CLBlast/src/database/kernels/pad/pad_32.hpp

102 lines
5.1 KiB
C++
Raw Normal View History

// =================================================================================================
// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
// is auto-generated by the 'scripts/database/database.py' Python script.
//
// This file populates the database with best-found tuning parameters for the 'Pad32' kernels.
//
// =================================================================================================
namespace clblast {
namespace database {
const DatabaseEntry PadSingle = {
"Pad", Precision::kSingle, {"PAD_DIMX", "PAD_DIMY", "PAD_WPTX", "PAD_WPTY"}, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
{ "AMD Radeon R9 M370X Compute Engine", { 32, 8, 1, 1 } },
{ "ATI Radeon HD 6750M", { 8, 16, 2, 1 } },
{ "Ellesmere", { 32, 8, 2, 2 } },
{ "Fiji", { 16, 16, 1, 2 } },
{ "Hawaii", { 32, 8, 1, 4 } },
{ "Oland", { 8, 8, 1, 2 } },
{ "Pitcairn", { 32, 8, 1, 2 } },
{ "Tahiti", { 32, 8, 1, 2 } },
{ "Tonga", { 16, 16, 2, 2 } },
{ "Turks", { 32, 8, 2, 1 } },
{ "default", { 8, 16, 1, 2 } },
}
},
{ // ARM GPUs
kDeviceTypeGPU, "ARM", {
{ "Mali-T628", { 32, 8, 1, 4 } },
{ "default", { 32, 8, 1, 4 } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
{ "Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz", { 32, 32, 4, 4 } },
{ "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { 32, 16, 4, 1 } },
{ "Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz", { 32, 8, 2, 4 } },
{ "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { 16, 32, 4, 4 } },
{ "Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz", { 32, 16, 4, 4 } },
{ "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { 32, 8, 4, 1 } },
{ "default", { 32, 8, 4, 2 } },
}
},
{ // Intel GPUs
kDeviceTypeGPU, "Intel", {
{ "Intel(R) HD Graphics 530", { 32, 8, 2, 4 } },
{ "Intel(R) HD Graphics 5500 BroadWell U-Processor GT2", { 32, 8, 2, 4 } },
{ "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { 16, 8, 1, 2 } },
{ "Intel(R) HD Graphics IvyBridge M GT2", { 16, 8, 4, 1 } },
{ "Intel(R) HD Graphics Skylake ULT GT2", { 32, 8, 4, 2 } },
{ "Iris", { 32, 16, 2, 1 } },
{ "Iris Pro", { 16, 8, 2, 1 } },
{ "default", { 32, 8, 4, 2 } },
}
},
{ // Intel accelerators
kDeviceTypeAccelerator, "Intel", {
{ "Intel(R) Many Integrated Core Acceleration Card", { 32, 16, 2, 1 } },
{ "default", { 32, 16, 2, 1 } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
{ "GRID K520", { 32, 8, 2, 1 } },
{ "GeForce GT 650M", { 32, 16, 2, 2 } },
{ "GeForce GTX 1070", { 16, 8, 1, 1 } },
{ "GeForce GTX 1080", { 16, 8, 1, 1 } },
{ "GeForce GTX 480", { 32, 8, 1, 4 } },
{ "GeForce GTX 670", { 32, 8, 2, 2 } },
{ "GeForce GTX 680", { 16, 8, 4, 1 } },
{ "GeForce GTX 750", { 32, 16, 4, 2 } },
{ "GeForce GTX 750 Ti", { 16, 8, 4, 1 } },
{ "GeForce GTX 980", { 16, 8, 1, 1 } },
{ "GeForce GTX TITAN", { 32, 8, 2, 1 } },
{ "GeForce GTX TITAN Black", { 32, 8, 1, 2 } },
{ "GeForce GTX TITAN X", { 16, 16, 1, 1 } },
{ "TITAN X (Pascal)", { 16, 8, 1, 2 } },
{ "Tesla K20m", { 32, 8, 2, 1 } },
{ "Tesla K40m", { 32, 8, 1, 1 } },
{ "default", { 32, 8, 4, 1 } },
}
},
{ // QUALCOMM GPUs
kDeviceTypeGPU, "QUALCOMM", {
{ "QUALCOMM Adreno(TM)", { 32, 8, 2, 1 } },
{ "default", { 32, 8, 2, 1 } },
}
},
{ // Default
kDeviceTypeAll, "default", {
{ "default", { 32, 8, 2, 1 } },
}
},
}
};
} // namespace database
} // namespace clblast