[GHC] #8885: Add inline versions of clone array primops

[GHC]

#8885: Add inline versions of clone array primops
-------------------------------------+------------------------------------
        Reporter:  tibbe             |            Owner:  simonmar
            Type:  feature request   |           Status:  patch
        Priority:  normal            |        Milestone:
       Component:  Compiler          |          Version:  7.9
      Resolution:                    |         Keywords:
Operating System:  Unknown/Multiple  |     Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown      |       Difficulty:  Unknown
       Test Case:                    |       Blocked By:
        Blocking:                    |  Related Tickets:
-------------------------------------+------------------------------------

Comment (by tibbe):

 My intuition to use a pointer based copy loop instead of an index based
 loop seems to have been wrong. Changing the loop from:

 {{{
   dst_p = dst + SIZEOF_StgMutArrPtrs;
   src_p = src + SIZEOF_StgMutArrPtrs + WDS(offset);
 while:
   if (n != 0) {
     n = n - 1;
     W_[dst_p] = W_[src_p];
     dst_p = dst_p + WDS(1);
     src_p = src_p + WDS(1);
     goto while;
   }
 }}}

 to

 {{{
   dst_p = dst + SIZEOF_StgMutArrPtrs;
   src_p = src + SIZEOF_StgMutArrPtrs + WDS(offset);
   i = 0;
 while:
   if (i < n) {
     W_[dst_p + i] = W_[src_p + i];
     i = i + 1;
     goto while;
   }
 }}}

 improves performance on my benchmark (clone array of 17 elements) by 14%.

 Attached patch with this improvement.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8885#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler