From 0d13d814c21f1fdb3b0781b84bf12914aec5225a Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Thu, 14 Sep 2017 21:27:33 +0200 Subject: Added architecture layer in the tuning database for better performance on unseen devices --- src/database/kernels/xgemm/xgemm_64.hpp | 125 ++++++++++++++++++++++---------- 1 file changed, 88 insertions(+), 37 deletions(-) (limited to 'src/database/kernels/xgemm/xgemm_64.hpp') diff --git a/src/database/kernels/xgemm/xgemm_64.hpp b/src/database/kernels/xgemm/xgemm_64.hpp index 45d04e80..625920b7 100644 --- a/src/database/kernels/xgemm/xgemm_64.hpp +++ b/src/database/kernels/xgemm/xgemm_64.hpp @@ -14,63 +14,114 @@ const DatabaseEntry XgemmDouble = { "Xgemm", Precision::kDouble, {"KWG", "KWI", "MDIMA", "MDIMC", "MWG", "NDIMB", "NDIMC", "NWG", "SA", "SB", "STRM", "STRN", "VWM", "VWN"}, { { // AMD GPUs kDeviceTypeGPU, "AMD", { - { "AMD Radeon R9 M370X Compute Engine", { 32, 2, 16, 16, 64, 8, 8, 32, 0, 0, 0, 0, 4, 4 } }, - { "Ellesmere", { 32, 2, 16, 16, 32, 16, 16, 64, 1, 1, 0, 0, 2, 2 } }, - { "Fiji", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, - { "Hawaii", { 16, 8, 32, 8, 128, 8, 8, 32, 0, 1, 0, 0, 1, 4 } }, - { "Oland", { 16, 2, 8, 16, 64, 16, 8, 16, 0, 0, 1, 1, 1, 1 } }, - { "Pitcairn", { 32, 2, 32, 16, 64, 8, 16, 32, 0, 0, 0, 0, 1, 2 } }, - { "Tahiti", { 32, 2, 16, 8, 16, 8, 8, 32, 0, 0, 0, 1, 1, 4 } }, - { "Tonga", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, - { "default", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, + { "Ellesmere", { + { "AMD Radeon RX 480", { 32, 2, 16, 16, 32, 16, 16, 64, 1, 1, 0, 0, 2, 2 } }, + { "default", { 32, 2, 16, 16, 32, 16, 16, 64, 1, 1, 0, 0, 2, 2 } }, + } }, + { "Fiji", { + { "AMD Radeon R9 Fury X", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, + { "AMD Radeon R9 M370X Compute Engine", { 32, 2, 16, 16, 64, 8, 8, 32, 0, 0, 0, 0, 4, 4 } }, + { "default", { 32, 2, 8, 8, 16, 32, 32, 32, 1, 1, 0, 0, 1, 1 } }, + } }, + { "Hawaii", { + { "AMD Radeon R9 290X", { 16, 8, 32, 8, 128, 8, 8, 32, 0, 1, 0, 0, 1, 4 } }, + { "default", { 16, 8, 32, 8, 128, 8, 8, 32, 0, 1, 0, 0, 1, 4 } }, + } }, + { "Oland", { + { "Oland", { 16, 2, 8, 16, 64, 16, 8, 16, 0, 0, 1, 1, 1, 1 } }, + { "default", { 16, 2, 8, 16, 64, 16, 8, 16, 0, 0, 1, 1, 1, 1 } }, + } }, + { "Pitcairn", { + { "AMD Radeon R9 270X", { 32, 2, 32, 16, 64, 8, 16, 32, 0, 0, 0, 0, 1, 2 } }, + { "default", { 32, 2, 32, 16, 64, 8, 16, 32, 0, 0, 0, 0, 1, 2 } }, + } }, + { "Tahiti", { + { "AMD Radeon HD 7970", { 32, 2, 16, 8, 16, 8, 8, 32, 0, 0, 0, 1, 1, 4 } }, + { "default", { 32, 2, 16, 8, 16, 8, 8, 32, 0, 0, 0, 1, 1, 4 } }, + } }, + { "Tonga", { + { "AMD Radeon R9 380", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, + { "default", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, + } }, + { "default", { + { "default", { 32, 2, 16, 16, 32, 16, 16, 32, 1, 1, 0, 0, 2, 2 } }, + } }, } }, { // ARM GPUs kDeviceTypeGPU, "ARM", { - { "Mali-T628", { 32, 2, 8, 8, 64, 8, 8, 16, 0, 1, 1, 0, 8, 2 } }, - { "default", { 32, 2, 8, 8, 64, 8, 8, 16, 0, 1, 1, 0, 8, 2 } }, + { "default", { + { "Mali-T628", { 32, 2, 8, 8, 64, 8, 8, 16, 0, 1, 1, 0, 8, 2 } }, + { "default", { 32, 2, 8, 8, 64, 8, 8, 16, 0, 1, 1, 0, 8, 2 } }, + } }, } }, { // Intel CPUs kDeviceTypeCPU, "Intel", { - { "Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz", { 16, 2, 32, 8, 128, 16, 16, 128, 1, 1, 1, 1, 2, 8 } }, - { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { 32, 2, 32, 16, 128, 16, 16, 64, 0, 1, 1, 0, 1, 2 } }, - { "Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz", { 32, 2, 32, 16, 128, 16, 16, 128, 0, 0, 1, 0, 1, 2 } }, - { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { 32, 2, 16, 8, 128, 8, 8, 64, 1, 0, 0, 1, 2, 8 } }, - { "Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz", { 32, 2, 16, 8, 128, 8, 8, 128, 1, 0, 0, 0, 2, 8 } }, - { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { 32, 2, 8, 16, 128, 16, 8, 128, 0, 0, 1, 1, 1, 8 } }, - { "default", { 32, 2, 16, 16, 64, 8, 8, 64, 1, 1, 0, 0, 1, 4 } }, + { "default", { + { "Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz", { 16, 2, 32, 8, 128, 16, 16, 128, 1, 1, 1, 1, 2, 8 } }, + { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { 32, 2, 32, 16, 128, 16, 16, 64, 0, 1, 1, 0, 1, 2 } }, + { "Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz", { 32, 2, 32, 16, 128, 16, 16, 128, 0, 0, 1, 0, 1, 2 } }, + { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { 32, 2, 16, 8, 128, 8, 8, 64, 1, 0, 0, 1, 2, 8 } }, + { "Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz", { 32, 2, 16, 8, 128, 8, 8, 128, 1, 0, 0, 0, 2, 8 } }, + { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { 32, 2, 8, 16, 128, 16, 8, 128, 0, 0, 1, 1, 1, 8 } }, + { "default", { 32, 2, 16, 16, 64, 8, 8, 64, 1, 1, 0, 0, 1, 4 } }, + } }, } }, { // Intel accelerators kDeviceTypeAccelerator, "Intel", { - { "Intel(R) Many Integrated Core Acceleration Card", { 32, 8, 8, 16, 16, 16, 16, 128, 0, 0, 1, 0, 1, 4 } }, - { "default", { 32, 8, 8, 16, 16, 16, 16, 128, 0, 0, 1, 0, 1, 4 } }, + { "default", { + { "Intel(R) Many Integrated Core Acceleration Card", { 32, 8, 8, 16, 16, 16, 16, 128, 0, 0, 1, 0, 1, 4 } }, + { "default", { 32, 8, 8, 16, 16, 16, 16, 128, 0, 0, 1, 0, 1, 4 } }, + } }, } }, { // NVIDIA GPUs kDeviceTypeGPU, "NVIDIA", { - { "GRID K520", { 16, 2, 8, 8, 16, 8, 8, 32, 1, 0, 0, 1, 2, 2 } }, - { "GeForce GTX 1070", { 16, 2, 8, 16, 32, 8, 8, 64, 0, 0, 1, 1, 2, 8 } }, - { "GeForce GTX 1080", { 32, 2, 16, 16, 32, 16, 16, 64, 0, 0, 0, 0, 2, 4 } }, - { "GeForce GTX 480", { 16, 2, 8, 16, 32, 32, 8, 64, 1, 1, 1, 0, 1, 2 } }, - { "GeForce GTX 670", { 32, 8, 16, 32, 128, 16, 8, 32, 0, 1, 1, 0, 1, 1 } }, - { "GeForce GTX 680", { 32, 8, 8, 8, 32, 16, 32, 128, 1, 0, 0, 1, 2, 4 } }, - { "GeForce GTX 750", { 32, 8, 16, 32, 64, 16, 8, 128, 0, 0, 0, 1, 2, 1 } }, - { "GeForce GTX 750 Ti", { 32, 2, 8, 8, 32, 16, 16, 32, 0, 0, 0, 0, 4, 2 } }, - { "GeForce GTX 980", { 32, 8, 16, 8, 64, 32, 32, 128, 0, 0, 1, 0, 2, 4 } }, - { "GeForce GTX TITAN", { 16, 8, 16, 8, 32, 16, 32, 128, 1, 1, 1, 1, 2, 2 } }, - { "GeForce GTX TITAN Black", { 16, 2, 16, 8, 16, 16, 8, 16, 1, 1, 1, 0, 1, 1 } }, - { "GeForce GTX TITAN X", { 16, 8, 16, 16, 16, 16, 16, 16, 0, 0, 0, 0, 1, 1 } }, - { "TITAN X (Pascal)", { 32, 2, 32, 32, 32, 16, 16, 32, 0, 0, 0, 0, 1, 2 } }, - { "Tesla K20m", { 16, 2, 32, 8, 32, 16, 16, 64, 1, 0, 0, 0, 1, 1 } }, - { "Tesla K40m", { 32, 2, 16, 8, 64, 16, 32, 128, 1, 0, 1, 1, 2, 4 } }, - { "default", { 32, 2, 16, 16, 32, 16, 16, 64, 0, 0, 0, 0, 2, 4 } }, + { "SM2.0", { + { "GeForce GTX 480", { 16, 2, 8, 16, 32, 32, 8, 64, 1, 1, 1, 0, 1, 2 } }, + { "default", { 16, 2, 8, 16, 32, 32, 8, 64, 1, 1, 1, 0, 1, 2 } }, + } }, + { "SM3.0", { + { "GRID K520", { 16, 2, 8, 8, 16, 8, 8, 32, 1, 0, 0, 1, 2, 2 } }, + { "GeForce GTX 670", { 32, 8, 16, 32, 128, 16, 8, 32, 0, 1, 1, 0, 1, 1 } }, + { "GeForce GTX 680", { 32, 8, 8, 8, 32, 16, 32, 128, 1, 0, 0, 1, 2, 4 } }, + { "default", { 16, 2, 16, 32, 128, 16, 32, 128, 0, 0, 0, 0, 1, 1 } }, + } }, + { "SM3.5", { + { "GeForce GTX TITAN", { 16, 8, 16, 8, 32, 16, 32, 128, 1, 1, 1, 1, 2, 2 } }, + { "GeForce GTX TITAN Black", { 16, 2, 16, 8, 16, 16, 8, 16, 1, 1, 1, 0, 1, 1 } }, + { "Tesla K20m", { 16, 2, 32, 8, 32, 16, 16, 64, 1, 0, 0, 0, 1, 1 } }, + { "Tesla K40m", { 32, 2, 16, 8, 64, 16, 32, 128, 1, 0, 1, 1, 2, 4 } }, + { "default", { 16, 2, 16, 8, 16, 16, 16, 128, 1, 0, 0, 0, 1, 1 } }, + } }, + { "SM5.0", { + { "GeForce GTX 750", { 32, 8, 16, 32, 64, 16, 8, 128, 0, 0, 0, 1, 2, 1 } }, + { "GeForce GTX 750 Ti", { 32, 2, 8, 8, 32, 16, 16, 32, 0, 0, 0, 0, 4, 2 } }, + { "default", { 32, 2, 16, 32, 32, 16, 16, 128, 0, 0, 0, 0, 2, 1 } }, + } }, + { "SM5.2", { + { "GeForce GTX 980", { 32, 8, 16, 8, 64, 32, 32, 128, 0, 0, 1, 0, 2, 4 } }, + { "GeForce GTX TITAN X", { 16, 8, 16, 16, 16, 16, 16, 16, 0, 0, 0, 0, 1, 1 } }, + { "default", { 16, 8, 16, 16, 16, 16, 16, 128, 0, 0, 0, 0, 1, 1 } }, + } }, + { "SM6.1", { + { "GeForce GTX 1070", { 16, 2, 8, 16, 32, 8, 8, 64, 0, 0, 1, 1, 2, 8 } }, + { "GeForce GTX 1080", { 32, 2, 16, 16, 32, 16, 16, 64, 0, 0, 0, 0, 2, 4 } }, + { "TITAN X (Pascal)", { 32, 2, 32, 32, 32, 16, 16, 32, 0, 0, 0, 0, 1, 2 } }, + { "default", { 32, 2, 16, 16, 32, 16, 16, 64, 0, 0, 0, 0, 2, 4 } }, + } }, + { "default", { + { "default", { 32, 2, 16, 16, 32, 16, 16, 64, 0, 0, 0, 0, 2, 4 } }, + } }, } }, { // Default kDeviceTypeAll, "default", { - { "default", { 32, 2, 8, 8, 32, 8, 8, 64, 0, 0, 0, 0, 4, 4 } }, + { "default", { + { "default", { 32, 2, 8, 8, 32, 8, 8, 64, 0, 0, 0, 0, 4, 4 } }, + } }, } }, } -- cgit v1.2.3