From 5b0da4b778d8f51e21dd7c35f6f7397c38c6be03 Mon Sep 17 00:00:00 2001 From: Benoit Jacob Date: Tue, 24 Jun 2008 18:08:35 +0000 Subject: [PATCH] make use of ei_pmadd in dot-product: will further improve performance on architectures having a packed-mul-add assembly instruction. --- Eigen/src/Core/Dot.h | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-) diff --git a/Eigen/src/Core/Dot.h b/Eigen/src/Core/Dot.h index bf487c462..8f38ee946 100644 --- a/Eigen/src/Core/Dot.h +++ b/Eigen/src/Core/Dot.h @@ -108,8 +108,9 @@ struct ei_dot_vec_unroller inline static PacketScalar run(const Derived1& v1, const Derived2& v2) { - return ei_padd( - ei_pmul(v1.template packet(row1, col1), v2.template packet(row2, col2)), + return ei_pmadd( + v1.template packet(row1, col1), + v2.template packet(row2, col2), ei_dot_vec_unroller::size, Stop>::run(v1, v2) ); } @@ -192,12 +193,10 @@ struct ei_dot_impl const int col1 = rowVector1 ? index : 0; const int row2 = rowVector2 ? 0 : index; const int col2 = rowVector2 ? index : 0; - packet_res = ei_padd( - packet_res, - ei_pmul( - v1.template packet(row1, col1), - v2.template packet(row2, col2) - ) + packet_res = ei_pmadd( + v1.template packet(row1, col1), + v2.template packet(row2, col2), + packet_res ); } res = ei_predux(packet_res);