Generalize first_aligned to take the requested alignment as a template parameter, and add a first_default_aligned variante calling first_aligned with the requirement of the largest packet for the given scalar type.

This commit is contained in:
Gael Guennebaud
2015-08-06 17:52:01 +02:00
parent 1f5024332e
commit 2afdef6a54
14 changed files with 69 additions and 48 deletions

View File

@@ -221,12 +221,13 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
{
const Index size = mat.size();
const Index packetSize = packet_traits<Scalar>::size;
const Index alignedStart = internal::first_aligned(mat.nestedExpression());
const Index packetSize = packet_traits<Scalar>::size;
const int packetBytes = int(packetSize*sizeof(Scalar));
enum {
alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(sizeof(Scalar)*packetSize) : int(Unaligned), // FIXME take into account alignment requirement
alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetBytes) : int(Unaligned), // FIXME take into account alignment requirement
alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
};
const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
const Index alignedEnd2 = alignedStart + alignedSize2;