*fix bug: Error in Base implementation of SynetMergedConvolution32f (…

…type=CDC, add=1).
luofalei · Dec 7, 2020 · f56db34 · f56db34
1 parent 9d20d80
commit f56db34
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 10 deletions.
diff --git a/docs/2021.html b/docs/2021.html
@@ -40,6 +40,10 @@ <h5>New features</h5>
  <li>Base implementation, SSE2, AVX2, AVX-512F and NEON optimizations of function SynetMish32f.</li>
  <li>Support of Mish activation function in Convolution32f framework.</li>
 </ul>
+<h5>Bug fixing</h5>
+<ul>
+ <li>Error in Base implementation of SynetMergedConvolution32f (type=CDC, add=1).</li>
+</ul>
 
 <h4>Test framework</h4>
 <h5>New features</h5>

diff --git a/src/Simd/SimdBaseSynetMergedConvolution32f.cpp b/src/Simd/SimdBaseSynetMergedConvolution32f.cpp
@@ -68,8 +68,16 @@ namespace Simd
                             }
                         }
                     }
-                    for (size_t dc = 0; dc < dstC; ++dc)
-                        Update<update>(dst + dc, Activate<type>(buf[dc], params, dc));
+                    if (update == UpdateAdd)
+                    {
+                        for (size_t dc = 0; dc < dstC; ++dc)
+                            dst[dc] = Activate<type>(dst[dc] + buf[dc], params, dc);
+                    }
+                    else
+                    {
+                        for (size_t dc = 0; dc < dstC; ++dc)
+                            dst[dc] = Activate<type>(buf[dc], params, dc);
+                    }
                     dst += p.dstC;
                 }
             }

diff --git a/src/Test/TestSynetMergedConvolution32f.cpp b/src/Test/TestSynetMergedConvolution32f.cpp
@@ -204,8 +204,9 @@ namespace Test
     {
         bool result = true;
         const SimdBool t = SimdTrue, f = SimdFalse;
-        const ::SimdConvolutionActivationType a0 = ::SimdConvolutionActivationHswish, a1 = ::SimdConvolutionActivationIdentity, a2 = ::SimdConvolutionActivationPrelu;
+        //const ::SimdConvolutionActivationType a0 = ::SimdConvolutionActivationHswish, a1 = ::SimdConvolutionActivationIdentity, a2 = ::SimdConvolutionActivationPrelu;
         //const ::SimdConvolutionActivationType a0 = ::SimdConvolutionActivationPrelu, a1 = ::SimdConvolutionActivationRestrictRange, a2 = ::SimdConvolutionActivationHswish;
+        const ::SimdConvolutionActivationType a0 = ::SimdConvolutionActivationHswish, a1 = ::SimdConvolutionActivationPrelu, a2 = ::SimdConvolutionActivationRelu;
 #ifdef NDEBUG
 #if 0
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 3, 384, 389), Cnv(a0, 3, 2, 32), Cnv(a1, 3, 1), Cnv(a2, 1, 1, 16), f), f1, f2);
@@ -238,10 +239,6 @@ namespace Test
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 48, 70, 81), Cnv(a0, 3, 2), Cnv(a1, 1, 1, 64)), f1, f2);
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 3, 320, 320), Cnv(a0, 3, 2, 16), Cnv(a1, 3, 1)), f1, f2);
 #endif
-#if 0
-        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 19, 63, 81), Cnv(a0, 1, 1, 51), Cnv(a1, 3, 2)), f1, f2);
-        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 1280, 12, 21), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 2)), f1, f2);
-#endif
 #if 0
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 3, 320, 180), Cnv(a0, 3, 2, 16), Cnv(a1, 3, 1)), f1, f2);
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 16, 160, 90), Cnv(a0, 1, 1, 32), Cnv(a1, 3, 2)), f1, f2);
@@ -266,13 +263,17 @@ namespace Test
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 256, 10, 6), Cnv(a0, 3, 1), Cnv(a1, 1, 1, 4)), f1, f2);
 #endif
 #if 1
-        //result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 256, 10, 6), Cnv(a0, 1, 1, 64), Cnv(a1, 3, 2), Cnv(a2, 1, 1, 256), f), f1, f2);
-        //result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 64, 10, 6), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 1), Cnv(a2, 1, 1, 64), t), f1, f2);
+        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 19, 63, 81), Cnv(a0, 1, 1, 51), Cnv(a1, 3, 2)), f1, f2);
+        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 1280, 12, 21), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 2)), f1, f2);
+#endif
+#if 1
+        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 64, 10, 6), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 1), Cnv(a2, 1, 1, 64), t), f1, f2);
+        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 256, 10, 6), Cnv(a0, 1, 1, 64), Cnv(a1, 3, 2), Cnv(a2, 1, 1, 256), f), f1, f2);
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 128, 20, 12), Cnv(a0, 3, 1), Cnv(a1, 1, 1, 20)), f1, f2);
         result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 128, 20, 12), Cnv(a0, 3, 1), Cnv(a1, 1, 1, 128)), f1, f2);
 #endif
 #else
-        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 1280, 12, 21), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 2)), f1, f2);
+        result = result && SynetMergedConvolution32fForwardAutoTest(eps, Param(Shp(1, 64, 10, 6), Cnv(a0, 1, 1, 256), Cnv(a1, 3, 1), Cnv(a2, 1, 1, 64), t), f1, f2);
 #endif
         return result;
     }