[GHC] #7830: Error: operand out of range

[GHC]

#7830: Error: operand out of range
-------------------------------------+-------------------------------------
        Reporter:  erikd             |                   Owner:
            Type:  bug               |                  Status:  new
        Priority:  highest           |               Milestone:  8.0.1
       Component:  Compiler          |                 Version:  7.8.1-rc1
      Resolution:                    |                Keywords:
Operating System:  Linux             |            Architecture:  powerpc
 Type of failure:  Installing GHC    |               Test Case:
  failed                             |
      Blocked By:                    |                Blocking:
 Related Tickets:                    |  Differential Revisions:
-------------------------------------+-------------------------------------

Comment (by erikd):

 Compiling git HEAD (currently c8d438fb02 but anything after commit
 5d57087e31 is basically the same) fails on PowerPC due to the issue in
 this ticket (comment 19 and later). When it fails, the generated assembler
 for `compiler/cmm/PprC.hs` is about 300000 lines of code, but if I compile
 the tree at commit bcfae08c0b (the one before 5d57087e31) the generated
 assemble for this file is only 135297 lines of code.

 Ignoring the stage1 compiler and looking at compiling
 `compiler/cmm/PprC.hs` with ghc-7.8.4 I see:

 {{{
 revision             wc -l PprC.s         lwz_count        stw_count
 ---------------------------------------------------------------------
 commit bcfae08c0b       135297               27751           30921
 HEAD (c8d438fb02)       279333               70407           77666
 }}}

 where
 {{{
    lwz_count = grep lwz PprC.s | wc -l
    stw_count = grep stw PprC.s | wc -l
 }}}


 Just out of curiosity I decided to do the same test on x86_64/linux and
 found:

 {{{
 revision             wc -l PprC.s
 ---------------------------------
 commit bcfae08c0b       119246
 HEAD (c8d438fb02)       246768
 }}}


 a near doubling of the code size.

 Is it possible we are inlining something we shouldn't?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/7830#comment:28>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler