Added support for xHER, xHPR, xSYR, and xSPR routines

4 files changed, 196 insertions, 0 deletions

diff --git a/include/internal/routines/level2/xher.h b/include/internal/routines/level2/xher.h
new file mode 100644
index 00000000..6322265b
--- /dev/null
+++ b/include/internal/routines/level2/xher.h
@@ -0,0 +1,61 @@
+
+// =================================================================================================
+// 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 the Xher routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XHER_H_
+#define CLBLAST_ROUTINES_XHER_H_
+
+#include "internal/routine.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T, typename U>
+class Xher: public Routine<T> {
+ public:
+
+  // Members and methods from the base class
+  using Routine<T>::db_;
+  using Routine<T>::source_string_;
+  using Routine<T>::queue_;
+  using Routine<T>::GetProgramFromCache;
+  using Routine<T>::TestVectorX;
+  using Routine<T>::TestMatrixA;
+  using Routine<T>::TestMatrixAP;
+  using Routine<T>::RunKernel;
+  using Routine<T>::ErrorIn;
+
+  // Constructor
+  Xher(Queue &queue, Event &event, const std::string &name = "HER");
+
+  // Translates alpha of type 'U' into type 'T'
+  T GetAlpha(const U alpha);
+
+  // Templated-precision implementation of the routine
+  StatusCode DoHer(const Layout layout, const Triangle triangle,
+                   const size_t n,
+                   const U alpha,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+                   const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
+                   const bool packed = false);
+
+ private:
+  // Static variable to get the precision
+  const static Precision precision_;
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XHER_H_
+#endif
diff --git a/include/internal/routines/level2/xhpr.h b/include/internal/routines/level2/xhpr.h
new file mode 100644
index 00000000..a0c3cb92
--- /dev/null
+++ b/include/internal/routines/level2/xhpr.h
@@ -0,0 +1,45 @@
+
+// =================================================================================================
+// 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 the Xhpr routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XHPR_H_
+#define CLBLAST_ROUTINES_XHPR_H_
+
+#include "internal/routines/level2/xher.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T, typename U>
+class Xhpr: public Xher<T,U> {
+ public:
+
+  // Uses the regular Xher routine
+  using Xher<T,U>::DoHer;
+
+  // Constructor
+  Xhpr(Queue &queue, Event &event, const std::string &name = "HPR");
+
+  // Templated-precision implementation of the routine
+  StatusCode DoHpr(const Layout layout, const Triangle triangle,
+                   const size_t n,
+                   const U alpha,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+                   const Buffer<T> &ap_buffer, const size_t ap_offset);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XHPR_H_
+#endif
diff --git a/include/internal/routines/level2/xspr.h b/include/internal/routines/level2/xspr.h
new file mode 100644
index 00000000..5b01d2cb
--- /dev/null
+++ b/include/internal/routines/level2/xspr.h
@@ -0,0 +1,45 @@
+
+// =================================================================================================
+// 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 the Xspr routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XSPR_H_
+#define CLBLAST_ROUTINES_XSPR_H_
+
+#include "internal/routines/level2/xher.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xspr: public Xher<T,T> {
+ public:
+
+  // Uses the regular Xher routine
+  using Xher<T,T>::DoHer;
+
+  // Constructor
+  Xspr(Queue &queue, Event &event, const std::string &name = "SPR");
+
+  // Templated-precision implementation of the routine
+  StatusCode DoSpr(const Layout layout, const Triangle triangle,
+                   const size_t n,
+                   const T alpha,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+                   const Buffer<T> &ap_buffer, const size_t ap_offset);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XSPR_H_
+#endif
diff --git a/include/internal/routines/level2/xsyr.h b/include/internal/routines/level2/xsyr.h
new file mode 100644
index 00000000..9704a881
--- /dev/null
+++ b/include/internal/routines/level2/xsyr.h
@@ -0,0 +1,45 @@
+
+// =================================================================================================
+// 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 the Xsyr routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XSYR_H_
+#define CLBLAST_ROUTINES_XSYR_H_
+
+#include "internal/routines/level2/xher.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xsyr: public Xher<T,T> {
+ public:
+
+  // Uses the regular Xher routine
+  using Xher<T,T>::DoHer;
+
+  // Constructor
+  Xsyr(Queue &queue, Event &event, const std::string &name = "SYR");
+
+  // Templated-precision implementation of the routine
+  StatusCode DoSyr(const Layout layout, const Triangle triangle,
+                   const size_t n,
+                   const T alpha,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+                   const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XSYR_H_
+#endif