Fw: [PATCH] Fix invalid code generation (regression) in pass-by-value on PPC32

[Andy Gibbs]

Great, thanks.  This patch should probably also be marked for inclusion in the next releases of the 3.5 and 3.6 lines, but I don't know who is in charge of that.

  ----- Original Message ----- 
  From: Reid Kleckner 
  To: Andy Gibbs 
  Cc: Reid Kleckner ; cfe-commits cfe 
  Sent: Monday, May 18, 2015 8:08 PM
  Subject: Re: Fw: [PATCH] Fix invalid code generation (regression) in pass-by-value on PPC32


  On Sun, May 17, 2015 at 1:26 PM, Andy Gibbs <andyg1001 at hotmail.co.uk> wrote:

    Reid,

    I'm forwarding this onto you since you are the original committer for r208733
    which I wish to partially revert.  The patch is attached and full(er) details
    can be found at the bug report (link below).  Sorry for sending it to two
    email addresses, but I don't know which is your active one.

    I think you simply factored out code block too many, where (if I understand
    your intentions correctly) you were factoring out the common code for
    *ABI*::classifyReturnType but unintentionally included
    DefaultABIInfo::classifyArgumentType, breaking ppc32 code generation.

    Please can you review my patch to be sure I haven't done anything stupid
    before I commit it.



  Yeah, it looks like the refactor went awry. I'm concerned that PPC32 is relying on DefaultABIInfo and that no test caught this. I'll go ahead and apply that, thanks for the patch!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150519/16933785/attachment.html>