diff options
Diffstat (limited to 'src/tuning/xgemm.cc')
-rw-r--r-- | src/tuning/xgemm.cc | 126 |
1 files changed, 126 insertions, 0 deletions
diff --git a/src/tuning/xgemm.cc b/src/tuning/xgemm.cc new file mode 100644 index 00000000..aba56810 --- /dev/null +++ b/src/tuning/xgemm.cc @@ -0,0 +1,126 @@ + +// ================================================================================================= +// 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 file implements an auto-tuner to tune the Xgemm OpenCL kernel. It uses the CLTune library. +// Note that this tuner uses random-search: running it multiple times or with a larger fraction +// argument might be neccessary to obtain good results. +// +// ================================================================================================= + +#include <string> +#include <vector> +#include <stdexcept> + +#include "internal/utilities.h" +#include "internal/tuning.h" + +namespace clblast { +// ================================================================================================= + +// The Xgemm auto-tuner +template <typename T> +void XgemmTune(const Arguments<T> &args, + const std::vector<T> &a_mat, const std::vector<T> &b_mat, std::vector<T> &c_mat, + cltune::Tuner &tuner) { + + // This points to the Xgemm kernel as found in the CLBlast library and its golden reference + std::string common_source = + #include "../src/kernels/common.opencl" + std::string kernel_source = + #include "../src/kernels/xgemm.opencl" + auto sources = common_source + kernel_source; + auto id = tuner.AddKernelFromString(sources, "Xgemm", {args.m, args.n}, {1, 1}); + tuner.SetReferenceFromString(sources, "Xgemm", {args.m, args.n}, {8, 8}); + + // Sets the tunable parameters and their possible values + tuner.AddParameter(id, "MWG", {16, 32, 64, 128}); + tuner.AddParameter(id, "NWG", {16, 32, 64, 128}); + tuner.AddParameter(id, "KWG", {16, 32}); + tuner.AddParameter(id, "MDIMC", {8, 16, 32}); + tuner.AddParameter(id, "NDIMC", {8, 16, 32}); + tuner.AddParameter(id, "MDIMA", {8, 16, 32}); + tuner.AddParameter(id, "NDIMB", {8, 16, 32}); + tuner.AddParameter(id, "KWI", {2, 8}); + tuner.AddParameter(id, "VWM", {1, 2, 4, 8}); + tuner.AddParameter(id, "VWN", {1, 2, 4, 8}); + tuner.AddParameter(id, "STRM", {0, 1}); + tuner.AddParameter(id, "STRN", {0, 1}); + tuner.AddParameter(id, "SA", {0, 1}); + tuner.AddParameter(id, "SB", {0, 1}); + + // Tests for a specific precision + tuner.AddParameter(id, "PRECISION", {static_cast<size_t>(args.precision)}); + tuner.AddParameterReference("PRECISION", static_cast<size_t>(args.precision)); + + // Sets the helper functions to implement the constraints below + auto MultipleOfX = [] (std::vector<size_t> v) { return IsMultiple(v[0], v[1]); }; + auto MultipleOfXMulY = [] (std::vector<size_t> v) { return IsMultiple(v[0], v[1]*v[2]); }; + auto MultipleOfXMulYDivZ = [] (std::vector<size_t> v) { return IsMultiple(v[0], (v[1]*v[2])/v[3]); }; + + // Sets constraints: Requirement for unrolling the KWG loop + tuner.AddConstraint(id, MultipleOfX, {"KWG", "KWI"}); + + // Sets constraints: Required for integer MWI and NWI + tuner.AddConstraint(id, MultipleOfXMulY, {"MWG", "MDIMC", "VWM"}); + tuner.AddConstraint(id, MultipleOfXMulY, {"NWG", "NDIMC", "VWN"}); + + // Sets constraints: Required for integer MWIA and NWIB + tuner.AddConstraint(id, MultipleOfXMulY, {"MWG", "MDIMA", "VWM"}); + tuner.AddConstraint(id, MultipleOfXMulY, {"NWG", "NDIMB", "VWN"}); + + // Sets constraints: KWG has to be a multiple of KDIMA = ((MDIMC*NDIMC)/(MDIMA)) and KDIMB = (...) + tuner.AddConstraint(id, MultipleOfXMulYDivZ, {"KWG", "MDIMC", "NDIMC", "MDIMA"}); + tuner.AddConstraint(id, MultipleOfXMulYDivZ, {"KWG", "MDIMC", "NDIMC", "NDIMB"}); + + // Sets the constraints for local memory size limitations + auto LocalMemorySize = [args] (std::vector<size_t> v) { + return (((v[0]*v[1]*v[2]/v[3]) + (v[4]*v[5]*v[6]/v[7]))*GetBytes(args.precision)); + }; + tuner.SetLocalMemoryUsage(id, LocalMemorySize, {"SA", "KWG", "MWG", "VWM", + "SB", "KWG", "NWG", "VWN"}); + + // Modifies the thread-sizes (both global and local) based on the parameters + tuner.MulLocalSize(id, {"MDIMC", "NDIMC"}); + tuner.MulGlobalSize(id, {"MDIMC", "NDIMC"}); + tuner.DivGlobalSize(id, {"MWG", "NWG"}); + + // Sets the function's arguments + tuner.AddArgumentScalar(static_cast<int>(args.m)); + tuner.AddArgumentScalar(static_cast<int>(args.n)); + tuner.AddArgumentScalar(static_cast<int>(args.k)); + tuner.AddArgumentScalar(args.alpha); + tuner.AddArgumentScalar(args.beta); + tuner.AddArgumentInput(a_mat); + tuner.AddArgumentInput(b_mat); + tuner.AddArgumentOutput(c_mat); +} + +// ================================================================================================= + +// Main function which calls the common client code with the routine-specific function as argument. +void TunerXgemm(int argc, char *argv[]) { + switch(GetPrecision(argc, argv)) { + case Precision::kHalf: throw std::runtime_error("Unsupported precision mode"); + case Precision::kSingle: TunerABC<float>(argc, argv, XgemmTune<float>); break; + case Precision::kDouble: TunerABC<double>(argc, argv, XgemmTune<double>); break; + case Precision::kComplexSingle: TunerABC<float2>(argc, argv, XgemmTune<float2>); break; + case Precision::kComplexDouble: TunerABC<double2>(argc, argv, XgemmTune<double2>); break; + } +} + +// ================================================================================================= +} // namespace clblast + +// Main function (not within the clblast namespace) +int main(int argc, char *argv[]) { + clblast::TunerXgemm(argc, argv); + return 0; +} + +// ================================================================================================= |