+add Base implementation of class SynetConvolution16bNhwcSpecV2 (Forward, part 1).

Author: ermig1979

src/Simd/SimdBaseSynetConvolution16bNhwcSpecV2.cpp

@@ -194,10 +194,7 @@ namespace Simd
             {
                 size_t macroD = Simd::Min(p.dstC, mad + a.macroD) - mad;
                 const uint16_t* weight = _weight.data + mad * a.K;
-        //        for (size_t mac = 0, mao = 0; mac < a.srcC; mac += a.macroC, mao += a.macroO)
-        //        {
-        //            size_t macroC = Simd::Min(a.srcC, mac + a.macroC) - mac;
-        //            size_t nK = DivHi(macroC, a.microC) * a.kA;
+
         //            for (size_t dyBeg = 0, dyN = 0; dyBeg < dstH; dyN++)
         //            {
         //                size_t dyEnd = Simd::Min(dyBeg + a.macroH, dstH);
@@ -236,12 +233,11 @@ namespace Simd
         //                }
         //                dyBeg = dyEnd;
         //            }
-        //            weight += macroC * a.kA * a.F;
-        //        }
-        //        bias += macroD;
-        //        if (p.activation == ::SimdConvolutionActivationPrelu)
-        //            params += macroD;
-        //        dst += macroD * _elemD;
+
+                bias += macroD;
+                if (p.activation == ::SimdConvolutionActivationPrelu)
+                    params += macroD;
+                dst += macroD * _elemD;
             }
         }
 

src/Test/TestSynetConvolution16b.cpp

@@ -530,6 +530,13 @@ namespace Test
         //result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 512, 32, 32, 512, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
         //result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 1024, 32, 32, 1024, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
 #endif
+#if 1
+        //result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 64, 64, 64, 64, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
+        result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 128, 32, 32, 1024, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
+        result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 256, 16, 16, 256, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
+        //result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 512, 32, 32, 512, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
+        //result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 1024, 32, 32, 1024, _3, _1, _1, _1, _1, 1, aPr, tT, b16, b16), c, f1, f2);
+#endif
 #else
         result = result && SynetConvolution16bForwardAutoTest(eps, Param(1, 16, 320, 320, 32, _2, _1, _1, _0, _1, 1, aId, tT, b16, f32), c, f1, f2);
 #endif