gcc-2.95.3 internal error on head/MachOp.hs

[Simon Marlow]

On 15 September 2004 08:45, Donald Bruce Stewart wrote:

> The problem seems to be in the if-goto code sequence marked below. In
> the old code generator we had a giant switch, but now we have an
> if-goto block, which branches to a switch for some cases. gcc seems
> to get confused by the if-goto construct, in this case, and it is
> dependent on the number of branches in the function somehow:
> 
>         IF_(s1Ee_ret) {
>                 W_ _c1Es;
>                 FB_
>                 _c1Es = (W_)((*((StgWord16*)((*R1.p) + (-0x2)))));
>               >>>>>
>                 if (_c1Es < 0x5) goto _c1EB;
>                 if (_c1Es < 0x7) goto _c1EC;
>                 if (_c1Es < 0x8) goto _c1ED;
>                 if (_c1Es != 0x8) goto _c1EE;
>               <<<<<
>                 R1.p = (P_)(W_)GHCziBase_True_closure;
>                 Sp=Sp+1;
>                 JMP_(*((P_)((*Sp) + (-0x14))));
>         ...
>             _c1EB:
>                 switch (_c1Es) {
>                     case 0x0: goto _c1EF;
>                     case 0x1: goto _c1EG;
>                     case 0x2: goto _c1EH;
>                     case 0x3: goto _c1EI;
>                     case 0x4: goto _c1EE;
>                 }

Regardless of the gcc bug, that sequence of ifs looks really dumb.  Can
we see the STG fragment that gives rise to this code?

Cheers,
	Simon