diff --git a/paddle/fluid/operators/fill_constant_batch_size_like_op.cc b/paddle/fluid/operators/fill_constant_batch_size_like_op.cc index 087463df91d18985707f3d6e6748cd924da03af9..ccd4a3628783780c5b8cd76c97c54dcac49539e5 100644 --- a/paddle/fluid/operators/fill_constant_batch_size_like_op.cc +++ b/paddle/fluid/operators/fill_constant_batch_size_like_op.cc @@ -38,6 +38,8 @@ class FillConstantBatchSizeLikeOpMaker : public BatchSizeLikeOpMaker { .SetDefault(framework::proto::VarType::FP32); AddAttr("value", "default 0. The value to be filled") .SetDefault(0.0f); + AddAttr("str_value", "default empty. The value to be filled") + .SetDefault(""); AddAttr("force_cpu", "(bool, default false) Force fill output variable to cpu " "memory. Otherwise, fill output variable to the running " diff --git a/paddle/fluid/operators/fill_constant_batch_size_like_op.h b/paddle/fluid/operators/fill_constant_batch_size_like_op.h index f915f37feab5df29506b347f89eefd40082b4c24..e8a35d22277076ee355f3a489056f49beb9ace80 100644 --- a/paddle/fluid/operators/fill_constant_batch_size_like_op.h +++ b/paddle/fluid/operators/fill_constant_batch_size_like_op.h @@ -13,6 +13,7 @@ See the License for the specific language governing permissions and limitations under the License. */ #pragma once +#include #include "paddle/fluid/framework/op_registry.h" #include "paddle/fluid/operators/math/math_function.h" @@ -22,14 +23,15 @@ namespace operators { template class FillConstantBatchSizeLikeOpKernel : public framework::OpKernel { public: - void Compute(const framework::ExecutionContext& ctx) const override { + void Compute(const framework::ExecutionContext &ctx) const override { auto data_type = static_cast(ctx.Attr("dtype")); - auto value = ctx.Attr("value"); + auto float_value = ctx.Attr("value"); + auto str_value = ctx.Attr("str_value"); auto force_cpu = ctx.Attr("force_cpu"); - auto* out = ctx.Output("Out"); - auto* in = ctx.Input("Input"); + auto *out = ctx.Output("Out"); + auto *in = ctx.Input("Input"); if (in->lod().size() && ctx.Attr("input_dim_idx") == 0) { // set the correct batch size for the LoDTensor. auto odims = out->dims(); @@ -38,15 +40,39 @@ class FillConstantBatchSizeLikeOpKernel : public framework::OpKernel { out->mutable_data(odims, ctx.GetPlace()); } - if (force_cpu) { - out->mutable_data(platform::CPUPlace(), data_type); + T value; + if (str_value.empty()) { + value = static_cast(float_value); } else { - out->mutable_data(ctx.GetPlace(), data_type); + std::stringstream convert_stream(str_value); + if (std::is_same::value) { + int64_t tmp_value; + convert_stream >> tmp_value; + value = static_cast(tmp_value); + } else { + double tmp_value; + convert_stream >> tmp_value; + value = static_cast(tmp_value); + } } - platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance(); - auto& dev_ctx = *pool.Get(ctx.GetPlace()); - math::set_constant(dev_ctx, out, value); + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto &dev_ctx = *pool.Get(ctx.GetPlace()); + bool cpu_place = force_cpu || ctx.GetPlace() == platform::CPUPlace(); + if (cpu_place) { + math::SetConstant functor; + out->mutable_data(platform::CPUPlace(), data_type); + functor(reinterpret_cast(dev_ctx), + out, static_cast(value)); + } +#ifdef PADDLE_WITH_CUDA + if (!cpu_place) { + math::SetConstant functor; + out->mutable_data(ctx.GetPlace(), data_type); + functor(reinterpret_cast(dev_ctx), + out, static_cast(value)); + } +#endif } }; diff --git a/python/paddle/fluid/layers/tensor.py b/python/paddle/fluid/layers/tensor.py index c4f247236286e4d74a37f67def80ab38dc7c46c8..cb2913150f8f33daf9fcfb987c7a4a3bfaad45f4 100644 --- a/python/paddle/fluid/layers/tensor.py +++ b/python/paddle/fluid/layers/tensor.py @@ -668,18 +668,23 @@ def fill_constant_batch_size_like(input, """ helper = LayerHelper("fill_constant_batch_size_like", **locals()) out = helper.create_variable_for_type_inference(dtype=dtype) + attrs = { + 'shape': shape, + 'dtype': out.dtype, + 'value': float(value), + 'input_dim_idx': input_dim_idx, + 'output_dim_idx': output_dim_idx, + 'force_cpu': force_cpu or force_init_on_cpu() + } + if convert_dtype(dtype) in ['int64', 'int32']: + attrs['str_value'] = str(int(value)) + else: + attrs['str_value'] = str(float(value)) helper.append_op( type='fill_constant_batch_size_like', inputs={'Input': input}, outputs={'Out': [out]}, - attrs={ - 'shape': shape, - 'dtype': out.dtype, - 'value': float(value), - 'input_dim_idx': input_dim_idx, - 'output_dim_idx': output_dim_idx, - 'force_cpu': force_cpu or force_init_on_cpu() - }) + attrs=attrs) out.stop_gradient = True return out diff --git a/python/paddle/fluid/tests/unittests/test_fill_constant_batch_size_like_op.py b/python/paddle/fluid/tests/unittests/test_fill_constant_batch_size_like_op.py index fdc8a118e56f4473da5ed60169daebec14c7c33c..273d7e070c0ef555485aa3d30063707ee5311f8d 100644 --- a/python/paddle/fluid/tests/unittests/test_fill_constant_batch_size_like_op.py +++ b/python/paddle/fluid/tests/unittests/test_fill_constant_batch_size_like_op.py @@ -16,6 +16,7 @@ from __future__ import print_function import unittest import numpy as np +import paddle.fluid as fluid from op_test import OpTest @@ -52,6 +53,20 @@ class TestFillConstantBatchSizeLikeWhenSecondDimIsBatchSize(OpTest): self.check_output() +class TestFillConstantBatchSizeLikeInt64(OpTest): + def setUp(self): + self.op_type = "fill_constant_batch_size_like" + self.inputs = {'Input': np.random.random((219, 232)).astype("int64")} + self.attrs = {'value': 5894589485094, 'shape': [-1, 132, 7]} + + out = np.random.random((219, 132, 7)).astype("int64") + out.fill(5894589485094) + self.outputs = {'Out': out} + + def test_check_output(self): + self.check_output() + + class TestFillConstantBatchSizeLikeWithLoDTensor(OpTest): def setUp(self): self.op_type = "fill_constant_batch_size_like" @@ -74,5 +89,20 @@ class TestFillConstantBatchSizeLikeWithLoDTensor(OpTest): self.check_output() +# Test python API +class TestFillConstantBatchSizeLikeAPI(OpTest): + def test_api(self): + like = fluid.layers.fill_constant( + shape=[1, 200], value=10, dtype='int64') + out = fluid.layers.fill_constant_batch_size_like( + input=like, shape=[2, 300], value=1315454564656, dtype='int64') + exe = fluid.Executor(place=fluid.CPUPlace()) + res, = exe.run(fluid.default_main_program(), fetch_list=[out]) + + assert np.array_equal( + res[0], np.full( + [300], 1315454564656, dtype="int64")) + + if __name__ == "__main__": unittest.main() diff --git a/python/paddle/fluid/tests/unittests/test_layers.py b/python/paddle/fluid/tests/unittests/test_layers.py index ca7efebfb79943127d536e83502ef81c585e3b0c..a06cefef5cbe3ff43674199896cc410c029f983e 100644 --- a/python/paddle/fluid/tests/unittests/test_layers.py +++ b/python/paddle/fluid/tests/unittests/test_layers.py @@ -2682,6 +2682,14 @@ class TestBook(LayerTest): x, axes=axes, starts=starts, ends=ends, strides=strides) return out + def test_fill_constant_batch_size_like(self): + with self.static_graph(): + like = fluid.layers.fill_constant( + shape=[1, 200], value=10, dtype='int64') + out = layers.fill_constant_batch_size_like( + input=like, shape=[2, 3300], value=1315454564656, dtype='int64') + return out + def test_psroi_pool(self): # TODO(minqiyang): dygraph do not support lod now with self.static_graph():