co-authored-by: Yoh <wangpeizhou@bilibili.com>
Co-authored-by: Yoh-Z <Yoh-Z@users.noreply.github.com>

* optimize x86 avx exp log sincos fma

* optimize avx&sse_mathfun exp,log,cos,sin,sincos and add fnmadd to x86_usability

Co-authored-by: Yoh-Z <Yoh-Z@users.noreply.github.com>

* lets accept the float div vs recp error

Co-authored-by: MegEngine <megengine@megvii.com>
Co-authored-by: tpoisonooo <tpoisonooo@users.noreply.github.com>

Co-authored-by: Yoh-Z <Yoh-Z@users.noreply.github.com>