Converting all sycl buffers to uninitialised device only buffers; adding memcpyHostToDevice and memcpyDeviceToHost on syclDevice; modifying all examples to obey the new rules; moving sycl queue creating to the device based on Benoit suggestion; removing the sycl specefic condition for returning m_result in TensorReduction.h according to Benoit suggestion.

1 files changed, 23 insertions, 21 deletions

diff --git a/unsupported/test/cxx11_tensor_forced_eval_sycl.cpp b/unsupported/test/cxx11_tensor_forced_eval_sycl.cpp
index 182ec7fa8..5690da723 100644
--- a/unsupported/test/cxx11_tensor_forced_eval_sycl.cpp
+++ b/unsupported/test/cxx11_tensor_forced_eval_sycl.cpp
@@ -22,18 +22,7 @@
 
 using Eigen::Tensor;
 
-void test_forced_eval_sycl() {
-	cl::sycl::gpu_selector s;
-  cl::sycl::queue q(s, [=](cl::sycl::exception_list l) {
-    for (const auto& e : l) {
-      try {
-        std::rethrow_exception(e);
-      } catch (cl::sycl::exception e) {
-        std::cout << e.what() << std::endl;
-      }
-    }
-  });
-	SyclDevice sycl_device(q);
+void test_forced_eval_sycl(const Eigen::SyclDevice &sycl_device) {
 
   int sizeDim1 = 100;
   int sizeDim2 = 200;
@@ -43,17 +32,22 @@ void test_forced_eval_sycl() {
   Eigen::Tensor<float, 3> in2(tensorRange);
   Eigen::Tensor<float, 3> out(tensorRange);
 
+  float * gpu_in1_data  = static_cast<float*>(sycl_device.allocate(in1.dimensions().TotalSize()*sizeof(float)));
+  float * gpu_in2_data  = static_cast<float*>(sycl_device.allocate(in2.dimensions().TotalSize()*sizeof(float)));
+  float * gpu_out_data =  static_cast<float*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(float)));
+
   in1 = in1.random() + in1.constant(10.0f);
   in2 = in2.random() + in2.constant(10.0f);
 
-	// creating TensorMap from tensor
-  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in1(in1.data(), tensorRange);
-  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in2(in2.data(), tensorRange);
-  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_out(out.data(), tensorRange);
-
+  // creating TensorMap from tensor
+  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in1(gpu_in1_data, tensorRange);
+  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in2(gpu_in2_data, tensorRange);
+  Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_out(gpu_out_data, tensorRange);
+  sycl_device.memcpyHostToDevice(gpu_in1_data, in1.data(),(in1.dimensions().TotalSize())*sizeof(float));
+  sycl_device.memcpyHostToDevice(gpu_in2_data, in2.data(),(in1.dimensions().TotalSize())*sizeof(float));
   /// c=(a+b)*b
-	gpu_out.device(sycl_device) =(gpu_in1 + gpu_in2).eval() * gpu_in2;
-	sycl_device.deallocate(out.data());
+  gpu_out.device(sycl_device) =(gpu_in1 + gpu_in2).eval() * gpu_in2;
+  sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
   for (int i = 0; i < sizeDim1; ++i) {
     for (int j = 0; j < sizeDim2; ++j) {
       for (int k = 0; k < sizeDim3; ++k) {
@@ -62,7 +56,15 @@ void test_forced_eval_sycl() {
       }
     }
   }
-	printf("(a+b)*b Test Passed\n");
+  printf("(a+b)*b Test Passed\n");
+  sycl_device.deallocate(gpu_in1_data);
+  sycl_device.deallocate(gpu_in2_data);
+  sycl_device.deallocate(gpu_out_data);
+
 }
 
-void test_cxx11_tensor_forced_eval_sycl() { CALL_SUBTEST(test_forced_eval_sycl()); }
+void test_cxx11_tensor_forced_eval_sycl() {
+  cl::sycl::gpu_selector s;
+  Eigen::SyclDevice sycl_device(s);
+  CALL_SUBTEST(test_forced_eval_sycl(sycl_device));
+}