builtin memcpy behaviour compared to gcc when destinatio​n unknown...

[Quuxplusone]

Bugzilla Link	PR12909
Status	NEW
Importance	P enhancement
Reported by	viswabramana.rajesh@gmail.com
Reported on	2012-05-22 03:57:38 -0700
Last modified on	2012-05-24 09:45:56 -0700
Version	3.0
Hardware	PC Windows XP
CC	geek4civic@gmail.com, llvm-bugs@lists.llvm.org
Fixed by commit(s)
Attachments	libcalls.c (643 bytes, text/plain)
Blocks
Blocked by
See also

Created attachment 8615
library calls file

__builtin_memcpy_chk function optimization O123 behaviour in clang.

I have noticed in clang for builtin __memcpy_chk function when destination
buffer unknown than, builtin calls optimized to direct memcpy.
Where as if destination has known pointer at compilation time than it will call
builtin mem_chk function which in turn calls memcpy.

Test code:

typedef __SIZE_TYPE__ size_t;
extern void abort (void);
extern void *memcpy (void *, const void *, size_t);
#define memcpy(dst, src, len) \
  __builtin___memcpy_chk (dst, src, len, __builtin_object_size (dst, 0))

int l1 = 1;
int chk_calls;
int main ()
{
    char *s3 = "FGH";
    char *r;
    char buf1[10], buf2[10];
    chk_calls = 0;
    memcpy(buf1, s3, l1); // Destination is known(buf1), length is not known calls memcpy_chk
    r = l1 == 1 ? buf1 : buf2;
    memcpy(r, s3, l1 + 1); // Destination not known, clang optimizing call to memcpy
    if (chk_calls != 2)
        abort ();
}
Above test code along with attached libcalls.c,when run in gcc with
optimizaiton O123 test code passes.
Same set of code when run in Clang(O123) its aborted, since second builtin
memcpy call optimized to direct memcpy function and chk_calls not incremented
to 2

Optimizing builtin memcpy function when destination buffer is unknown, Is it
Accepted behaviour?

-viswa

[Quuxplusone]

Attached libcalls.c (643 bytes, text/plain): library calls file