Add SVE2 optimizations for SynetConvolution32fNhwcGroupedBlock1x2 (#809)

Port the NhwcGroupedBlock1x2 algorithm to SVE2, wire Sve2 dispatch,
extend AutoTest coverage for grouped convolution shapes, and document
the change in release 7.2.165.

Co-authored-by: Cursor Agent <cursoragent@cursor.com>

Author: ermig1979

docs/2026.html

@@ -45,6 +45,7 @@ <h5>New features</h5>
  <li>SVE2 optimizations of class SynetConvolution32fGemmNT.</li>
  <li>SVE2 optimizations of class SynetConvolution32fWinograd.</li>
  <li>NEON optimizations of class SynetConvolution32fNhwcGroupedBlock1x2.</li>
+ <li>SVE2 optimizations of class SynetConvolution32fNhwcGroupedBlock1x2.</li>
 </ul>
 
 <a href="#HOME">Home</a>

prj/vs2022/Sve2.vcxproj

@@ -107,6 +107,7 @@
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution8i.cpp" />
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution32f.cpp" />
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution32fGemm.cpp" />
+    <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution32fNhwcGrouped.cpp" />
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetConversion.cpp" />
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetDeconvolution32f.cpp" />
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetMergedConvolution32f.cpp" />

prj/vs2022/Sve2.vcxproj.filters

@@ -560,6 +560,9 @@
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution32fGemm.cpp">
       <Filter>Sve2\Synet\Convolution</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\src\Simd\SimdSve2SynetConvolution32fNhwcGrouped.cpp">
+      <Filter>Sve2\Synet\Convolution</Filter>
+    </ClCompile>
     <ClCompile Include="..\..\src\Simd\SimdSve2SynetScale.cpp">
       <Filter>Sve2\Synet\Other</Filter>
     </ClCompile>

src/Simd/SimdSve2SynetConvolution32f.cpp

@@ -48,8 +48,8 @@ namespace Simd
                 return new Neon::SynetConvolution32fNhwcDirect(param);
             else if (Neon::SynetConvolution32fNhwcDepthwise::Preferable(param))
                 return new Neon::SynetConvolution32fNhwcDepthwise(param);
-            else if (Base::SynetConvolution32fNhwcGroupedBlock1x2::Preferable(param))
-                return new Base::SynetConvolution32fNhwcGroupedBlock1x2(param);
+            else if (SynetConvolution32fNhwcGroupedBlock1x2::Preferable(param))
+                return new SynetConvolution32fNhwcGroupedBlock1x2(param);
             else
                 return new SynetConvolution32fGemmNN(param);
         }

src/Simd/SimdSve2SynetConvolution32fNhwcGrouped.cpp

@@ -753,6 +753,13 @@ namespace Simd
             static bool Preferable(const ConvParam & p);
         };
 
+        class SynetConvolution32fNhwcGroupedBlock1x2 : public Base::SynetConvolution32fNhwcGroupedBlock1x2
+        {
+        public:
+            SynetConvolution32fNhwcGroupedBlock1x2(const ConvParam& p);
+            virtual String Ext() const { return "Sve2"; }
+        };
+
         void * SynetConvolution32fInit(size_t batch, const SimdConvolutionParameters * conv);
     }
 #endif//SIMD_SVE2_ENABLE

src/Simd/SimdSynetConvolution32f.h

@@ -278,6 +278,8 @@ namespace Test
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 32, 4, 16, 32, Size(1, 5), _1, _1, Size(0, 2), Size(0, 2), 1, aId, tT), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 1280, 32, 32, 256, _1, _1, _1, _0, _0, 1, aRe, t), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 384, 7, 7, 768, _3, _1, _1, _1, _1, 384, a, t), f1, f2);
+        result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 192, 14, 14, 384, _7, _1, _2, _3, _3, 192, a, t), f1, f2);
+        result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 96, 14, 14, 192, _3, _1, _1, _1, _1, 96, a, t), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 152, 32, 32, 152, _7, _1, _1, _3, _3, 152, aRe, t), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 49, 29, 29, 98, _7, _1, _2, _3, _3, 49, a, t), f1, f2);
 #endif
@@ -294,6 +296,7 @@ namespace Test
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 48, 4, 16, 48, Size(1, 3), _1, _1, Size(0, 1), Size(0, 1), 1, aId, tT), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 16, 9, 9, 16, _3, Size(2, 2), _1, _2, _2, 1, aRe, t), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 16, 9, 9, 16, _3, Size(2, 2), _2, _2, _2, 1, aRe, tF), f1, f2);
+        result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 96, 14, 14, 192, _3, _1, _1, _1, _1, 96, a, t), f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 49, 29, 29, 98, _7, _1, _2, _3, _3, 49, a, t), f1, f2);
 #endif
         return result;