llvm dev - Apr 2012 - [LLVMdev] Adding a Yasm ASM printer

Hi,
  I'm in the process of adding a X86 ASM Printer for outputting the asm in
a style that Yasm can take without any post processing. Yasm's format is
very close to that of intel.  I was wondering if my approach is correct:

I've made a custom MCAsmStreamer to remove some of the COFF symbol
directives that yasm doesn't understand.

I've made a X86YasmInstPrinter, and added it to the x86.td, as a new
variant, and due to it having a new Variant number, I'm having to edit all
the X86*.td files and add the extra variant's descriptions in the { | }
blocks.

I've also added a new ASMInfo class with the required changes for yasm.


Now what is worrying me is the number of .td files I'm having to touch, and
most of the changes are adding a variant with exactly the same code as the
intel variant, there are just a few cases that a slightly different.

I'm about 1/4 way through the .td files, and already the asm is starting to
look close to something yasm could handle, so I must be at least on the
right track, but that size of the changes to the .td files is alarming.

Any pointers, or suggestions welcome :-)

Cheers

Matthew.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20120403/f9e3549f/attachment.html>

Hi Matthew,

Whether this is the right approach depends on the details of the problem. If
it's mainly directives that are the issue, then a new variant is huge
overkill. If, on the other hand, most/many of the instructions themselves need
different syntax, then it's exactly the right approach. From your
description, in particular, "Yasm's format is very close to that of
intel," I suspect you may be able to get away with some custom handling in
the current streamers and maybe a few tweaks to problematic instructions. For
example, the current streamers conditionally use CFI directives based on whether
the target environment supports them. You may be able to do something similar.

A new streamer (or customized behavior on the current one) and a new variant are
somewhat orthogonal. I suggest solving the issues for each independently to
avoid having too many moving parts at once.

-Jim



On Apr 2, 2012, at 1:46 PM, Matthew Huck <matthew.huck at gmail.com>
wrote:
> Hi,
>   I'm in the process of adding a X86 ASM Printer for outputting the asm
in a style that Yasm can take without any post processing. Yasm's format is
very close to that of intel.  I was wondering if my approach is correct:
> 
> I've made a custom MCAsmStreamer to remove some of the COFF symbol
directives that yasm doesn't understand.
> 
> I've made a X86YasmInstPrinter, and added it to the x86.td, as a new
variant, and due to it having a new Variant number, I'm having to edit all
the X86*.td files and add the extra variant's descriptions in the { | }
blocks.
> 
> I've also added a new ASMInfo class with the required changes for yasm.
> 
> 
> Now what is worrying me is the number of .td files I'm having to touch,
and most of the changes are adding a variant with exactly the same code as the
intel variant, there are just a few cases that a slightly different.
> 
> I'm about 1/4 way through the .td files, and already the asm is
starting to look close to something yasm could handle, so I must be at least on
the right track, but that size of the changes to the .td files is alarming.
> 
> Any pointers, or suggestions welcome :-)
> 
> Cheers
> 
> Matthew.
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20120403/65dbb211/attachment.html>