From 66f65ccc364034150bdc47333e05ebbde29825e5 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Fri, 2 Dec 2016 22:41:26 +0100
Subject: Ease compiler job to generate clean and efficient code in mat*vec.

---
 Eigen/src/Core/GeneralProduct.h | 83 ++++++++++++++++++++++++-----------------
 1 file changed, 49 insertions(+), 34 deletions(-)

(limited to 'Eigen/src/Core/GeneralProduct.h')

diff --git a/Eigen/src/Core/GeneralProduct.h b/Eigen/src/Core/GeneralProduct.h
index f62ecfcbe..0f16cd8e3 100644
--- a/Eigen/src/Core/GeneralProduct.h
+++ b/Eigen/src/Core/GeneralProduct.h
@@ -224,50 +224,65 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
       // on, the other hand it is good for the cache to pack the vector anyways...
       EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = (ActualDest::InnerStrideAtCompileTime!=1) || ComplexByReal
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
     };
 
-    gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
-    const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
-    const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+    if(!MightCannotUseDest)
+    {
+      // shortcut if we are sure to be able to use dest directly,
+      // this ease the compiler to generate cleaner and more optimzized code for most common cases
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          dest.data(), 1,
+          compatibleAlpha);
+    }
+    else
+    {
+      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
 
-    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
+      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
 
-    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
-                                                  evalToDest ? dest.data() : static_dest.data());
+      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                    evalToDest ? dest.data() : static_dest.data());
 
-    if(!evalToDest)
-    {
-      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      Index size = dest.size();
-      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      #endif
-      if(!alphaIsCompatible)
+      if(!evalToDest)
       {
-        MappedDest(actualDestPtr, dest.size()).setZero();
-        compatibleAlpha = RhsScalar(1);
+        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        Index size = dest.size();
+        EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        #endif
+        if(!alphaIsCompatible)
+        {
+          MappedDest(actualDestPtr, dest.size()).setZero();
+          compatibleAlpha = RhsScalar(1);
+        }
+        else
+          MappedDest(actualDestPtr, dest.size()) = dest;
       }
-      else
-        MappedDest(actualDestPtr, dest.size()) = dest;
-    }
 
-    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
-    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
-    general_matrix_vector_product
-        <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-        RhsMapper(actualRhs.data(), actualRhs.innerStride()),
-        actualDestPtr, 1,
-        compatibleAlpha);
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          actualDestPtr, 1,
+          compatibleAlpha);
 
-    if (!evalToDest)
-    {
-      if(!alphaIsCompatible)
-        dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
-      else
-        dest = MappedDest(actualDestPtr, dest.size());
+      if (!evalToDest)
+      {
+        if(!alphaIsCompatible)
+          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        else
+          dest = MappedDest(actualDestPtr, dest.size());
+      }
     }
   }
 };
-- 
cgit v1.2.3