[Diffusion] rL237954: Resubmit r237708 (MIR Serialization: print and parse LLVM IR using MIR format).

[Alex L]

2015-05-22 15:09 GMT-07:00 Duncan P. N. Exon Smith <dexonsmith at apple.com>:

>
> > On 2015-May-22, at 14:11, Alex L <arphaman at gmail.com> wrote:
> >
> >
> >
> > 2015-05-22 11:00 GMT-07:00 Duncan P. N. Exon Smith <dexonsmith at apple.com
> >:
> > Why not move the machine IR state into lib/CodeGen/MIR?  Then it
> > won't depend on lib/CodeGen.
> >
> > While moving the cpp files for MachineFunction, MachineBasicBlock, etc.
> to MIR
> > would be easy, moving the headers will require changes to all the
> targets. At the same time the headers
> > can stay under CodeGen just like the header file for AsmPrinter, but I
> would prefer either to move
> > everything or move nothing at all. I attached a patch below that moves
> the necessary files from CodeGen
> > to a new library called MIR, but I don't think this kind of change will
> be accepted.
>
> Why not?  This patch makes sense to me.  I think it's a good
> cleanup *anyway*.
>
> (Deserves its own thread though, since as you indicate it's a
> fairly invasive change.)
>

I'm just worried about the timeframe. I'll bring it up in on llvm-dev with
a short proposal and a patch and we'll see
what happens.


>
> > I think the best approach here would be to move everything from
> CodeGen/MIR to CodeGen so that
> > the serialization will be inside of a CodeGen library. The approach that
> I outlined in my previous email where
> > only the MIRParser is a separate library should also work.
>
> Having a separate library for the parser sounds like a good idea
> regardless.  This makes sense to me:
>
>   - lib/MIR: machine IR
>   - lib/MIR/MIRParser: machine IR parser


> >
> > Cheers,
> > Alex.
> >
> >
> > > On 2015 May 22, at 09:38, Alex L <arphaman at gmail.com> wrote:
> > >
> > > Hi,
> > >
> > > I'm not sure if it would possible to break this circular dependency.
> > >
> > > LLVMMIR has to depend on LLVMCodeGen as it needs to access all the
> machine IR state.
> > > But at the same time LLVMCodeGen has to connect to LLVMMIR to create a
> printing pass and
> > > then later to parse the machine functions, as llc does all the pass
> setup with LLVMCodeGen.
> > >
> > > If this dependency is a problem maybe LLVMMIR should be a part of
> LLVMCodeGen?
> > >
> > > Alex.
> > >
> > > 2015-05-22 7:48 GMT-07:00 NAKAMURA Takumi <geek4civic at gmail.com>:
> > > Reverted in r238007. Could you reorganize MIR to dissolve dependencies
> between LLVMCodeGen and LLVMMIR?
> > >
> > >
> > > /llvm/trunk/lib/CodeGen/MIR/MIRPrintingPass.cpp:18 This depends on
> LLVMCodeGen.
> > > /llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp:153 This depends on
> LLVMMIR.
> > > /llvm/trunk/include/llvm/CodeGen/Passes.h:379 I suggest it might be
> logically in include/llvm/CodeGen/MIR.
> > >
> > > USERS
> > >   arphaman (Author)
> > >
> > > http://reviews.llvm.org/rL237954
> > >
> > > EMAIL PREFERENCES
> > >   http://reviews.llvm.org/settings/panel/emailpreferences/
> > >
> > >
> > >
> > > _______________________________________________
> > > llvm-commits mailing list
> > > llvm-commits at cs.uiuc.edu
> > > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> >
> >
> > <mir-move.patch>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150522/74693b1d/attachment.html>