[LLVMdev] [Patch] Let MC/ELF generate Thumb/Thumb-2 are properly

[Koan-Sin Tan]

On Wed, May 18, 2011 at 12:05 AM, Rafael Avila de Espindola
<rafael.espindola at gmail.com> wrote:
>> Regarding elf-thumbfunc-reloc.ll, it seems to me that current ARMAsmParser
>> doesn't recognize "(PLT)", so something like  "bl foo(PLT)" doesn't work
>> consequently. And I don't know how to write .s to test this without (PLT).
>
> NP. Can you just add that as a FIXME in elf-thumbfunc-reloc.ll? That way we
> know why it is there and can port it once more of the asm parser works.
>

Done!

>> Regarding elf-xscale-attribute.ll, those attributes are for
>> ".eabi_attribute" lines
>> in .s files and attributes in .aeabi section in .o files. Both .ll->.s
>> and .ll->.o
>> should be tested. Surely, not .s ->  .o.  I didn't write .ll ->  .s
>> because
>> ARMAsmParser ignore ".eabi_attribute" for now
>
> Since this is something that happens in Codegen (.ll -> .s), I think this
> should probably be tested by looking at the .s produced. The fact that
> llvm-mc is still not able to process .eabi_attribute is a different issues
> which should get its own tests once it is able to.

.ll -> .s test added. As far as I can tell .ll -> .s and .ll -> .o go
different paths, so I have both .ll->.s and .ll -> .o in the test case

The revised patches (with tests in .diff) are attached.

patch-2-3.diff:
  fixes target address tBL and tBLX and sets relocation type
  of tBL/tBLX to R_ARM_THM_CALL (ARM ELF 4.7.1.6)

patch-4.diff:
  adds some attributes to attribute section when cpu is "xscale"
  (this is what used in Android NDK, when architecture is ARMv5)

Thanks,
// koan-sin tan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mc-elf-thumb.tgz
Type: application/x-gzip
Size: 1973 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20110518/ca2bd16e/attachment.bin>