Added prototype for xASUM routines

4 files changed, 157 insertions, 0 deletions

diff --git a/test/correctness/routines/level1/xasum.cc b/test/correctness/routines/level1/xasum.cc
new file mode 100644
index 00000000..5ec20596
--- /dev/null
+++ b/test/correctness/routines/level1/xasum.cc
@@ -0,0 +1,28 @@
+
+// =================================================================================================
+// 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/xasum.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::TestXasum<float>, float, float>(argc, argv, false, "SASUM");
+  clblast::RunTests<clblast::TestXasum<double>, double, double>(argc, argv, true, "DASUM");
+  clblast::RunTests<clblast::TestXasum<float2>, float2, float2>(argc, argv, true, "ScASUM");
+  clblast::RunTests<clblast::TestXasum<double2>, double2, double2>(argc, argv, true, "DzASUM");
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/performance/routines/level1/xasum.cc b/test/performance/routines/level1/xasum.cc
new file mode 100644
index 00000000..2680966e
--- /dev/null
+++ b/test/performance/routines/level1/xasum.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/xasum.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::TestXasum<float>, float, float>(argc, argv); break;
+    case clblast::Precision::kDouble:
+      clblast::RunClient<clblast::TestXasum<double>, double, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle:
+      clblast::RunClient<clblast::TestXasum<float2>, float2, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble:
+      clblast::RunClient<clblast::TestXasum<double2>, double2, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/wrapper_cblas.h b/test/wrapper_cblas.h
index dec272b0..af0eec9b 100644
--- a/test/wrapper_cblas.h
+++ b/test/wrapper_cblas.h
@@ -345,6 +345,32 @@ void cblasXnrm2(const size_t n,
                                          reinterpret_cast<const double*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
 }
 
+// Forwards the Netlib BLAS calls for SASUM/DASUM/ScASUM/DzASUM
+void cblasXasum(const size_t n,
+                std::vector<float>& asum_buffer, const size_t asum_offset,
+                const std::vector<float>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  asum_buffer[asum_offset] = cblas_sasum(n,
+                                         &x_buffer[x_offset], static_cast<int>(x_inc));
+}
+void cblasXasum(const size_t n,
+                std::vector<double>& asum_buffer, const size_t asum_offset,
+                const std::vector<double>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  asum_buffer[asum_offset] = cblas_dasum(n,
+                                         &x_buffer[x_offset], static_cast<int>(x_inc));
+}
+void cblasXasum(const size_t n,
+                std::vector<float2>& asum_buffer, const size_t asum_offset,
+                const std::vector<float2>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  asum_buffer[asum_offset] = cblas_scasum(n,
+                                         reinterpret_cast<const float*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
+}
+void cblasXasum(const size_t n,
+                std::vector<double2>& asum_buffer, const size_t asum_offset,
+                const std::vector<double2>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  asum_buffer[asum_offset] = cblas_dzasum(n,
+                                         reinterpret_cast<const double*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
+}
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 89b708b8..09b3310b 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -558,6 +558,74 @@ clblasStatus clblasXnrm2<double2>(const size_t n,
                      num_queues, queues, num_wait_events, wait_events, events);
 }
 
+// Forwards the clBLAS calls for SASUM/DASUM/ScASUM/DzASUM
+template <typename T>
+clblasStatus clblasXasum(const size_t n,
+                         cl_mem asum_buffer, const size_t asum_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 clblasXasum<float>(const size_t n,
+                                cl_mem asum_buffer, const size_t asum_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);
+  return clblasSasum(n,
+                     asum_buffer, asum_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXasum<double>(const size_t n,
+                                 cl_mem asum_buffer, const size_t asum_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);
+  return clblasDasum(n,
+                     asum_buffer, asum_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXasum<float2>(const size_t n,
+                                 cl_mem asum_buffer, const size_t asum_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<float2>(context, n);
+  return clblasScasum(n,
+                     asum_buffer, asum_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXasum<double2>(const size_t n,
+                                  cl_mem asum_buffer, const size_t asum_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<double2>(context, n);
+  return clblasDzasum(n,
+                     asum_buffer, asum_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
 // =================================================================================================