Added XOMATCOPY routines to perform out-of-place matrix scaling, copying, and/or transposing

3 files changed, 61 insertions, 9 deletions

diff --git a/test/performance/client.cc b/test/performance/client.cc
index 5a7226df..4c0c79a6 100644
--- a/test/performance/client.cc
+++ b/test/performance/client.cc
@@ -42,8 +42,10 @@ Client<T,U>::Client(const Routine run_routine,
 // applicable, but are searched for anyway to be able to create one common argument parser. All
 // arguments have a default value in case they are not found.
 template <typename T, typename U>
-Arguments<U> Client<T,U>::ParseArguments(int argc, char *argv[], const GetMetric default_a_ld,
-                                         const GetMetric default_b_ld, const GetMetric default_c_ld) {
+Arguments<U> Client<T,U>::ParseArguments(int argc, char *argv[], const size_t level,
+                                         const GetMetric default_a_ld,
+                                         const GetMetric default_b_ld,
+                                         const GetMetric default_c_ld) {
   auto args = Arguments<U>{};
   auto help = std::string{"\n* Options given/available:\n"};
 
@@ -116,6 +118,17 @@ Arguments<U> Client<T,U>::ParseArguments(int argc, char *argv[], const GetMetric
   // which is thus always displayed (unless silence is specified).
   if (!args.silent) { fprintf(stdout, "%s\n", help.c_str()); }
 
+  // Comparison against a non-BLAS routine is not supported
+  if (level == 4) { // level-4 == level-X
+    if (args.compare_clblas != 0 || args.compare_cblas != 0) {
+      if (!args.silent) {
+        fprintf(stdout, "* Disabling clBLAS and CPU BLAS comparisons for this non-BLAS routine\n\n");
+      }
+    }
+    args.compare_clblas = 0;
+    args.compare_cblas = 0;
+  }
+
   // Comparison against clBLAS or a CPU BLAS library is not supported in case of half-precision
   if (args.precision == Precision::kHalf) {
     if (args.compare_clblas != 0 || args.compare_cblas != 0) {
diff --git a/test/performance/client.h b/test/performance/client.h
index 8d0597d7..493a7aed 100644
--- a/test/performance/client.h
+++ b/test/performance/client.h
@@ -53,8 +53,10 @@ class Client {
 
   // Parses all command-line arguments, filling in the arguments structure. If no command-line
   // argument is given for a particular argument, it is filled in with a default value.
-  Arguments<U> ParseArguments(int argc, char *argv[], const GetMetric default_a_ld,
-                              const GetMetric default_b_ld, const GetMetric default_c_ld);
+  Arguments<U> ParseArguments(int argc, char *argv[], const size_t level,
+                              const GetMetric default_a_ld,
+                              const GetMetric default_b_ld,
+                              const GetMetric default_c_ld);
 
   // The main client function, setting-up arguments, matrices, OpenCL buffers, etc. After set-up, it
   // calls the client routines.
@@ -97,14 +99,14 @@ void RunClient(int argc, char *argv[]) {
 
   // Sets the reference to test against
   #ifdef CLBLAST_REF_CLBLAS
-    const auto reference1 = C::RunReference1; // clBLAS when available
+    auto reference1 = C::RunReference1; // clBLAS when available
   #else
-    const auto reference1 = ReferenceNotAvailable<T,U>;
+    auto reference1 = ReferenceNotAvailable<T,U>;
   #endif
   #ifdef CLBLAST_REF_CBLAS
-    const auto reference2 = C::RunReference2; // CBLAS when available
+    auto reference2 = C::RunReference2; // CBLAS when available
   #else
-    const auto reference2 = ReferenceNotAvailable<T,U>;
+    auto reference2 = ReferenceNotAvailable<T,U>;
   #endif
 
   // Creates a new client
@@ -112,7 +114,8 @@ void RunClient(int argc, char *argv[]) {
                             C::GetFlops, C::GetBytes);
 
   // Simple command line argument parser with defaults
-  auto args = client.ParseArguments(argc, argv, C::DefaultLDA, C::DefaultLDB, C::DefaultLDC);
+  auto args = client.ParseArguments(argc, argv, C::BLASLevel(),
+                                    C::DefaultLDA, C::DefaultLDB, C::DefaultLDC);
   if (args.print_help) { return; }
 
   // Runs the client
diff --git a/test/performance/routines/levelx/xomatcopy.cc b/test/performance/routines/levelx/xomatcopy.cc
new file mode 100644
index 00000000..851f6ee1
--- /dev/null
+++ b/test/performance/routines/levelx/xomatcopy.cc
@@ -0,0 +1,36 @@
+
+// =================================================================================================
+// 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/levelx/xomatcopy.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:
+      clblast::RunClient<clblast::TestXomatcopy<half>, half, half>(argc, argv); break;
+    case clblast::Precision::kSingle:
+      clblast::RunClient<clblast::TestXomatcopy<float>, float, float>(argc, argv); break;
+    case clblast::Precision::kDouble:
+      clblast::RunClient<clblast::TestXomatcopy<double>, double, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle:
+      clblast::RunClient<clblast::TestXomatcopy<float2>, float2, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble:
+      clblast::RunClient<clblast::TestXomatcopy<double2>, double2, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================