diff options
Diffstat (limited to 'src/database')
-rw-r--r-- | src/database/database.cpp | 6 | ||||
-rw-r--r-- | src/database/database.hpp | 2 | ||||
-rw-r--r-- | src/database/kernel_selection.hpp | 129 | ||||
-rw-r--r-- | src/database/kernels/xgemm.hpp | 4 | ||||
-rw-r--r-- | src/database/kernels/xgemm_direct.hpp | 136 |
5 files changed, 274 insertions, 3 deletions
diff --git a/src/database/database.cpp b/src/database/database.cpp index 34c44a29..df9ac373 100644 --- a/src/database/database.cpp +++ b/src/database/database.cpp @@ -21,10 +21,12 @@ #include "database/kernels/xgemv_fast_rot.hpp" #include "database/kernels/xger.hpp" #include "database/kernels/xgemm.hpp" +#include "database/kernels/xgemm_direct.hpp" #include "database/kernels/copy.hpp" #include "database/kernels/pad.hpp" #include "database/kernels/transpose.hpp" #include "database/kernels/padtranspose.hpp" +#include "database/kernel_selection.hpp" namespace clblast { // ================================================================================================= @@ -38,10 +40,12 @@ const std::vector<Database::DatabaseEntry> Database::database = { XgemvFastRotHalf, XgemvFastRotSingle, XgemvFastRotDouble, XgemvFastRotComplexSingle, XgemvFastRotComplexDouble, XgerHalf, XgerSingle, XgerDouble, XgerComplexSingle, XgerComplexDouble, XgemmHalf, XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble, + XgemmDirectHalf, XgemmDirectSingle, XgemmDirectDouble, XgemmDirectComplexSingle, XgemmDirectComplexDouble, CopyHalf, CopySingle, CopyDouble, CopyComplexSingle, CopyComplexDouble, PadHalf, PadSingle, PadDouble, PadComplexSingle, PadComplexDouble, TransposeHalf, TransposeSingle, TransposeDouble, TransposeComplexSingle, TransposeComplexDouble, - PadtransposeHalf, PadtransposeSingle, PadtransposeDouble, PadtransposeComplexSingle, PadtransposeComplexDouble + PadtransposeHalf, PadtransposeSingle, PadtransposeDouble, PadtransposeComplexSingle, PadtransposeComplexDouble, + KernelSelectionHalf, KernelSelectionSingle, KernelSelectionDouble, KernelSelectionComplexSingle, KernelSelectionComplexDouble }; // ================================================================================================= diff --git a/src/database/database.hpp b/src/database/database.hpp index a6ab49c5..912f0f15 100644 --- a/src/database/database.hpp +++ b/src/database/database.hpp @@ -75,10 +75,12 @@ class Database { static const DatabaseEntry XgemvFastRotHalf, XgemvFastRotSingle, XgemvFastRotDouble, XgemvFastRotComplexSingle, XgemvFastRotComplexDouble; static const DatabaseEntry XgerHalf, XgerSingle, XgerDouble, XgerComplexSingle, XgerComplexDouble; static const DatabaseEntry XgemmHalf, XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble; + static const DatabaseEntry XgemmDirectHalf, XgemmDirectSingle, XgemmDirectDouble, XgemmDirectComplexSingle, XgemmDirectComplexDouble; static const DatabaseEntry CopyHalf, CopySingle, CopyDouble, CopyComplexSingle, CopyComplexDouble; static const DatabaseEntry PadHalf, PadSingle, PadDouble, PadComplexSingle, PadComplexDouble; static const DatabaseEntry TransposeHalf, TransposeSingle, TransposeDouble, TransposeComplexSingle, TransposeComplexDouble; static const DatabaseEntry PadtransposeHalf, PadtransposeSingle, PadtransposeDouble, PadtransposeComplexSingle, PadtransposeComplexDouble; + static const DatabaseEntry KernelSelectionHalf, KernelSelectionSingle, KernelSelectionDouble, KernelSelectionComplexSingle, KernelSelectionComplexDouble; static const std::vector<DatabaseEntry> database; // The constructor with a user-provided database overlay (potentially an empty vector) diff --git a/src/database/kernel_selection.hpp b/src/database/kernel_selection.hpp new file mode 100644 index 00000000..c9462c7a --- /dev/null +++ b/src/database/kernel_selection.hpp @@ -0,0 +1,129 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren <www.cedricnugteren.nl> +// +// This determines when to switch between the direct (for small sizes) and in-direct GEMM kernel +// with pre/post-processing kernels (for larger sizes). These can be set in a similar way as for the +// regular kernel tuning parameters: they can be specific for a certain vendor or device or can use +// some common default values. +// +// ================================================================================================= + +namespace clblast { +// ================================================================================================= + +const Database::DatabaseEntry Database::KernelSelectionHalf = { + "KernelSelection", Precision::kHalf, { + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",384*384*384} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",768*768*768} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",512*512*512} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::KernelSelectionSingle = { + "KernelSelection", Precision::kSingle, { + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",384*384*384} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",768*768*768} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",512*512*512} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::KernelSelectionComplexSingle = { + "KernelSelection", Precision::kComplexSingle, { + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",384*384*384} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",768*768*768} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",512*512*512} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::KernelSelectionDouble = { + "KernelSelection", Precision::kDouble, { + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",384*384*384} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",768*768*768} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",512*512*512} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::KernelSelectionComplexDouble = { + "KernelSelection", Precision::kComplexDouble, { + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",384*384*384} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",768*768*768} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"XGEMM_MIN_INDIRECT_SIZE",512*512*512} } }, + } + }, + } +}; + +// ================================================================================================= +} // namespace clblast diff --git a/src/database/kernels/xgemm.hpp b/src/database/kernels/xgemm.hpp index d19c55b5..e289c542 100644 --- a/src/database/kernels/xgemm.hpp +++ b/src/database/kernels/xgemm.hpp @@ -59,8 +59,8 @@ const Database::DatabaseEntry Database::XgemmSingle = { { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"KWG",16}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",8}, {"NDIMC",16}, {"NWG",128}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",1}, {"VWM",2}, {"VWN",4} } }, { "Intel(R) HD Graphics Skylake ULT GT2", { {"KWG",32}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",16}, {"NWG",128}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",1}, {"VWM",1}, {"VWN",8} } }, { "Iris", { {"KWG",16}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",128}, {"NDIMB",32}, {"NDIMC",16}, {"NWG",64}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",1}, {"VWM",4}, {"VWN",1} } }, - { "Iris Pro", { {"KWG",32}, {"KWI",8}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",1}, {"SB",0}, {"STRM",1}, {"STRN",0}, {"VWM",4}, {"VWN",4} } }, - { "default", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, + { "Iris Pro", { {"KWG",16}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",8}, {"MWG",64}, {"NDIMB",32}, {"NDIMC",32}, {"NWG",128}, {"SA",1}, {"SB",1}, {"STRM",1}, {"STRN",0}, {"VWM",4}, {"VWN",4} } }, + { "default", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",8}, {"NDIMC",16}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } }, } }, { // Intel accelerators diff --git a/src/database/kernels/xgemm_direct.hpp b/src/database/kernels/xgemm_direct.hpp new file mode 100644 index 00000000..202deb1f --- /dev/null +++ b/src/database/kernels/xgemm_direct.hpp @@ -0,0 +1,136 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Database generator <database.py> +// +// This file populates the database with best-found tuning parameters for the 'Xgemm_Direct' kernels. +// +// ================================================================================================= + +namespace clblast { +// ================================================================================================= + +const Database::DatabaseEntry Database::XgemmDirectHalf = { + "XgemmDirect", Precision::kHalf, { + { // Default + kDeviceTypeAll, "default", { + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",4}, {"WGD",32} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::XgemmDirectSingle = { + "XgemmDirect", Precision::kSingle, { + { // AMD GPUs + kDeviceTypeGPU, "AMD", { + { "AMD Radeon R9 M370X Compute Engine", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + } + }, + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "Iris Pro", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",4}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",4}, {"WGD",32} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",4}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",4}, {"WGD",32} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",4}, {"VWND",4}, {"WGD",32} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::XgemmDirectComplexSingle = { + "XgemmDirect", Precision::kComplexSingle, { + { // AMD GPUs + kDeviceTypeGPU, "AMD", { + { "AMD Radeon R9 M370X Compute Engine", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + } + }, + { // Intel GPUs + kDeviceTypeGPU, "Intel", { + { "Iris Pro", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",2}, {"WGD",16} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::XgemmDirectDouble = { + "XgemmDirect", Precision::kDouble, { + { // AMD GPUs + kDeviceTypeGPU, "AMD", { + { "AMD Radeon R9 M370X Compute Engine", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",8}, {"MDIMCD",8}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",2}, {"VWND",2}, {"WGD",32} } }, + } + }, + { // Default + kDeviceTypeAll, "default", { + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + } + }, + } +}; + +// ================================================================================================= + +const Database::DatabaseEntry Database::XgemmDirectComplexDouble = { + "XgemmDirect", Precision::kComplexDouble, { + { // AMD GPUs + kDeviceTypeGPU, "AMD", { + { "AMD Radeon R9 M370X Compute Engine", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + { "default", { {"KWID",2}, {"MDIMAD",16}, {"MDIMCD",16}, {"NDIMBD",16}, {"NDIMCD",16}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",16} } }, + } + }, + { // NVIDIA GPUs + kDeviceTypeGPU, "NVIDIA", { + { "GeForce GTX 750 Ti", { {"KWID",2}, {"MDIMAD",32}, {"MDIMCD",32}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"WGD",32} } }, + { "default", { {"KWID",2}, {"MDIMAD",32}, {"MDIMCD",32}, {"NDIMBD",8}, {"NDIMCD",8}, {"PADA",1}, {"PADB",1}, {"VWMD",1}, {"VWND",1}, {"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",16} } }, + } + }, + } +}; + +// ================================================================================================= +} // namespace clblast |