Calculate correct section sizes for llvm-size

[khemant]

Summary

llvm-size adds sections such as symbol table etc. into size totals. This is incorrect. BSD utilty does not:
$llvm-size -A a.o
a.o :
section size addr
.text 77 0
.data 0 0
.bss 0 0
.comment 44 0
.note.GNU-stack 0 0
.eh_frame 248 0
.rela.eh_frame 168 0
.shstrtab 84 0
.symtab 432 0
.strtab 52 0
Total 1105

$size -A a.o
a.o :
section size addr
.text 77 0
.data 0 0
.bss 0 0
.comment 44 0
.note.GNU-stack 0 0
.eh_frame 248 0
Total 369

[mgrang]

tools/llvm-size/llvm-size.cpp

124

Could you write this as a switch-case? Would make future maintenance easier.

[davide]

What's the output of binutils size?

tools/llvm-size/llvm-size.cpp

22

Unsorted I guess?

120

No need to call getType() each time.

[khemant]

Binutils size outputs similar to BSD. It discounts these sections as well

tools/llvm-size/llvm-size.cpp

22

Correct. Will be putting in right place.

120

I will be changing this to switch as mgrang mentioned.

[davide]

Looks good. Thanks for the testcase/fix.

[khemant]

Committed r259578

test/tools/llvm-size/basic.test

 RUN: llvm-size %t.blah 2>&1 | FileCheck --check-prefix=ENOENT %s
+#1.o source
+#int x ; int y ; int z;
+#
+#int main () {
+#  x = 1;
+#  y = 1;
+#  z = 1;
+#  return x + y + z;
+#}
+#Compilation: clang -c -O0
+RUN: llvm-size -A %p/Inputs/1.o | FileCheck --check-prefix="SYSV" %s
+RUN: llvm-size -B %p/Inputs/1.o | FileCheck --check-prefix="BSD" %s
+
 ENOENT: {{.*}}llvm-size{{(\.EXE|\.exe)?}}: {{.*}}.blah: {{[Nn]}}o such file or directory
+SYSV:    {{[ -\(\)_A-Za-z0-9.\\/:]+}}  :
+SYSV-NEXT:    section             size   addr
+SYSV-NEXT:    .text                 67      0
+SYSV-NEXT:    .data                  0      0
+SYSV-NEXT:    .bss                   0      0
+SYSV-NEXT:    .comment              46      0
+SYSV-NEXT:    .note.GNU-stack        0      0
+SYSV-NEXT:    .eh_frame             56      0
+SYSV-NEXT:    Total                169
+
+BSD:        text    data     bss     dec     hex filename
+BSD-NEXT:     67      56       0     123      7b {{[ -\(\)_A-Za-z0-9.\\/:]+}}
+

tools/llvm-size/llvm-size.cpp

 // This program is a utility that works like traditional Unix "size",
 // that is, it prints out the size of each section, and the total size of all
 // sections.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/APInt.h"
 #include "llvm/Object/Archive.h"
+#include "llvm/Object/ELFObjectFile.h"
 #include "llvm/Object/MachO.h"
 #include "llvm/Object/MachOUniversal.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ELF.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/PrettyStackTrace.h"
 #include "llvm/Support/Signals.h"
 #include "llvm/Support/raw_ostream.h"
 #include <algorithm>
static const char *getRadixFmt() {
   case decimal:
     return PRIu64;
   case hexadecimal:
     return PRIx64;
   }
   return nullptr;
 }
 
+/// @brief Remove unneeded ELF sections from calculation
+static bool ConsiderForSize(ObjectFile *Obj, SectionRef Section) {
+  if (Obj->isELF()) {
+    switch (static_cast<ELFSectionRef>(Section).getType()) {
+    case ELF::SHT_NULL:
+    case ELF::SHT_SYMTAB:
+    case ELF::SHT_STRTAB:
+    case ELF::SHT_REL:
+    case ELF::SHT_RELA:
+      return false;
+    }
+  }
+  return true;
+}
 /// @brief Print the size of each Mach-O segment and section in @p MachO.
 ///
 /// This is when used when @c OutputFormat is darwin and produces the same
 /// output as darwin's size(1) -m output.
 static void PrintDarwinSectionSizes(MachOObjectFile *MachO) {
   std::string fmtbuf;
   raw_string_ostream fmt(fmtbuf);
   const char *radix_fmt = getRadixFmt();
else if (MachO && OutputFormat == berkeley)
     PrintDarwinSegmentSizes(MachO);
   else if (OutputFormat == sysv) {
     // Run two passes over all sections. The first gets the lengths needed for
     // formatting the output. The second actually does the output.
     std::size_t max_name_len = strlen("section");
     std::size_t max_size_len = strlen("size");
     std::size_t max_addr_len = strlen("addr");
     for (const SectionRef &Section : Obj->sections()) {
+      if (!ConsiderForSize(Obj, Section))
+        continue;
       uint64_t size = Section.getSize();
       total += size;
 
       StringRef name;
       if (error(Section.getName(name)))
         return;
       uint64_t addr = Section.getAddress();
       max_name_len = std::max(max_name_len, name.size());
else if (OutputFormat == sysv) {
 
     // Setup per section format.
     fmt << "%-" << max_name_len << "s "
         << "%#" << max_size_len << radix_fmt << " "
         << "%#" << max_addr_len << radix_fmt << "\n";
 
     // Print each section.
     for (const SectionRef &Section : Obj->sections()) {
+      if (!ConsiderForSize(Obj, Section))
+        continue;
       StringRef name;
       if (error(Section.getName(name)))
         return;
       uint64_t size = Section.getSize();
       uint64_t addr = Section.getAddress();
       std::string namestr = name;
 
       outs() << format(fmt.str().c_str(), namestr.c_str(), size, addr);