|
|
@@ -481,65 +481,69 @@ implementation
|
|
|
{ the / operator is handled later }
|
|
|
(nodetype<>slashn) then
|
|
|
begin
|
|
|
- { convert constants to u32bit }
|
|
|
-{$ifndef cardinalmulfix}
|
|
|
- if (porddef(ld)^.typ<>u32bit) then
|
|
|
- begin
|
|
|
- { s32bit will be used for when the other is also s32bit }
|
|
|
-
|
|
|
- { the following line doesn't make any sense: it's the same as }
|
|
|
- { if ((porddef(rd)^.typ=u32bit) or (porddef(ld)^.typ=u32bit)) and }
|
|
|
- { (porddef(ld)^.typ<>u32bit) and (porddef(rd)^.typ=s32bit) then }
|
|
|
- { which can be simplified to }
|
|
|
- { if ((porddef(rd)^.typ=u32bit) and (porddef(rd)^.typ=s32bit) then }
|
|
|
- { which can never be true (JM) }
|
|
|
- if (porddef(rd)^.typ=s32bit) and (lt<>ordconstn) then
|
|
|
- left:=gentypeconvnode(left,s32bitdef)
|
|
|
- else
|
|
|
- left:=gentypeconvnode(left,u32bitdef);
|
|
|
- firstpass(left);
|
|
|
- end;
|
|
|
- if (porddef(rd)^.typ<>u32bit) then
|
|
|
- begin
|
|
|
- { s32bit will be used for when the other is also s32bit }
|
|
|
- if (porddef(ld)^.typ=s32bit) and (rt<>ordconstn) then
|
|
|
- right:=gentypeconvnode(right,s32bitdef)
|
|
|
- else
|
|
|
- right:=gentypeconvnode(right,u32bitdef);
|
|
|
- firstpass(right);
|
|
|
- end;
|
|
|
-{$else cardinalmulfix}
|
|
|
- { only do a conversion if the nodes have different signs }
|
|
|
- if (porddef(rd)^.typ=u32bit) xor (porddef(ld)^.typ=u32bit) then
|
|
|
- if (porddef(rd)^.typ=u32bit) then
|
|
|
- begin
|
|
|
- { can we make them both unsigned? }
|
|
|
- if (porddef(ld)^.typ in [u8bit,u16bit]) or
|
|
|
- (is_constintnode(left) and
|
|
|
- (nodetype <> subn) and
|
|
|
- (left.value > 0)) then
|
|
|
- left:=gentypeconvnode(left,u32bitdef)
|
|
|
- else
|
|
|
- left:=gentypeconvnode(left,s32bitdef);
|
|
|
- firstpass(left);
|
|
|
- end
|
|
|
- else {if (porddef(ld)^.typ=u32bit) then}
|
|
|
- begin
|
|
|
- { can we make them both unsigned? }
|
|
|
- if (porddef(rd)^.typ in [u8bit,u16bit]) or
|
|
|
- (is_constintnode(right) and
|
|
|
- (right.value > 0)) then
|
|
|
- right:=gentypeconvnode(right,u32bitdef)
|
|
|
- else
|
|
|
- right:=gentypeconvnode(right,s32bitdef);
|
|
|
- firstpass(right);
|
|
|
- end;
|
|
|
-{$endif cardinalmulfix}
|
|
|
- calcregisters(self,1,0,0);
|
|
|
+ if is_signed(ld) and
|
|
|
+ { then rd = u32bit }
|
|
|
+ { convert positive constants to u32bit }
|
|
|
+ not(is_constintnode(left) and
|
|
|
+ (tordconstnode(left).value >= 0)) and
|
|
|
+ { range/overflow checking on mixed signed/cardinal expressions }
|
|
|
+ { is only possible if you convert everything to 64bit (JM) }
|
|
|
+ ((aktlocalswitches * [cs_check_overflow,cs_check_range] <> []) and
|
|
|
+ (nodetype in [addn,subn,muln])) then
|
|
|
+ begin
|
|
|
+ { perform the operation in 64bit }
|
|
|
+ CGMessage(type_w_mixed_signed_unsigned);
|
|
|
+ left := gentypeconvnode(left,cs64bitdef);
|
|
|
+ firstpass(left);
|
|
|
+ right := gentypeconvnode(right,cs64bitdef);
|
|
|
+ firstpass(right);
|
|
|
+ end
|
|
|
+ else
|
|
|
+ begin
|
|
|
+ if is_signed(ld) and
|
|
|
+ not(is_constintnode(left) and
|
|
|
+ (tordconstnode(left).value >= 0)) and
|
|
|
+ (cs_check_range in aktlocalswitches) then
|
|
|
+ CGMessage(type_w_mixed_signed_unsigned2);
|
|
|
+ left := gentypeconvnode(left,u32bitdef);
|
|
|
+ firstpass(left);
|
|
|
+
|
|
|
+ if is_signed(rd) and
|
|
|
+ { then ld = u32bit }
|
|
|
+ { convert positive constants to u32bit }
|
|
|
+ not(is_constintnode(right) and
|
|
|
+ (tordconstnode(right).value >= 0)) and
|
|
|
+ ((aktlocalswitches * [cs_check_overflow,cs_check_range] <> []) and
|
|
|
+ (nodetype in [addn,subn,muln])) then
|
|
|
+ begin
|
|
|
+ { perform the operation in 64bit }
|
|
|
+ CGMessage(type_w_mixed_signed_unsigned);
|
|
|
+ left := gentypeconvnode(left,cs64bitdef);
|
|
|
+ firstpass(left);
|
|
|
+ right := gentypeconvnode(right,cs64bitdef);
|
|
|
+ firstpass(right);
|
|
|
+ end
|
|
|
+ else
|
|
|
+ begin
|
|
|
+ if is_signed(rd) and
|
|
|
+ not(is_constintnode(right) and
|
|
|
+ (tordconstnode(right).value >= 0)) and
|
|
|
+ (cs_check_range in aktlocalswitches) then
|
|
|
+ CGMessage(type_w_mixed_signed_unsigned2);
|
|
|
+ right := gentypeconvnode(right,u32bitdef);
|
|
|
+ firstpass(right);
|
|
|
+ end;
|
|
|
+ end;
|
|
|
+ { did we convert things to 64bit? }
|
|
|
+ if porddef(left.resulttype)^.typ = s64bit then
|
|
|
+ calcregisters(self,2,0,0)
|
|
|
+ else
|
|
|
+ begin
|
|
|
+ calcregisters(self,1,0,0);
|
|
|
{ for unsigned mul we need an extra register }
|
|
|
-{ registers32:=left.registers32+right.registers32; }
|
|
|
- if nodetype=muln then
|
|
|
- inc(registers32);
|
|
|
+ if nodetype=muln then
|
|
|
+ inc(registers32);
|
|
|
+ end;
|
|
|
convdone:=true;
|
|
|
end;
|
|
|
end
|
|
|
@@ -1198,37 +1202,9 @@ implementation
|
|
|
resulttype:=left.resulttype;
|
|
|
end;
|
|
|
end;
|
|
|
-{$ifdef cardinalmulfix}
|
|
|
- muln:
|
|
|
- { if we multiply an unsigned with a signed number, the result is signed }
|
|
|
- { in the other cases, the result remains signed or unsigned depending on }
|
|
|
- { the multiplication factors (JM) }
|
|
|
- if (left.resulttype^.deftype = orddef) and
|
|
|
- (right.resulttype^.deftype = orddef) and
|
|
|
- is_signed(right.resulttype) then
|
|
|
- resulttype := right.resulttype
|
|
|
- else resulttype := left.resulttype;
|
|
|
-(*
|
|
|
- subn:
|
|
|
- { if we substract a u32bit from a positive constant, the result becomes }
|
|
|
- { s32bit as well (JM) }
|
|
|
- begin
|
|
|
- if (right.resulttype^.deftype = orddef) and
|
|
|
- (left.resulttype^.deftype = orddef) and
|
|
|
- (porddef(right.resulttype)^.typ = u32bit) and
|
|
|
- is_constintnode(left) and
|
|
|
-{ (porddef(left.resulttype)^.typ <> u32bit) and}
|
|
|
- (left.value > 0) then
|
|
|
- begin
|
|
|
- left := gentypeconvnode(left,u32bitdef);
|
|
|
- firstpass(left);
|
|
|
- end;
|
|
|
- resulttype:=left.resulttype;
|
|
|
- end;
|
|
|
-*)
|
|
|
-{$endif cardinalmulfix}
|
|
|
else
|
|
|
- resulttype:=left.resulttype;
|
|
|
+ if not assigned(resulttype) then
|
|
|
+ resulttype:=left.resulttype;
|
|
|
end;
|
|
|
end;
|
|
|
|
|
|
@@ -1237,7 +1213,13 @@ begin
|
|
|
end.
|
|
|
{
|
|
|
$Log$
|
|
|
- Revision 1.18 2000-11-29 00:30:31 florian
|
|
|
+ Revision 1.19 2000-12-16 15:55:32 jonas
|
|
|
+ + warning when there is a chance to get a range check error because of
|
|
|
+ automatic type conversion to u32bit
|
|
|
+ * arithmetic operations with a cardinal and a signed operand are carried
|
|
|
+ out in 64bit when range checking is on ("merged" from fixes branch)
|
|
|
+
|
|
|
+ Revision 1.18 2000/11/29 00:30:31 florian
|
|
|
* unused units removed from uses clause
|
|
|
* some changes for widestrings
|
|
|
|
Jonas Maebe