Added non-absolute counter-parts xSUM and IxMAX of the BLAS routines xASUM and IxAMAX

4 files changed, 107 insertions, 9 deletions

diff --git a/include/clblast.h b/include/clblast.h
index f3b74f6e..57fca119 100644
--- a/include/clblast.h
+++ b/include/clblast.h
@@ -188,10 +188,10 @@ StatusCode Asum(const size_t n,
                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                 cl_command_queue* queue, cl_event* event = nullptr);
 
-// Sum of values in a vector: SSUM/DSUM/ScSUM/DzSUM
+// Sum of values in a vector (non-BLAS function): SSUM/DSUM/ScSUM/DzSUM
 template <typename T>
 StatusCode Sum(const size_t n,
-               cl_mem asum_buffer, const size_t asum_offset,
+               cl_mem sum_buffer, const size_t sum_offset,
                const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                cl_command_queue* queue, cl_event* event = nullptr);
 
@@ -202,7 +202,7 @@ StatusCode Amax(const size_t n,
                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                 cl_command_queue* queue, cl_event* event = nullptr);
 
-// Index of maximum value in a vector: iSMAX/iDMAX/iCMAX/iZMAX
+// Index of maximum value in a vector (non-BLAS function): iSMAX/iDMAX/iCMAX/iZMAX
 template <typename T>
 StatusCode Max(const size_t n,
                cl_mem imax_buffer, const size_t imax_offset,
diff --git a/include/clblast_c.h b/include/clblast_c.h
index 2f692b66..e23f0305 100644
--- a/include/clblast_c.h
+++ b/include/clblast_c.h
@@ -296,21 +296,21 @@ StatusCode PUBLIC_API CLBlastDzasum(const size_t n,
                                    const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                    cl_command_queue* queue, cl_event* event);
 
-// Sum of values in a vector: SSUM/DSUM/ScSUM/DzSUM
+// Sum of values in a vector (non-BLAS function): SSUM/DSUM/ScSUM/DzSUM
 StatusCode PUBLIC_API CLBlastSsum(const size_t n,
-                                  cl_mem asum_buffer, const size_t asum_offset,
+                                  cl_mem sum_buffer, const size_t sum_offset,
                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                   cl_command_queue* queue, cl_event* event);
 StatusCode PUBLIC_API CLBlastDsum(const size_t n,
-                                  cl_mem asum_buffer, const size_t asum_offset,
+                                  cl_mem sum_buffer, const size_t sum_offset,
                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                   cl_command_queue* queue, cl_event* event);
 StatusCode PUBLIC_API CLBlastScsum(const size_t n,
-                                  cl_mem asum_buffer, const size_t asum_offset,
+                                  cl_mem sum_buffer, const size_t sum_offset,
                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                   cl_command_queue* queue, cl_event* event);
 StatusCode PUBLIC_API CLBlastDzsum(const size_t n,
-                                  cl_mem asum_buffer, const size_t asum_offset,
+                                  cl_mem sum_buffer, const size_t sum_offset,
                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                   cl_command_queue* queue, cl_event* event);
 
@@ -332,7 +332,7 @@ StatusCode PUBLIC_API CLBlastiZamax(const size_t n,
                                    const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                    cl_command_queue* queue, cl_event* event);
 
-// Index of maximum value in a vector: iSMAX/iDMAX/iCMAX/iZMAX
+// Index of maximum value in a vector (non-BLAS function): iSMAX/iDMAX/iCMAX/iZMAX
 StatusCode PUBLIC_API CLBlastiSmax(const size_t n,
                                   cl_mem imax_buffer, const size_t imax_offset,
                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
diff --git a/include/internal/routines/level1/xmax.h b/include/internal/routines/level1/xmax.h
new file mode 100644
index 00000000..860a043b
--- /dev/null
+++ b/include/internal/routines/level1/xmax.h
@@ -0,0 +1,49 @@
+
+// =================================================================================================
+// 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 Xmax routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XMAX_H_
+#define CLBLAST_ROUTINES_XMAX_H_
+
+#include "internal/routine.h"
+#include "internal/routines/level1/xamax.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xmax: public Xamax<T> {
+ public:
+
+  // Members and methods from the base class
+  using Xamax<T>::DoAmax;
+
+  // Constructor
+  Xmax(Queue &queue, EventPointer event, const std::string &name = "MAX"):
+    Xamax<T>(queue, event, name) {
+  }
+
+  // Forwards to the regular absolute version. The implementation difference is realised in the
+  // kernel through a pre-processor macro based on the name of the routine.
+  StatusCode DoMax(const size_t n,
+                   const Buffer<T> &imax_buffer, const size_t imax_offset,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) {
+    return DoAmax(n, imax_buffer, imax_offset, x_buffer, x_offset, x_inc);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XMAX_H_
+#endif
diff --git a/include/internal/routines/level1/xsum.h b/include/internal/routines/level1/xsum.h
new file mode 100644
index 00000000..2f633b52
--- /dev/null
+++ b/include/internal/routines/level1/xsum.h
@@ -0,0 +1,49 @@
+
+// =================================================================================================
+// 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 Xsum routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XSUM_H_
+#define CLBLAST_ROUTINES_XSUM_H_
+
+#include "internal/routine.h"
+#include "internal/routines/level1/xasum.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xsum: public Xasum<T> {
+ public:
+
+  // Members and methods from the base class
+  using Xasum<T>::DoAsum;
+
+  // Constructor
+  Xsum(Queue &queue, EventPointer event, const std::string &name = "SUM"):
+    Xasum<T>(queue, event, name) {
+  }
+
+  // Forwards to the regular absolute version. The implementation difference is realised in the
+  // kernel through a pre-processor macro based on the name of the routine.
+  StatusCode DoSum(const size_t n,
+                   const Buffer<T> &sum_buffer, const size_t sum_offset,
+                   const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) {
+    return DoAsum(n, sum_buffer, sum_offset, x_buffer, x_offset, x_inc);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XSUM_H_
+#endif