llvm dev - Oct 2010 - [LLVMdev] Is it possible to map an LLVM instruction to x86 assembly instructions?

Dear folks,
    If I want to setup a mapping between an LLVM instruction and the x86
assembly instructions it generates, is this possible?
    Or, an equavalent question is, when LLVM is emitting assebmly
instructions, will it transform a set of LLVM instructions to a set of
assembly instructions, or transform each LLVM instruction to assembly
instructions independently?

-- 
Regards,
Heming Cui
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20101025/13cf27a0/attachment.html>

On Mon, Oct 25, 2010 at 12:39 PM, Heming Cui <hc2428 at columbia.edu>
wrote:
> Dear folks,
>     If I want to setup a mapping between an LLVM instruction and the x86
> assembly instructions it generates, is this possible?
>     Or, an equavalent question is, when LLVM is emitting assebmly
> instructions, will it transform a set of LLVM instructions to a set of
> assembly instructions, or transform each LLVM instruction to assembly
> instructions independently?
Essentially, it's from a set to a set; if you're compiling without
optimizations, the correlation will likely be close, but not exact.
With optimizations, the end result might not be particularly close.
You can come up with an approximation using debug information, though.

-Eli

On Mon, Oct 25, 2010 at 4:54 PM, Eli Friedman <eli.friedman at gmail.com>
wrote:
> On Mon, Oct 25, 2010 at 12:39 PM, Heming Cui <hc2428 at columbia.edu>
wrote:
>> Dear folks,
>>     If I want to setup a mapping between an LLVM instruction and the
x86
>> assembly instructions it generates, is this possible?
>>     Or, an equavalent question is, when LLVM is emitting assebmly
>> instructions, will it transform a set of LLVM instructions to a set of
>> assembly instructions, or transform each LLVM instruction to assembly
>> instructions independently?
>
> Essentially, it's from a set to a set; if you're compiling without
> optimizations, the correlation will likely be close, but not exact.
> With optimizations, the end result might not be particularly close.
> You can come up with an approximation using debug information, though.
Does the pass 'X86DAGToDAGISel' added by
'X86TargetMachine::addInstSelector' define such mappings? although it
maps DAG to X86 code.
>
> -Eli
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>


-- 
Jianzhou