Fix 128bit packet size assumptions in unit tests.

3 files changed, 49 insertions, 32 deletions

diff --git a/test/dynalloc.cpp b/test/dynalloc.cpp
index 8bbda1c94..c98cc80f0 100644
--- a/test/dynalloc.cpp
+++ b/test/dynalloc.cpp
@@ -10,11 +10,13 @@
 #include "main.h"
 
 #if EIGEN_ALIGN
-#define ALIGNMENT 16
+#define ALIGNMENT EIGEN_ALIGN_BYTES
 #else
 #define ALIGNMENT 1
 #endif
 
+typedef Matrix<float,8,1> Vector8f;
+
 void check_handmade_aligned_malloc()
 {
   for(int i = 1; i < 1000; i++)
@@ -68,7 +70,7 @@ struct MyStruct
 {
   EIGEN_MAKE_ALIGNED_OPERATOR_NEW
   char dummychar;
-  Vector4f avec;
+  Vector8f avec;
 };
 
 class MyClassA
@@ -76,15 +78,19 @@ class MyClassA
   public:
     EIGEN_MAKE_ALIGNED_OPERATOR_NEW
     char dummychar;
-    Vector4f avec;
+    Vector8f avec;
 };
 
 template<typename T> void check_dynaligned()
 {
-  T* obj = new T;
-  VERIFY(T::NeedsToAlign==1);
-  VERIFY(size_t(obj)%ALIGNMENT==0);
-  delete obj;
+  // TODO have to be updated once we support multiple alignment values
+  if(T::SizeAtCompileTime % ALIGNMENT == 0)
+  {
+    T* obj = new T;
+    VERIFY(T::NeedsToAlign==1);
+    VERIFY(size_t(obj)%ALIGNMENT==0);
+    delete obj;
+  }
 }
 
 void test_dynalloc()
@@ -102,6 +108,7 @@ void test_dynalloc()
     CALL_SUBTEST(check_dynaligned<Matrix4f>() );
     CALL_SUBTEST(check_dynaligned<Vector4d>() );
     CALL_SUBTEST(check_dynaligned<Vector4i>() );
+    CALL_SUBTEST(check_dynaligned<Vector8f>() );
   }
   
   // check static allocation, who knows ?
diff --git a/test/unalignedassert.cpp b/test/unalignedassert.cpp
index 601dbf214..d8815263a 100644
--- a/test/unalignedassert.cpp
+++ b/test/unalignedassert.cpp
@@ -9,6 +9,8 @@
 
 #include "main.h"
 
+typedef Matrix<float,8,1> Vector8f;
+
 struct TestNew1
 {
   MatrixXd m; // good: m will allocate its own array, taking care of alignment.
@@ -69,7 +71,7 @@ void construct_at_boundary(int boundary)
 {
   char buf[sizeof(T)+256];
   size_t _buf = reinterpret_cast<size_t>(buf);
-  _buf += (16 - (_buf % 16)); // make 16-byte aligned
+  _buf += (EIGEN_ALIGN_BYTES - (_buf % EIGEN_ALIGN_BYTES)); // make 16/32-byte aligned
   _buf += boundary; // make exact boundary-aligned
   T *x = ::new(reinterpret_cast<void*>(_buf)) T;
   x[0].setZero(); // just in order to silence warnings
@@ -85,18 +87,18 @@ void unalignedassert()
   construct_at_boundary<Vector4f>(16);
   construct_at_boundary<Matrix2f>(16);
   construct_at_boundary<Matrix3f>(4);
-  construct_at_boundary<Matrix4f>(16);
+  construct_at_boundary<Matrix4f>(EIGEN_ALIGN_BYTES);
 
   construct_at_boundary<Vector2d>(16);
   construct_at_boundary<Vector3d>(4);
-  construct_at_boundary<Vector4d>(16);
-  construct_at_boundary<Matrix2d>(16);
+  construct_at_boundary<Vector4d>(EIGEN_ALIGN_BYTES);
+  construct_at_boundary<Matrix2d>(EIGEN_ALIGN_BYTES);
   construct_at_boundary<Matrix3d>(4);
-  construct_at_boundary<Matrix4d>(16);
+  construct_at_boundary<Matrix4d>(EIGEN_ALIGN_BYTES);
 
   construct_at_boundary<Vector2cf>(16);
   construct_at_boundary<Vector3cf>(4);
-  construct_at_boundary<Vector2cd>(16);
+  construct_at_boundary<Vector2cd>(EIGEN_ALIGN_BYTES);
   construct_at_boundary<Vector3cd>(16);
   #endif
 
@@ -110,14 +112,21 @@ void unalignedassert()
   check_unalignedassert_good<Depends<true> >();
 
 #if EIGEN_ALIGN_STATICALLY
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4f>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4f>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2d>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4d>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix2d>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4d>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cf>(8));
-  VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cd>(8));
+  if(EIGEN_ALIGN_BYTES==16)
+  {
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4f>(8));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2d>(8));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cf>(8));
+  }
+  for(int b=8; b<EIGEN_ALIGN_BYTES; b+=8)
+  {
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector8f>(b));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4f>(b));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4d>(b));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix2d>(b));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4d>(b));
+    VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cd>(b));
+  }
 #endif
 }
 
diff --git a/test/vectorization_logic.cpp b/test/vectorization_logic.cpp
index aee68a87f..09b46660b 100644
--- a/test/vectorization_logic.cpp
+++ b/test/vectorization_logic.cpp
@@ -90,8 +90,8 @@ template<typename Scalar, bool Enable = internal::packet_traits<Scalar>::Vectori
     typedef Matrix<Scalar,2*PacketSize,2*PacketSize> Matrix22;
     typedef Matrix<Scalar,(Matrix11::Flags&RowMajorBit)?16:4*PacketSize,(Matrix11::Flags&RowMajorBit)?4*PacketSize:16> Matrix44;
     typedef Matrix<Scalar,(Matrix11::Flags&RowMajorBit)?16:4*PacketSize,(Matrix11::Flags&RowMajorBit)?4*PacketSize:16,DontAlign|EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION> Matrix44u;
-    typedef Matrix<Scalar,4*PacketSize,16,ColMajor> Matrix44c;
-    typedef Matrix<Scalar,4*PacketSize,16,RowMajor> Matrix44r;
+    typedef Matrix<Scalar,4*PacketSize,4*PacketSize,ColMajor> Matrix44c;
+    typedef Matrix<Scalar,4*PacketSize,4*PacketSize,RowMajor> Matrix44r;
 
     typedef Matrix<Scalar,
         (PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
@@ -149,15 +149,15 @@ template<typename Scalar, bool Enable = internal::packet_traits<Scalar>::Vectori
         LinearTraversal,CompleteUnrolling));
       VERIFY(test_assign(Matrix33c().col(0),Matrix33c().col(1)+Matrix33c().col(1),
         LinearTraversal,CompleteUnrolling));
-      
+              
       VERIFY(test_assign(Matrix3(),Matrix3().cwiseQuotient(Matrix3()),
         LinearVectorizedTraversal,CompleteUnrolling));
-
+        
       VERIFY(test_assign(Matrix<Scalar,17,17>(),Matrix<Scalar,17,17>()+Matrix<Scalar,17,17>(),
         LinearTraversal,NoUnrolling));
-
-      VERIFY(test_assign(Matrix11(),Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(2,3)+Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(10,4),
-      DefaultTraversal,CompleteUnrolling));
+        
+      VERIFY(test_assign(Matrix11(),Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(2,3)+Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(8,4),
+        DefaultTraversal,PacketSize>4?InnerUnrolling:CompleteUnrolling));
     }
     
     VERIFY(test_redux(Matrix3(),
@@ -174,18 +174,19 @@ template<typename Scalar, bool Enable = internal::packet_traits<Scalar>::Vectori
 
     VERIFY(test_redux(Matrix44r().template block<1,2*PacketSize>(2,1),
       LinearVectorizedTraversal,CompleteUnrolling));
-    
+
     VERIFY((test_assign<
             Map<Matrix22, Aligned, OuterStride<3*PacketSize> >,
             Matrix22
             >(InnerVectorizedTraversal,CompleteUnrolling)));
 
     VERIFY((test_assign<
-            Map<Matrix22, Aligned, InnerStride<3*PacketSize> >,
-            Matrix22
+            Map<Matrix<Scalar,EIGEN_PLAIN_ENUM_MAX(2,PacketSize),EIGEN_PLAIN_ENUM_MAX(2,PacketSize)>, Aligned, InnerStride<3*PacketSize> >,
+            Matrix<Scalar,EIGEN_PLAIN_ENUM_MAX(2,PacketSize),EIGEN_PLAIN_ENUM_MAX(2,PacketSize)>
             >(DefaultTraversal,CompleteUnrolling)));
 
-    VERIFY((test_assign(Matrix11(), Matrix11()*Matrix11(), InnerVectorizedTraversal, CompleteUnrolling)));
+    VERIFY((test_assign(Matrix11(), Matrix<Scalar,PacketSize,EIGEN_PLAIN_ENUM_MIN(2,PacketSize)>()*Matrix<Scalar,EIGEN_PLAIN_ENUM_MIN(2,PacketSize),PacketSize>(),
+                        PacketSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD?DefaultTraversal:InnerVectorizedTraversal, CompleteUnrolling)));
     #endif
 
     VERIFY(test_assign(MatrixXX(10,10),MatrixXX(20,20).block(10,10,2,3),