[OpenCL] Fixes scatter_nd_op_test

lukeiwanski · Apr 27, 2017 · 5cbaab5 · 5cbaab5
1 parent 4704240
commit 5cbaab5
Show file tree

Hide file tree

Showing 2 changed files with 42 additions and 22 deletions.
diff --git a/tensorflow/core/kernels/scatter_nd_op.cc b/tensorflow/core/kernels/scatter_nd_op.cc
@@ -27,6 +27,10 @@ limitations under the License.
 #include "tensorflow/core/platform/types.h"
 #include "tensorflow/core/util/util.h"
 
+#ifdef TENSORFLOW_USE_SYCL
+#include "tensorflow/core/common_runtime/sycl/sycl_util.h"
+#endif // TENSORFLOW_USE_SYCL
+
 namespace tensorflow {
 
 typedef Eigen::ThreadPoolDevice CPUDevice;
@@ -141,6 +145,40 @@ static void PrepareAndValidateInputs(OpKernelContext* c,
   *num_updates = indices_shape.num_elements() / safe_slice_dim;
 }
 
+template <typename Device, typename Index>
+class IndexFlattener {
+public:
+  inline typename TTypes<Index, 2>::ConstTensor
+  operator()(OpKernelContext*, const Tensor& indices) {
+    return indices.flat_inner_dims<Index>();
+  }
+};
+
+#ifdef TENSORFLOW_USE_SYCL
+template <typename Index>
+class IndexFlattener<SYCLDevice, Index> {
+public:
+  IndexFlattener() { indices_host_ = nullptr; }
+  ~IndexFlattener() { delete[] indices_host_; }
+
+  inline typename TTypes<Index, 2>::ConstTensor
+  operator()(OpKernelContext* c, const Tensor& indices) {
+    size_t num_indices = indices.NumElements();
+    indices_host_ = new Index[num_indices];
+    auto device = c->eigen_sycl_device();
+    auto size = sizeof(Index) * num_indices;
+    auto src_ptr = GetBase(&indices);
+    device.memcpyDeviceToHost(indices_host_, static_cast<const Index*>(src_ptr),
+                              size);
+    return typename TTypes<Index, 2>::ConstTensor(indices_host_,
+           indices.shape().AsEigenDSizes<2>());
+  }
+
+private:
+  Index* indices_host_;
+};
+#endif
+
 template <typename Device, typename T, typename Index>
 class ScatterNdOp : public OpKernel {
  public:
@@ -169,7 +207,8 @@ class ScatterNdOp : public OpKernel {
                                     &num_updates, &slice_size);
     if (!c->status().ok()) return;
 
-    auto indices_flat = indices.flat_inner_dims<Index>();
+    IndexFlattener<Device, Index> index_flattener;
+    auto indices_flat = index_flattener(c, indices);
     auto updates_flat = updates.shaped<T, 2>({num_updates, slice_size});
 
     Tensor* out = nullptr;
@@ -265,7 +304,8 @@ class ScatterNdUpdateOp : public OpKernel {
                                     &slice_dim, &num_updates, &slice_size);
     if (!c->status().ok()) return;
 
-    auto indices_flat = indices.flat_inner_dims<Index>();
+    IndexFlattener<Device, Index> index_flattener;
+    auto indices_flat = index_flattener(c, indices);
     auto updates_flat = updates.shaped<T, 2>({num_updates, slice_size});
     auto params_matrix = params.template shaped<T, 2>(
         {params_shape.num_elements() / slice_size, slice_size});

diff --git a/tensorflow/python/kernel_tests/scatter_nd_ops_test.py b/tensorflow/python/kernel_tests/scatter_nd_ops_test.py
@@ -95,10 +95,6 @@ def _VariableRankTest(self,
                         vtype,
                         itype,
                         repeat_indices=False):
-    # Currently not implemented for OpenCL
-    if test_util.is_sycl_enabled():
-      return
-
     np.random.seed(8)
     ref_shapes = [(3, 6), (3, 6), (3, 6, 9), (3, 6, 9), (3, 6, 9), (3, 6, 9)]
     indices_shapes = [(2,), (2, 2), (2,), (2, 2), (2, 3), (2, 3, 3)]
@@ -149,10 +145,6 @@ def _VariableRankTests(self, np_scatter, tf_scatter):
         self._VariableRankTest(np_scatter, tf_scatter, vtype, itype)
 
   def testSimple(self):
-    # Currently not implemented for OpenCL
-    if test_util.is_sycl_enabled():
-      return
-
     indices = constant_op.constant([[4], [3], [1], [7]], dtype=dtypes.int32)
     updates = constant_op.constant([9, 10, 11, 12], dtype=dtypes.float32)
     ref = variables.Variable([0, 0, 0, 0, 0, 0, 0, 0], dtype=dtypes.float32)
@@ -166,10 +158,6 @@ def testSimple(self):
       self.assertAllClose(result, expected)
 
   def testSimple2(self):
-    # Currently not implemented for OpenCL
-    if test_util.is_sycl_enabled():
-      return
-
     indices = constant_op.constant([[1, 0], [1, 1]], dtype=dtypes.int32)
     updates = constant_op.constant([11., 12.], dtype=dtypes.float32)
     ref = variables.Variable(
@@ -184,10 +172,6 @@ def testSimple2(self):
       self.assertAllClose(result, expected)
 
   def testSimple3(self):
-    # Currently not implemented for OpenCL
-    if test_util.is_sycl_enabled():
-      return
-
     indices = constant_op.constant([[1]], dtype=dtypes.int32)
     updates = constant_op.constant([[11., 12.]], dtype=dtypes.float32)
     ref = variables.Variable(
@@ -420,10 +404,6 @@ def testGradientsRank3SliceUpdate(self):
       self.assertAllEqual(expected_grads, grads.eval())
 
   def testConcurrentUpdates(self):
-    # Currently not implemented for OpenCL
-    if test_util.is_sycl_enabled():
-      return
-
     num_updates = 10000
     update_values = np.random.rand(num_updates)
     ref = variables.Variable(np.zeros([2, 2]), dtype=dtypes.float64)