9 files changed, 163 insertions, 77 deletions
diff --git a/test/routines/level3/xgemm.hpp b/test/routines/level3/xgemm.hpp
index 1c430c1c..8444c1c3 100644
--- a/test/routines/level3/xgemm.hpp
+++ b/test/routines/level3/xgemm.hpp
@@ -90,15 +90,25 @@ class TestXgemm {
                                                       {{"XGEMM_MIN_INDIRECT_SIZE", switch_threshold}});
       if (override_status != StatusCode::kSuccess) { return override_status; }
     }
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Gemm(args.layout, args.a_transpose, args.b_transpose,
-                       args.m, args.n, args.k, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld,
-                       buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
-                       buffers.c_mat(), args.c_offset, args.c_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Gemm(args.layout, args.a_transpose, args.b_transpose,
+                         args.m, args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Gemm(args.layout, args.a_transpose, args.b_transpose,
+                         args.m, args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xhemm.hpp b/test/routines/level3/xhemm.hpp
index a89617b5..3b70d3f1 100644
--- a/test/routines/level3/xhemm.hpp
+++ b/test/routines/level3/xhemm.hpp
@@ -83,15 +83,25 @@ class TestXhemm {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Hemm(args.layout, args.side, args.triangle,
-                       args.m, args.n, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld,
-                       buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
-                       buffers.c_mat(), args.c_offset, args.c_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Hemm(args.layout, args.side, args.triangle,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Hemm(args.layout, args.side, args.triangle,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xher2k.hpp b/test/routines/level3/xher2k.hpp
index 55e6d894..6c4e12f1 100644
--- a/test/routines/level3/xher2k.hpp
+++ b/test/routines/level3/xher2k.hpp
@@ -81,16 +81,26 @@ class TestXher2k {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<U> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
     auto alpha2 = T{args.alpha, args.alpha};
-    auto status = Her2k(args.layout, args.triangle, args.a_transpose,
-                        args.n, args.k, alpha2,
-                        buffers.a_mat(), args.a_offset, args.a_ld,
-                        buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
-                        buffers.c_mat(), args.c_offset, args.c_ld,
-                        &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Her2k(args.layout, args.triangle, args.a_transpose,
+                          args.n, args.k, alpha2,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                          buffers.c_mat(), args.c_offset, args.c_ld,
+                          &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Her2k(args.layout, args.triangle, args.a_transpose,
+                          args.n, args.k, alpha2,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                          buffers.c_mat(), args.c_offset, args.c_ld,
+                          queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xherk.hpp b/test/routines/level3/xherk.hpp
index 3e1e7e02..c1bb7a0b 100644
--- a/test/routines/level3/xherk.hpp
+++ b/test/routines/level3/xherk.hpp
@@ -74,14 +74,23 @@ class TestXherk {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<U> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Herk(args.layout, args.triangle, args.a_transpose,
-                       args.n, args.k, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
-                       buffers.c_mat(), args.c_offset, args.c_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Herk(args.layout, args.triangle, args.a_transpose,
+                         args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Herk(args.layout, args.triangle, args.a_transpose,
+                         args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xsymm.hpp b/test/routines/level3/xsymm.hpp
index 5d840d40..90cc1888 100644
--- a/test/routines/level3/xsymm.hpp
+++ b/test/routines/level3/xsymm.hpp
@@ -83,15 +83,25 @@ class TestXsymm {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Symm(args.layout, args.side, args.triangle,
-                       args.m, args.n, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld,
-                       buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
-                       buffers.c_mat(), args.c_offset, args.c_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Symm(args.layout, args.side, args.triangle,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Symm(args.layout, args.side, args.triangle,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xsyr2k.hpp b/test/routines/level3/xsyr2k.hpp
index 4a4a2f10..6b29aff7 100644
--- a/test/routines/level3/xsyr2k.hpp
+++ b/test/routines/level3/xsyr2k.hpp
@@ -81,15 +81,25 @@ class TestXsyr2k {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Syr2k(args.layout, args.triangle, args.a_transpose,
-                        args.n, args.k, args.alpha,
-                        buffers.a_mat(), args.a_offset, args.a_ld,
-                        buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
-                        buffers.c_mat(), args.c_offset, args.c_ld,
-                        &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Syr2k(args.layout, args.triangle, args.a_transpose,
+                          args.n, args.k, args.alpha,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                          buffers.c_mat(), args.c_offset, args.c_ld,
+                          &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Syr2k(args.layout, args.triangle, args.a_transpose,
+                          args.n, args.k, args.alpha,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                          buffers.c_mat(), args.c_offset, args.c_ld,
+                          queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xsyrk.hpp b/test/routines/level3/xsyrk.hpp
index 90e46727..b7782176 100644
--- a/test/routines/level3/xsyrk.hpp
+++ b/test/routines/level3/xsyrk.hpp
@@ -74,14 +74,23 @@ class TestXsyrk {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Syrk(args.layout, args.triangle, args.a_transpose,
-                       args.n, args.k, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
-                       buffers.c_mat(), args.c_offset, args.c_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Syrk(args.layout, args.triangle, args.a_transpose,
+                         args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Syrk(args.layout, args.triangle, args.a_transpose,
+                         args.n, args.k, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld, args.beta,
+                         buffers.c_mat(), args.c_offset, args.c_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xtrmm.hpp b/test/routines/level3/xtrmm.hpp
index acc00e01..62d0f573 100644
--- a/test/routines/level3/xtrmm.hpp
+++ b/test/routines/level3/xtrmm.hpp
@@ -74,14 +74,23 @@ class TestXtrmm {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Trmm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
-                       args.m, args.n, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld,
-                       buffers.b_mat(), args.b_offset, args.b_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Trmm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Trmm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/level3/xtrsm.hpp b/test/routines/level3/xtrsm.hpp
index d63c9d79..9ce1f09c 100644
--- a/test/routines/level3/xtrsm.hpp
+++ b/test/routines/level3/xtrsm.hpp
@@ -85,14 +85,23 @@ class TestXtrsm {
 
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    auto queue_plain = queue();
-    auto event = cl_event{};
-    auto status = Trsm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
-                       args.m, args.n, args.alpha,
-                       buffers.a_mat(), args.a_offset, args.a_ld,
-                       buffers.b_mat(), args.b_offset, args.b_ld,
-                       &queue_plain, &event);
-    if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #ifdef OPENCL_API
+      auto queue_plain = queue();
+      auto event = cl_event{};
+      auto status = Trsm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld,
+                         &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Trsm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
+                         args.m, args.n, args.alpha,
+                         buffers.a_mat(), args.a_offset, args.a_ld,
+                         buffers.b_mat(), args.b_offset, args.b_ld,
+                         queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }