I think the inliner assumes temporaries are single-assignment. Could
that be the problem?
Cheers,
Simon
On 30/03/11 18:25, Edward Z. Yang wrote:
I believe I've run into an inliner bug. (This is the "new codegen
starts segfaulting stage2 GHC after I started inlining across comments").
The following code:
// outOfLine should follow:
_c21P::I32 = _s1uR::I32 == _s1uS::I32;
// emitReturn: Sequel: Assign
_s20i::I32 = I32[GHC.Types.Bool_closure_tbl + (_c21P::I32<< 2)];
_c21Q::I32 = _s20i::I32& 3;
if (_c21Q::I32>= 2) goto c21k;
// outOfLine should follow:
_c21u::I32 = _s1uR::I32 + 1;
// emitReturn: Sequel: Assign
_s20j::I32 = _c21u::I32;
_c21y::I32 = _s1uX::I32 * 128;
// emitReturn: Sequel: Assign
_s20k::I32 = _c21y::I32;
_c21C::I32 = %MO_UU_Conv_W8_W32(I8[_s1uU::I32 + (_s1uR::I32<< 0)]);
// emitReturn: Sequel: Assign
_s20l::I32 = _c21C::I32;
_c21G::I32 = _s20l::I32;
// emitReturn: Sequel: Assign
_s20m::I32 = _c21G::I32;
_c21K::I32 = _s20m::I32 + _s20k::I32;
// emitReturn: Sequel: Assign
_s20n::I32 = _c21K::I32;
_c21O::I32 = %MO_S_Rem_W32(_s20n::I32, 4091);
// emitReturn: Sequel: Assign
_s20o::I32 = _c21O::I32;
_s1uR::I32 = _s20j::I32;
_s1uX::I32 = _s20o::I32;
goto c21a;
Gets optimized into:
// outOfLine should follow:
// emitReturn: Sequel: Assign
if (I32[((_s1uR::I32 == _s1uS::I32)<< 2) + GHC.Types.Bool_closure_tbl]&
3>= 2) goto c21k;
// outOfLine should follow:
// emitReturn: Sequel: Assign
// emitReturn: Sequel: Assign
// emitReturn: Sequel: Assign
// emitReturn: Sequel: Assign
// emitReturn: Sequel: Assign
// emitReturn: Sequel: Assign
_s1uR::I32 = _s1uR::I32 + 1;
_s1uX::I32 = %MO_S_Rem_W32(%MO_UU_Conv_W8_W32(I8[_s1uU::I32 + _s1uR::I32]) +
(_s1uX::I32<< 7),
4091);
For whatever reason, _s1uR has been inlined into an immediate increment, which
is definitely
incorrect since the dereference at _c21C is still working off the old version.
I'll look more closely at the inliner tomorrow and see if I can figure out
where the bug is.
Edward
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
