提交 458b16f4 编写于 作者: J Jacek Czaja 提交者: jacek.czaja

Rebase of seqpool-max optimization

test=develop

- Added rough profiling

- Profiled maxpool itself

- First draft of max seqpool optimization (is_test added)

- Added unit tests to seqpool

- Cosmetic fixes

- Fix to UT of Seq pool

Disabled grad checking for sequence max pool when is_test is set to True

-Cosmetic fix to comment

test=develop

- Fix to GPU build

test=develop

- yet another GPU fix for sequence max pool

- Fix to comment

test=develop

- Change to API of sequence_pool

test=develop

- Yet another API spec change

test=develop
上级 79da263b
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_incubate/lite delete_paddle_tiny_install delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/123malin/netifaces github/fork/123malin/tdm_abacus github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LielinJiang/visual-dl-cb github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/fix-example-code-for-hapi-Model github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/fix_ps_profiler github/fork/MrChengmo/update_ps_heter github/fork/PWhiddy/patch-1 github/fork/Shixiaowei02/dev/save_load_upgrade github/fork/TCChenlong/fix_hapi github/fork/TCChenlong/fix_inden github/fork/Thunderbrook/xpu_slice github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_2 github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/XieYunshen/timeout_20S_ut github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chalsliu/set_timeout github/fork/chen-zhiyu/develop github/fork/chenwhql/ci/try_to_find_test_buffer_shared_memory_reuse_pass_error github/fork/chenwhql/dygraph/remove_scale_loss_and_apply_collective_grads github/fork/chenwhql/saveload/add_get_inference_program github/fork/chenwhql/saveload/remove_save_load_config github/fork/cryoco/pass-compatibility-trt github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/hbwx24/error_msg/cuda_kernel_error_msg github/fork/heavengate/cherry_yolo_box github/fork/heavengate/update_yolo_box github/fork/iclementine/rnn_fix github/fork/iducn/testestse github/fork/jczaja/prv-25537-fix github/fork/jeff41404/release/1.8 github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/lfchener/sample_code github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/luotao1/profiler_ut github/fork/mapingshuo/add_wait github/fork/mapingshuo/doc_2.0 github/fork/mapingshuo/zero-0.5 github/fork/miraiwk/dev github/fork/pangyoki/add-Categorical-class-branch github/fork/pangyoki/add-multinomial-op-branch github/fork/pangyoki/fix-test_distritbution-CI github/fork/qjing666/doublegrad github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/sandyhouse/pipeline_exe_run github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_errors_fix github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0924 github/fork/smallv0221/yxp0925 github/fork/swtkiwi/del-matplotlib github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/bert_fuse github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_coverage_build_sh github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wojtuss/wojtuss/fusion_gru_quantization github/fork/wojtuss/wojtuss/quantization-with-shift github/fork/wzzju/fix_err_info github/fork/wzzju/pure_fp16 github/fork/xiemoyuan/op_error_message github/fork/xiemoyuan/optimize_error_message github/fork/yaoxuefeng6/fix_doc github/fork/yaoxuefeng6/mod_dataset_v2 github/fork/yongqiangma/lod github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/conv_filter_grad github/fork/zhangting2020/is_compile_with_cuda github/fork/zhangting2020/place_doc github/fork/zhangting2020/program github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhhsplendid/refine_api2_test github/fork/zhhsplendid/refine_api_test_ptb_lm github/fork/zhhsplendid/refine_api_test_resnet github/fork/zhhsplendid/refine_api_test_simnet github/fork/zhiqiu/dev/refine_initializer github/fork/zhiqiu/dev/remove_inplace_argument github/fork/zlsh80826/nvinfer_plugin_var_len_cuda11 improve_sccache incubate/infrt incubate/lite inplace_addto make_flag_adding_easier move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc numel paddle_tiny_install paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 pten_tensor_refactor release/1.2 release/1.3 release/1.4 release/1.5 release/1.6 release/1.7 release/1.8 release/2.0 release/2.0-alpha release/2.0-beta release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 release/lite-0.1 revert-24981-add_device_attr_for_regulization revert-26856-strategy_example2 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment rocm_dev_0217 support_weight_transpose test_benchmark_ci test_feature_precision_test_c test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0 v2.0.0-beta0 v2.0.0-alpha0 v1.8.5 v1.8.4 v1.8.3 v1.8.2 v1.8.1 v1.8.0 v1.7.2 v1.7.1 v1.7.0 v1.6.3 v1.6.2 v1.6.1 v1.6.0 v1.6.0-rc0 v1.5.2 v1.5.1 v1.5.0 v1.4.1 v1.4.0 v1.3.2 v1.3.1 v1.3.0 v1.2.1 v1.2.0 lite-v0.1
无相关合并请求
......@@ -64,7 +64,7 @@ paddle.fluid.layers.chunk_eval ArgSpec(args=['input', 'label', 'chunk_scheme', '
paddle.fluid.layers.sequence_conv ArgSpec(args=['input', 'num_filters', 'filter_size', 'filter_stride', 'padding', 'bias_attr', 'param_attr', 'act', 'name'], varargs=None, keywords=None, defaults=(3, 1, None, None, None, None, None))
paddle.fluid.layers.conv2d ArgSpec(args=['input', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None, None))
paddle.fluid.layers.conv3d ArgSpec(args=['input', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None, None))
paddle.fluid.layers.sequence_pool ArgSpec(args=['input', 'pool_type'], varargs=None, keywords=None, defaults=None)
paddle.fluid.layers.sequence_pool ArgSpec(args=['input', 'pool_type', 'is_test'], varargs=None, keywords=None, defaults=(False,))
paddle.fluid.layers.sequence_softmax ArgSpec(args=['input', 'use_cudnn', 'name'], varargs=None, keywords=None, defaults=(False, None))
paddle.fluid.layers.softmax ArgSpec(args=['input', 'use_cudnn', 'name'], varargs=None, keywords=None, defaults=(True, None))
paddle.fluid.layers.pool2d ArgSpec(args=['input', 'pool_size', 'pool_type', 'pool_stride', 'pool_padding', 'global_pooling', 'use_cudnn', 'ceil_mode', 'name'], varargs=None, keywords=None, defaults=(-1, 'max', 1, 0, False, True, False, None))
......
......@@ -31,7 +31,7 @@ template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
template <typename T>
template <typename T, bool is_test>
class MaxSeqPoolFunctor {
public:
void operator()(const platform::CPUDeviceContext& context,
......@@ -70,7 +70,41 @@ class MaxSeqPoolFunctor {
}
}
};
// Instantisation of Max Sequence Pooling for test phase eg. no need to fill
// index buffer
template <typename T>
class MaxSeqPoolFunctor<T, true> {
public:
void operator()(const platform::CPUDeviceContext& context,
const framework::LoDTensor& input, framework::Tensor* output,
framework::Tensor* index) {
auto in_dims = input.dims();
auto out_dims = output->dims();
PADDLE_ENFORCE_GT(in_dims.size(), 1);
PADDLE_ENFORCE_GT(out_dims.size(), 1);
for (int64_t i = 1; i < in_dims.size(); ++i) {
PADDLE_ENFORCE_EQ(in_dims[i], out_dims[i]);
}
auto starts = input.lod()[0];
const T* in_data = input.data<T>();
T* out_data = output->data<T>();
int64_t num_seq = out_dims[0];
int64_t dim = output->numel() / num_seq;
for (int64_t i = 0; i < num_seq; ++i) {
std::memcpy(&out_data[i * dim], &in_data[starts[i] * dim],
dim * sizeof(T));
for (size_t j = starts[i] + 1; j < starts[i + 1]; ++j) {
for (int64_t k = 0; k < dim; ++k) {
if (in_data[j * dim + k] > out_data[i * dim + k]) {
out_data[i * dim + k] = in_data[j * dim + k];
}
}
}
}
}
};
template <typename T>
class MaxSeqPoolGradFunctor {
public:
......@@ -188,11 +222,16 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
/* max pool has index output */
void operator()(const platform::CPUDeviceContext& context,
const std::string pooltype, const framework::LoDTensor& input,
framework::Tensor* output,
framework::Tensor* output, bool is_test,
framework::Tensor* index = nullptr) {
if (pooltype == "MAX") {
math::MaxSeqPoolFunctor<T> max_pool;
max_pool(context, input, output, index);
if (is_test) {
math::MaxSeqPoolFunctor<T, true> max_pool;
max_pool(context, input, output, index);
} else {
math::MaxSeqPoolFunctor<T, false> max_pool;
max_pool(context, input, output, index);
}
return;
}
if (pooltype == "LAST") {
......@@ -200,6 +239,7 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
last_pool(context, input, output);
return;
}
if (pooltype == "FIRST") {
math::FirstSeqPoolFunctor<T> first_pool;
first_pool(context, input, output);
......
......@@ -133,7 +133,7 @@ class SequencePoolFunctor<platform::CUDADeviceContext, T> {
public:
void operator()(const platform::CUDADeviceContext& context,
const std::string pooltype, const framework::LoDTensor& input,
framework::Tensor* output,
framework::Tensor* output, bool is_test,
framework::Tensor* index = nullptr) {
auto& lod = input.lod()[0];
const size_t item_dim = output->numel() / output->dims()[0];
......
......@@ -28,7 +28,7 @@ class SequencePoolFunctor {
/* max pool has index output */
void operator()(const DeviceContext& context, const std::string pooltype,
const framework::LoDTensor& input, framework::Tensor* output,
framework::Tensor* index = nullptr);
bool is_test = false, framework::Tensor* index = nullptr);
};
template <typename DeviceContext, typename T>
......
......@@ -47,6 +47,7 @@ class SequencePoolOpMaker : public framework::OpProtoAndCheckerMaker {
"(Tensor<int>) This tensor is used for the sequence max-pooling "
"to record the max indexes.")
.AsIntermediate();
AddAttr<bool>("is_test", "").SetDefault(false);
AddAttr<std::string>(
"pooltype",
"(string, default 'AVERAGE') the pooling pooltype of SequencePoolOp.")
......
......@@ -32,10 +32,6 @@ class SequencePoolKernel : public framework::OpKernel<T> {
auto* in = context.Input<LoDTensor>("X");
auto* out = context.Output<Tensor>("Out");
std::string pooltype = context.Attr<std::string>("pooltype");
Tensor* index = nullptr;
if (pooltype == "MAX") {
index = context.Output<Tensor>("MaxIndex");
}
auto dims = in->dims();
auto lod = in->lod();
......@@ -48,13 +44,22 @@ class SequencePoolKernel : public framework::OpKernel<T> {
dims[0] = lod[0].size() - 1;
out->Resize({dims});
out->mutable_data<T>(context.GetPlace());
if (pooltype == "MAX") {
Tensor* index = nullptr;
const bool is_test = context.Attr<bool>("is_test");
// Do not create index buffer for inference (is_test) mode
// TODO(jczaja): Skip index buffer creation for other devices eg. GPU
if (pooltype == "MAX" &&
(is_test == false ||
platform::is_cpu_place(context.GetPlace()) == false)) {
index = context.Output<Tensor>("MaxIndex");
index->Resize({dims});
index->mutable_data<int>(context.GetPlace());
}
math::SequencePoolFunctor<DeviceContext, T> pool;
pool(context.template device_context<DeviceContext>(), pooltype, *in, out,
index);
is_test, index);
}
};
......
......@@ -1823,7 +1823,7 @@ def conv3d(input,
return helper.append_activation(pre_act)
def sequence_pool(input, pool_type):
def sequence_pool(input, pool_type, is_test=False):
"""
This function add the operator for sequence pooling.
It pools features of all time-steps of each instance, and is applied
......@@ -1860,6 +1860,7 @@ def sequence_pool(input, pool_type):
input(variable): The input variable which is a LoDTensor.
pool_type (string): The pooling type of sequence_pool.
It supports average, sum, sqrt and max.
is_test(bool, Default False): Used distinguish training from scoring mode.
Returns:
The sequence pooling variable which is a Tensor.
......@@ -1887,7 +1888,8 @@ def sequence_pool(input, pool_type):
inputs={"X": input},
outputs={"Out": pool_out,
"MaxIndex": max_index},
attrs={"pooltype": pool_type.upper()})
attrs={"pooltype": pool_type.upper(),
"is_test": is_test})
# when pool_type is max, variable max_index is initialized,
# so we stop the gradient explicitly here
......
......@@ -184,6 +184,20 @@ class TestSeqMaxPool2D(TestSeqAvgPool2D):
out[i] = np.reshape(np.amax(sub_x, axis=0), (3, 11))
class TestSeqMaxPool2DInference(TestSeqMaxPool2D):
def compute(self, x, offset, out):
self.attrs = {'pooltype': "MAX", 'is_test': True}
for i in range(len(offset[0]) - 1):
sub_x = np.reshape(x[offset[0][i]:offset[0][i + 1], :],
(-1, 3 * 11))
out[i] = np.reshape(np.amax(sub_x, axis=0), (3, 11))
def test_check_grad(self):
"""Grad computation does not apply to Sequence MAX
Pool executed when is_test is true """
return
class TestSeqLastPool2D(TestSeqAvgPool2D):
def compute(self, x, offset, out):
self.attrs = {'pooltype': "LAST"}
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册
反馈
建议
客服 返回
顶部