Moved all headers into the source tree, changed headers to .hpp extension

9 files changed, 1223 insertions, 0 deletions

diff --git a/src/tuning/kernels/copy_fast.cc b/src/tuning/kernels/copy_fast.cc
new file mode 100644
index 00000000..34269bc7
--- /dev/null
+++ b/src/tuning/kernels/copy_fast.cc
@@ -0,0 +1,122 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the copy OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TuneCopy {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "copy"; }
+  static std::string KernelName() { return "CopyMatrixFast"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level3/level3.opencl"
+      #include "../src/kernels/level3/copy_fast.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgM, kArgN, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeY(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "COPY_DIMX", {8, 16, 32});
+    tuner.AddParameter(id, "COPY_DIMY", {8, 16, 32});
+    tuner.AddParameter(id, "COPY_WPT", {1, 2, 4, 8});
+    tuner.AddParameter(id, "COPY_VW", {1, 2, 4, 8});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &, const size_t, const Arguments<T> &) { }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"COPY_DIMX", "COPY_DIMY"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"COPY_VW", "COPY_WPT"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentOutput(b_mat);
+    tuner.AddArgumentInput(alpha_buffer);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 2 * args.m * args.n * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneCopy<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneCopy<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneCopy<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneCopy<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneCopy<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/copy_pad.cc b/src/tuning/kernels/copy_pad.cc
new file mode 100644
index 00000000..1e0dccd3
--- /dev/null
+++ b/src/tuning/kernels/copy_pad.cc
@@ -0,0 +1,130 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the pad OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TunePad {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "pad"; }
+  static std::string KernelName() { return "CopyPadMatrix"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level3/level3.opencl"
+      #include "../src/kernels/level3/copy_pad.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgM, kArgN, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeY(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "PAD_DIMX", {8, 16, 32});
+    tuner.AddParameter(id, "PAD_DIMY", {8, 16, 32});
+    tuner.AddParameter(id, "PAD_WPTX", {1, 2, 4});
+    tuner.AddParameter(id, "PAD_WPTY", {1, 2, 4});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &, const size_t, const Arguments<T> &) { }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"PAD_DIMX", "PAD_DIMY"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"PAD_WPTX", "PAD_WPTY"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentOutput(b_mat);
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(0);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 2 * args.m * args.n * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TunePad<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TunePad<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TunePad<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TunePad<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TunePad<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/transpose_fast.cc b/src/tuning/kernels/transpose_fast.cc
new file mode 100644
index 00000000..7ac19cb6
--- /dev/null
+++ b/src/tuning/kernels/transpose_fast.cc
@@ -0,0 +1,127 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the transpose OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TuneTranspose {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "transpose"; }
+  static std::string KernelName() { return "TransposeMatrixFast"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level3/level3.opencl"
+      #include "../src/kernels/level3/transpose_fast.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgM, kArgN, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeY(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "TRA_DIM", {4, 8, 16, 32, 64});
+    tuner.AddParameter(id, "TRA_WPT", {1, 2, 4, 8, 16});
+    tuner.AddParameter(id, "TRA_PAD", {0, 1});
+    tuner.AddParameter(id, "TRA_SHUFFLE", {0, 1});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &tuner, const size_t id, const Arguments<T> &args) {
+    auto LocalMemorySize = [args] (std::vector<size_t> v) {
+      return ((v[0]*v[1]*(v[0]*v[1]+v[2]))*GetBytes(args.precision));
+    };
+    tuner.SetLocalMemoryUsage(id, LocalMemorySize, {"TRA_DIM", "TRA_WPT", "TRA_PAD"});
+  }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"TRA_DIM", "TRA_DIM"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"TRA_WPT", "TRA_WPT"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentOutput(b_mat);
+    tuner.AddArgumentInput(alpha_buffer);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 2 * args.m * args.n * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneTranspose<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneTranspose<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneTranspose<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneTranspose<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneTranspose<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/transpose_pad.cc b/src/tuning/kernels/transpose_pad.cc
new file mode 100644
index 00000000..63274415
--- /dev/null
+++ b/src/tuning/kernels/transpose_pad.cc
@@ -0,0 +1,134 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the padtranspose OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TunePadTranspose {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "padtranspose"; }
+  static std::string KernelName() { return "TransposePadMatrix"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level3/level3.opencl"
+      #include "../src/kernels/level3/transpose_pad.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgM, kArgN, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeY(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "PADTRA_TILE", {8, 16, 32, 64});
+    tuner.AddParameter(id, "PADTRA_WPT", {1, 2, 4, 8, 16});
+    tuner.AddParameter(id, "PADTRA_PAD", {0, 1});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &tuner, const size_t id, const Arguments<T> &args) {
+    auto LocalMemorySize = [args] (std::vector<size_t> v) {
+      return ((v[0]*v[1]*(v[0]*v[1]+v[2]))*GetBytes(args.precision));
+    };
+    tuner.SetLocalMemoryUsage(id, LocalMemorySize, {"PADTRA_TILE", "PADTRA_WPT", "PADTRA_PAD"});
+  }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"PADTRA_TILE", "PADTRA_TILE"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"PADTRA_WPT", "PADTRA_WPT"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentOutput(b_mat);
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(0);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 2 * args.m * args.n * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TunePadTranspose<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TunePadTranspose<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TunePadTranspose<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TunePadTranspose<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TunePadTranspose<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/xaxpy.cc b/src/tuning/kernels/xaxpy.cc
new file mode 100644
index 00000000..88d12c1f
--- /dev/null
+++ b/src/tuning/kernels/xaxpy.cc
@@ -0,0 +1,125 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the xaxpy OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TuneXaxpy {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "xaxpy"; }
+  static std::string KernelName() { return "XaxpyFast"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level1/level1.opencl"
+      #include "../src/kernels/level1/xaxpy.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgN, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &args) {
+    if (!IsMultiple(args.n, 64)) {
+      throw std::runtime_error("'XaxpyFast' requires 'n' to be a multiple of WGS*WPT*VW");
+    }
+  }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1; } // N/A for this kernel
+  static size_t DefaultN() { return 4096*1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeY(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeA(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeB(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "WGS", {64, 128, 256, 512, 1024, 2048});
+    tuner.AddParameter(id, "WPT", {1, 2, 4, 8});
+    tuner.AddParameter(id, "VW", {1, 2, 4, 8});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &, const size_t, const Arguments<T> &) { }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1}; }
+  static std::vector<size_t> LocalSizeRef() { return {64}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"WGS"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"WPT"},{"VW"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &x_vec, std::vector<T> &y_vec,
+                           std::vector<T> &, std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentInput(x_vec);
+    tuner.AddArgumentOutput(y_vec);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 3 * args.n * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXaxpy<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXaxpy<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXaxpy<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXaxpy<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXaxpy<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/xdot.cc b/src/tuning/kernels/xdot.cc
new file mode 100644
index 00000000..1581e13f
--- /dev/null
+++ b/src/tuning/kernels/xdot.cc
@@ -0,0 +1,137 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the xdot OpenCL kernels. Note that the results are
+// not verified, since the result is not final and depends on the WGS2 parameter.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T, int V>
+class TuneXdot {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "xdot_"+std::to_string(V); }
+  static std::string KernelName() { return (V==1) ? "Xdot" : "XdotEpilogue"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level1/xdot.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgN}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1; } // N/A for this kernel
+  static size_t DefaultN() { return 2*1024*1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeY(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeA(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeB(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &args) { return args.n; } // Worst case
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "WGS"+std::to_string(V), {32, 64, 128, 256, 512, 1024});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &, const size_t, const Arguments<T> &) { }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &) { return (V==1) ? std::vector<size_t>{2*64} : std::vector<size_t>{1}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &) { return (V==1) ? std::vector<size_t>{2*64*64} : std::vector<size_t>{64}; }
+  static std::vector<size_t> LocalSize() { return {1}; }
+  static std::vector<size_t> LocalSizeRef() { return {64}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return (V==1) ? TransformVector{{"WGS1"}} : TransformVector{{"WGS2"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return (V==1) ? TransformVector{{"WGS1"}} : TransformVector{{"WGS2"}}; }
+  static TransformVector DivGlobal() { return {}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &x_vec, std::vector<T> &y_vec,
+                           std::vector<T> &, std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &temp) {
+    if (V == 1) {
+      tuner.AddArgumentScalar(static_cast<int>(args.n));
+      tuner.AddArgumentInput(x_vec);
+      tuner.AddArgumentScalar(0);
+      tuner.AddArgumentScalar(1);
+      tuner.AddArgumentInput(y_vec);
+      tuner.AddArgumentScalar(0);
+      tuner.AddArgumentScalar(1);
+      tuner.AddArgumentInput(temp); // No output checking for the result - size varies
+      tuner.AddArgumentScalar(static_cast<int>(false));
+    }
+    else {
+      tuner.AddArgumentInput(temp);
+      tuner.AddArgumentInput(x_vec); // No output checking for the result - store somewhere
+      tuner.AddArgumentScalar(0);
+    }
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return (V==1) ? (2*args.n + 1) * GetBytes(args.precision) : 1 * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return (V==1) ? "GB/s" : "N/A"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Function to tune a specific variation V (not within the clblast namespace)
+template <int V>
+void StartVariation(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXdot<half, V>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXdot<float, V>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXdot<double, V>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXdot<float2, V>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXdot<double2, V>, double2>(argc, argv); break;
+  }
+}
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  StartVariation<1>(argc, argv);
+  StartVariation<2>(argc, argv);
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/xgemm.cc b/src/tuning/kernels/xgemm.cc
new file mode 100644
index 00000000..4b1efdef
--- /dev/null
+++ b/src/tuning/kernels/xgemm.cc
@@ -0,0 +1,162 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the xgemm OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TuneXgemm {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "xgemm"; }
+  static std::string KernelName() { return "Xgemm"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level3/xgemm_part1.opencl"
+      #include "../src/kernels/level3/xgemm_part2.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() {
+    return {kArgM, kArgN, kArgK, kArgAlpha, kArgBeta, kArgFraction};
+  }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1024; }
+  static double DefaultFraction() { return 2048.0; }
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeY(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.k; }
+  static size_t GetSizeB(const Arguments<T> &args) { return args.n * args.k; }
+  static size_t GetSizeC(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    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});
+  }
+
+  // Sets the constraints
+  static void SetConstraints(cltune::Tuner &tuner, const size_t id) {
+    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]); };
+    // Requirement for unrolling the KWG loop
+    tuner.AddConstraint(id, MultipleOfX, {"KWG", "KWI"});
+    // Required for integer MWI and NWI
+    tuner.AddConstraint(id, MultipleOfXMulY, {"MWG", "MDIMC", "VWM"});
+    tuner.AddConstraint(id, MultipleOfXMulY, {"NWG", "NDIMC", "VWN"});
+    // Required for integer MWIA and NWIB
+    tuner.AddConstraint(id, MultipleOfXMulY, {"MWG", "MDIMA", "VWM"});
+    tuner.AddConstraint(id, MultipleOfXMulY, {"NWG", "NDIMB", "VWN"});
+    // 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 local memory size
+  static void SetLocalMemorySize(cltune::Tuner &tuner, const size_t id, const Arguments<T> &args) {
+    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"});
+  }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"MDIMC", "NDIMC"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {{"MDIMC", "NDIMC"}}; }
+  static TransformVector DivGlobal() { return {{"MWG", "NWG"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &c_mat,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    auto beta_buffer = std::vector<T>{args.beta};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentScalar(static_cast<int>(args.k));
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentInput(beta_buffer);
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentInput(b_mat);
+    tuner.AddArgumentOutput(c_mat);
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return 2 * args.m * args.n * args.k;
+  }
+  static std::string PerformanceUnit() { return "GFLOPS"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXgemm<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXgemm<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXgemm<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXgemm<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXgemm<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/xgemv.cc b/src/tuning/kernels/xgemv.cc
new file mode 100644
index 00000000..d42155ae
--- /dev/null
+++ b/src/tuning/kernels/xgemv.cc
@@ -0,0 +1,156 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the xgemv OpenCL kernels. Three variants are tuned:
+// 1: The full version of the kernel
+// 2: The fast version for non-transposed matrices
+// 3: The fast version for transposed matrices
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T, int V>
+class TuneXgemv {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "xgemv_"+std::to_string(V); }
+  static std::string KernelName() { return (V==1) ? "Xgemv" : ((V==2) ? "XgemvFast" : "XgemvFastRot"); }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level2/xgemv.opencl"
+      #include "../src/kernels/level2/xgemv_fast.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgM, kArgN, kArgAlpha, kArgBeta}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 2048; }
+  static size_t DefaultN() { return 2048; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeY(const Arguments<T> &args) { return args.m; }
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "WGS"+std::to_string(V), {64, 128, 256});
+    tuner.AddParameter(id, "WPT"+std::to_string(V), {1, 2, 4});
+    if (V==2 || V==3) { tuner.AddParameter(id, "VW"+std::to_string(V), {1, 2, 4, 8}); }
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &tuner, const size_t id) {
+    auto MultipleOfX = [] (std::vector<size_t> v) { return IsMultiple(v[0], v[1]); };
+    if (V==2 || V==3) {
+      tuner.AddConstraint(id, MultipleOfX, {"WPT"+std::to_string(V), "VW"+std::to_string(V)});
+    }
+  }
+  static void SetLocalMemorySize(cltune::Tuner &tuner, const size_t id, const Arguments<T> &args) {
+    auto LocalMemorySize = [args] (std::vector<size_t> v) { return v[0]*GetBytes(args.precision); };
+    tuner.SetLocalMemoryUsage(id, LocalMemorySize, {"WGS"+std::to_string(V)});
+  }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1}; }
+  static std::vector<size_t> LocalSizeRef() { return {64}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"WGS"+std::to_string(V)}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"WPT"+std::to_string(V)}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &x_vec, std::vector<T> &y_vec,
+                           std::vector<T> &a_mat, std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    auto beta_buffer = std::vector<T>{args.beta};
+    auto a_rotated = (V==3) ? 1 : 0;
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentInput(beta_buffer);
+    tuner.AddArgumentScalar(static_cast<int>(a_rotated));
+    tuner.AddArgumentInput(a_mat);
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentInput(x_vec);
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentScalar(1);
+    tuner.AddArgumentOutput(y_vec);
+    tuner.AddArgumentScalar(0);
+    tuner.AddArgumentScalar(1);
+    tuner.AddArgumentScalar(0); // Conjugate transpose
+    tuner.AddArgumentScalar(0); // Additional parameter
+    tuner.AddArgumentScalar(0); // Banded 'kl'
+    tuner.AddArgumentScalar(0); // Banded 'ku'
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return (args.m*args.n + 2*args.m + args.n) * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Function to tune a specific variation V (not within the clblast namespace)
+template <int V>
+void StartVariation(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXgemv<half,V>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXgemv<float,V>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXgemv<double,V>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXgemv<float2,V>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXgemv<double2,V>, double2>(argc, argv); break;
+  }
+}
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  StartVariation<1>(argc, argv);
+  StartVariation<2>(argc, argv);
+  StartVariation<3>(argc, argv);
+  return 0;
+}
+
+// =================================================================================================
diff --git a/src/tuning/kernels/xger.cc b/src/tuning/kernels/xger.cc
new file mode 100644
index 00000000..d2590c53
--- /dev/null
+++ b/src/tuning/kernels/xger.cc
@@ -0,0 +1,130 @@
+
+// =================================================================================================
+// 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 uses the CLTune auto-tuner to tune the xger OpenCL kernels.
+//
+// =================================================================================================
+
+#include <string>
+#include <vector>
+
+#include "utilities.hpp"
+#include "tuning/tuning.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TuneXger {
+ public:
+
+  // The representative kernel and the source code
+  static std::string KernelFamily() { return "xger"; }
+  static std::string KernelName() { return "Xger"; }
+  static std::string GetSources() {
+    return
+      #include "../src/kernels/common.opencl"
+      #include "../src/kernels/level2/level2.opencl"
+      #include "../src/kernels/level2/xger.opencl"
+    ;
+  }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() { return {kArgN, kArgM, kArgAlpha}; }
+
+  // Tests for valid arguments
+  static void TestValidArguments(const Arguments<T> &) { }
+
+  // Sets the default values for the arguments
+  static size_t DefaultM() { return 1024; }
+  static size_t DefaultN() { return 1024; }
+  static size_t DefaultK() { return 1; } // N/A for this kernel
+  static double DefaultFraction() { return 1.0; } // N/A for this kernel
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) { return args.m; }
+  static size_t GetSizeY(const Arguments<T> &args) { return args.n; }
+  static size_t GetSizeA(const Arguments<T> &args) { return args.m * args.n; }
+  static size_t GetSizeB(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeC(const Arguments<T> &) { return 1; } // N/A for this kernel
+  static size_t GetSizeTemp(const Arguments<T> &) { return 1; } // N/A for this kernel
+
+  // Sets the tuning parameters and their possible values
+  static void SetParameters(cltune::Tuner &tuner, const size_t id) {
+    tuner.AddParameter(id, "WGS1", {4, 8, 16, 32, 64, 128, 256, 512});
+    tuner.AddParameter(id, "WGS2", {1, 2, 4, 8, 16, 32, 64, 128, 256});
+    tuner.AddParameter(id, "WPT", {1, 2, 4});
+  }
+
+  // Sets the constraints and local memory size
+  static void SetConstraints(cltune::Tuner &, const size_t) { }
+  static void SetLocalMemorySize(cltune::Tuner &, const size_t, const Arguments<T> &) { }
+
+  // Sets the base thread configuration
+  static std::vector<size_t> GlobalSize(const Arguments<T> &args) { return {args.m, args.n}; }
+  static std::vector<size_t> GlobalSizeRef(const Arguments<T> &args) { return GlobalSize(args); }
+  static std::vector<size_t> LocalSize() { return {1, 1}; }
+  static std::vector<size_t> LocalSizeRef() { return {8, 8}; }
+
+  // Transforms the thread configuration based on the parameters
+  using TransformVector = std::vector<std::vector<std::string>>;
+  static TransformVector MulLocal() { return {{"WGS1", "WGS2"}}; }
+  static TransformVector DivLocal() { return {}; }
+  static TransformVector MulGlobal() { return {}; }
+  static TransformVector DivGlobal() { return {{"WPT", "WPT"}}; }
+
+  // Sets the kernel's arguments
+  static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
+                           std::vector<T> &x_vec, std::vector<T> &y_vec,
+                           std::vector<T> &a_mat, std::vector<T> &, std::vector<T> &,
+                           std::vector<T> &) {
+    auto alpha_buffer = std::vector<T>{args.alpha};
+    tuner.AddArgumentScalar(static_cast<int>(args.m));
+    tuner.AddArgumentScalar(static_cast<int>(args.n));
+    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentInput(x_vec);
+    tuner.AddArgumentScalar(0); // x_offset
+    tuner.AddArgumentScalar(1); // x_increment
+    tuner.AddArgumentInput(y_vec);
+    tuner.AddArgumentScalar(0); // y_offset
+    tuner.AddArgumentScalar(1); // y_increment
+    tuner.AddArgumentOutput(a_mat);
+    tuner.AddArgumentScalar(0); // a_offset
+    tuner.AddArgumentScalar(static_cast<int>(args.m)); // a_ld
+    tuner.AddArgumentScalar(0); // a_is_rowmajor
+  }
+
+  // Describes how to compute the performance metrics
+  static size_t GetMetric(const Arguments<T> &args) {
+    return (2*args.m*args.n + args.m + args.n) * GetBytes(args.precision);
+  }
+  static std::string PerformanceUnit() { return "GB/s"; }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXger<half>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXger<float>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXger<double>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXger<float2>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXger<double2>, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================