[PATCH] Implement the ldr-pseudo opcode for ARM assembly

[Renato Golin]

Hi David,

Sorry it's taking so long, many other things flying around...

Can you attach your patch to phabricator? First the original one, than this
one, so it's easier to see what's the difference between the original
proposal and the current. Otherwise, we'll have to go through the whole
patch every time...

cheers,
--renato


On 22 November 2013 18:25, David Peixotto <dpeixott at codeaurora.org> wrote:

> ping
>
>
>
> *From:* llvm-commits-bounces at cs.uiuc.edu [mailto:
> llvm-commits-bounces at cs.uiuc.edu] *On Behalf Of *David Peixotto
> *Sent:* Wednesday, November 20, 2013 11:51 AM
>
> *To:* 'Jim Grosbach'
> *Cc:* 'Commit Messages and Patches for LLVM'
> *Subject:* RE: [PATCH] Implement the ldr-pseudo opcode for ARM assembly
>
>
>
> I've attached updated patches that contain the corresponding tests for
> darwin. I also modifed the patch to use MCContext::CreateTempSymbol()
> instead of creating the symbols explicitly.
>
>
>
> Please help to review this change.
>
>
>
> *From:* llvm-commits-bounces at cs.uiuc.edu [
> mailto:llvm-commits-bounces at cs.uiuc.edu <llvm-commits-bounces at cs.uiuc.edu>]
> *On Behalf Of *David Peixotto
> *Sent:* Tuesday, November 19, 2013 7:39 PM
> *To:* 'Jim Grosbach'
> *Cc:* 'Commit Messages and Patches for LLVM'
> *Subject:* RE: [PATCH] Implement the ldr-pseudo opcode for ARM assembly
>
>
>
> I've converted the ARM tests in ldr-pseudo.s to run for darwin and they
> are passing. The darwin tests for arm are attached. I have not converted
> the thumb1 and thumb2 tests yet.
>
>
>
> Compared to the linux tests, I made the following changes:
>
>
>
> 1. Adjust CHECK lines for the address of the constant pool entries. Needed
> because with llvm-objdump -d, new sections do not start at address zero,
> but continue from end of previous section. However, llvm-objdump -r still
> prints relocation addresses as offset from the section.
>
> 2. Adjust CHECK lines for the type of relocation: R_ARM_ABS32(linx) vs.
> ARM_RELOC_VANILLA(darwin)
>
> 3. Adjust the CHECK lines for the reloc value in the constant pools for
> known symbols. For darwin it contains the address in the section rather
> than the offset from the symbol (e.g. see test in section h that has a
> relocation with value 3c in section d rather than offset 0 from symbol f5).
>
>
>
> None of the pc-relative loads in the generated ldr instructions needed to
> change. For the linux tests, I had gcc to compare against for the correct
> output. It is more difficult for me to verify the results on darwin. The
> changes look reasonable to me, but I would appreciate any feedback/help on
> the test.
>
>
>
> Thanks,
>
> -David
>
>
>
> *From:* David Peixotto [mailto:dpeixott at codeaurora.org<dpeixott at codeaurora.org>]
>
> *Sent:* Tuesday, November 19, 2013 6:01 PM
> *To:* 'Jim Grosbach'
> *Cc:* 'Commit Messages and Patches for LLVM'
> *Subject:* RE: [PATCH] Implement the ldr-pseudo opcode for ARM assembly
>
>
>
> Hi Jim,
>
>
>
> I had thought that these set of tests would be ok because I was using the
> machine independent assembler apis, but I can see that I have made a bad
> assumption. I will add the corresponding tests for darwin.
>
>
>
> (As an aside, I just tried a quick test with the first example from
> ldr-pseudo.s and it did not complain about the labels. I'm used the same
> commands, but switched the target to armv7-darwin-apple and updated the
> .section syntax. In any case I will fix the labels to use real temporaries,
> but let me know if there is a better way to test for darwin).
>
>
>
> As for the subsections-via-symbols, I had looked for a test that I could
> follow, but did not see anything. Based on our discussion it seems that
> pc-relative loads across symbols is undefined behavior. So even the
> existing ldr <label> instructions should have a problem, right? Is there
> some test you can point me to that checks for this or some testing guidance
> you can provide? I'm not quite sure what the correct behavior should be
> here.
>
>
>
> -David
>
>
>
>
>
> *From:* Jim Grosbach [mailto:grosbach at apple.com <grosbach at apple.com>]
> *Sent:* Tuesday, November 19, 2013 4:56 PM
> *To:* David Peixotto
> *Cc:* Commit Messages and Patches for LLVM
> *Subject:* Re: [PATCH] Implement the ldr-pseudo opcode for ARM assembly
>
>
>
> Hi David,
>
>
>
> As far as I can tell, these patches don’t handle the
> subsections-via-symbols or anything else Darwin related at all. Am I
> missing something? For a trivial example, the label names are not going to
> work at all on Darwin.
>
>
>
> -Jim
>
>
>
> On Nov 15, 2013, at 6:00 PM, David Peixotto <dpeixott at codeaurora.org>
> wrote:
>
>
>
> Attached are 3 patches that together implement the ldr pseudo for arm and
> the .ltorg directive. Please help to review these patches.
>
>
>
> Thanks,
>
> David
>
>
>
> -- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> hosted by The Linux Foundation
>
>
>
>
>
>
> <0001-Add-a-finishParse-callback-to-the-targer-asm-parser.patch><0002-Implement-the-ldr-pseudo-opcode-for-ARM-assembly.patch><0003-Implement-the-.ltorg-directive-for-ARM-assembly.patch>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131125/4c12a44e/attachment.html>