2
0
mirror of git://gcc.gnu.org/git/gcc.git synced 2025-03-24 13:11:44 +08:00

Revert unwanted change in previous commit.

From-SVN: r145925
This commit is contained in:
Arnaud Charlet 2009-04-10 18:02:23 +02:00
parent fad0600db0
commit 1147c70401

@ -4799,7 +4799,7 @@ package body Exp_Ch4 is
-- Check for 64-bit division available, or long shifts if the divisor
-- is a small power of 2 (since such divides will be converted into
-- long shifts.
-- long shifts).
if Esize (Ltyp) > 32
and then not Support_64_Bit_Divides_On_Target
@ -5715,7 +5715,7 @@ package body Exp_Ch4 is
-- Otherwise we have to introduce conversions (conversions are also
-- required in the universal cases, since the runtime routine is
-- typed using one of the standard types.
-- typed using one of the standard types).
else
Rewrite (N,
@ -7936,72 +7936,6 @@ package body Exp_Ch4 is
or else
(Is_Fixed_Point_Type (Target_Type) and then Conversion_OK (N)))
then
-- Handle case in which type conversions from real types to integers
-- are truncated instead of rounded. For example, in the .NET target
-- the only instructions available for conversion from float types to
-- integers truncate the result. That is, the result of Integer (3.9)
-- is 3 instead of 4. The frontend expansion done here to handle also
-- negative values is the following composition of conditional
-- expressions:
-- (if Abs (Operand - Float(Integer(Operand))) >= 0.5 then
-- (if Operand >= 0.0 then
-- Integer(Operand) + 1
-- else
-- Integer(Operand) - 1)
-- else
-- Integer(Operand))
if Integer_Truncation_On_Target and then Comes_From_Source (N) then
declare
Conv_Node : Node_Id;
begin
-- This code is weird, why are we doing all these copy tree
-- operations, instead of just capturing Integer(Operand)
-- once and then reusing the value instead of forcing this
-- conversion to be done four times! ???
-- There should be no New_Copy_Tree operations in the below
-- code at all???
Conv_Node := New_Copy_Tree (N);
Set_Parent (Conv_Node, Parent (N));
Set_Comes_From_Source (Conv_Node, False);
Analyze_And_Resolve (Conv_Node, Target_Type);
Rewrite (N,
Make_Conditional_Expression (Loc,
Expressions => New_List (
Make_Op_Ge (Loc,
Left_Opnd =>
Make_Op_Abs (Loc,
Make_Op_Subtract (Loc,
New_Copy_Tree (Operand),
Make_Type_Conversion (Loc,
New_Reference_To (Etype (Operand), Loc),
New_Copy_Tree (Conv_Node)))),
Right_Opnd => Make_Real_Literal (Loc, Ureal_Half)),
Make_Conditional_Expression (Loc,
Expressions => New_List (
Make_Op_Ge (Loc,
Left_Opnd => New_Copy_Tree (Operand),
Right_Opnd => Make_Real_Literal (Loc, Ureal_0)),
Make_Op_Add (Loc,
New_Copy_Tree (Conv_Node),
Make_Integer_Literal (Loc, 1)),
Make_Op_Subtract (Loc,
New_Copy_Tree (Conv_Node),
Make_Integer_Literal (Loc, 1)))),
New_Copy_Tree (Conv_Node))));
Analyze_And_Resolve (N, Target_Type);
return;
end;
end if;
-- One more check here, gcc is still not able to do conversions of
-- this type with proper overflow checking, and so gigi is doing an
-- approximation of what is required by doing floating-point compares