arm-elf-size

Greetings,

I have a problem understanding how the arm-elf-size command works:

When I use 'arm-elf-size -A' I get:

arm-elf-size -B Output/Exe/Bootstrapper.elf
    text    data     bss     dec     hex filename
   18068       0    5508   23576    5c18 Output/Exe/Bootstrapper.elf

When I use 'arm-elf-size -B' I get:

arm-elf-size -A Output/Exe/Bootstrapper.elf
Output/Exe/Bootstrapper.elf  :
section             size      addr
.startup              80   1048576
.text              16892   1048656
.data               1096   2097152
.bss                5508   2098248
.debug_abbrev      22956         0
.debug_info       102031         0
.debug_line        19194         0
.debug_frame        9912         0
.debug_loc         44707         0
.debug_pubnames     8590         0
.debug_aranges      4768         0
.debug_ranges       2104         0
.debug_str         17160         0
.comment            1188         0
Total             256186


Question is :

Why does '-A' not show how much .data is used ?

When trying to optimize the code I need to know how much flash-space and 
how much RAM is used.
Using option '-A' I cannot see the .data-sections size so this is 
useless here.
Using option '-B' I can see how much space is used in each segment, but 
I have to manually calculate the used space as in "Flash = .startup + 
.text + .data" and "RAM = .data + .bss". The '-B' option is the must 
usefull one but not very good.
Is there a way / tool that can produce an output that simply tells me 
how much code space and RAM space I have used ???

Cheers
RaceMouse

P.S. This is using both GCC-4.1.0 and GCC-4.1.1
P.P.S. This is a crosspost from gnu.gcc.help, but there seems to be a 
lot more activity here - leading to an increasing the chance for an 
answer :-)

RaceMouse wrote:

> When I use 'arm-elf-size -A' I get:

Actually, no. You used the -B option here:

> arm-elf-size -B Output/Exe/Bootstrapper.elf
>    text    data     bss     dec     hex filename
>   18068       0    5508   23576    5c18 Output/Exe/Bootstrapper.elf
> 
> When I use 'arm-elf-size -B' I get:

... and -A here:

> arm-elf-size -A Output/Exe/Bootstrapper.elf
> Output/Exe/Bootstrapper.elf  :
> section             size      addr
> .startup              80   1048576
> .text              16892   1048656
> .data               1096   2097152
> .bss                5508   2098248
[...]

> Total             256186

> Why does '-A' not show how much .data is used ?

First of all because it's -B, not -A, that fails to show it.

Second reason is: because.  The -B format is the traditional default 
output format of this tool, i.e. it's governed by history, not maximum 
possible usefulness.

Third, possible reason is that your linker script is slightly strange, 
putting data into a segment where 'size -B' doesn't know to look for it. 
  For comparison, some other platform has

	text = .text + .rdata
	data = .data + .idata

Yours appears to have neither .rdata nor .idata.

The fourth reason could be that your program just happens to have no 
initialized data at all, only constants and zero-initialized data. 
Quite impossible to tell without seeing the details section lists
(objdump -s).

Hans-Bernhard Br�ker wrote:

> First of all because it's -B, not -A, that fails to show it.
> 
> Second reason is: because.  The -B format is the traditional default 
> output format of this tool, i.e. it's governed by history, not maximum 
> possible usefulness.

I will send a feature request to the devs of arm-elf-size. A new option 
"-C" wich will produce a usable output...

Until then : Do you have any idea how I can work around this and get me 
a usable output ?

> 
> Third, possible reason is that your linker script is slightly strange, 
> putting data into a segment where 'size -B' doesn't know to look for it. 
>  For comparison, some other platform has
> 
>     text = .text + .rdata
>     data = .data + .idata
> 

I use -ffunction-sections and -fdata-sections when compiling. This puts 
all functions into their own (.text.*) segment and ofcourse all data 
segments into their own (.data.*) - not (.data). This explains why 
arm-elf-size -B can't see it. Thanks for the tip.

> Yours appears to have neither .rdata nor .idata.
> 
> The fourth reason could be that your program just happens to have no 
> initialized data at all, only constants and zero-initialized data. Quite 
> impossible to tell without seeing the details section lists
> (objdump -s).

RaceMouse wrote:
> Hans-Bernhard Br�ker wrote:

> I will send a feature request to the devs of arm-elf-size. A new option 
> "-C" wich will produce a usable output...

The -A output is quite usable as it is --- it has more information than 
you need.

> Until then : Do you have any idea how I can work around this and get me 
> a usable output ?

Use the tools like they're supposed to be: as parts of a toolset, not as 
single be-all-and-do-everything gizmos.  In the case at hands, that 
means to write a little script in a suitable scripting language (shell, 
awk, perl, python, whatever floats your boat) that parses 'size -A' 
output and computes whatever you want from the output.

  > I use -ffunction-sections and -fdata-sections when compiling. This puts
> all functions into their own (.text.*) segment and ofcourse all data 
> segments into their own (.data.*) - not (.data). 

But the linker joins those segments back together into a single '.data' 
segment, as you can see in the 'size -A' output you posted originally.