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

8 files changed, 0 insertions, 396 deletions

diff --git a/include/internal/routines/level3/xgemm.h b/include/internal/routines/level3/xgemm.h
deleted file mode 100644
index c0541eef..00000000
--- a/include/internal/routines/level3/xgemm.h
+++ /dev/null
@@ -1,48 +0,0 @@
-
-// =================================================================================================
-// 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 Xgemm routine. The precision is implemented using a template argument.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XGEMM_H_
-#define CLBLAST_ROUTINES_XGEMM_H_
-
-#include "internal/routine.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xgemm: public Routine {
- public:
-
-  // Constructor
-  Xgemm(Queue &queue, EventPointer event, const std::string &name = "GEMM");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoGemm(const Layout layout, const Transpose a_transpose, const Transpose b_transpose,
-                    const size_t m, const size_t n, const size_t k,
-                    const T alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
-                    const T beta,
-                    const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-
- protected:
-  // Static variable to get the precision
-  const static Precision precision_;
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XGEMM_H_
-#endif
diff --git a/include/internal/routines/level3/xhemm.h b/include/internal/routines/level3/xhemm.h
deleted file mode 100644
index e0f35669..00000000
--- a/include/internal/routines/level3/xhemm.h
+++ /dev/null
@@ -1,54 +0,0 @@
-
-// =================================================================================================
-// 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 Xhemm routine. It is based on the generalized matrix multiplication
-// routine (Xgemm). The implementation is very similar to the Xsymm routine.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XHEMM_H_
-#define CLBLAST_ROUTINES_XHEMM_H_
-
-#include "internal/routines/level3/xgemm.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xhemm: public Xgemm<T> {
- public:
-
-  // Uses methods and variables the regular Xgemm routine
-  using Xgemm<T>::precision_;
-  using Xgemm<T>::routine_name_;
-  using Xgemm<T>::queue_;
-  using Xgemm<T>::context_;
-  using Xgemm<T>::device_;
-  using Xgemm<T>::db_;
-  using Xgemm<T>::DoGemm;
-
-  // Constructor
-  Xhemm(Queue &queue, EventPointer event, const std::string &name = "HEMM");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoHemm(const Layout layout, const Side side, const Triangle triangle,
-                    const size_t m, const size_t n,
-                    const T alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
-                    const T beta,
-                    const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XHEMM_H_
-#endif
diff --git a/include/internal/routines/level3/xher2k.h b/include/internal/routines/level3/xher2k.h
deleted file mode 100644
index b7764e18..00000000
--- a/include/internal/routines/level3/xher2k.h
+++ /dev/null
@@ -1,46 +0,0 @@
-
-// =================================================================================================
-// 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 Xher2k routine. The precision is implemented using the template argument
-// 'T', whereas the alpha/beta arguments are of type 'U'. The implementation is very similar to the
-// Xsyr2k routine.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XHER2K_H_
-#define CLBLAST_ROUTINES_XHER2K_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 Xher2k: public Routine {
- public:
-
-  // Constructor
-  Xher2k(Queue &queue, EventPointer event, const std::string &name = "HER2K");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoHer2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose,
-                     const size_t n, const size_t k,
-                     const T alpha,
-                     const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                     const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
-                     const U beta,
-                     const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XHER2K_H_
-#endif
diff --git a/include/internal/routines/level3/xherk.h b/include/internal/routines/level3/xherk.h
deleted file mode 100644
index abcf4c1a..00000000
--- a/include/internal/routines/level3/xherk.h
+++ /dev/null
@@ -1,45 +0,0 @@
-
-// =================================================================================================
-// 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 Xherk routine. The precision is implemented using the template argument
-// 'T', whereas the alpha/beta arguments are of type 'U'. The implementation is very similar to the
-// Xsyrk routine.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XHERK_H_
-#define CLBLAST_ROUTINES_XHERK_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 Xherk: public Routine {
- public:
-
-  // Constructor
-  Xherk(Queue &queue, EventPointer event, const std::string &name = "HERK");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoHerk(const Layout layout, const Triangle triangle, const Transpose a_transpose,
-                    const size_t n, const size_t k,
-                    const U alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const U beta,
-                    const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XHERK_H_
-#endif
diff --git a/include/internal/routines/level3/xsymm.h b/include/internal/routines/level3/xsymm.h
deleted file mode 100644
index 889abfb7..00000000
--- a/include/internal/routines/level3/xsymm.h
+++ /dev/null
@@ -1,56 +0,0 @@
-
-// =================================================================================================
-// 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 Xsymm routine. It is based on the generalized matrix multiplication
-// routine (Xgemm). The Xsymm class inherits from the templated class Xgemm, allowing it to call the
-// "DoGemm" function directly. The "DoSymm" function first preprocesses the symmetric matrix by
-// transforming it into a general matrix, and then calls the regular GEMM code.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XSYMM_H_
-#define CLBLAST_ROUTINES_XSYMM_H_
-
-#include "internal/routines/level3/xgemm.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xsymm: public Xgemm<T> {
- public:
-
-  // Uses methods and variables the regular Xgemm routine
-  using Xgemm<T>::precision_;
-  using Xgemm<T>::routine_name_;
-  using Xgemm<T>::queue_;
-  using Xgemm<T>::context_;
-  using Xgemm<T>::device_;
-  using Xgemm<T>::db_;
-  using Xgemm<T>::DoGemm;
-
-  // Constructor
-  Xsymm(Queue &queue, EventPointer event, const std::string &name = "SYMM");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoSymm(const Layout layout, const Side side, const Triangle triangle,
-                    const size_t m, const size_t n,
-                    const T alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
-                    const T beta,
-                    const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XSYMM_H_
-#endif
diff --git a/include/internal/routines/level3/xsyr2k.h b/include/internal/routines/level3/xsyr2k.h
deleted file mode 100644
index f75c91e5..00000000
--- a/include/internal/routines/level3/xsyr2k.h
+++ /dev/null
@@ -1,46 +0,0 @@
-
-// =================================================================================================
-// 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 Xsyr2k routine. The precision is implemented using a template argument.
-// The implementation is very similar to Xsyrk (see header for details), except for the fact that
-// the main XgemmUpper/XgemmLower kernel is called twice: C = AB^T + C and C = BA^T + C.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XSYR2K_H_
-#define CLBLAST_ROUTINES_XSYR2K_H_
-
-#include "internal/routine.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xsyr2k: public Routine {
- public:
-
-  // Constructor
-  Xsyr2k(Queue &queue, EventPointer event, const std::string &name = "SYR2K");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoSyr2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose,
-                     const size_t n, const size_t k,
-                     const T alpha,
-                     const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                     const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
-                     const T beta,
-                     const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XSYR2K_H_
-#endif
diff --git a/include/internal/routines/level3/xsyrk.h b/include/internal/routines/level3/xsyrk.h
deleted file mode 100644
index 0710fa74..00000000
--- a/include/internal/routines/level3/xsyrk.h
+++ /dev/null
@@ -1,47 +0,0 @@
-
-// =================================================================================================
-// 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 Xsyrk routine. The precision is implemented using a template argument.
-// The implementation is based on the regular Xgemm routine and kernel, but with two main changes:
-// 1) The final unpad(transpose) kernel updates only the upper/lower triangular part.
-// 2) The main Xgemm kernel masks workgroups not contributing to usefull data. This is only for
-//    performance reasons, as the actual masking is done later (see the first point).
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XSYRK_H_
-#define CLBLAST_ROUTINES_XSYRK_H_
-
-#include "internal/routine.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xsyrk: public Routine {
- public:
-
-  // Constructor
-  Xsyrk(Queue &queue, EventPointer event, const std::string &name = "SYRK");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoSyrk(const Layout layout, const Triangle triangle, const Transpose a_transpose,
-                    const size_t n, const size_t k,
-                    const T alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const T beta,
-                    const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XSYRK_H_
-#endif
diff --git a/include/internal/routines/level3/xtrmm.h b/include/internal/routines/level3/xtrmm.h
deleted file mode 100644
index e18ad17a..00000000
--- a/include/internal/routines/level3/xtrmm.h
+++ /dev/null
@@ -1,54 +0,0 @@
-
-// =================================================================================================
-// 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 Xtrmm routine. The implementation is based on first transforming the
-// upper/lower unit/non-unit triangular matrix into a regular matrix and then calling the GEMM
-// routine. Therefore, this class inherits from the Xgemm class.
-//
-// =================================================================================================
-
-#ifndef CLBLAST_ROUTINES_XTRMM_H_
-#define CLBLAST_ROUTINES_XTRMM_H_
-
-#include "internal/routines/level3/xgemm.h"
-
-namespace clblast {
-// =================================================================================================
-
-// See comment at top of file for a description of the class
-template <typename T>
-class Xtrmm: public Xgemm<T> {
- public:
-
-  // Uses methods and variables the regular Xgemm routine
-  using Xgemm<T>::precision_;
-  using Xgemm<T>::routine_name_;
-  using Xgemm<T>::queue_;
-  using Xgemm<T>::context_;
-  using Xgemm<T>::device_;
-  using Xgemm<T>::db_;
-  using Xgemm<T>::DoGemm;
-
-  // Constructor
-  Xtrmm(Queue &queue, EventPointer event, const std::string &name = "TRMM");
-
-  // Templated-precision implementation of the routine
-  StatusCode DoTrmm(const Layout layout, const Side side, const Triangle triangle,
-                    const Transpose a_transpose, const Diagonal diagonal,
-                    const size_t m, const size_t n,
-                    const T alpha,
-                    const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
-                    const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld);
-};
-
-// =================================================================================================
-} // namespace clblast
-
-// CLBLAST_ROUTINES_XTRMM_H_
-#endif