Added the HBMV routine

5 files changed, 76 insertions, 12 deletions

diff --git a/src/routines/level2/xgbmv.cc b/src/routines/level2/xgbmv.cc
index 8657c254..d8815542 100644
--- a/src/routines/level2/xgbmv.cc
+++ b/src/routines/level2/xgbmv.cc
@@ -52,7 +52,7 @@ StatusCode Xgbmv<T>::DoGbmv(const Layout layout, const Transpose a_transpose,
                 x_buffer, x_offset, x_inc, beta,
                 y_buffer, y_offset, y_inc,
                 fast_kernels, fast_kernels,
-                false, kl_real, ku_real);
+                0, kl_real, ku_real);
 }
 
 // =================================================================================================
diff --git a/src/routines/level2/xgemv.cc b/src/routines/level2/xgemv.cc
index 6e2303c0..0c312756 100644
--- a/src/routines/level2/xgemv.cc
+++ b/src/routines/level2/xgemv.cc
@@ -55,7 +55,7 @@ StatusCode Xgemv<T>::DoGemv(const Layout layout, const Transpose a_transpose,
                 x_buffer, x_offset, x_inc, beta,
                 y_buffer, y_offset, y_inc,
                 true, true,
-                false, 0, 0); // N/A for this routine
+                0, 0, 0); // N/A for this routine
 }
 
 // =================================================================================================
@@ -69,7 +69,7 @@ StatusCode Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
                             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,
-                            bool fast_kernel, bool fast_kernel_rot, bool reversed,
+                            bool fast_kernel, bool fast_kernel_rot, const size_t parameter,
                             const size_t kl, const size_t ku) {
 
   // Makes sure all dimensions are larger than zero
@@ -151,9 +151,9 @@ StatusCode Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
     kernel.SetArgument(12, static_cast<int>(y_offset));
     kernel.SetArgument(13, static_cast<int>(y_inc));
     kernel.SetArgument(14, static_cast<int>(a_conjugate));
-    kernel.SetArgument(15, static_cast<int>(reversed)); // only used for SYMV/HEMV routines
-    kernel.SetArgument(16, static_cast<int>(kl)); // only used for GBMV routines
-    kernel.SetArgument(17, static_cast<int>(ku)); // only used for GBMV routines
+    kernel.SetArgument(15, static_cast<int>(parameter)); // extra parameter used for symm/herm
+    kernel.SetArgument(16, static_cast<int>(kl)); // only used for banded matrices
+    kernel.SetArgument(17, static_cast<int>(ku)); // only used for banded matrices
 
     // Launches the kernel
     auto global = std::vector<size_t>{global_size};
diff --git a/src/routines/level2/xhbmv.cc b/src/routines/level2/xhbmv.cc
new file mode 100644
index 00000000..90c0c787
--- /dev/null
+++ b/src/routines/level2/xhbmv.cc
@@ -0,0 +1,64 @@
+
+// =================================================================================================
+// 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 Xhbmv class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xhbmv.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xhbmv<T>::Xhbmv(Queue &queue, Event &event, const std::string &name):
+    Xgemv<T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xhbmv<T>::DoHbmv(const Layout layout, const Triangle triangle,
+                            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> &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) {
+
+  // The data is either in the upper or lower triangle
+  size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
+                     (triangle == Triangle::kLower && layout == Layout::kRowMajor));
+
+  // Runs the generic matrix-vector multiplication, disabling the use of fast vectorized kernels.
+  // The specific hermitian banded matrix-accesses are implemented in the kernel guarded by the
+  // ROUTINE_HBMV define.
+  bool fast_kernels = false;
+  return MatVec(layout, Transpose::kNo,
+                n, n, alpha,
+                a_buffer, a_offset, a_ld,
+                x_buffer, x_offset, x_inc, beta,
+                y_buffer, y_offset, y_inc,
+                fast_kernels, fast_kernels,
+                is_upper, k, 0);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xhbmv<float2>;
+template class Xhbmv<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xhemv.cc b/src/routines/level2/xhemv.cc
index 917bf9b6..efad829b 100644
--- a/src/routines/level2/xhemv.cc
+++ b/src/routines/level2/xhemv.cc
@@ -38,8 +38,8 @@ StatusCode Xhemv<T>::DoHemv(const Layout layout, const Triangle triangle,
                             const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) {
 
   // The data is either in the upper or lower triangle
-  bool reversed = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
-                   (triangle == Triangle::kLower && layout == Layout::kRowMajor));
+  size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
+                     (triangle == Triangle::kLower && layout == Layout::kRowMajor));
 
   // Runs the generic matrix-vector multiplication, disabling the use of fast vectorized kernels.
   // The specific hermitian matrix-accesses are implemented in the kernel guarded by the
@@ -51,7 +51,7 @@ StatusCode Xhemv<T>::DoHemv(const Layout layout, const Triangle triangle,
                 x_buffer, x_offset, x_inc, beta,
                 y_buffer, y_offset, y_inc,
                 fast_kernels, fast_kernels,
-                reversed, 0, 0);
+                is_upper, 0, 0);
 }
 
 // =================================================================================================
diff --git a/src/routines/level2/xsymv.cc b/src/routines/level2/xsymv.cc
index 15c91f47..458d7385 100644
--- a/src/routines/level2/xsymv.cc
+++ b/src/routines/level2/xsymv.cc
@@ -38,8 +38,8 @@ StatusCode Xsymv<T>::DoSymv(const Layout layout, const Triangle triangle,
                             const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) {
 
   // The data is either in the upper or lower triangle
-  bool reversed = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
-                   (triangle == Triangle::kLower && layout == Layout::kRowMajor));
+  size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
+                     (triangle == Triangle::kLower && layout == Layout::kRowMajor));
 
   // Runs the generic matrix-vector multiplication, disabling the use of fast vectorized kernels.
   // The specific symmetric matrix-accesses are implemented in the kernel guarded by the
@@ -51,7 +51,7 @@ StatusCode Xsymv<T>::DoSymv(const Layout layout, const Triangle triangle,
                 x_buffer, x_offset, x_inc, beta,
                 y_buffer, y_offset, y_inc,
                 fast_kernels, fast_kernels,
-                reversed, 0, 0);
+                is_upper, 0, 0);
 }
 
 // =================================================================================================