From 1c33af6eabdc1bc7be7b311fa29c0599cebbf0e6 Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sun, 23 Apr 2017 17:58:56 +0200 Subject: Re-added Titan X (Pascal) tuning results based on more averaging when tuning --- src/database/kernels/copy.hpp | 12 ++++++------ src/database/kernels/pad.hpp | 12 ++++++------ src/database/kernels/padtranspose.hpp | 6 +++--- src/database/kernels/transpose.hpp | 6 +++--- src/database/kernels/xaxpy.hpp | 10 +++++----- src/database/kernels/xdot.hpp | 8 ++++---- src/database/kernels/xgemm.hpp | 14 +++++++------- src/database/kernels/xgemm_direct.hpp | 12 ++++++------ src/database/kernels/xgemv.hpp | 4 ++-- src/database/kernels/xgemv_fast.hpp | 2 +- src/database/kernels/xgemv_fast_rot.hpp | 4 ++-- src/database/kernels/xger.hpp | 10 +++++----- 12 files changed, 50 insertions(+), 50 deletions(-) (limited to 'src/database') diff --git a/src/database/kernels/copy.hpp b/src/database/kernels/copy.hpp index bd316233..89d5a3a6 100644 --- a/src/database/kernels/copy.hpp +++ b/src/database/kernels/copy.hpp @@ -104,10 +104,10 @@ const Database::DatabaseEntry CopySingle = { { "GeForce GTX TITAN", { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",4} } }, { "GeForce GTX TITAN Black", { {"COPY_DIMX",8}, {"COPY_DIMY",32}, {"COPY_VW",4}, {"COPY_WPT",8} } }, { "GeForce GTX TITAN X", { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",2} } }, - { "TITAN X (Pascal)", { {"COPY_DIMX",32}, {"COPY_DIMY",32}, {"COPY_VW",4}, {"COPY_WPT",1} } }, + { "TITAN X (Pascal)", { {"COPY_DIMX",8}, {"COPY_DIMY",32}, {"COPY_VW",4}, {"COPY_WPT",1} } }, { "Tesla K20m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",4}, {"COPY_WPT",4} } }, { "Tesla K40m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",4}, {"COPY_WPT",2} } }, - { "default", { {"COPY_DIMX",8}, {"COPY_DIMY",16}, {"COPY_VW",4}, {"COPY_WPT",1} } }, + { "default", { {"COPY_DIMX",8}, {"COPY_DIMY",32}, {"COPY_VW",4}, {"COPY_WPT",1} } }, } }, { // Default @@ -176,7 +176,7 @@ const Database::DatabaseEntry CopyComplexSingle = { { "GeForce GTX 980", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, { "GeForce GTX TITAN Black", { {"COPY_DIMX",16}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, { "GeForce GTX TITAN X", { {"COPY_DIMX",16}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, - { "TITAN X (Pascal)", { {"COPY_DIMX",32}, {"COPY_DIMY",16}, {"COPY_VW",1}, {"COPY_WPT",1} } }, + { "TITAN X (Pascal)", { {"COPY_DIMX",8}, {"COPY_DIMY",16}, {"COPY_VW",2}, {"COPY_WPT",1} } }, { "Tesla K20m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",4} } }, { "Tesla K40m", { {"COPY_DIMX",16}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, { "default", { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, @@ -242,10 +242,10 @@ const Database::DatabaseEntry CopyDouble = { { "GeForce GTX TITAN", { {"COPY_DIMX",16}, {"COPY_DIMY",32}, {"COPY_VW",2}, {"COPY_WPT",2} } }, { "GeForce GTX TITAN Black", { {"COPY_DIMX",16}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",8} } }, { "GeForce GTX TITAN X", { {"COPY_DIMX",32}, {"COPY_DIMY",16}, {"COPY_VW",1}, {"COPY_WPT",1} } }, - { "TITAN X (Pascal)", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",1} } }, + { "TITAN X (Pascal)", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",2} } }, { "Tesla K20m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",1} } }, { "Tesla K40m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",2} } }, - { "default", { {"COPY_DIMX",32}, {"COPY_DIMY",16}, {"COPY_VW",2}, {"COPY_WPT",1} } }, + { "default", { {"COPY_DIMX",32}, {"COPY_DIMY",32}, {"COPY_VW",2}, {"COPY_WPT",1} } }, } }, { // Default @@ -308,7 +308,7 @@ const Database::DatabaseEntry CopyComplexDouble = { { "GeForce GTX TITAN", { {"COPY_DIMX",16}, {"COPY_DIMY",16}, {"COPY_VW",1}, {"COPY_WPT",1} } }, { "GeForce GTX TITAN Black", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",2} } }, { "GeForce GTX TITAN X", { {"COPY_DIMX",16}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, - { "TITAN X (Pascal)", { {"COPY_DIMX",32}, {"COPY_DIMY",32}, {"COPY_VW",1}, {"COPY_WPT",1} } }, + { "TITAN X (Pascal)", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",2} } }, { "Tesla K20m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",2} } }, { "Tesla K40m", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, { "default", { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",1} } }, diff --git a/src/database/kernels/pad.hpp b/src/database/kernels/pad.hpp index 84845349..c3bce0e3 100644 --- a/src/database/kernels/pad.hpp +++ b/src/database/kernels/pad.hpp @@ -104,10 +104,10 @@ const Database::DatabaseEntry PadSingle = { { "GeForce GTX TITAN", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",2}, {"PAD_WPTY",1} } }, { "GeForce GTX TITAN Black", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "GeForce GTX TITAN X", { {"PAD_DIMX",16}, {"PAD_DIMY",16}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, - { "TITAN X (Pascal)", { {"PAD_DIMX",16}, {"PAD_DIMY",16}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, + { "TITAN X (Pascal)", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "Tesla K20m", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",2}, {"PAD_WPTY",1} } }, { "Tesla K40m", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, - { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",2}, {"PAD_WPTY",2} } }, + { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",4} } }, } }, { // Default @@ -184,7 +184,7 @@ const Database::DatabaseEntry PadComplexSingle = { { "GeForce GTX TITAN", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",2}, {"PAD_WPTY",1} } }, { "GeForce GTX TITAN Black", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "GeForce GTX TITAN X", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, - { "TITAN X (Pascal)", { {"PAD_DIMX",16}, {"PAD_DIMY",16}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, + { "TITAN X (Pascal)", { {"PAD_DIMX",32}, {"PAD_DIMY",32}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "Tesla K20m", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "Tesla K40m", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, @@ -192,7 +192,7 @@ const Database::DatabaseEntry PadComplexSingle = { }, { // Default kDeviceTypeAll, "default", { - { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, + { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, } }, } @@ -250,7 +250,7 @@ const Database::DatabaseEntry PadDouble = { { "GeForce GTX TITAN", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "GeForce GTX TITAN Black", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "GeForce GTX TITAN X", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, - { "TITAN X (Pascal)", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, + { "TITAN X (Pascal)", { {"PAD_DIMX",8}, {"PAD_DIMY",32}, {"PAD_WPTX",4}, {"PAD_WPTY",1} } }, { "Tesla K20m", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "Tesla K40m", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "default", { {"PAD_DIMX",32}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, @@ -316,7 +316,7 @@ const Database::DatabaseEntry PadComplexDouble = { { "GeForce GTX TITAN", { {"PAD_DIMX",8}, {"PAD_DIMY",32}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "GeForce GTX TITAN Black", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",4} } }, { "GeForce GTX TITAN X", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, - { "TITAN X (Pascal)", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",2}, {"PAD_WPTY",1} } }, + { "TITAN X (Pascal)", { {"PAD_DIMX",8}, {"PAD_DIMY",16}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "Tesla K20m", { {"PAD_DIMX",8}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",2} } }, { "Tesla K40m", { {"PAD_DIMX",8}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, { "default", { {"PAD_DIMX",16}, {"PAD_DIMY",8}, {"PAD_WPTX",1}, {"PAD_WPTY",1} } }, diff --git a/src/database/kernels/padtranspose.hpp b/src/database/kernels/padtranspose.hpp index 325df121..94bac195 100644 --- a/src/database/kernels/padtranspose.hpp +++ b/src/database/kernels/padtranspose.hpp @@ -184,7 +184,7 @@ const Database::DatabaseEntry PadtransposeComplexSingle = { { "GeForce GTX TITAN", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN Black", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN X", { {"PADTRA_PAD",1}, {"PADTRA_TILE",32}, {"PADTRA_WPT",1} } }, - { "TITAN X (Pascal)", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, + { "TITAN X (Pascal)", { {"PADTRA_PAD",1}, {"PADTRA_TILE",8}, {"PADTRA_WPT",1} } }, { "Tesla K20m", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "Tesla K40m", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "default", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, @@ -250,7 +250,7 @@ const Database::DatabaseEntry PadtransposeDouble = { { "GeForce GTX TITAN", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN Black", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN X", { {"PADTRA_PAD",1}, {"PADTRA_TILE",32}, {"PADTRA_WPT",1} } }, - { "TITAN X (Pascal)", { {"PADTRA_PAD",0}, {"PADTRA_TILE",8}, {"PADTRA_WPT",2} } }, + { "TITAN X (Pascal)", { {"PADTRA_PAD",0}, {"PADTRA_TILE",8}, {"PADTRA_WPT",1} } }, { "Tesla K20m", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "Tesla K40m", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "default", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, @@ -316,7 +316,7 @@ const Database::DatabaseEntry PadtransposeComplexDouble = { { "GeForce GTX TITAN", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN Black", { {"PADTRA_PAD",0}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "GeForce GTX TITAN X", { {"PADTRA_PAD",1}, {"PADTRA_TILE",32}, {"PADTRA_WPT",1} } }, - { "TITAN X (Pascal)", { {"PADTRA_PAD",0}, {"PADTRA_TILE",8}, {"PADTRA_WPT",1} } }, + { "TITAN X (Pascal)", { {"PADTRA_PAD",1}, {"PADTRA_TILE",8}, {"PADTRA_WPT",1} } }, { "Tesla K20m", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "Tesla K40m", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, { "default", { {"PADTRA_PAD",1}, {"PADTRA_TILE",16}, {"PADTRA_WPT",1} } }, diff --git a/src/database/kernels/transpose.hpp b/src/database/kernels/transpose.hpp index ebe1db8f..8567c725 100644 --- a/src/database/kernels/transpose.hpp +++ b/src/database/kernels/transpose.hpp @@ -104,7 +104,7 @@ const Database::DatabaseEntry TransposeSingle = { { "GeForce GTX TITAN", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, { "GeForce GTX TITAN Black", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, { "GeForce GTX TITAN X", { {"TRA_DIM",16}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, - { "TITAN X (Pascal)", { {"TRA_DIM",8}, {"TRA_PAD",0}, {"TRA_SHUFFLE",1}, {"TRA_WPT",4} } }, + { "TITAN X (Pascal)", { {"TRA_DIM",8}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, { "Tesla K20m", { {"TRA_DIM",8}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, { "Tesla K40m", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, { "default", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",4} } }, @@ -178,7 +178,7 @@ const Database::DatabaseEntry TransposeComplexSingle = { { "GeForce GTX TITAN", { {"TRA_DIM",16}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, { "GeForce GTX TITAN Black", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, { "GeForce GTX TITAN X", { {"TRA_DIM",32}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, - { "TITAN X (Pascal)", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, + { "TITAN X (Pascal)", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, { "Tesla K20m", { {"TRA_DIM",16}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, { "Tesla K40m", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, { "default", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, @@ -244,7 +244,7 @@ const Database::DatabaseEntry TransposeDouble = { { "GeForce GTX TITAN", { {"TRA_DIM",8}, {"TRA_PAD",0}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, { "GeForce GTX TITAN Black", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, { "GeForce GTX TITAN X", { {"TRA_DIM",32}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",1} } }, - { "TITAN X (Pascal)", { {"TRA_DIM",8}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, + { "TITAN X (Pascal)", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, { "Tesla K20m", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",0}, {"TRA_WPT",2} } }, { "Tesla K40m", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",1}, {"TRA_WPT",2} } }, { "default", { {"TRA_DIM",16}, {"TRA_PAD",1}, {"TRA_SHUFFLE",1}, {"TRA_WPT",2} } }, diff --git a/src/database/kernels/xaxpy.hpp b/src/database/kernels/xaxpy.hpp index 18f63c4c..f31e2f6e 100644 --- a/src/database/kernels/xaxpy.hpp +++ b/src/database/kernels/xaxpy.hpp @@ -104,7 +104,7 @@ const Database::DatabaseEntry XaxpySingle = { { "GeForce GTX TITAN", { {"VW",4}, {"WGS",256}, {"WPT",1} } }, { "GeForce GTX TITAN Black", { {"VW",4}, {"WGS",128}, {"WPT",4} } }, { "GeForce GTX TITAN X", { {"VW",1}, {"WGS",64}, {"WPT",1} } }, - { "TITAN X (Pascal)", { {"VW",1}, {"WGS",128}, {"WPT",1} } }, + { "TITAN X (Pascal)", { {"VW",4}, {"WGS",128}, {"WPT",1} } }, { "Tesla K20m", { {"VW",4}, {"WGS",128}, {"WPT",1} } }, { "Tesla K40m", { {"VW",4}, {"WGS",128}, {"WPT",1} } }, { "default", { {"VW",4}, {"WGS",256}, {"WPT",1} } }, @@ -184,7 +184,7 @@ const Database::DatabaseEntry XaxpyComplexSingle = { { "GeForce GTX TITAN", { {"VW",1}, {"WGS",256}, {"WPT",1} } }, { "GeForce GTX TITAN Black", { {"VW",1}, {"WGS",128}, {"WPT",2} } }, { "GeForce GTX TITAN X", { {"VW",1}, {"WGS",512}, {"WPT",1} } }, - { "TITAN X (Pascal)", { {"VW",1}, {"WGS",64}, {"WPT",1} } }, + { "TITAN X (Pascal)", { {"VW",2}, {"WGS",512}, {"WPT",1} } }, { "Tesla K20m", { {"VW",1}, {"WGS",128}, {"WPT",1} } }, { "Tesla K40m", { {"VW",1}, {"WGS",128}, {"WPT",1} } }, { "default", { {"VW",1}, {"WGS",256}, {"WPT",1} } }, @@ -250,10 +250,10 @@ const Database::DatabaseEntry XaxpyDouble = { { "GeForce GTX TITAN", { {"VW",2}, {"WGS",1024}, {"WPT",1} } }, { "GeForce GTX TITAN Black", { {"VW",2}, {"WGS",128}, {"WPT",1} } }, { "GeForce GTX TITAN X", { {"VW",1}, {"WGS",512}, {"WPT",1} } }, - { "TITAN X (Pascal)", { {"VW",2}, {"WGS",256}, {"WPT",4} } }, + { "TITAN X (Pascal)", { {"VW",2}, {"WGS",512}, {"WPT",1} } }, { "Tesla K20m", { {"VW",2}, {"WGS",128}, {"WPT",1} } }, { "Tesla K40m", { {"VW",2}, {"WGS",128}, {"WPT",1} } }, - { "default", { {"VW",2}, {"WGS",1024}, {"WPT",1} } }, + { "default", { {"VW",1}, {"WGS",128}, {"WPT",1} } }, } }, { // Default @@ -316,7 +316,7 @@ const Database::DatabaseEntry XaxpyComplexDouble = { { "GeForce GTX TITAN", { {"VW",1}, {"WGS",64}, {"WPT",4} } }, { "GeForce GTX TITAN Black", { {"VW",1}, {"WGS",128}, {"WPT",4} } }, { "GeForce GTX TITAN X", { {"VW",1}, {"WGS",1024}, {"WPT",1} } }, - { "TITAN X (Pascal)", { {"VW",1}, {"WGS",128}, {"WPT",1} } }, + { "TITAN X (Pascal)", { {"VW",1}, {"WGS",256}, {"WPT",2} } }, { "Tesla K20m", { {"VW",1}, {"WGS",64}, {"WPT",1} } }, { "Tesla K40m", { {"VW",1}, {"WGS",64}, {"WPT",1} } }, { "default", { {"VW",1}, {"WGS",64}, {"WPT",1} } }, diff --git a/src/database/kernels/xdot.hpp b/src/database/kernels/xdot.hpp index c3f86cf9..2bc7eeba 100644 --- a/src/database/kernels/xdot.hpp +++ b/src/database/kernels/xdot.hpp @@ -87,9 +87,9 @@ const Database::DatabaseEntry XdotSingle = { { "GeForce GTX 980", { {"WGS1",256}, {"WGS2",32} } }, { "GeForce GTX TITAN Black", { {"WGS1",512}, {"WGS2",64} } }, { "GeForce GTX TITAN X", { {"WGS1",256}, {"WGS2",32} } }, - { "TITAN X (Pascal)", { {"WGS1",256}, {"WGS2",512} } }, + { "TITAN X (Pascal)", { {"WGS1",1024}, {"WGS2",32} } }, { "Tesla K20m", { {"WGS1",1024}, {"WGS2",32} } }, - { "default", { {"WGS1",256}, {"WGS2",512} } }, + { "default", { {"WGS1",256}, {"WGS2",64} } }, } }, { // Default @@ -198,7 +198,7 @@ const Database::DatabaseEntry XdotDouble = { { "GeForce GTX 980", { {"WGS1",128}, {"WGS2",32} } }, { "GeForce GTX TITAN Black", { {"WGS1",128}, {"WGS2",64} } }, { "GeForce GTX TITAN X", { {"WGS1",256}, {"WGS2",32} } }, - { "TITAN X (Pascal)", { {"WGS1",256}, {"WGS2",32} } }, + { "TITAN X (Pascal)", { {"WGS1",128}, {"WGS2",32} } }, { "Tesla K20m", { {"WGS1",512}, {"WGS2",32} } }, { "default", { {"WGS1",128}, {"WGS2",128} } }, } @@ -247,7 +247,7 @@ const Database::DatabaseEntry XdotComplexDouble = { { "GeForce GTX 980", { {"WGS1",64}, {"WGS2",32} } }, { "GeForce GTX TITAN Black", { {"WGS1",128}, {"WGS2",32} } }, { "GeForce GTX TITAN X", { {"WGS1",128}, {"WGS2",32} } }, - { "TITAN X (Pascal)", { {"WGS1",128}, {"WGS2",512} } }, + { "TITAN X (Pascal)", { {"WGS1",128}, {"WGS2",64} } }, { "Tesla K20m", { {"WGS1",128}, {"WGS2",32} } }, { "default", { {"WGS1",128}, {"WGS2",64} } }, } diff --git a/src/database/kernels/xgemm.hpp b/src/database/kernels/xgemm.hpp index 768aa817..49e86edc 100644 --- a/src/database/kernels/xgemm.hpp +++ b/src/database/kernels/xgemm.hpp @@ -103,10 +103,10 @@ const Database::DatabaseEntry XgemmSingle = { { "GeForce GTX TITAN", { {"KWG",16}, {"KWI",8}, {"MDIMA",32}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",2}, {"VWN",2} } }, { "GeForce GTX TITAN Black", { {"KWG",16}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",4}, {"VWN",1} } }, { "GeForce GTX TITAN X", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",16}, {"MWG",128}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",128}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",1}, {"VWM",4}, {"VWN",8} } }, - { "TITAN X (Pascal)", { {"KWG",16}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",16}, {"MWG",128}, {"NDIMB",32}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",4}, {"VWN",2} } }, + { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",4}, {"VWN",1} } }, { "Tesla K20m", { {"KWG",16}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",2}, {"VWN",4} } }, { "Tesla K40m", { {"KWG",16}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",128}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",2}, {"VWN",4} } }, - { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",4}, {"VWN",4} } }, + { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",4}, {"VWN",2} } }, } }, { // Default @@ -183,10 +183,10 @@ const Database::DatabaseEntry XgemmComplexSingle = { { "GeForce GTX TITAN", { {"KWG",16}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",32}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, { "GeForce GTX TITAN Black", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",0}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",1}, {"VWN",2} } }, { "GeForce GTX TITAN X", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",1}, {"SB",0}, {"STRM",1}, {"STRN",1}, {"VWM",1}, {"VWN",4} } }, - { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",32}, {"NDIMC",16}, {"NWG",32}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",1}, {"VWM",1}, {"VWN",1} } }, + { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",32}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",2}, {"VWN",4} } }, { "Tesla K20m", { {"KWG",32}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",4} } }, { "Tesla K40m", { {"KWG",16}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",32}, {"MWG",32}, {"NDIMB",32}, {"NDIMC",8}, {"NWG",64}, {"SA",0}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, - { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, + { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",16}, {"NDIMB",32}, {"NDIMC",32}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, } }, { // Default @@ -249,7 +249,7 @@ const Database::DatabaseEntry XgemmDouble = { { "GeForce GTX TITAN", { {"KWG",16}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",32}, {"NWG",128}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",1}, {"VWM",2}, {"VWN",2} } }, { "GeForce GTX TITAN Black", { {"KWG",16}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",16}, {"NDIMB",16}, {"NDIMC",8}, {"NWG",16}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, { "GeForce GTX TITAN X", { {"KWG",16}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",16}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",16}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, - { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",32}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",1}, {"STRN",0}, {"VWM",4}, {"VWN",1} } }, + { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",32}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",2} } }, { "Tesla K20m", { {"KWG",16}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, { "Tesla K40m", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",32}, {"NWG",128}, {"SA",1}, {"SB",0}, {"STRM",1}, {"STRN",1}, {"VWM",2}, {"VWN",4} } }, { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",2}, {"VWN",4} } }, @@ -317,12 +317,12 @@ const Database::DatabaseEntry XgemmComplexDouble = { { "TITAN X (Pascal)", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",16}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",16}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, { "Tesla K20m", { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",1}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, { "Tesla K40m", { {"KWG",16}, {"KWI",8}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",32}, {"NDIMC",16}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",1}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, - { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",16}, {"NDIMB",32}, {"NDIMC",32}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",2}, {"VWN",1} } }, + { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",32}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, } }, { // Default kDeviceTypeAll, "default", { - { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",16}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",2} } }, + { "default", { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",32}, {"MWG",32}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",1}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, } }, } diff --git a/src/database/kernels/xgemm_direct.hpp b/src/database/kernels/xgemm_direct.hpp index 081589f8..29a3fb19 100644 --- a/src/database/kernels/xgemm_direct.hpp +++ b/src/database/kernels/xgemm_direct.hpp @@ -70,13 +70,13 @@ const Database::DatabaseEntry XgemmDirectSingle = { { "GeForce GTX 1080", { {"KWID",16}, {"MDIMAD",16}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",32} } }, { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",2}, {"WGD",32} } }, { "GeForce GTX TITAN Black", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",2}, {"WGD",32} } }, - { "TITAN X (Pascal)", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",32} } }, - { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",2}, {"WGD",32} } }, + { "TITAN X (Pascal)", { {"KWID",8}, {"MDIMAD",32}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",2}, {"WGD",32} } }, } }, { // Default kDeviceTypeAll, "default", { - { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",2}, {"WGD",32} } }, } }, } @@ -114,7 +114,7 @@ const Database::DatabaseEntry XgemmDirectComplexSingle = { { "GeForce GTX 1080", { {"KWID",8}, {"MDIMAD",8}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, { "GeForce GTX 750 Ti", { {"KWID",16}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",1}, {"WGD",16} } }, { "GeForce GTX TITAN Black", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, - { "TITAN X (Pascal)", { {"KWID",16}, {"MDIMAD",16}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + { "TITAN X (Pascal)", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, } }, @@ -150,7 +150,7 @@ const Database::DatabaseEntry XgemmDirectDouble = { { "GeForce GTX 1080", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",4}, {"WGD",32} } }, { "GeForce GTX TITAN Black", { {"KWID",8}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",8}, {"PADA",1}, {"PADB",0}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, - { "TITAN X (Pascal)", { {"KWID",8}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",1}, {"WGD",16} } }, + { "TITAN X (Pascal)", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, } }, @@ -186,7 +186,7 @@ const Database::DatabaseEntry XgemmDirectComplexDouble = { { "GeForce GTX 1080", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",32}, {"MDIMCD",32}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",32} } }, { "GeForce GTX TITAN Black", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",8} } }, - { "TITAN X (Pascal)", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",16}, {"NDIMCD",8}, {"PADA",1}, {"PADB",0}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + { "TITAN X (Pascal)", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, } }, diff --git a/src/database/kernels/xgemv.hpp b/src/database/kernels/xgemv.hpp index 0a72a68c..d56f9acc 100644 --- a/src/database/kernels/xgemv.hpp +++ b/src/database/kernels/xgemv.hpp @@ -100,7 +100,7 @@ const Database::DatabaseEntry XgemvSingle = { { "TITAN X (Pascal)", { {"WGS1",32}, {"WPT1",1} } }, { "Tesla K20m", { {"WGS1",128}, {"WPT1",1} } }, { "Tesla K40m", { {"WGS1",256}, {"WPT1",1} } }, - { "default", { {"WGS1",128}, {"WPT1",1} } }, + { "default", { {"WGS1",256}, {"WPT1",1} } }, } }, { // Default @@ -233,7 +233,7 @@ const Database::DatabaseEntry XgemvDouble = { }, { // Default kDeviceTypeAll, "default", { - { "default", { {"WGS1",64}, {"WPT1",1} } }, + { "default", { {"WGS1",128}, {"WPT1",1} } }, } }, } diff --git a/src/database/kernels/xgemv_fast.hpp b/src/database/kernels/xgemv_fast.hpp index d63aa3f3..2ab6265e 100644 --- a/src/database/kernels/xgemv_fast.hpp +++ b/src/database/kernels/xgemv_fast.hpp @@ -97,7 +97,7 @@ const Database::DatabaseEntry XgemvFastSingle = { { "GeForce GTX TITAN", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } }, { "GeForce GTX TITAN Black", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } }, { "GeForce GTX TITAN X", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } }, - { "TITAN X (Pascal)", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } }, + { "TITAN X (Pascal)", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } }, { "Tesla K20m", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } }, { "Tesla K40m", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } }, { "default", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } }, diff --git a/src/database/kernels/xgemv_fast_rot.hpp b/src/database/kernels/xgemv_fast_rot.hpp index 6b25e0d6..b234e27f 100644 --- a/src/database/kernels/xgemv_fast_rot.hpp +++ b/src/database/kernels/xgemv_fast_rot.hpp @@ -75,7 +75,7 @@ const Database::DatabaseEntry XgemvFastRotSingle = { { "GeForce GTX 750 Ti", { {"VW3",8}, {"WGS3",32}, {"WPT3",32} } }, { "GeForce GTX TITAN", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } }, { "GeForce GTX TITAN Black", { {"VW3",4}, {"WGS3",128}, {"WPT3",16} } }, - { "TITAN X (Pascal)", { {"VW3",8}, {"WGS3",32}, {"WPT3",32} } }, + { "TITAN X (Pascal)", { {"VW3",8}, {"WGS3",64}, {"WPT3",32} } }, { "default", { {"VW3",8}, {"WGS3",32}, {"WPT3",32} } }, } }, @@ -153,7 +153,7 @@ const Database::DatabaseEntry XgemvFastRotDouble = { { "GeForce GTX 750 Ti", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } }, { "GeForce GTX TITAN", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } }, { "GeForce GTX TITAN Black", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } }, - { "TITAN X (Pascal)", { {"VW3",8}, {"WGS3",32}, {"WPT3",16} } }, + { "TITAN X (Pascal)", { {"VW3",8}, {"WGS3",32}, {"WPT3",32} } }, { "default", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } }, } }, diff --git a/src/database/kernels/xger.hpp b/src/database/kernels/xger.hpp index 70832f6f..5a61f0df 100644 --- a/src/database/kernels/xger.hpp +++ b/src/database/kernels/xger.hpp @@ -94,8 +94,8 @@ const Database::DatabaseEntry XgerSingle = { { "GeForce GTX 750 Ti", { {"WGS1",64}, {"WGS2",1}, {"WPT",2} } }, { "GeForce GTX TITAN", { {"WGS1",32}, {"WGS2",4}, {"WPT",2} } }, { "GeForce GTX TITAN Black", { {"WGS1",32}, {"WGS2",4}, {"WPT",2} } }, - { "TITAN X (Pascal)", { {"WGS1",128}, {"WGS2",4}, {"WPT",1} } }, - { "default", { {"WGS1",256}, {"WGS2",1}, {"WPT",2} } }, + { "TITAN X (Pascal)", { {"WGS1",512}, {"WGS2",2}, {"WPT",1} } }, + { "default", { {"WGS1",128}, {"WGS2",1}, {"WPT",2} } }, } }, { // Default @@ -217,8 +217,8 @@ const Database::DatabaseEntry XgerDouble = { { "GeForce GTX 750 Ti", { {"WGS1",32}, {"WGS2",16}, {"WPT",1} } }, { "GeForce GTX TITAN", { {"WGS1",16}, {"WGS2",8}, {"WPT",2} } }, { "GeForce GTX TITAN Black", { {"WGS1",32}, {"WGS2",4}, {"WPT",2} } }, - { "TITAN X (Pascal)", { {"WGS1",32}, {"WGS2",16}, {"WPT",1} } }, - { "default", { {"WGS1",64}, {"WGS2",2}, {"WPT",2} } }, + { "TITAN X (Pascal)", { {"WGS1",32}, {"WGS2",2}, {"WPT",1} } }, + { "default", { {"WGS1",128}, {"WGS2",1}, {"WPT",2} } }, } }, { // Default @@ -272,7 +272,7 @@ const Database::DatabaseEntry XgerComplexDouble = { { "GeForce GTX 750 Ti", { {"WGS1",32}, {"WGS2",8}, {"WPT",2} } }, { "GeForce GTX TITAN", { {"WGS1",32}, {"WGS2",4}, {"WPT",2} } }, { "GeForce GTX TITAN Black", { {"WGS1",16}, {"WGS2",16}, {"WPT",2} } }, - { "TITAN X (Pascal)", { {"WGS1",64}, {"WGS2",4}, {"WPT",1} } }, + { "TITAN X (Pascal)", { {"WGS1",4}, {"WGS2",8}, {"WPT",1} } }, { "default", { {"WGS1",16}, {"WGS2",8}, {"WPT",2} } }, } }, -- cgit v1.2.3