This patch uses _mm_cvtsi128_si{32,64} intrinsics when returning value
in float2int8() to reduce unnecessary memory accesses.

Resolves TODO "use _mm_cvtsi128_si64 on 64bit target".

Signed-off-by: Kagurazaka Kotori <kagurazakakotori@gmail.com>

* x86/avx_mathfun.h: Remove fallback warnings

This patch removes warning messages indicating falling back to SSE2
when AVX2 support is disabled as suggested. Also reorders non-AVX2
macros for readability and faster preprocessing.

Suggested-by: nihui <shuizhuyuanluo@126.com>
Signed-off-by: Kagurazaka Kotori <kagurazakakotori@gmail.com>

* apply code-format changes

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

* avx-vnni and avx512-vnni optimization for convolution int8 gemm and 3x3 winograd pack8to4/pack8to1

* some sse 4.1 optimization

* sse2/avx2 optimization for convolution 3x3 winograd42 int8 pack8to4/pack8to1

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

Co-authored-by: ElvisYu <elvisyuovo@gmail.com>
Co-authored-by: 余浩文 <m18107220188@163.com>
Co-authored-by: nihui <nihui@users.noreply.github.com>
Co-authored-by: Zr2223 <67497651+Zr2223@users.noreply.github.com>
Co-authored-by: Zr2223 <Zr2223@users.noreply.github.com>

Co-authored-by: ElvisYu <elvisyuovo@gmail.com>
Co-authored-by: 余浩文 <m18107220188@163.com>
Co-authored-by: Zr2223 <67497651+Zr2223@users.noreply.github.com>

* added a number of optimized sse layers, specifically to increase performance of mobilenet style networks