Added benchmarks for type casting of float16

2 files changed, 4 insertions, 4 deletions

diff --git a/bench/tensors/tensor_benchmarks.h b/bench/tensors/tensor_benchmarks.h
index b208a401a..131d056b4 100644
--- a/bench/tensors/tensor_benchmarks.h
+++ b/bench/tensors/tensor_benchmarks.h
@@ -48,12 +48,12 @@ template <typename Device, typename T> class BenchmarkSuite {
     Eigen::array<TensorIndex, 2> sizes;
     sizes[0] = m_;
     sizes[1] = k_;
-    const TensorMap<Tensor<T, 2, 0, TensorIndex>, Eigen::Aligned> A(a_, sizes);
-    TensorMap<Tensor<int, 2, 0, TensorIndex>, Eigen::Aligned> B((int*)b_, sizes);
+    const TensorMap<Tensor<int, 2, 0, TensorIndex>, Eigen::Aligned> A((int*)a_, sizes);
+    TensorMap<Tensor<T, 2, 0, TensorIndex>, Eigen::Aligned> B(b_, sizes);
 
     StartBenchmarkTiming();
     for (int iter = 0; iter < num_iters; ++iter) {
-      B.device(device_) = A.template cast<int>();
+      B.device(device_) = A.template cast<T>();
     }
     // Record the number of values copied per second
     finalizeBenchmark(static_cast<int64_t>(m_) * k_ * num_iters);
diff --git a/bench/tensors/tensor_benchmarks_fp16_gpu.cu b/bench/tensors/tensor_benchmarks_fp16_gpu.cu
index d841bcdac..49f75472a 100644
--- a/bench/tensors/tensor_benchmarks_fp16_gpu.cu
+++ b/bench/tensors/tensor_benchmarks_fp16_gpu.cu
@@ -19,7 +19,7 @@
   BENCHMARK_RANGE(BM_##FUNC, 10, 5000);
 
 BM_FuncGPU(memcpy);
-//BM_FuncGPU(typeCasting);
+BM_FuncGPU(typeCasting);
 //BM_FuncGPU(random);
 BM_FuncGPU(slicing);
 BM_FuncGPU(rowChip);