Added prototypes for SNRM2/DNRM2 routines

3 files changed, 97 insertions, 0 deletions

diff --git a/test/correctness/routines/level1/xnrm2.cc b/test/correctness/routines/level1/xnrm2.cc
new file mode 100644
index 00000000..8238e868
--- /dev/null
+++ b/test/correctness/routines/level1/xnrm2.cc
@@ -0,0 +1,26 @@
+
+// =================================================================================================
+// 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>
+//
+// =================================================================================================
+
+#include "correctness/testblas.h"
+#include "routines/level1/xnrm2.h"
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  clblast::RunTests<clblast::TestXnrm2<float>, float, float>(argc, argv, false, "SNRM2");
+  clblast::RunTests<clblast::TestXnrm2<double>, double, double>(argc, argv, true, "DNRM2");
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/performance/routines/level1/xnrm2.cc b/test/performance/routines/level1/xnrm2.cc
new file mode 100644
index 00000000..d5ae348b
--- /dev/null
+++ b/test/performance/routines/level1/xnrm2.cc
@@ -0,0 +1,33 @@
+
+// =================================================================================================
+// 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>
+//
+// =================================================================================================
+
+#include "performance/client.h"
+#include "routines/level1/xnrm2.h"
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv, clblast::Precision::kSingle)) {
+    case clblast::Precision::kHalf: throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kSingle:
+      clblast::RunClient<clblast::TestXnrm2<float>, float, float>(argc, argv); break;
+    case clblast::Precision::kDouble:
+      clblast::RunClient<clblast::TestXnrm2<double>, double, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kComplexDouble: throw std::runtime_error("Unsupported precision mode");
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 23a02a45..501f0bc5 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -350,6 +350,44 @@ clblasStatus clblasXdotc<double2>(const size_t n,
                      num_queues, queues, num_wait_events, wait_events, events);
 }
 
+// Forwards the clBLAS calls for SNRM2/DNRM2
+template <typename T>
+clblasStatus clblasXnrm2(const size_t n,
+                         cl_mem nrm2_buffer, const size_t nrm2_offset,
+                         const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                         cl_uint num_queues, cl_command_queue *queues,
+                         cl_uint num_wait_events, const cl_event *wait_events, cl_event *events);
+template <>
+clblasStatus clblasXnrm2<float>(const size_t n,
+                                cl_mem nrm2_buffer, const size_t nrm2_offset,
+                                const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                cl_uint num_queues, cl_command_queue *queues,
+                                cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<float>(context, n*x_inc + x_offset);
+  return clblasSnrm2(n,
+                     nrm2_buffer, nrm2_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXnrm2<double>(const size_t n,
+                                 cl_mem nrm2_buffer, const size_t nrm2_offset,
+                                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                 cl_uint num_queues, cl_command_queue *queues,
+                                 cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<double>(context, n*x_inc + x_offset);
+  return clblasDnrm2(n,
+                     nrm2_buffer, nrm2_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================