Add malloc/new profiling on linux

src/LinuxMallocProfiling.cpp

Index: src/LinuxMallocProfiling.cpp
diff --git a/src/LinuxMallocProfiling.cpp b/src/LinuxMallocProfiling.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..25782cde7574749ed34c64ec161975d04e87cba6
--- /dev/null
+++ b/src/LinuxMallocProfiling.cpp
@@ -0,0 +1,92 @@
+//===- subzero/src/LinuxMallocProfiling.cpp - malloc/new tracing  ---------===//
+//
+//                        The Subzero Code Generator
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief malloc/new/...caller tracing.
+///
+//===----------------------------------------------------------------------===//
+
+#include "LinuxMallocProfiling.h"
+
+#ifdef ALLOW_LINUX_MALLOC_PROFILE

[Jim Stichnoth, 2016/03/15 15:36:50]

I don't like all the #ifdef here.

I think it would be better to do something like this:

#ifdef ALLOW_LINUX_MALLOC_PROFILE
... full implementation ...
#else // !ALLOW_LINUX_MALLOC_PROFILE
... minimal skeleton implementation
#endif // !ALLOW_LINUX_MALLOC_PROFILE

[sehr, 2016/03/17 16:47:14]

Done.

+#include <dlfcn.h>
+#include <malloc.h>
+#include <iostream>

[Jim Stichnoth, 2016/03/15 15:36:50]

LLVM sez iostream is forbidden.

http://llvm.org/docs/CodingStandards.html#include-iostream-is-forbidden

Can you use GlobalContext::StrDump instead?

[sehr, 2016/03/17 16:47:14]

I plumbed through the Ostream that was already used in this file.  PTAL.

+#include <unordered_map>
+
+extern "C" void *__libc_malloc(size_t size);
+
+namespace {
+// The Callers structure allocates memory, which would perturb the tracing.
+// InAllocatorFunction is true when we are tracing a new/malloc/...
+bool InAllocatorFunction = false;
+
+// Keep track of the number of times a particular call site address invoked
+// an allocator. NOTE: this is not thread safe, so the user must invoke with
+// --threads=0 to enable profilin.

[John, 2016/03/15 15:19:06]

profiling

[sehr, 2016/03/17 16:47:14]

Done.

+using MallocMap = std::unordered_map<void*, uint64_t>;
+MallocMap *Callers;
+
+void *internalAllocator(size_t size, void *caller) {
+  if (Callers != nullptr && !InAllocatorFunction) {
+    InAllocatorFunction = true;
+    ++(*Callers)[caller];
+    InAllocatorFunction = false;
+  }
+  return __libc_malloc(size);
+}
+} // end of anonymous namespace
+
+void *operator new (size_t size) {

[John, 2016/03/15 15:19:06]

can you add a comment explaining why this works?

[sehr, 2016/03/17 16:47:14]

Done.

+  void *caller = __builtin_return_address(0);
+  return internalAllocator(size, caller);
+}
+
+void *operator new[] (size_t size) {
+  void *caller = __builtin_return_address(0);
+  return internalAllocator(size, caller);
+}
+
+extern "C" void *malloc(size_t size) {
+  void *caller = __builtin_return_address(0);
+  return internalAllocator(size, caller);
+}
+#endif // ALLOW_LINUX_MALLOC_PROFILE
+
+namespace Ice {
+
+void LinuxMallocProfiling::start() {
+#ifdef ALLOW_LINUX_MALLOC_PROFILE
+  Callers = new MallocMap();
+#endif // ALLOW_LINUX_MALLOC_PROFILE
+}
+
+void LinuxMallocProfiling::end() {
+#ifdef ALLOW_LINUX_MALLOC_PROFILE
+  if (Callers == nullptr) {
+    return;
+  }
+  for (auto &C : *Callers) {

[John, 2016/03/15 15:19:06]

does

const auto &

work?

[sehr, 2016/03/17 16:47:14]

Yes.  Done.

+    Dl_info dli;
+    dladdr(C.first, &dli);
+
+    std::cout << C.second << " ";
+    if (dli.dli_sname == NULL) {
+      std::cout << C.first;
+    } else {
+      std::cout << dli.dli_sname;
+    }
+    std::cout << "\n";
+  }
+  delete Callers;
+  Callers = nullptr;
+#endif // ALLOW_LINUX_MALLOC_PROFILE
+}
+
+} // end of namespace Ice