Merge pull request #347 from ermig1979/dev

*fix bug in Error in AVX-512BW optimizations of function SynetSoftmax…

Author: ermig1979

docs/2026.html

@@ -87,7 +87,7 @@ <h5>Improving</h5>
 <h5>Bug fixing</h5>
 <ul>
  <li>Error in NEON optimization of function TransformImage (BGR, T0 transform).</li>
- <li>Error in AVX-512BW optimizations of function SynetSoftmax16b (MSVS, Debug, Win32).</li>
+ <li>Error in AVX-512BW optimizations of function SynetSoftmax16b (MSVS, Win32).</li>
  <li>Possible aligned load of unaligned memory in AVX2 optimizations of function AbsDifferenceSums3 (Windows 7, x64, gcc 8.1.0).</li>
  <li>Possible aligned store in file SimdExtract.h for SSE4.1, AVX2, AVX-512BW.</li>
  <li>Wrong assert condition in function Simd::Uyvy422ToBgr.</li>

src/Simd/SimdAvx512bwSynetPoolingMax16b.cpp

@@ -122,7 +122,7 @@ namespace Simd
         {
             if (format == SimdTensorFormatNhwc)
             {
-#if (defined(_MSC_VER) && !defined(NDEBUG) && defined(SIMD_X86_ENABLE)) || defined(__clang__)
+#if (defined(_MSC_VER) && defined(SIMD_X86_ENABLE)) || defined(__clang__)
                 Avx2::SynetPoolingMax16b(src, srcC, srcH, srcW, kernelY, kernelX, strideY, strideX, padY, padX, dst, dstH, dstW, format);
 #else
                 Array32f max(srcC);