#7308: Better code for top-level indirections
---------------------------------+------------------------------------------
Reporter: simonpj | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: Compiler | Version: 7.6.1
Keywords: | Os: Unknown/Multiple
Architecture: Unknown/Multiple | Failure: None/Unknown
Difficulty: Unknown | Testcase:
Blockedby: | Blocking:
Related: |
---------------------------------+------------------------------------------
GHC generates lots of top-level indirections looking like this:
{{{
foo :: T
foo = bar `cast` (some coercion)
}}}
There are good reasons for this. Moreover `foo` will almost never be
evaluated, because GHC will inline any call to `foo` at its call site.
'''But''' GHC generates rather a lot of code for this one (almost-vacuous)
bindings:
{{{
.align 4,0x90
.long _Swx_srt-(_Foo_zdfTypeableT_info)+4
.long 0
.long 65558
.globl _Foo_zdfTypeableT_info
_Foo_zdfTypeableT_info:
.LcwI:
movl %esi,%eax
leal -8(%ebp),%ecx
cmpl 84(%ebx),%ecx
jb .LcwP
addl $8,%edi
cmpl 92(%ebx),%edi
ja .LcwR
movl $_stg_CAF_BLACKHOLE_info,-4(%edi)
movl 100(%ebx),%ecx
movl %ecx,0(%edi)
leal -4(%edi),%ecx
pushl %ecx
pushl %eax
pushl %ebx
movl %eax,76(%esp)
call _newCAF
addl $12,%esp
testl %eax,%eax
je .LcwK
movl $_stg_bh_upd_frame_info,-8(%ebp)
leal -4(%edi),%eax
movl %eax,-4(%ebp)
movl $_Foo_zdfTypeableTzuzdctypeOf_closure+1,%esi
addl $-8,%ebp
jmp *(%ebp)
.LcwK:
movl 64(%esp),%eax
jmp *(%eax)
.LcwR:
movl $8,116(%ebx)
.LcwP:
movl %eax,%esi
jmp *-12(%ebx)
}}}
Whereas all we really need for `foo` is code that says "jump to `bar`".
No need to update that top-level thunk. It's just an indirection and (if
you have `-O` on) one that will never even be used.
In short, I think that top-level indirections could do with a tiny special
case in the code generator.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7308>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
