Modified test interfaces such that they support either OpenCL or CUDA

5 files changed, 95 insertions, 45 deletions

diff --git a/test/routines/levelx/xaxpybatched.hpp b/test/routines/levelx/xaxpybatched.hpp
index 4a8fc564..e9715f4e 100644
--- a/test/routines/levelx/xaxpybatched.hpp
+++ b/test/routines/levelx/xaxpybatched.hpp
@@ -83,14 +83,23 @@ class TestXaxpyBatched {
 
   // 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 = AxpyBatched(args.n, args.alphas.data(),
-                              buffers.x_vec(), args.x_offsets.data(), args.x_inc,
-                              buffers.y_vec(), args.y_offsets.data(), args.y_inc,
-                              args.batch_count,
-                              &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 = AxpyBatched(args.n, args.alphas.data(),
+                                buffers.x_vec(), args.x_offsets.data(), args.x_inc,
+                                buffers.y_vec(), args.y_offsets.data(), args.y_inc,
+                                args.batch_count,
+                                &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = AxpyBatched(args.n, args.alphas.data(),
+                                buffers.x_vec(), args.x_offsets.data(), args.x_inc,
+                                buffers.y_vec(), args.y_offsets.data(), args.y_inc,
+                                args.batch_count,
+                                queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/levelx/xgemmbatched.hpp b/test/routines/levelx/xgemmbatched.hpp
index 704d0578..2a8bd9d4 100644
--- a/test/routines/levelx/xgemmbatched.hpp
+++ b/test/routines/levelx/xgemmbatched.hpp
@@ -108,8 +108,6 @@ class TestXgemmBatched {
 
   // 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{};
     // Relaxed requirement on ld_a and ld_b within the library, this is here to match clBLAS
     auto a_rotated = (args.layout == Layout::kColMajor && args.a_transpose != Transpose::kNo) ||
                      (args.layout == Layout::kRowMajor && args.a_transpose == Transpose::kNo);
@@ -119,14 +117,27 @@ class TestXgemmBatched {
     auto b_one = (!b_rotated) ? args.k : args.n;
     if (args.a_ld < a_one) { return StatusCode::kInvalidLeadDimA; }
     if (args.b_ld < b_one) { return StatusCode::kInvalidLeadDimB; }
-    auto status = GemmBatched(args.layout, args.a_transpose, args.b_transpose,
-                              args.m, args.n, args.k, args.alphas.data(),
-                              buffers.a_mat(), args.a_offsets.data(), args.a_ld,
-                              buffers.b_mat(), args.b_offsets.data(), args.b_ld, args.betas.data(),
-                              buffers.c_mat(), args.c_offsets.data(), args.c_ld,
-                              args.batch_count,
-                              &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 = GemmBatched(args.layout, args.a_transpose, args.b_transpose,
+                                args.m, args.n, args.k, args.alphas.data(),
+                                buffers.a_mat(), args.a_offsets.data(), args.a_ld,
+                                buffers.b_mat(), args.b_offsets.data(), args.b_ld, args.betas.data(),
+                                buffers.c_mat(), args.c_offsets.data(), args.c_ld,
+                                args.batch_count,
+                                &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = GemmBatched(args.layout, args.a_transpose, args.b_transpose,
+                                args.m, args.n, args.k, args.alphas.data(),
+                                buffers.a_mat(), args.a_offsets.data(), args.a_ld,
+                                buffers.b_mat(), args.b_offsets.data(), args.b_ld, args.betas.data(),
+                                buffers.c_mat(), args.c_offsets.data(), args.c_ld,
+                                args.batch_count,
+                                queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/levelx/xim2col.hpp b/test/routines/levelx/xim2col.hpp
index 4124190f..ebffe85e 100644
--- a/test/routines/levelx/xim2col.hpp
+++ b/test/routines/levelx/xim2col.hpp
@@ -84,17 +84,29 @@ public:
 
   // 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 = Im2col<T>(args.channels, args.height, args.width,
-                            args.kernel_h, args.kernel_w,
-                            args.pad_h, args.pad_w,
-                            args.stride_h, args.stride_w,
-                            args.dilation_h, args.dilation_w,
-                            buffers.a_mat(), args.a_offset,
-                            buffers.b_mat(), args.b_offset,
-                            &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 = Im2col<T>(args.channels, args.height, args.width,
+                              args.kernel_h, args.kernel_w,
+                              args.pad_h, args.pad_w,
+                              args.stride_h, args.stride_w,
+                              args.dilation_h, args.dilation_w,
+                              buffers.a_mat(), args.a_offset,
+                              buffers.b_mat(), args.b_offset,
+                              &queue_plain, &event);
+      if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+    #elif CUDA_API
+      auto status = Im2col<T>(args.channels, args.height, args.width,
+                              args.kernel_h, args.kernel_w,
+                              args.pad_h, args.pad_w,
+                              args.stride_h, args.stride_w,
+                              args.dilation_h, args.dilation_w,
+                              buffers.a_mat(), args.a_offset,
+                              buffers.b_mat(), args.b_offset,
+                              queue.GetContext()(), queue.GetDevice()());
+      cuStreamSynchronize(queue());
+    #endif
     return status;
   }
 
diff --git a/test/routines/levelx/xinvert.hpp b/test/routines/levelx/xinvert.hpp
index cc02a88b..3df1e2b0 100644
--- a/test/routines/levelx/xinvert.hpp
+++ b/test/routines/levelx/xinvert.hpp
@@ -164,14 +164,23 @@ class TestXinvert {
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
     try {
-      auto event = cl_event{};
-      auto inverter = Xinvert<T>(queue, &event);
-      inverter.InvertMatrixDiagonalBlocks(args.layout, args.triangle, args.diagonal,
-                                          args.n, args.m,
-                                          buffers.a_mat, args.a_offset, args.a_ld,
-                                          buffers.b_mat);
-      clWaitForEvents(1, &event);
-      clReleaseEvent(event);
+      #ifdef OPENCL_API
+        auto event = cl_event{};
+        auto inverter = Xinvert<T>(queue, &event);
+        inverter.InvertMatrixDiagonalBlocks(args.layout, args.triangle, args.diagonal,
+                                            args.n, args.m,
+                                            buffers.a_mat, args.a_offset, args.a_ld,
+                                            buffers.b_mat);
+        clWaitForEvents(1, &event);
+        clReleaseEvent(event);
+      #elif CUDA_API
+        auto inverter = Xinvert<T>(queue, nullptr);
+        inverter.InvertMatrixDiagonalBlocks(args.layout, args.triangle, args.diagonal,
+                                            args.n, args.m,
+                                            buffers.a_mat, args.a_offset, args.a_ld,
+                                            buffers.b_mat);
+        cuStreamSynchronize(queue());
+      #endif
     } catch (...) { return DispatchException(); }
     return StatusCode::kSuccess;
   }
diff --git a/test/routines/levelx/xomatcopy.hpp b/test/routines/levelx/xomatcopy.hpp
index 2736cf75..70bda452 100644
--- a/test/routines/levelx/xomatcopy.hpp
+++ b/test/routines/levelx/xomatcopy.hpp
@@ -126,14 +126,23 @@ class TestXomatcopy {
 
   // 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 = Omatcopy<T>(args.layout, args.a_transpose,
-                              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 = Omatcopy<T>(args.layout, args.a_transpose,
+                                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 = Omatcopy<T>(args.layout, args.a_transpose,
+                                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;
   }