GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-24 02:29:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

make use of ei_pmadd in dot-product: will further improve performance

Browse Source 
on architectures having a packed-mul-add assembly instruction.
This commit is contained in:
Benoit Jacob 2008-06-24 18:08:35 +00:00
parent 3b94436d2f
commit 5b0da4b778
1 changed files with 7 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Eigen/src/Core/Dot.h
View File

@ -108,8 +108,9 @@ struct ei_dot_vec_unroller
inline static PacketScalar run(const Derived1& v1, const Derived2& v2) inline static PacketScalar run(const Derived1& v1, const Derived2& v2)
{ {
return ei_padd( return ei_pmadd(
ei_pmul(v1.template packet<Aligned>(row1, col1), v2.template packet<Aligned>(row2, col2)), v1.template packet<Aligned>(row1, col1),
v2.template packet<Aligned>(row2, col2),
ei_dot_vec_unroller<Derived1, Derived2, Index+ei_packet_traits<Scalar>::size, Stop>::run(v1, v2) ei_dot_vec_unroller<Derived1, Derived2, Index+ei_packet_traits<Scalar>::size, Stop>::run(v1, v2)
); );
} }
@ -192,12 +193,10 @@ struct ei_dot_impl<Derived1, Derived2, LinearVectorization, NoUnrolling>
const int col1 = rowVector1 ? index : 0; const int col1 = rowVector1 ? index : 0;
const int row2 = rowVector2 ? 0 : index; const int row2 = rowVector2 ? 0 : index;
const int col2 = rowVector2 ? index : 0; const int col2 = rowVector2 ? index : 0;
packet_res = ei_padd( packet_res = ei_pmadd(
packet_res, v1.template packet<Aligned>(row1, col1),
ei_pmul( v2.template packet<Aligned>(row2, col2),
v1.template packet<Aligned>(row1, col1), packet_res
v2.template packet<Aligned>(row2, col2)
)
); );
} }
res = ei_predux(packet_res); res = ei_predux(packet_res);