| #include "caffe2/operators/batch_matmul_op.h" |
| |
| #include "caffe2/core/context_gpu.h" |
| |
| namespace caffe2 { |
| |
| template <> |
| bool BatchMatMulOp<CUDAContext, DefaultEngine>::RunOnDevice() { |
| return DispatchHelper<TensorTypes<float, float16>>::call(this, Input(0)); |
| } |
| |
| REGISTER_CUDA_OPERATOR(BatchMatMul, BatchMatMulOp<CUDAContext>); |
| |
| #if CUDA_VERSION >= 9000 |
| |
| template <> |
| bool BatchMatMulOp<CUDAContext, TensorCoreEngine>::RunOnDevice() { |
| return DispatchHelper<TensorTypes<float, float16>>::call(this, Input(0)); |
| } |
| |
| REGISTER_CUDA_OPERATOR_WITH_ENGINE( |
| BatchMatMul, |
| TENSORCORE, |
| BatchMatMulOp<CUDAContext, TensorCoreEngine>); |
| #endif |
| |
| } // namespace caffe2 |
