Added initial naive version of Xgemv kernel

1 files changed, 89 insertions, 0 deletions

diff --git a/src/tuning/xgemv.cc b/src/tuning/xgemv.cc
new file mode 100644
index 00000000..1ee7c7bf
--- /dev/null
+++ b/src/tuning/xgemv.cc
@@ -0,0 +1,89 @@
+
+// =================================================================================================
+// 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 Xgemv OpenCL kernel. It uses the CLTune library.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+#include <stdexcept>
+
+#include "internal/utilities.h"
+#include "internal/tuning.h"
+
+namespace clblast {
+// =================================================================================================
+
+// The Xgemv auto-tuner
+template <typename T>
+void XgemvTune(const Arguments<T> &args,
+               const std::vector<T> &a_mat, const std::vector<T> &x_vec, std::vector<T> &y_vec,
+               cltune::Tuner &tuner) {
+
+  // This points to the Xgemv kernel as found in the CLBlast library
+  std::string common_source =
+  #include "../src/kernels/common.opencl"
+  std::string kernel_source =
+  #include "../src/kernels/xgemv.opencl"
+  auto sources = common_source + kernel_source;
+  auto id = tuner.AddKernelFromString(sources, "Xgemv", {args.m}, {1});
+  tuner.SetReferenceFromString(sources, "Xgemv", {args.m}, {64});
+
+  // Sets the tunable parameters and their possible values
+  tuner.AddParameter(id, "WGS", {64, 128});
+  tuner.AddParameter(id, "WPT", {1});
+  tuner.AddParameter(id, "VW", {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));
+
+  // Modifies the thread-sizes (local) based on the parameters
+  tuner.MulLocalSize(id, {"WGS"});
+  tuner.DivGlobalSize(id, {"WPT"});
+  tuner.DivGlobalSize(id, {"VW"});
+
+  // Sets the function's arguments
+  tuner.AddArgumentScalar(static_cast<int>(args.m));
+  tuner.AddArgumentScalar(static_cast<int>(args.n));
+  tuner.AddArgumentScalar(args.alpha);
+  tuner.AddArgumentScalar(args.beta);
+  tuner.AddArgumentInput(a_mat);
+  tuner.AddArgumentInput(x_vec);
+  tuner.AddArgumentScalar(0);
+  tuner.AddArgumentScalar(1);
+  tuner.AddArgumentOutput(y_vec);
+  tuner.AddArgumentScalar(0);
+  tuner.AddArgumentScalar(1);
+}
+
+// =================================================================================================
+
+// Main function which calls the common client code with the routine-specific function as argument.
+void TunerXgemv(int argc, char *argv[]) {
+  switch(GetPrecision(argc, argv)) {
+    case Precision::kHalf: throw std::runtime_error("Unsupported precision mode");
+    case Precision::kSingle: TunerAXY<float>(argc, argv, XgemvTune<float>); break;
+    case Precision::kDouble: TunerAXY<double>(argc, argv, XgemvTune<double>); break;
+    case Precision::kComplexSingle: TunerAXY<float2>(argc, argv, XgemvTune<float2>); break;
+    case Precision::kComplexDouble: TunerAXY<double2>(argc, argv, XgemvTune<double2>); break;
+  }
+}
+
+// =================================================================================================
+} // namespace clblast
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  clblast::TunerXgemv(argc, argv);
+  return 0;
+}
+
+// =================================================================================================