This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFT/RFA] Fix AIX fallout from PR/19653 patch

David Edelsohn wrote: 
	The fundamental cause of all this problem seems to be the initial
RTL generation of:

(insn 38 37 39 3 (set (mem:SI (plus:SI (reg/f:SI 117 virtual-outgoing-args)
                (const_int 32 [0x20])) [0 S4 A32])
        (subreg:SI (reg:DF 141) 4)) -1 (nil)
    (nil))

despite CANNOT_CHANGE_MODE_CLASS returning true for DFmode.  It's probably
most useful to try to generate better initial RTL, avoiding reload trying
to do something unnatural.

This is true (indeed, if we compare 4.2 pre-Dale-patch, and post-Dale-patch with my fix, the produced code is slightly better but still I would not call it decent). However, I would also like to hear from you or Geoff about my analysis, and especially to know if the suspicious pieces I pointed out in his TOC patches are really oversights, or just false problems.

Paolo

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]