Re: [avr-gcc-list] Re: Avoiding copy_data in config/avr/avr.c

[Ben Jackson]

On Thu, Jul 28, 2005 at 10:43:45PM +0200, Bj??rn Haase wrote:
> IIUC, there should be an easy workaround that, however, could result in 
> slightly bloated assembler source files.

Ok, I took this idea and tried it out.  Here are some details:

The *_SECTION_ASM_OPs are already in avr.h, so you can just replace them.
The .bss section is slightly different than your suggestion:

#define DATA_SECTION_ASM_OP "\t.global __do_copy_data\n\t.data"
#define BSS_SECTION_ASM_OP "\t.global __do_clear_bss\n\t.section .bss"

However, most bss comes from .comm and .lcomm, so you have to change
two other macros as well:

Make the following change to each of ASM_OUTPUT_COMMON and ASM_OUTPUT_LOCAL:

#define ASM_OUTPUT_COMMON(STREAM, NAME, SIZE, ROUNDED)                     \
do {                                                                       \
     static char done;                                                     \
                                                                           \
     if (!done) {                                                          \
        done = 1;                                                          \
        fputs("\t.global __do_clear_bss\n", (STREAM));                     \
     }                                                                     \
        ...

> it should be possible to remove the two lines
> 
>   fputs ("\t.global __do_copy_data\n", asm_out_file);
>   fputs ("\t.global __do_clear_bss\n", asm_out_file);

With that it seems to work.  There are a few more issues that might crop up:

1)  You'll want to rebuild all of your libraries so they don't have
unnecessary .global in every C .o pulling in the copy/zero functions,

2)  People may have gotten lazy in asm libraries and declared data or bss
without the .global to bring them in.  That's a downside to doing it
this way instead of in the linker.  That would not prevent your application
from building but it would have odd behavior at runtime.

-- 
Ben Jackson
<address@hidden>
http://www.ben.com/