[GHC] #14251: LLVM Code Gen messes up registers

[GHC]

#14251: LLVM Code Gen messes up registers
-------------------------------------+-------------------------------------
        Reporter:  angerman          |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  highest           |            Milestone:  8.6.1
       Component:  Compiler (LLVM)   |              Version:  8.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Incorrect result  |  Unknown/Multiple
  at runtime                         |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by awson):

 AFAIUI, the most recent takes on this problem are:

 1. https://phabricator.haskell.org/D4003 by @angerman
 2. https://gist.github.com/kavon/566fc6c21ff51803538884b79dc1d841 by
 @kavon (referred from the former)

 My understanding is that "1" doesn't quite work (generates wrong code for
 handwritten `cmm`).

 Also my understanding is that "2" should work (haven't tested it), but
 isn't so much aesthetically pleasant, since all padding arguments are of
 `FloatReg` type regardless of which type they are at the call-site.

 A year has passed since. Neither these two approaches, no other suggested
 in the email thread started from https://mail.haskell.org/pipermail/ghc-
 devs/2017-September/014700.html, were elaborated further.

 I wonder then what are chances of e.g. "2" to be upstreamed if properly
 polished?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14251#comment:11>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler