Added SYMV routine

3 files changed, 144 insertions, 1 deletions

diff --git a/src/clblast.cc b/src/clblast.cc
index eddb8022..c04462f3 100644
--- a/src/clblast.cc
+++ b/src/clblast.cc
@@ -22,6 +22,7 @@
 
 // BLAS level-2 includes
 #include "internal/routines/level2/xgemv.h"
+#include "internal/routines/level2/xsymv.h"
 
 // BLAS level-3 includes
 #include "internal/routines/level3/xgemm.h"
@@ -36,6 +37,7 @@
 namespace clblast {
 // =================================================================================================
 // BLAS level-1 (vector-vector) routines
+// =================================================================================================
 
 // AXPY
 template <typename T>
@@ -75,6 +77,7 @@ template StatusCode Axpy<double2>(const size_t, const double2,
 
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
+// =================================================================================================
 
 // GEMV
 template <typename T>
@@ -125,7 +128,46 @@ template StatusCode Gemv<double2>(const Layout, const Transpose,
                                   cl_command_queue*, cl_event*);
 
 // =================================================================================================
+
+// SYMV
+template <typename T>
+StatusCode Symv(const Layout layout, const Triangle triangle,
+                const size_t n, const T alpha,
+                const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                const cl_mem x_buffer, const size_t x_offset, const size_t x_inc, const T beta,
+                cl_mem y_buffer, const size_t y_offset, const size_t y_inc,
+                cl_command_queue* queue, cl_event* event) {
+
+  auto queue_cpp = Queue(*queue);
+  auto event_cpp = Event(*event);
+  auto routine = Xsymv<T>(queue_cpp, event_cpp);
+
+  // Compiles the routine's device kernels
+  auto status = routine.SetUp();
+  if (status != StatusCode::kSuccess) { return status; }
+
+  // Runs the routine
+  return routine.DoSymv(layout, triangle, n, alpha,
+                        Buffer<T>(a_buffer), a_offset, a_ld,
+                        Buffer<T>(x_buffer), x_offset, x_inc, beta,
+                        Buffer<T>(y_buffer), y_offset, y_inc);
+}
+template StatusCode Symv<float>(const Layout, const Triangle,
+                                const size_t, const float,
+                                const cl_mem, const size_t, const size_t,
+                                const cl_mem, const size_t, const size_t, const float,
+                                cl_mem, const size_t, const size_t,
+                                cl_command_queue*, cl_event*);
+template StatusCode Symv<double>(const Layout, const Triangle,
+                                 const size_t, const double,
+                                 const cl_mem, const size_t, const size_t,
+                                 const cl_mem, const size_t, const size_t, const double,
+                                 cl_mem, const size_t, const size_t,
+                                 cl_command_queue*, cl_event*);
+
+// =================================================================================================
 // BLAS level-3 (matrix-matrix) routines
+// =================================================================================================
 
 // GEMM
 template <typename T>
diff --git a/src/routines/level2/xgemv.cc b/src/routines/level2/xgemv.cc
index 75219b63..7505e9f7 100644
--- a/src/routines/level2/xgemv.cc
+++ b/src/routines/level2/xgemv.cc
@@ -30,8 +30,9 @@ template <> const Precision Xgemv<double2>::precision_ = Precision::kComplexDoub
 // Constructor: forwards to base class constructor
 template <typename T>
 Xgemv<T>::Xgemv(Queue &queue, Event &event):
-    Routine<T>(queue, event, "GEMV", {"Xgemv"}, precision_) {
+    Routine<T>(queue, event, "GEMV", {"Pad", "Xgemv"}, precision_) {
   source_string_ =
+    #include "../../kernels/pad.opencl" // For SymmUpperToSquared and SymmLowerToSquared (for SYMV)
     #include "../../kernels/xgemv.opencl"
   ;
 }
diff --git a/src/routines/level2/xsymv.cc b/src/routines/level2/xsymv.cc
new file mode 100644
index 00000000..1b0ab6d7
--- /dev/null
+++ b/src/routines/level2/xsymv.cc
@@ -0,0 +1,100 @@
+
+// =================================================================================================
+// 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 Xsymv class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xsymv.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xsymv<T>::Xsymv(Queue &queue, Event &event):
+    Xgemv<T>(queue, event) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xsymv<T>::DoSymv(const Layout layout, const Triangle triangle,
+                            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> &x_buffer, const size_t x_offset, const size_t x_inc,
+                            const T beta,
+                            const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) {
+
+  // Makes sure all dimensions are larger than zero
+  if (n == 0) { return StatusCode::kInvalidDimension; }
+
+  // Checks for validity of the squared A matrix
+  auto status = TestMatrixA(n, n, a_buffer, a_offset, a_ld, sizeof(T));
+  if (ErrorIn(status)) { return status; }
+
+  // Determines which kernel to run based on the layout (the Xgemv kernel assumes column-major as
+  // default) and on whether we are dealing with an upper or lower triangle of the symmetric matrix
+  bool is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
+                   (triangle == Triangle::kLower && layout == Layout::kRowMajor));
+  auto kernel_name = (is_upper) ? "SymmUpperToSquared" : "SymmLowerToSquared";
+
+  // Temporary buffer for a copy of the symmetric matrix
+  try {
+    auto temp_symm = Buffer<T>(context_, n*n);
+
+    // Creates a general matrix from the symmetric matrix to be able to run the regular Xgemv
+    // routine afterwards
+    try {
+      auto& program = GetProgramFromCache();
+      auto kernel = Kernel(program, kernel_name);
+
+      // Sets the arguments for the symmetric-to-squared kernel
+      kernel.SetArgument(0, static_cast<int>(n));
+      kernel.SetArgument(1, static_cast<int>(a_ld));
+      kernel.SetArgument(2, static_cast<int>(a_offset));
+      kernel.SetArgument(3, a_buffer());
+      kernel.SetArgument(4, static_cast<int>(n));
+      kernel.SetArgument(5, static_cast<int>(n));
+      kernel.SetArgument(6, static_cast<int>(0));
+      kernel.SetArgument(7, temp_symm());
+
+      // Uses the common padding kernel's thread configuration. This is allowed, since the
+      // symmetric-to-squared kernel uses the same parameters.
+      auto global = std::vector<size_t>{Ceil(CeilDiv(n, db_["PAD_WPTX"]), db_["PAD_DIMX"]),
+                                        Ceil(CeilDiv(n, db_["PAD_WPTY"]), db_["PAD_DIMY"])};
+      auto local = std::vector<size_t>{db_["PAD_DIMX"], db_["PAD_DIMY"]};
+      status = RunKernel(kernel, global, local);
+      if (ErrorIn(status)) { return status; }
+
+      // Runs the regular Xgemv code
+      status = DoGemv(layout, Transpose::kNo, n, n, alpha,
+                      temp_symm, 0, n,
+                      x_buffer, x_offset, x_inc, beta,
+                      y_buffer, y_offset, y_inc);
+
+      // Return the status of the Xgemv routine
+      return status;
+    } catch (...) { return StatusCode::kInvalidKernel; }
+  } catch (...) { return StatusCode::kTempBufferAllocFailure; }
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xsymv<float>;
+template class Xsymv<double>;
+
+// =================================================================================================
+} // namespace clblast