Test broadcasting on OpenCL devices with 64 bit indexing

1 files changed, 52 insertions, 47 deletions

diff --git a/unsupported/test/cxx11_tensor_broadcast_sycl.cpp b/unsupported/test/cxx11_tensor_broadcast_sycl.cpp
index c4798d42c..6d6d762ad 100644
--- a/unsupported/test/cxx11_tensor_broadcast_sycl.cpp
+++ b/unsupported/test/cxx11_tensor_broadcast_sycl.cpp
@@ -14,7 +14,7 @@
 #define EIGEN_TEST_NO_LONGDOUBLE
 #define EIGEN_TEST_NO_COMPLEX
 #define EIGEN_TEST_FUNC cxx11_tensor_broadcast_sycl
-#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
+#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int64_t
 #define EIGEN_USE_SYCL
 
 #include "main.h"
@@ -25,47 +25,47 @@ using Eigen::SyclDevice;
 using Eigen::Tensor;
 using Eigen::TensorMap;
 
-template <typename DataType, int DataLayout>
+template <typename DataType, int DataLayout, typename IndexType>
 static void test_broadcast_sycl_fixed(const Eigen::SyclDevice &sycl_device){
 
   // BROADCAST test:
-  int inDim1=2;
-  int inDim2=3;
-  int inDim3=5;
-  int inDim4=7;
-  int bDim1=2;
-  int bDim2=3;
-  int bDim3=1;
-  int bDim4=4;
-  array<int, 4> in_range   = {{inDim1, inDim2, inDim3, inDim4}};
-  array<int, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
-  array<int, 4> out_range;  // = in_range * broadcasts
+  IndexType inDim1=2;
+  IndexType inDim2=3;
+  IndexType inDim3=5;
+  IndexType inDim4=7;
+  IndexType bDim1=2;
+  IndexType bDim2=3;
+  IndexType bDim3=1;
+  IndexType bDim4=4;
+  array<IndexType, 4> in_range   = {{inDim1, inDim2, inDim3, inDim4}};
+  array<IndexType, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
+  array<IndexType, 4> out_range;  // = in_range * broadcasts
   for (size_t i = 0; i < out_range.size(); ++i)
     out_range[i] = in_range[i] * broadcasts[i];
 
-  Tensor<DataType, 4, DataLayout>  input(in_range);
-  Tensor<DataType, 4, DataLayout> out(out_range);
+  Tensor<DataType, 4, DataLayout, IndexType>  input(in_range);
+  Tensor<DataType, 4, DataLayout, IndexType> out(out_range);
 
   for (size_t i = 0; i < in_range.size(); ++i)
     VERIFY_IS_EQUAL(out.dimension(i), out_range[i]);
 
 
-  for (int i = 0; i < input.size(); ++i)
+  for (IndexType i = 0; i < input.size(); ++i)
     input(i) = static_cast<DataType>(i);
 
   DataType * gpu_in_data  = static_cast<DataType*>(sycl_device.allocate(input.dimensions().TotalSize()*sizeof(DataType)));
   DataType * gpu_out_data  = static_cast<DataType*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(DataType)));
 
-  TensorMap<TensorFixedSize<DataType, Sizes<2, 3, 5, 7>, DataLayout>> gpu_in(gpu_in_data, in_range);
-  TensorMap<Tensor<DataType, 4, DataLayout>> gpu_out(gpu_out_data, out_range);
+  TensorMap<TensorFixedSize<DataType, Sizes<2, 3, 5, 7>, DataLayout, IndexType>> gpu_in(gpu_in_data, in_range);
+  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu_out(gpu_out_data, out_range);
   sycl_device.memcpyHostToDevice(gpu_in_data, input.data(),(input.dimensions().TotalSize())*sizeof(DataType));
   gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
   sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(DataType));
 
-  for (int i = 0; i < inDim1*bDim1; ++i) {
-    for (int j = 0; j < inDim2*bDim2; ++j) {
-      for (int k = 0; k < inDim3*bDim3; ++k) {
-        for (int l = 0; l < inDim4*bDim4; ++l) {
+  for (IndexType i = 0; i < inDim1*bDim1; ++i) {
+    for (IndexType j = 0; j < inDim2*bDim2; ++j) {
+      for (IndexType k = 0; k < inDim3*bDim3; ++k) {
+        for (IndexType l = 0; l < inDim4*bDim4; ++l) {
           VERIFY_IS_APPROX(input(i%2,j%3,k%5,l%7), out(i,j,k,l));
         }
       }
@@ -76,47 +76,47 @@ static void test_broadcast_sycl_fixed(const Eigen::SyclDevice &sycl_device){
   sycl_device.deallocate(gpu_out_data);
 }
 
-template <typename DataType, int DataLayout>
+template <typename DataType, int DataLayout, typename IndexType>
 static void test_broadcast_sycl(const Eigen::SyclDevice &sycl_device){
 
   // BROADCAST test:
-  int inDim1=2;
-  int inDim2=3;
-  int inDim3=5;
-  int inDim4=7;
-  int bDim1=2;
-  int bDim2=3;
-  int bDim3=1;
-  int bDim4=4;
-  array<int, 4> in_range   = {{inDim1, inDim2, inDim3, inDim4}};
-  array<int, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
-  array<int, 4> out_range;  // = in_range * broadcasts
+  IndexType inDim1=2;
+  IndexType inDim2=3;
+  IndexType inDim3=5;
+  IndexType inDim4=7;
+  IndexType bDim1=2;
+  IndexType bDim2=3;
+  IndexType bDim3=1;
+  IndexType bDim4=4;
+  array<IndexType, 4> in_range   = {{inDim1, inDim2, inDim3, inDim4}};
+  array<IndexType, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
+  array<IndexType, 4> out_range;  // = in_range * broadcasts
   for (size_t i = 0; i < out_range.size(); ++i)
     out_range[i] = in_range[i] * broadcasts[i];
 
-  Tensor<DataType, 4, DataLayout>  input(in_range);
-  Tensor<DataType, 4, DataLayout> out(out_range);
+  Tensor<DataType, 4, DataLayout, IndexType>  input(in_range);
+  Tensor<DataType, 4, DataLayout, IndexType> out(out_range);
 
   for (size_t i = 0; i < in_range.size(); ++i)
     VERIFY_IS_EQUAL(out.dimension(i), out_range[i]);
 
 
-  for (int i = 0; i < input.size(); ++i)
+  for (IndexType i = 0; i < input.size(); ++i)
     input(i) = static_cast<DataType>(i);
 
   DataType * gpu_in_data  = static_cast<DataType*>(sycl_device.allocate(input.dimensions().TotalSize()*sizeof(DataType)));
   DataType * gpu_out_data  = static_cast<DataType*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(DataType)));
 
-  TensorMap<Tensor<DataType, 4, DataLayout>>  gpu_in(gpu_in_data, in_range);
-  TensorMap<Tensor<DataType, 4, DataLayout>> gpu_out(gpu_out_data, out_range);
+  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>>  gpu_in(gpu_in_data, in_range);
+  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu_out(gpu_out_data, out_range);
   sycl_device.memcpyHostToDevice(gpu_in_data, input.data(),(input.dimensions().TotalSize())*sizeof(DataType));
   gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
   sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(DataType));
 
-  for (int i = 0; i < inDim1*bDim1; ++i) {
-    for (int j = 0; j < inDim2*bDim2; ++j) {
-      for (int k = 0; k < inDim3*bDim3; ++k) {
-        for (int l = 0; l < inDim4*bDim4; ++l) {
+  for (IndexType i = 0; i < inDim1*bDim1; ++i) {
+    for (IndexType j = 0; j < inDim2*bDim2; ++j) {
+      for (IndexType k = 0; k < inDim3*bDim3; ++k) {
+        for (IndexType l = 0; l < inDim4*bDim4; ++l) {
           VERIFY_IS_APPROX(input(i%inDim1,j%inDim2,k%inDim3,l%inDim4), out(i,j,k,l));
         }
       }
@@ -130,10 +130,15 @@ static void test_broadcast_sycl(const Eigen::SyclDevice &sycl_device){
 template<typename DataType, typename dev_Selector> void sycl_broadcast_test_per_device(dev_Selector s){
   QueueInterface queueInterface(s);
   auto sycl_device = Eigen::SyclDevice(&queueInterface);
-  test_broadcast_sycl_fixed<DataType, RowMajor>(sycl_device);
-  test_broadcast_sycl<DataType, RowMajor>(sycl_device);
-  test_broadcast_sycl_fixed<DataType, ColMajor>(sycl_device);
-  test_broadcast_sycl<DataType, ColMajor>(sycl_device);
+  test_broadcast_sycl_fixed<DataType, RowMajor, int>(sycl_device);
+  test_broadcast_sycl<DataType, RowMajor, int>(sycl_device);
+  test_broadcast_sycl_fixed<DataType, ColMajor, int>(sycl_device);
+  test_broadcast_sycl<DataType, ColMajor, int>(sycl_device);
+
+  test_broadcast_sycl_fixed<DataType, RowMajor, int64_t>(sycl_device);
+  test_broadcast_sycl<DataType, RowMajor, int64_t>(sycl_device);
+  test_broadcast_sycl_fixed<DataType, ColMajor, int64_t>(sycl_device);
+  test_broadcast_sycl<DataType, ColMajor, int64_t>(sycl_device);
 }
 
 void test_cxx11_tensor_broadcast_sycl() {