GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 19:59:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

let redux use the new ByOuterInner accessors

Browse Source
This commit is contained in:
Benoit Jacob 2010-02-26 21:46:43 -05:00
parent 4927841cba
commit 814e40c72a
1 changed files with 16 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
Eigen/src/Core/Redux.h
View File

@ -40,7 +40,7 @@ struct ei_redux_traits
private: private:
enum { enum {
PacketSize = ei_packet_traits<typename Derived::Scalar>::size, PacketSize = ei_packet_traits<typename Derived::Scalar>::size,
InnerMaxSize = int(Derived::Flags)&RowMajorBit InnerMaxSize = int(Derived::IsRowMajor)
? Derived::MaxColsAtCompileTime ? Derived::MaxColsAtCompileTime
: Derived::MaxRowsAtCompileTime : Derived::MaxRowsAtCompileTime
}; };
@ -100,15 +100,15 @@ template<typename Func, typename Derived, int Start>
struct ei_redux_novec_unroller<Func, Derived, Start, 1> struct ei_redux_novec_unroller<Func, Derived, Start, 1>
{ {
enum { enum {
col = Start / Derived::RowsAtCompileTime, outer = Start / Derived::InnerSizeAtCompileTime,
row = Start % Derived::RowsAtCompileTime inner = Start % Derived::InnerSizeAtCompileTime
}; };
typedef typename Derived::Scalar Scalar; typedef typename Derived::Scalar Scalar;
EIGEN_STRONG_INLINE static Scalar run(const Derived &mat, const Func&) EIGEN_STRONG_INLINE static Scalar run(const Derived &mat, const Func&)
{ {
return mat.coeff(row, col); return mat.coeffByOuterInner(outer, inner);
} }
}; };
@ -148,12 +148,8 @@ struct ei_redux_vec_unroller<Func, Derived, Start, 1>
{ {
enum { enum {
index = Start * ei_packet_traits<typename Derived::Scalar>::size, index = Start * ei_packet_traits<typename Derived::Scalar>::size,
row = int(Derived::Flags)&RowMajorBit outer = index / int(Derived::InnerSizeAtCompileTime),
? index / int(Derived::ColsAtCompileTime) inner = index % int(Derived::InnerSizeAtCompileTime),
: index % Derived::RowsAtCompileTime,
col = int(Derived::Flags)&RowMajorBit
? index % int(Derived::ColsAtCompileTime)
: index / Derived::RowsAtCompileTime,
alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned
}; };
@ -162,7 +158,7 @@ struct ei_redux_vec_unroller<Func, Derived, Start, 1>
EIGEN_STRONG_INLINE static PacketScalar run(const Derived &mat, const Func&) EIGEN_STRONG_INLINE static PacketScalar run(const Derived &mat, const Func&)
{ {
return mat.template packet<alignment>(row, col); return mat.template packetByOuterInner<alignment>(outer, inner);
} }
}; };
@ -184,12 +180,12 @@ struct ei_redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
{ {
ei_assert(mat.rows()>0 && mat.cols()>0 && "you are using a non initialized matrix"); ei_assert(mat.rows()>0 && mat.cols()>0 && "you are using a non initialized matrix");
Scalar res; Scalar res;
res = mat.coeff(0, 0); res = mat.coeffByOuterInner(0, 0);
for(int i = 1; i < mat.rows(); ++i) for(int i = 1; i < mat.innerSize(); ++i)
res = func(res, mat.coeff(i, 0)); res = func(res, mat.coeffByOuterInner(0, i));
for(int j = 1; j < mat.cols(); ++j) for(int i = 1; i < mat.outerSize(); ++i)
for(int i = 0; i < mat.rows(); ++i) for(int j = 0; j < mat.innerSize(); ++j)
res = func(res, mat.coeff(i, j)); res = func(res, mat.coeffByOuterInner(i, j));
return res; return res;
} }
}; };
@ -253,8 +249,7 @@ struct ei_redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
const int innerSize = mat.innerSize(); const int innerSize = mat.innerSize();
const int outerSize = mat.outerSize(); const int outerSize = mat.outerSize();
enum { enum {
packetSize = ei_packet_traits<Scalar>::size, packetSize = ei_packet_traits<Scalar>::size
isRowMajor = Derived::Flags&RowMajorBit?1:0
}; };
const int packetedInnerSize = ((innerSize)/packetSize)*packetSize; const int packetedInnerSize = ((innerSize)/packetSize)*packetSize;
Scalar res; Scalar res;
@ -263,13 +258,12 @@ struct ei_redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
PacketScalar packet_res = mat.template packet<Unaligned>(0,0); PacketScalar packet_res = mat.template packet<Unaligned>(0,0);
for(int j=0; j<outerSize; ++j) for(int j=0; j<outerSize; ++j)
for(int i=0; i<packetedInnerSize; i+=int(packetSize)) for(int i=0; i<packetedInnerSize; i+=int(packetSize))
packet_res = func.packetOp(packet_res, mat.template packet<Unaligned> packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned>(j,i));
(isRowMajor?j:i, isRowMajor?i:j));
res = func.predux(packet_res); res = func.predux(packet_res);
for(int j=0; j<outerSize; ++j) for(int j=0; j<outerSize; ++j)
for(int i=packetedInnerSize; i<innerSize; ++i) for(int i=packetedInnerSize; i<innerSize; ++i)
res = func(res, mat.coeff(isRowMajor?j:i, isRowMajor?i:j)); res = func(res, mat.coeffByOuterInner(j,i));
} }
else // too small to vectorize anything. else // too small to vectorize anything.
// since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize. // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.