Merge branch 'level1_routines' into development

2 files changed, 41 insertions, 5 deletions

diff --git a/test/performance/client.cc b/test/performance/client.cc
index ebfad3a6..17f54231 100644
--- a/test/performance/client.cc
+++ b/test/performance/client.cc
@@ -136,14 +136,14 @@ void Client<T,U>::PerformanceTest(Arguments<U> &args, const SetMetric set_sizes)
     std::vector<T> b_source(args.b_size);
     std::vector<T> c_source(args.c_size);
     std::vector<T> ap_source(args.ap_size);
-    std::vector<T> dot_source(args.dot_size);
+    std::vector<T> scalar_source(args.scalar_size);
     PopulateVector(x_source);
     PopulateVector(y_source);
     PopulateVector(a_source);
     PopulateVector(b_source);
     PopulateVector(c_source);
     PopulateVector(ap_source);
-    PopulateVector(dot_source);
+    PopulateVector(scalar_source);
 
     // Creates the matrices on the device
     auto x_vec = Buffer<T>(context, args.x_size);
@@ -152,15 +152,15 @@ void Client<T,U>::PerformanceTest(Arguments<U> &args, const SetMetric set_sizes)
     auto b_mat = Buffer<T>(context, args.b_size);
     auto c_mat = Buffer<T>(context, args.c_size);
     auto ap_mat = Buffer<T>(context, args.ap_size);
-    auto dot = Buffer<T>(context, args.dot_size);
+    auto scalar = Buffer<T>(context, args.scalar_size);
     x_vec.Write(queue, args.x_size, x_source);
     y_vec.Write(queue, args.y_size, y_source);
     a_mat.Write(queue, args.a_size, a_source);
     b_mat.Write(queue, args.b_size, b_source);
     c_mat.Write(queue, args.c_size, c_source);
     ap_mat.Write(queue, args.ap_size, ap_source);
-    dot.Write(queue, args.dot_size, dot_source);
-    auto buffers = Buffers<T>{x_vec, y_vec, a_mat, b_mat, c_mat, ap_mat, dot};
+    scalar.Write(queue, args.scalar_size, scalar_source);
+    auto buffers = Buffers<T>{x_vec, y_vec, a_mat, b_mat, c_mat, ap_mat, scalar};
 
     // Runs the routines and collects the timings
     auto timings = std::vector<std::pair<std::string, double>>();
@@ -267,6 +267,7 @@ void Client<T,U>::PrintTableRow(const Arguments<U>& args,
     else if (o == kArgCOffset) {  integers.push_back(args.c_offset); }
     else if (o == kArgAPOffset) { integers.push_back(args.ap_offset); }
     else if (o == kArgDotOffset) {integers.push_back(args.dot_offset); }
+    else if (o == kArgNrm2Offset){integers.push_back(args.nrm2_offset); }
   }
   auto strings = std::vector<std::string>{};
   for (auto &o: options_) {
diff --git a/test/performance/routines/level1/xnrm2.cc b/test/performance/routines/level1/xnrm2.cc
new file mode 100644
index 00000000..db6ec9ad
--- /dev/null
+++ b/test/performance/routines/level1/xnrm2.cc
@@ -0,0 +1,35 @@
+
+// =================================================================================================
+// 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:
+      clblast::RunClient<clblast::TestXnrm2<float2>, float2, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble:
+      clblast::RunClient<clblast::TestXnrm2<double2>, double2, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================