#4065: Inconsistent loop performance
---------------------------------+------------------------------------------
Reporter: rl | Owner:
Type: bug | Status: new
Priority: normal | Milestone: 7.2.1
Component: Compiler | Version: 6.13
Keywords: | Testcase:
Blockedby: | Difficulty:
Os: Unknown/Multiple | Blocking:
Architecture: Unknown/Multiple | Failure: Runtime performance bug
---------------------------------+------------------------------------------
Comment(by daniel.is.fischer):
I just ran the benchmark with 7.0.3, I get the reverse outcome with that,
foo takes about 48ms and bar about 42ms. Via C (-O2 -fvia-C -optc-O3),
both take 40-41ms (bar is typically a bit faster).
So, unless it's platform-dependent, the new code generator seems to have
turned things around.
The core looks fine for both, basically
{{{
case n of
0# -> ...
_ -> ...
case n <=# 0# of
True -> ...
False -> ...
}}}
for the workers. I don't understand cmm, so I'm guessing, the relevant
parts seem to be
{{{
cnO:
_sn9::I32 = I32[Sp + 0];
if (_sn9::I32 != 0) goto cnS;
R1 = I32[Sp + 4];
Sp = Sp + 8;
jump (I32[Sp + 0]) ();
cnS:
_snI::I32 = I32[Sp + 4] + 1;
_snJ::I32 = _sn9::I32 - 1;
I32[Sp + 4] = _snI::I32;
I32[Sp + 0] = _snJ::I32;
jump FooBar_zdwbar_info ();
}}}
for bar and
{{{
coF:
_coI::I32 = %MO_S_Le_W32(I32[Sp + 0], 0);
;
if (_coI::I32 >= 1) goto coL;
_soy::I32 = I32[Sp + 4] + 1;
_soz::I32 = I32[Sp + 0] - 1;
I32[Sp + 4] = _soy::I32;
I32[Sp + 0] = _soz::I32;
jump FooBar_zdwfoo_info ();
coL:
R1 = I32[Sp + 4];
Sp = Sp + 8;
jump (I32[Sp + 0]) ();
}}}
for foo. The asm:
{{{
FooBar_zdwbar_info:
.LcnT:
movl 0(%ebp),%eax
testl %eax,%eax
jne .LcnX
movl 4(%ebp),%esi
addl $8,%ebp
jmp *0(%ebp)
.LcnX:
decl %eax
incl 4(%ebp)
movl %eax,0(%ebp)
jmp FooBar_zdwbar_info
}}}
resp.
{{{
FooBar_zdwfoo_info:
.Lcpe:
cmpl $0,0(%ebp)
jle .Lcph
movl 0(%ebp),%eax
decl %eax
incl 4(%ebp)
movl %eax,0(%ebp)
jmp FooBar_zdwfoo_info
.Lcph:
movl 4(%ebp),%esi
addl $8,%ebp
jmp *0(%ebp)
}}}
doesn't tell me either why bar is faster than foo.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/4065#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
