Added SBMV and SPMV routines

4 files changed, 206 insertions, 43 deletions

diff --git a/src/clblast.cc b/src/clblast.cc
index 85c63442..f7baf5e8 100644
--- a/src/clblast.cc
+++ b/src/clblast.cc
@@ -33,6 +33,8 @@
 #include "internal/routines/level2/xhbmv.h"
 #include "internal/routines/level2/xhpmv.h"
 #include "internal/routines/level2/xsymv.h"
+#include "internal/routines/level2/xsbmv.h"
+#include "internal/routines/level2/xspmv.h"
 
 // BLAS level-3 includes
 #include "internal/routines/level3/xgemm.h"
@@ -546,15 +548,26 @@ template StatusCode Symv<double>(const Layout, const Triangle,
 
 // Symmetric banded matrix-vector multiplication: SSBMV/DSBMV
 template <typename T>
-StatusCode Sbmv(const Layout, const Triangle,
-                const size_t, const size_t,
-                const T,
-                const cl_mem, const size_t, const size_t,
-                const cl_mem, const size_t, const size_t,
-                const T,
-                cl_mem, const size_t, const size_t,
-                cl_command_queue*, cl_event*) {
-  return StatusCode::kNotImplemented;
+StatusCode Sbmv(const Layout layout, const Triangle triangle,
+                const size_t n, const size_t k,
+                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 = Xsbmv<T>(queue_cpp, event_cpp);
+  auto status = routine.SetUp();
+  if (status != StatusCode::kSuccess) { return status; }
+  return routine.DoSbmv(layout, triangle,
+                        n, k,
+                        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 Sbmv<float>(const Layout, const Triangle,
                                 const size_t, const size_t,
@@ -575,15 +588,26 @@ template StatusCode Sbmv<double>(const Layout, const Triangle,
 
 // Symmetric packed matrix-vector multiplication: SSPMV/DSPMV
 template <typename T>
-StatusCode Spmv(const Layout, const Triangle,
-                const size_t,
-                const T,
-                const cl_mem, const size_t,
-                const cl_mem, const size_t, const size_t,
-                const T,
-                cl_mem, const size_t, const size_t,
-                cl_command_queue*, cl_event*) {
-  return StatusCode::kNotImplemented;
+StatusCode Spmv(const Layout layout, const Triangle triangle,
+                const size_t n,
+                const T alpha,
+                const cl_mem ap_buffer, const size_t ap_offset,
+                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 = Xspmv<T>(queue_cpp, event_cpp);
+  auto status = routine.SetUp();
+  if (status != StatusCode::kSuccess) { return status; }
+  return routine.DoSpmv(layout, triangle,
+                        n,
+                        alpha,
+                        Buffer<T>(ap_buffer), ap_offset,
+                        Buffer<T>(x_buffer), x_offset, x_inc,
+                        beta,
+                        Buffer<T>(y_buffer), y_offset, y_inc);
 }
 template StatusCode Spmv<float>(const Layout, const Triangle,
                                 const size_t,
diff --git a/src/kernels/level2/xgemv.opencl b/src/kernels/level2/xgemv.opencl
index f6c4476e..ab7802e5 100644
--- a/src/kernels/level2/xgemv.opencl
+++ b/src/kernels/level2/xgemv.opencl
@@ -92,31 +92,39 @@ inline real LoadMatrixA(const __global real* restrict agm, const int x, const in
     if (x >= y-ku && x < y+kl+1) { result = agm[a_ld*y + k + x + a_offset]; }
     else { SetToZero(result); }
 
-  // For hermitian matrices
-  #elif defined(ROUTINE_HEMV)
+  // For symmetric/hermitian matrices
+  #elif defined(ROUTINE_HEMV) || defined(ROUTINE_SYMV)
     if ((parameter == 0 && y <= x) || (parameter == 1 && x <= y)) {
       result = agm[a_ld*y + x + a_offset];
-      if (x == y) { result.y = ZERO; }
+      #if defined(ROUTINE_HEMV)
+        if (x == y) { result.y = ZERO; }
+      #endif
     }
     else {
       result = agm[a_ld*x + y + a_offset];
-      COMPLEX_CONJUGATE(result);
+      #if defined(ROUTINE_HEMV)
+        COMPLEX_CONJUGATE(result);
+      #endif
     }
 
-  // For hermitian banded matrices
-  #elif defined(ROUTINE_HBMV)
+  // For symmetric/hermitian banded matrices
+  #elif defined(ROUTINE_HBMV) || defined(ROUTINE_SBMV)
     if (parameter == 1) {
       if (x <= y) {
         const int m = kl - y;
         if (x >= y-kl && x <= y) { result = agm[a_ld*y + m + x + a_offset]; }
         else { SetToZero(result); }
-        if (x == y) { result.y = ZERO; }
+        #if defined(ROUTINE_HBMV)
+          if (x == y) { result.y = ZERO; }
+        #endif
       }
       else {
         const int m = kl - x;
         if (y >= x-kl && y <= x) { result = agm[a_ld*x + m + y + a_offset]; }
         else { SetToZero(result); }
-        COMPLEX_CONJUGATE(result);
+        #if defined(ROUTINE_HBMV)
+          COMPLEX_CONJUGATE(result);
+        #endif
       }
     }
     else {
@@ -124,48 +132,51 @@ inline real LoadMatrixA(const __global real* restrict agm, const int x, const in
         const int m = -y;
         if (x >= y && x < y+kl+1) { result = agm[a_ld*y + m + x + a_offset]; }
         else { SetToZero(result); }
-        if (x == y) { result.y = ZERO; }
+        #if defined(ROUTINE_HBMV)
+          if (x == y) { result.y = ZERO; }
+        #endif
       }
       else {
         const int m = -x;
         if (y >= x && y < x+kl+1) { result = agm[a_ld*x + m + y + a_offset]; }
         else { SetToZero(result); }
-        COMPLEX_CONJUGATE(result);
+        #if defined(ROUTINE_HBMV)
+          COMPLEX_CONJUGATE(result);
+        #endif
       }
     }
 
-  // For hermitian packed matrices
-  #elif defined(ROUTINE_HPMV)
+  // For symmetric/hermitian packed matrices
+  #elif defined(ROUTINE_HPMV) || defined(ROUTINE_SPMV)
     if (parameter == 1) {
       if (x <= y) {
         result = agm[((y+1)*y)/2 + x + a_offset];
-        if (x == y) { result.y = ZERO; }
+        #if defined(ROUTINE_HPMV)
+          if (x == y) { result.y = ZERO; }
+        #endif
       }
       else {
         result = agm[((x+1)*x)/2 + y + a_offset];
-        COMPLEX_CONJUGATE(result);
+        #if defined(ROUTINE_HPMV)
+          COMPLEX_CONJUGATE(result);
+        #endif
       }
     }
     else {
       if (x >= y) {
         result = agm[((2*a_ld-(y+1))*y)/2 + x + a_offset];
-        if (x == y) { result.y = ZERO; }
+        #if defined(ROUTINE_HPMV)
+          if (x == y) { result.y = ZERO; }
+        #endif
       }
       else {
         result = agm[((2*a_ld-(x+1))*x)/2 + y + a_offset];
-        COMPLEX_CONJUGATE(result);
+        #if defined(ROUTINE_HPMV)
+          COMPLEX_CONJUGATE(result);
+        #endif
       }
     }
 
-  // For symmetric matrices
-  #elif defined(ROUTINE_SYMV)
-    if ((parameter == 0 && y <= x) || (parameter == 1 && x <= y)) {
-      result = agm[y*a_ld + x + a_offset];
-    }
-    else {
-      result = agm[x*a_ld + y + a_offset];
-    }
-
   // For general matrices
   #else
     result = agm[a_ld*y + x + a_offset];
diff --git a/src/routines/level2/xsbmv.cc b/src/routines/level2/xsbmv.cc
new file mode 100644
index 00000000..457bd762
--- /dev/null
+++ b/src/routines/level2/xsbmv.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 Xsbmv class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xsbmv.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xsbmv<T>::Xsbmv(Queue &queue, Event &event, const std::string &name):
+    Xgemv<T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xsbmv<T>::DoSbmv(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 symmetric banded matrix-accesses are implemented in the kernel guarded by the
+  // ROUTINE_SBMV 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, false, k, 0);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xsbmv<float>;
+template class Xsbmv<double>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xspmv.cc b/src/routines/level2/xspmv.cc
new file mode 100644
index 00000000..4f1a9c61
--- /dev/null
+++ b/src/routines/level2/xspmv.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 Xspmv class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xspmv.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xspmv<T>::Xspmv(Queue &queue, Event &event, const std::string &name):
+    Xgemv<T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xspmv<T>::DoSpmv(const Layout layout, const Triangle triangle,
+                            const size_t n,
+                            const T alpha,
+                            const Buffer<T> &ap_buffer, const size_t ap_offset,
+                            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 symmetric packed matrix-accesses are implemented in the kernel guarded by the
+  // ROUTINE_SPMV define.
+  bool fast_kernels = false;
+  return MatVec(layout, Transpose::kNo,
+                n, n, alpha,
+                ap_buffer, ap_offset, n,
+                x_buffer, x_offset, x_inc, beta,
+                y_buffer, y_offset, y_inc,
+                fast_kernels, fast_kernels,
+                is_upper, true, 0, 0);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xspmv<float>;
+template class Xspmv<double>;
+
+// =================================================================================================
+} // namespace clblast