[docs][llvm-size] Write llvm-size documentation

[jhenderson]

Summary

Previously we only had a stub document.

[serge-sans-paille]

LGTM, that's super nice!

[MaskRay]

llvm/docs/CommandGuide/llvm-size.rst

14

for object files

Is there a concise term that refers to object files/executables/shared objects/archives collectively? I don't have particular suggestion here.

As a reference, FreeBSD size says:

The size utility lists the sizes of ELF sections, and optionally the total size, for each input file specified on the command line. The size utility can operate on ELF objects, on ar(1) archives containing ELF ob- jects, and on core dumps. If no file name is specified on the command-line, a.out is assumed.

We support Mach-O and PE/COFF so "ELF objects" is not accurate...

163

the total size

The Berkeley format prints more than 1 field, so GNU size refers to them as "totals".

GNU size:

Show totals of all objects listed

FreeBSD size:

Shows cumulative totals of section sizes from all objects.

178

Or simply Display the version of this program. as used by other utilities.

[jhenderson]

llvm/docs/CommandGuide/llvm-size.rst

14

TL;DR - Either we use "binary" (or possibly "archives and binaries"), or we go with the not-always-correct-but-commonly-used term "object file". Let me know if you have a specific preference.

---

So technically, "object files" for ELF refers to all ELF types, not just ET_REL objects. For example, the ELF gABI says:

e_type
This member identifies the object file type.

The FreeBSD spec seems to describe the same principle (note that it doesn't highlight executables separately from relocatable objects). On the other hand, wider usage seems to be "object file" = "relocatable object". Similarly, the terminology used in the PE/COFF file format calls executables "executables" and not objects (which is used for reloctable objects specifically). Mach-O uses the term binary. LLVM code seems to be inconsistent (e.g. the ObjectFile class can represent executables or reloctable objects of all supported formats).

We use the term "object files" to refer to any input (except archives) in llvm-symbolizer.rst, and llvm-readobj.rst. In llvm-nm.rst, we use "LLVM bitcode files, object files, and archives". llvm-objcopy uses "objects", so we should use the same terminology throughout the docs for these files. I'm happy to add a specific mention of archives though. I don't have a good way of grouping archives with object files into a single concise term, unless we go with the very imprecise term "binary".

178

I discussed this internally with a colleague and was planning on putting up a patch to change the docs for the other tools to match this version. The reasoning is that everywhere else, we refer to the program by its name e.g. "llvm-size exits with" or "llvm-size reads a file.." rather than "This program exits" etc. Additionally, as this doc isn't part of the executable, referring to the tool as "this program" isn't technically correct.

[jhenderson]

Address some review comments. Specifically, reworded the --totals description, added a note about archive inputs, and replaced "object files" with "binaries" etc, to be more correct.

[jhenderson]

Fix formatting of code blocks, so that the text following them are not considered part of the block.

llvm/docs/CommandGuide/llvm-size.rst

 llvm-size - print size information
 ==================================
 
 .. program:: llvm-size
 
 SYNOPSIS
 --------
 
-:program:`llvm-size` [*options*]
+:program:`llvm-size` [*options*] [*input...*]
 
 DESCRIPTION
 -----------
 
 :program:`llvm-size` is a tool that prints size information for object files.
-The goal is to make it a drop-in replacement for GNU's :program:`size`.
+It is intended to be a drop-in replacement for GNU's :program:`size`.
+
+The tool prints size information for each ``input`` specified. If no input is
+specified, the program prints size information for ``a.out``. If "``-``" is
+specified as an input file, :program:`llvm-size` reads a file from the standard
+input stream.
+
+OPTIONS
+-------
+
+.. option:: -A
+
+ Equivalent to :option:`--format` with a value of ``sysv``.
+
+.. option:: --arch=<arch>
+
+ Architecture(s) from Mach-O universal binaries to display information for.
+
+.. option:: -B
+
+ Equivalent to :option:`--format` with a value of ``berkeley``.
+
+.. option:: --common
+
+ Include ELF common symbol sizes in bss size for ``berkeley`` output format, or
+ as a separate section entry for ``sysv`` output. If not specified, these
+ symbols are not ignored.
+
+.. option:: -d
+
+ Equivalent to :option:`--radix` with a value of ``10``.
+
+.. option:: -l
+
+ Display verbose address and offset information for segments and sections in
+ Mach-O files in ``darwin`` format.
+
+.. option:: --format=<format>
+
+ Set the output format to the ``<format>`` specified. Available ``<format>``
+ options are ``berkeley`` (the default), ``sysv`` and ``darwin``.
+
+ Berkeley output summarises text, data and bss sizes in each file, as shown
+ below for a typical pair of ELF files:
+
+.. code-block:: console
+
+ $ llvm-size --format=berkeley test.o test2.o
+    text    data     bss     dec     hex filename
+     182      16       5     203      cb test.elf
+      82       8       1      91      5b test2.o
+
+ For Mach-O files, the output format is slightly different:
+
+.. code-block:: console
+
+ $ llvm-size --format=berkeley macho.obj macho2.obj
+ __TEXT  __DATA  __OBJC  others  dec     hex
+ 4       8       0       0       12      c       macho.obj
+ 16      32      0       0       48      30      macho2.obj
+
+ Sysv output displays size and address information for most sections, with each
+ file being listed separately:
+
+.. code-block:: console
+
+ $ llvm-size --format=sysv test.elf test2.o
+    test.elf  :
+    section       size      addr
+    .eh_frame       92   2097496
+    .text           90   2101248
+    .data           16   2105344
+    .bss             5   2105360
+    .comment       209         0
+    Total          412
+
+    test2.o  :
+    section             size   addr
+    .text                 26      0
+    .data                  8      0
+    .bss                   1      0
+    .comment             106      0
+    .note.GNU-stack        0      0
+    .eh_frame             56      0
+    .llvm_addrsig          2      0
+    Total                199
+
+ ``darwin`` format only affects Mach-O input files. If an object of a different
+ file format is specified, :program:`llvm-size` falls back to ``berkeley``
+ format. When producing ``darwin`` format, the tool displays information about
+ segments and sections:
+
+.. code-block:: console
+
+ $ llvm-size --format=darwin macho.obj macho2.obj
+    macho.obj:
+    Segment : 12
+            Section (__TEXT, __text): 4
+            Section (__DATA, __data): 8
+            total 12
+    total 12
+    macho2.obj:
+    Segment : 48
+            Section (__TEXT, __text): 16
+            Section (__DATA, __data): 32
+            total 48
+    total 48
+
+.. option:: --help, -h
+
+ Display a summary of command line options.
+
+.. option:: --help-list
+
+ Display an uncategorized summary of command line options.
+
+.. option:: -m
+
+ Equivalent to :option:`--format` with a value of ``darwin``.
+
+.. option:: -o
+
+ Equivalent to :option:`--radix` with a value of ``8``.
+
+.. option:: --radix=<value>
+
+ Display size information in the specified radix. Permitted values are ``8``,
+ ``10`` (the default) and ``16`` for octal, decimal and hexadecimal output
+ respectively.
+
+ Example:
+
+.. code-block:: console
+
+ $ llvm-size --radix=8 test.o
+    text    data     bss     oct     hex filename
+    0152      04      04     162      72 test.o
+
+ $ llvm-size --radix=10 test.o
+    text    data     bss     dec     hex filename
+     106       4       4     114      72 test.o
+
+ $ llvm-size --radix=16 test.o
+    text    data     bss     dec     hex filename
+    0x6a     0x4     0x4     114      72 test.o
+
+.. option:: --totals, -t
+
+ Applies only to ``berkeley`` output format. Display the total size for all
+ listed object files, in addition to the individual file listings.
+
+ Example:
+
+.. code-block:: console
+
+ $ llvm-size --totals test.elf test2.o
+    text    data     bss     dec     hex filename
+     182      16       5     203      cb test.elf
+      82       8       1      91      5b test2.o
+     264      24       6     294     126 (TOTALS)
+
+.. option:: --version
+
+ Display the version of the :program:`llvm-size` executable.
+
+.. option:: -x
+
+ Equivalent to :option:`--radix` with a value of ``16``.
+
+.. option:: @<FILE>
+
+ Read command-line options from response file ``<FILE>``.
+
+EXIT STATUS
+-----------
+
+:program:`llvm-size` exits with a non-zero exit code if there is an error.
+Otherwise, it exits with code 0.
+
+BUGS
+----
+
+To report bugs, please visit <http://llvm.org/bugs/>.