X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/d1fc2db12e159461fa3c6a1fb60a1f8e395e91f8..321102577020b194dfb7ba89e48687952816849e:/src/mc/inspect/ObjectInformation.cpp

diff --git a/src/mc/inspect/ObjectInformation.cpp b/src/mc/inspect/ObjectInformation.cpp
index 64adf52a8c..f7edbc5d72 100644
--- a/src/mc/inspect/ObjectInformation.cpp
+++ b/src/mc/inspect/ObjectInformation.cpp
@@ -1,8 +1,9 @@
-/* Copyright (c) 2014-2019. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2014-2022. The SimGrid Team. All rights reserved.          */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
+#include <algorithm>
 #include <cstdint>
 #include <sys/mman.h> // PROT_READ and friends
 #include <vector>
@@ -16,24 +17,21 @@
 namespace simgrid {
 namespace mc {
 
-/* For an executable object, addresses are virtual address
- * (there is no offset) i.e.
- * \f$\text{virtual address} = \{dwarf address}\f$
+/* For an executable object, addresses are virtual address (there is no offset) i.e.
+ *  \f$\text{virtual address} = \{dwarf address}\f$
  *
- * For a shared object, the addresses are offset from the begining
- * of the shared object (the base address of the mapped shared
- * object must be used as offset
+ * For a shared object, the addresses are offset from the beginning of the shared object (the base address of the
+ * mapped shared object must be used as offset
  * i.e. \f$\text{virtual address} = \text{shared object base address}
  *             + \text{dwarf address}\f$.
  */
 void* ObjectInformation::base_address() const
 {
-  // For an executable (more precisely for a ET_EXEC) the base it 0:
+  // For an executable (more precisely for an ET_EXEC) the base it 0:
   if (this->executable())
     return nullptr;
 
-  // For an a shared-object (ET_DYN, including position-independent executables)
-  // the base address is its lowest address:
+  // For an a shared-object (ET_DYN, including position-independent executables) the base address is its lowest address:
   void* result = this->start_exec;
   if (this->start_rw != nullptr && result > (void*)this->start_rw)
     result = this->start_rw;
@@ -42,8 +40,10 @@ void* ObjectInformation::base_address() const
   return result;
 }
 
-simgrid::mc::Frame* ObjectInformation::find_function(const void* ip) const
+Frame* ObjectInformation::find_function(const void* ip)
 {
+  ensure_dwarf_loaded();
+
   /* This is implemented by binary search on a sorted array.
    *
    * We do quite a lot of those so we want this to be cache efficient.
@@ -52,88 +52,44 @@ simgrid::mc::Frame* ObjectInformation::find_function(const void* ip) const
    * during the binary search (only at the end) so it is not included
    * in the index entry. We could use parallel arrays as well.
    *
-   * We cannot really use the std:: algorithm for this.
-   * We could use std::binary_search by including the high_pc inside
-   * the FunctionIndexEntry.
+   * Note the usage of reverse iterators to match the correct interval.
    */
-  const simgrid::mc::FunctionIndexEntry* base = this->functions_index.data();
-  int i                                       = 0;
-  int j                                       = this->functions_index.size() - 1;
-  while (j >= i) {
-    int k = i + ((j - i) / 2);
-
-    /* In most of the search, we do not dereference the base[k].function.
-     * This way the memory accesses are located in the base[k] array. */
-    if (ip < base[k].low_pc)
-      j = k - 1;
-    else if (k < j && ip >= base[k + 1].low_pc)
-      i = k + 1;
-
-    /* At this point, the search is over.
-     * Either we have found the correct function or we do not know
-     * any function corresponding to this instruction address.
-     * Only at the point do we dereference the function pointer. */
-    else if ((std::uint64_t)ip < base[k].function->range.end())
-      return base[k].function;
-    else
-      return nullptr;
-  }
-  return nullptr;
+  auto pos = std::lower_bound(this->functions_index.rbegin(), this->functions_index.rend(), ip,
+                              [](auto const& func, auto const* addr) { return func.low_pc > addr; });
+
+  /* At this point, the search is over.
+   * Either we have found the correct function or we do not know
+   * any function corresponding to this instruction address.
+   * Only at the point do we dereference the function pointer. */
+  return (pos != this->functions_index.rend() && reinterpret_cast<std::uint64_t>(ip) < pos->function->range.end())
+             ? pos->function
+             : nullptr;
 }
 
-const simgrid::mc::Variable* ObjectInformation::find_variable(const char* name) const
+const Variable* ObjectInformation::find_variable(const char* var_name)
 {
-  for (simgrid::mc::Variable const& variable : this->global_variables) {
-    if (variable.name == name)
-      return &variable;
-  }
-  return nullptr;
+  ensure_dwarf_loaded();
+
+  auto pos = std::lower_bound(this->global_variables.begin(), this->global_variables.end(), var_name,
+                              [](auto const& var, const char* name) { return var.name < name; });
+  return (pos != this->global_variables.end() && pos->name == var_name) ? &(*pos) : nullptr;
 }
 
-void ObjectInformation::remove_global_variable(const char* name)
+void ObjectInformation::remove_global_variable(const char* var_name)
 {
-  typedef std::vector<Variable>::size_type size_type;
-
-  if (this->global_variables.empty())
-    return;
-
   // Binary search:
-  size_type first = 0;
-  size_type last  = this->global_variables.size() - 1;
-
-  while (first <= last) {
-    size_type cursor                   = first + (last - first) / 2;
-    simgrid::mc::Variable& current_var = this->global_variables[cursor];
-    int cmp                            = current_var.name.compare(name);
-
-    if (cmp == 0) {
-
-      // Find the whole range:
-      first = cursor;
-      while (first != 0 && this->global_variables[first - 1].name == name)
-        first--;
-      size_type size = this->global_variables.size();
-      last           = cursor;
-      while (last != size - 1 && this->global_variables[last + 1].name == name)
-        last++;
-
-      // Remove the whole range:
-      this->global_variables.erase(this->global_variables.begin() + first, this->global_variables.begin() + last + 1);
-
-      return;
-    } else if (cmp < 0)
-      first = cursor + 1;
-    else if (cursor != 0)
-      last = cursor - 1;
-    else
-      break;
-  }
+  auto pos1 = std::lower_bound(this->global_variables.begin(), this->global_variables.end(), var_name,
+                               [](auto const& var, const char* name) { return var.name < name; });
+  // Find the whole range:
+  auto pos2 = std::upper_bound(pos1, this->global_variables.end(), var_name,
+                               [](const char* name, auto const& var) { return name < var.name; });
+  // Remove the whole range:
+  this->global_variables.erase(pos1, pos2);
 }
 
 /** Ignore a local variable in a scope
  *
- *  Ignore all instances of variables with a given name in
- *  any (possibly inlined) subprogram with a given namespaced
+ *  Ignore all instances of variables with a given name in any (possibly inlined) subprogram with a given namespaced
  *  name.
  *
  *  @param var_name        Name of the local variable to ignore
@@ -141,61 +97,45 @@ void ObjectInformation::remove_global_variable(const char* name)
  *  @param subprogram      (possibly inlined) Subprogram of the scope current scope
  *  @param scope           Current scope
  */
-static void remove_local_variable(simgrid::mc::Frame& scope, const char* var_name, const char* subprogram_name,
-                                  simgrid::mc::Frame const& subprogram)
+static void remove_local_variable(Frame& scope, const char* var_name, const char* subprogram_name,
+                                  Frame const& subprogram)
 {
-  typedef std::vector<Variable>::size_type size_type;
-
   // If the current subprogram matches the given name:
-  if ((subprogram_name == nullptr || (not subprogram.name.empty() && subprogram.name == subprogram_name)) &&
-      not scope.variables.empty()) {
-
+  if (subprogram_name == nullptr || (not subprogram.name.empty() && subprogram.name == subprogram_name)) {
     // Try to find the variable and remove it:
-    size_type start = 0;
-    size_type end   = scope.variables.size() - 1;
 
     // Binary search:
-    while (start <= end) {
-      size_type cursor                   = start + (end - start) / 2;
-      simgrid::mc::Variable& current_var = scope.variables[cursor];
-      int compare                        = current_var.name.compare(var_name);
-      if (compare == 0) {
-        // Variable found, remove it:
-        scope.variables.erase(scope.variables.begin() + cursor);
-        break;
-      } else if (compare < 0)
-        start = cursor + 1;
-      else if (cursor != 0)
-        end = cursor - 1;
-      else
-        break;
+    auto pos = std::lower_bound(scope.variables.begin(), scope.variables.end(), var_name,
+                                [](auto const& var, const char* name) { return var.name < name; });
+    if (pos != scope.variables.end() && pos->name == var_name) {
+      // Variable found, remove it:
+      scope.variables.erase(pos);
     }
   }
 
   // And recursive processing in nested scopes:
-  for (simgrid::mc::Frame& nested_scope : scope.scopes) {
+  for (Frame& nested_scope : scope.scopes) {
     // The new scope may be an inlined subroutine, in this case we want to use its
     // namespaced name in recursive calls:
-    simgrid::mc::Frame const& nested_subprogram =
-        nested_scope.tag == DW_TAG_inlined_subroutine ? nested_scope : subprogram;
+    Frame const& nested_subprogram = nested_scope.tag == DW_TAG_inlined_subroutine ? nested_scope : subprogram;
     remove_local_variable(nested_scope, var_name, subprogram_name, nested_subprogram);
   }
 }
 
 void ObjectInformation::remove_local_variable(const char* var_name, const char* subprogram_name)
 {
-  for (auto& entry : this->subprograms)
-    simgrid::mc::remove_local_variable(entry.second, var_name, subprogram_name, entry.second);
+  for (auto& [_, entry] : this->subprograms)
+    mc::remove_local_variable(entry, var_name, subprogram_name, entry);
 }
 
 /** @brief Fills the position of the segments (executable, read-only, read/write) */
 // TODO, use the ELF segment information for more robustness
-void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::mc::ObjectInformation* result)
+void find_object_address(std::vector<xbt::VmMap> const& maps, ObjectInformation* result)
 {
   const int PROT_RW = PROT_READ | PROT_WRITE;
   const int PROT_RX = PROT_READ | PROT_EXEC;
 
-  std::string name = simgrid::xbt::Path(result->file_name).get_base_name();
+  std::string name = xbt::Path(result->file_name).get_base_name();
 
   for (size_t i = 0; i < maps.size(); ++i) {
     simgrid::xbt::VmMap const& reg = maps[i];
@@ -236,7 +176,7 @@ void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::
     else if (reg.prot == PROT_READ) {
       xbt_assert(not result->start_ro,
                  "Multiple read-only segments for %s, not supported. Compiling with the following may help: "
-		 "-Wl,-znorelro -Wl,-znoseparate-code",
+                 "-g -Wl,-znorelro -Wl,-znoseparate-code",
                  maps[i].pathname.c_str());
       result->start_ro = (char*)reg.start_addr;
       result->end_ro   = (char*)reg.end_addr;