[Bug middle-end/19606] New: wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[Bug middle-end/19606] New: wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[heinrich dot brand at fujitsu-siemens dot com]

The value of test should be 0x7ffffffe and is 0xfffffffe; 

Flags: none

(Also in 3.3.5)

#include <stdio.h>
signed char a=-4;
int test(){
	return (((unsigned int)(signed int) a ) / 2LL) ;
}

int main(void){
	int r;
	r=test();
	printf("test output:  %#x == %d      %x %x\n",r,r
		,(r==0x7ffffffe),(r==0xfffffffe));
	if(r == ( ((unsigned int)(signed int) (signed char) -4 ) / 2LL ))
  		printf("test successful\n");
  	else
  		printf("test failed\n");	  
	return 0; 
}

-- 
           Summary: wrong code for arith.expr: (((unsigned int)(signed int)
                    a ) / 2LL) with signed char a=-4
           Product: gcc
           Version: 3.4.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: heinrich dot brand at fujitsu-siemens dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: Intel-Linux
  GCC host triplet: Intel-Linux
GCC target triplet: Intel-Linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606

[Bug middle-end/19606] wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[falk at debian dot org]

------- Additional Comments From falk at debian dot org  2005-01-24 16:20 -------
Confirmed on alpha-linux. Has been broken since at least 2.95.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
      Known to fail|                            |2.95.3 3.3.3 3.4.3 4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606

[Bug middle-end/19606] wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-24 16:41 -------
Also on powerpc-darwin.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
  GCC build triplet|Intel-Linux                 |
   GCC host triplet|Intel-Linux                 |
 GCC target triplet|Intel-Linux                 |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606

[Bug middle-end/19606] wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-02-11 22:40 -------
Fails on i686 and on AMD64 also.  And with both gcc and g++... 
 
In c_finish_return, retval is "(long long int) (a / 2)".  That is already 
wrong. 
 
In parser_build_binary_op we build TRUNC_DIV_EXPR <(unsigned int) a, 2LL>,  
which is also already wrong (lost the "signed int" cast). 
 
And so we dig deeper into the parser... 
 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606

[Bug middle-end/19606] wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-02-11 22:50 -------
And there was great rejoicing... 
 
In c-typeck.c: 
3253          ovalue = value; 
(gdb) p debug_generic_expr (value) 
(intD.0) aD.1454  // Good. 
(gdb) next 
3254          value = convert (type, value); 
(gdb) p debug_generic_expr (ovalue) 
(intD.0) aD.1454 
(gdb) next 
3257          if (TREE_CODE (value) == INTEGER_CST) 
(gdb) p debug_generic_expr (value) 
(unsigned intD.3) aD.1454 
(gdb) p debug_generic_expr(type) 
unsigned intD.3 
 
So "convert (<unsigned int>, <(int) a>)" results in "<(unsigned int) a>" which 
is wrong. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606

[Bug c/19606] wrong code for arith.expr: (((unsigned int)(signed int) a ) / 2LL) with signed char a=-4

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-02-11 23:42 -------
Ignore comment #4, it's wrong. 
 
An IRC debug session revealed that the call to get_narrower at c-typeck.c:7492 
is incorrect; it's not being fed values properly converted to result_type. 
 
A little archaeology shows that the offending code has been there since 
forever (ie. it is in old-gcc). 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19606