c/9475: Wrong ffs results

[Falk Hueffner]

>Number:         9475
>Category:       c
>Synopsis:       Wrong ffs results
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Jan 28 12:56:00 UTC 2003
>Closed-Date:
>Last-Modified:
>Originator:     Falk Hueffner
>Release:        3.4 20030128 (experimental)
>Organization:
>Environment:
System: Linux juist 2.5.59 #4 Sat Jan 18 12:46:41 CET 2003 alpha unknown unknown GNU/Linux
Architecture: alpha

	
host: alphaev68-unknown-linux-gnu
build: alphaev68-unknown-linux-gnu
target: alphaev68-unknown-linux-gnu
configured with: ../configure --enable-languages=c++ --disable-nls
>Description:
In certain contexts, the ffs builtin gives wrong results. It does not
happen when special purpose instructions are used instead of a libcall
(-mcpu=ev67). The problem is also not present in gcc 3.2, or in gcc
3.4 20030116 (so it looks like it has been introduced very recently).
>How-To-Repeat:
% cat ffs.c                                                
int printf(const char *format, ...);

int main(void) {
    int i;
    for (i = 0; i < 8; i++) {
        unsigned j = 1u << i;
        printf("%2d %10u %2d\n", i, j, __builtin_ffs(j));
    }
    return 0;
}
% gcc -O1 -W -Wall -mcpu=ev4 ffs.c  && ./a.out
 0          1  1
 1          2  0
 2          4  0
 3          8  0
 4         16  0
 5         32  0
 6         64  0
 7        128  0

The correct output would be:
 0          1  1
 1          2  2
 2          4  3
 3          8  4
 4         16  5
 5         32  6
 6         64  7
 7        128  8
	
>Fix:
	
>Release-Note:
>Audit-Trail:
>Unformatted: