remove boost::algorithm::ends_with ，boost macro and boost::lexical_ca…

…st apis (#34310) * replace boost::algorithm::ends_with with self define ends_with function * remove BOOST macro in certain operators * remove boost::lexical_cast * add test for string_helper * add more test case for string_helper * modify join_string func and test case * fix build_strategy_test failed bug * remove string_helper_test from parallel_UT_rule.py
PaddlePaddle · Aug 5, 2021 · bb7b4c0 · bb7b4c0
1 parent 911c859
commit bb7b4c0
Show file tree

Hide file tree

Showing 19 changed files with 333 additions and 226 deletions.
diff --git a/paddle/fluid/distributed/common/sparse_sharding_merge.h b/paddle/fluid/distributed/common/sparse_sharding_merge.h
@@ -21,7 +21,6 @@
 #include <vector>
 
 #include <ThreadPool.h>
-#include "boost/lexical_cast.hpp"
 #include "glog/logging.h"
 #include "paddle/fluid/distributed/common/utils.h"
 #include "paddle/fluid/framework/blocking_queue.h"
@@ -36,8 +35,6 @@ constexpr int Q_SIZE = 10000;
 constexpr int BUCKET = 10;
 constexpr char XEOF[] = "EOF";
 
-using boost::lexical_cast;
-
 inline double GetCurrentUS() {
   struct timeval time;
   gettimeofday(&time, NULL);
@@ -208,8 +205,10 @@ class ShardingMerge {
         for (int x = 0; x < embedding_dim; ++x) {
           float v = 0.0;
           try {
-            v = lexical_cast<float>(values_str[x]);
-          } catch (boost::bad_lexical_cast &e) {
+            v = std::stof(values_str[x]);
+          } catch (std::invalid_argument &e) {
+            VLOG(0) << " get unexpected line: " << line;
+          } catch (std::out_of_range &e) {
             VLOG(0) << " get unexpected line: " << line;
           }
           out->push_back(v);

diff --git a/paddle/fluid/distributed/index_dataset/index_wrapper.cc b/paddle/fluid/distributed/index_dataset/index_wrapper.cc
@@ -17,8 +17,6 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/framework/io/fs.h"
 
-#include <boost/algorithm/string.hpp>
-#include <boost/lexical_cast.hpp>
 #include "paddle/fluid/distributed/index_dataset/index_wrapper.h"
 
 namespace paddle {
@@ -65,7 +63,7 @@ int TreeIndex::Load(const std::string filename) {
     if (item.key() == ".tree_meta") {
       meta_.ParseFromString(item.value());
     } else {
-      auto code = boost::lexical_cast<uint64_t>(item.key());
+      auto code = std::stoull(item.key());
       IndexNode node;
       node.ParseFromString(item.value());
       PADDLE_ENFORCE_NE(node.id(), 0,

diff --git a/paddle/fluid/distributed/table/common_sparse_table.cc b/paddle/fluid/distributed/table/common_sparse_table.cc
@@ -15,7 +15,6 @@
 #include "paddle/fluid/distributed/table/common_sparse_table.h"
 #include <sstream>
 
-#include "boost/lexical_cast.hpp"
 #include "glog/logging.h"
 #include "paddle/fluid/platform/enforce.h"
 
@@ -50,8 +49,11 @@ void CommonSparseTable::ProcessALine(const std::vector<std::string>& columns,
       float v = 0.0;
 
       try {
-        v = lexical_cast<float>(va);
-      } catch (boost::bad_lexical_cast& e) {
+        v = std::stof(va);
+      } catch (std::invalid_argument& e) {
+        VLOG(0) << "id: " << id << " get unexpected value: " << va
+                << " and be reset to: 0.0";
+      } catch (std::out_of_range& e) {
         VLOG(0) << "id: " << id << " get unexpected value: " << va
                 << " and be reset to: 0.0";
       }
@@ -131,7 +133,7 @@ int64_t CommonSparseTable::LoadFromText(
 
   while (std::getline(file, line)) {
     auto values = paddle::string::split_string<std::string>(line, "\t");
-    auto id = lexical_cast<uint64_t>(values[0]);
+    auto id = std::stoull(values[0]);
 
     if (id % pserver_num != pserver_id) {
       VLOG(3) << "will not load " << values[0] << " from " << valuepath
@@ -150,10 +152,9 @@ int64_t CommonSparseTable::LoadFromText(
     VALUE* value_instant = block->GetValue(id);
 
     if (values.size() == 5) {
-      value_instant->count_ = lexical_cast<int>(values[1]);
-      value_instant->unseen_days_ = lexical_cast<int>(values[2]);
-      value_instant->is_entry_ =
-          static_cast<bool>(lexical_cast<int>(values[3]));
+      value_instant->count_ = std::stoi(values[1]);
+      value_instant->unseen_days_ = std::stoi(values[2]);
+      value_instant->is_entry_ = static_cast<bool>(std::stoi(values[3]));
     }
 
     std::vector<float*> block_values = block->Get(id, meta.names, meta.dims);

diff --git a/paddle/fluid/distributed/table/common_sparse_table.h b/paddle/fluid/distributed/table/common_sparse_table.h
@@ -33,7 +33,6 @@
 #include "paddle/fluid/string/string_helper.h"
 
 #define PSERVER_SAVE_SUFFIX ".shard"
-using boost::lexical_cast;
 
 namespace paddle {
 namespace distributed {

diff --git a/paddle/fluid/distributed/table/ssd_sparse_table.cc b/paddle/fluid/distributed/table/ssd_sparse_table.cc
@@ -310,7 +310,7 @@ int64_t SSDSparseTable::LoadFromText(
 
   while (std::getline(file, line)) {
     auto values = paddle::string::split_string<std::string>(line, "\t");
-    auto id = lexical_cast<uint64_t>(values[0]);
+    auto id = std::stoull(values[0]);
 
     if (id % pserver_num != pserver_id) {
       VLOG(3) << "will not load " << values[0] << " from " << valuepath
@@ -329,10 +329,9 @@ int64_t SSDSparseTable::LoadFromText(
     VALUE* value_instant = block->GetValue(id);
 
     if (values.size() == 5) {
-      value_instant->count_ = lexical_cast<int>(values[1]);
-      value_instant->unseen_days_ = lexical_cast<int>(values[2]);
-      value_instant->is_entry_ =
-          static_cast<bool>(lexical_cast<int>(values[3]));
+      value_instant->count_ = std::stoi(values[1]);
+      value_instant->unseen_days_ = std::stoi(values[2]);
+      value_instant->is_entry_ = static_cast<bool>(std::stoi(values[3]));
     }
 
     std::vector<float*> block_values = block->Get(id, meta.names, meta.dims);

diff --git a/paddle/fluid/framework/details/CMakeLists.txt b/paddle/fluid/framework/details/CMakeLists.txt
@@ -141,7 +141,7 @@ if(NOT APPLE AND NOT WIN32 AND (WITH_GPU OR WITH_ROCM))
 endif()
 cc_library(build_strategy SRCS build_strategy.cc DEPS pass_builder ${IR_PASS_DEPS})
 cc_test(build_strategy_test SRCS build_strategy_test.cc
-        DEPS build_strategy op_registry op_proto_maker graph)
+        DEPS build_strategy op_registry op_proto_maker graph string_helper)
 
 if (WITH_MKLDNN)
   target_link_libraries(build_strategy mkldnn_placement_pass)

diff --git a/paddle/fluid/framework/fleet/fleet_wrapper.cc b/paddle/fluid/framework/fleet/fleet_wrapper.cc
@@ -262,7 +262,7 @@ void FleetWrapper::HeterPushSparseVars(
     int64_t* ids = tensor->data<int64_t>();
     int slot = 0;
     if (dump_slot) {
-      slot = boost::lexical_cast<int>(sparse_key_names[i]);
+      slot = std::stoi(sparse_key_names[i]);
     }
     Variable* g_var = scope.FindVar(sparse_grad_names[i]);
     if (g_var == nullptr) {
@@ -915,12 +915,17 @@ void FleetWrapper::PushSparseVarsWithLabelAsync(
     int slot = 0;
     if (dump_slot) {
       try {
-        slot = boost::lexical_cast<int>(sparse_key_names[i]);
-      } catch (boost::bad_lexical_cast const& e) {
+        slot = std::stoi(sparse_key_names[i]);
+      } catch (std::invalid_argument const& e) {
         PADDLE_THROW(platform::errors::PreconditionNotMet(
             "sparse var's name: %s, doesn't support non-integer type name when "
             "dump_slot=True",
             sparse_key_names[i]));
+      } catch (std::out_of_range const& e) {
+        PADDLE_THROW(platform::errors::PreconditionNotMet(
+            "sparse var's name: %s, integer type name out of range when "
+            "dump_slot=True",
+            sparse_key_names[i]));
       }
     }
     Variable* g_var = scope.FindVar(sparse_grad_names[i]);
@@ -1121,7 +1126,7 @@ void FleetWrapper::PushSparseFromTensorWithLabelAsync(
         data[click_index] = static_cast<float>(fea_labels.at(input_idx));
       }
       if (dump_slot) {
-        int slot = boost::lexical_cast<int>(input_names[index]);
+        int slot = std::stoi(input_names[index]);
         data[0] = static_cast<float>(slot);
       }
       ++input_idx;

diff --git a/paddle/fluid/framework/ir/lock_free_optimize_pass.h b/paddle/fluid/framework/ir/lock_free_optimize_pass.h
@@ -17,10 +17,9 @@
 #include <string>
 #include <vector>
 
-#include <boost/algorithm/string/predicate.hpp>
-
 #include "paddle/fluid/framework/ir/graph.h"
 #include "paddle/fluid/framework/ir/pass.h"
+#include "paddle/fluid/string/string_helper.h"
 
 namespace paddle {
 namespace framework {
@@ -109,7 +108,7 @@ class LockFreeOptimizePass : public Pass {
                                 "Input argument node cannot be nullptr."));
 
     return node->NodeType() == Node::Type::kVariable &&
-           boost::algorithm::ends_with(node->Name(), name);
+           paddle::string::ends_with(node->Name(), name);
   }
 
   inline bool IsVarNameContains(ir::Node* node, const std::string& name) const {

diff --git a/paddle/fluid/operators/expand_as_op.h b/paddle/fluid/operators/expand_as_op.h
@@ -13,42 +13,12 @@ limitations under the License. */
 
 #include <vector>
 
-#include <boost/preprocessor/arithmetic/div.hpp>
-#include <boost/preprocessor/arithmetic/mod.hpp>
-#include <boost/preprocessor/comparison/greater.hpp>
-#include <boost/preprocessor/comparison/greater_equal.hpp>
-#include <boost/preprocessor/control/if.hpp>
-#include <boost/preprocessor/repetition/repeat.hpp>
 #include "paddle/fluid/framework/eigen.h"
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/operators/eigen/eigen_function.h"
 
 #define MAX_RANK_SUPPORTED 6
-// 1. BOOST_PP_REPEAT macro represents a fast horizontal repetition construct.
-//    Usage: BOOST_PP_REPEAT(count, macro, data).
-//    This macro expands to the sequence:
-//    macro(z, 0, data) macro(z, 1, data) ... macro(z, count - 1, data).
-// 2. As for our case, count = MAX_RANK_SUPPORTED(which is 6).
-//    So the range of n is 0-5(which is count-1).
-//    We want to generate case 1-6 instead of case 0-5.
-//    So we need to change n to n + 1.
-#define EXPAND_AS_TEMPLATE(z, n, data) \
-  case n + 1: {                        \
-    ExpandAs<n + 1>(context);          \
-    break;                             \
-  }
-#define REP_EXPAND_AS_TEMPLATE(n) BOOST_PP_REPEAT(n, EXPAND_AS_TEMPLATE, ~)
-#define COND(n) BOOST_PP_GREATER_EQUAL(n, BOOST_PP_MOD(n, MAX_RANK_SUPPORTED))
-#define EXPAND_AS_GRAD_CASE(n)                                           \
-  case n + 1: {                                                          \
-    ExpandAsBackward<n + 1>(context, reshape_dims_vec, reduce_dims_vec); \
-    break;                                                               \
-  }
-#define EXPAND_AS_GRAD_TEMPLATE(z, n, data) \
-  BOOST_PP_IF(COND(n), EXPAND_AS_GRAD_CASE(n), )
-#define REP_EXPAND_AS_GRAD_TEMPLATE(n) \
-  BOOST_PP_REPEAT(n, EXPAND_AS_GRAD_TEMPLATE, ~)
 
 namespace paddle {
 namespace operators {
@@ -67,7 +37,24 @@ class ExpandAsKernel : public framework::OpKernel<T> {
   void Compute(const framework::ExecutionContext& context) const override {
     auto rank = context.Input<Tensor>("X")->dims().size();
     switch (rank) {
-      REP_EXPAND_AS_TEMPLATE(MAX_RANK_SUPPORTED)
+      case 1:
+        ExpandAs<1>(context);
+        break;
+      case 2:
+        ExpandAs<2>(context);
+        break;
+      case 3:
+        ExpandAs<3>(context);
+        break;
+      case 4:
+        ExpandAs<4>(context);
+        break;
+      case 5:
+        ExpandAs<5>(context);
+        break;
+      case 6:
+        ExpandAs<6>(context);
+        break;
       default:
         PADDLE_THROW(platform::errors::InvalidArgument(
             "Only support tensor with rank being between 1 and 6. But received "
@@ -165,7 +152,24 @@ class ExpandAsGradKernel : public framework::OpKernel<T> {
                             "to %d, but the value received is %d.",
                             MAX_RANK_SUPPORTED, dims));
       switch (dims) {
-        REP_EXPAND_AS_GRAD_TEMPLATE(MAX_RANK_SUPPORTED)
+        case 1:
+          ExpandAsBackward<1>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 2:
+          ExpandAsBackward<2>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 3:
+          ExpandAsBackward<3>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 4:
+          ExpandAsBackward<4>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 5:
+          ExpandAsBackward<5>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 6:
+          ExpandAsBackward<6>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
         default:
           PADDLE_THROW(platform::errors::InvalidArgument(
               "Only support tensor with rank being between 1 and 6. But "

diff --git a/paddle/fluid/operators/expand_as_v2_op.h b/paddle/fluid/operators/expand_as_v2_op.h
@@ -14,42 +14,12 @@ limitations under the License. */
 #include <algorithm>
 #include <vector>
 
-#include <boost/preprocessor/arithmetic/div.hpp>
-#include <boost/preprocessor/arithmetic/mod.hpp>
-#include <boost/preprocessor/comparison/greater.hpp>
-#include <boost/preprocessor/comparison/greater_equal.hpp>
-#include <boost/preprocessor/control/if.hpp>
-#include <boost/preprocessor/repetition/repeat.hpp>
 #include "paddle/fluid/framework/eigen.h"
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/operators/eigen/eigen_function.h"
 
 #define MAX_RANK_SUPPORTED 6
-// 1. BOOST_PP_REPEAT macro represents a fast horizontal repetition construct.
-//    Usage: BOOST_PP_REPEAT(count, macro, data).
-//    This macro expands to the sequence:
-//    macro(z, 0, data) macro(z, 1, data) ... macro(z, count - 1, data).
-// 2. As for our case, count = MAX_RANK_SUPPORTED(which is 6).
-//    So the range of n is 0-5(which is count-1).
-//    We want to generate case 1-6 instead of case 0-5.
-//    So we need to change n to n + 1.
-#define EXPAND_AS_TEMPLATE(z, n, data) \
-  case n + 1: {                        \
-    ExpandAs<n + 1>(context);          \
-    break;                             \
-  }
-#define REP_EXPAND_AS_TEMPLATE(n) BOOST_PP_REPEAT(n, EXPAND_AS_TEMPLATE, ~)
-#define COND(n) BOOST_PP_GREATER_EQUAL(n, BOOST_PP_MOD(n, MAX_RANK_SUPPORTED))
-#define EXPAND_AS_GRAD_CASE(n)                                           \
-  case n + 1: {                                                          \
-    ExpandAsBackward<n + 1>(context, reshape_dims_vec, reduce_dims_vec); \
-    break;                                                               \
-  }
-#define EXPAND_AS_GRAD_TEMPLATE(z, n, data) \
-  BOOST_PP_IF(COND(n), EXPAND_AS_GRAD_CASE(n), )
-#define REP_EXPAND_AS_GRAD_TEMPLATE(n) \
-  BOOST_PP_REPEAT(n, EXPAND_AS_GRAD_TEMPLATE, ~)
 
 namespace paddle {
 namespace operators {
@@ -85,7 +55,26 @@ class ExpandAsV2Kernel : public framework::OpKernel<T> {
                           "expand_as_v2 op must be less than or equal to %d.",
                           target_rank, MAX_RANK_SUPPORTED));
 
-    switch (target_rank) { REP_EXPAND_AS_TEMPLATE(MAX_RANK_SUPPORTED) }
+    switch (target_rank) {
+      case 1:
+        ExpandAs<1>(context);
+        break;
+      case 2:
+        ExpandAs<2>(context);
+        break;
+      case 3:
+        ExpandAs<3>(context);
+        break;
+      case 4:
+        ExpandAs<4>(context);
+        break;
+      case 5:
+        ExpandAs<5>(context);
+        break;
+      case 6:
+        ExpandAs<6>(context);
+        break;
+    }
   }
 
  protected:
@@ -186,7 +175,24 @@ class ExpandAsV2GradKernel : public framework::OpKernel<T> {
                             "to %d, but the value received is %d.",
                             MAX_RANK_SUPPORTED, dims));
       switch (dims) {
-        REP_EXPAND_AS_GRAD_TEMPLATE(MAX_RANK_SUPPORTED)
+        case 1:
+          ExpandAsBackward<1>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 2:
+          ExpandAsBackward<2>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 3:
+          ExpandAsBackward<3>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 4:
+          ExpandAsBackward<4>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 5:
+          ExpandAsBackward<5>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
+        case 6:
+          ExpandAsBackward<6>(context, reshape_dims_vec, reduce_dims_vec);
+          break;
         default:
           PADDLE_THROW(platform::errors::InvalidArgument(
               "Only support tensor with rank being between 1 and 6. But "