gcc changes the logical order of the code.
I have a 16 byte structure on which I use a macro version of bzero() that looks
like this:
#define lipc_bzero( __buf, __size ) \
{ \
size_t __i = ( ( __size ) >> ( sizeof( long ) >> 1 ) ); \
while ( ( __i-- ) > 0 ) { \
( ( long * )__buf )[__i] = ( long )0; \
} \
}
and I use it like this:
/* > code */
struct header_t a;
my_bzero( &a, sizeof( a ) );
a.id = 0x4350494c;
a.rid = handle->rid; /* a 32bit random number */
a.st_size = sizeof( a );
a.size = 1024;
lipc_send( s, &a, sizeof( a ), MSG_NOSIGNAL, -1 );
/* < code */
if I compile the code with: "-c -ansi -g -std=gnu99 -fPIC -DPIC -D_REENTRANT
-Os -ffast-math" the asm code is:
;; [snip]-------------------------------------------
.loc 1 469 0
leal -28(%ebp), %ecx
movl $0, -4(%ecx,%eax,4)
decl %eax
jne .L46
.LBE9:
.loc 1 470 0
movl $1129335116, -28(%ebp)
.loc 1 477 0
subl $12, %esp
.loc 1 471 0
movl 8(%esi), %eax
.loc 1 472 0
movl $16, -20(%ebp)
.loc 1 473 0
movl %edi, -16(%ebp)
.loc 1 471 0
movl %eax, -24(%ebp)
.loc 1 477 0
leal 24(%ebp), %eax
movl 44(%esi), %edx
pushl %eax
movl $16, %eax
pushl $16384
subl %edx, %eax
pushl %eax
leal (%ecx,%edx), %edx
pushl %edx
pushl (%esi)
.LCFI23:
call [EMAIL PROTECTED]
;; [snip] ---------------------------------
and works as a charm, but if add "-funroll-loops" (to unroll the lipc_bzero()),
the asm is:
;; [snip]----------------------------
.LBB10:
.loc 1 470 0
movl $1129335116, -28(%ebp)
.loc 1 477 0
subl $12, %esp
.loc 1 471 0
movl 8(%esi), %ecx
.loc 1 477 0
leal 24(%ebp), %eax
.LVL26:
pushl %eax
pushl $16384
leal -28(%ebp,%edx), %eax
.loc 1 471 0
movl %ecx, -24(%ebp)
.loc 1 477 0
movl $16, %ecx
subl %edx, %ecx
pushl %ecx
pushl %eax
.loc 1 472 0
movl $16, -20(%ebp)
.loc 1 473 0
movl %edi, -16(%ebp)
.loc 1 477 0
pushl (%esi)
.LBB11:
.loc 1 469 0
movl $0, -16(%ebp)
movl $0, -20(%ebp)
movl $0, -24(%ebp)
movl $0, -28(%ebp)
.LBE11:
.loc 1 477 0
.LCFI23:
call [EMAIL PROTECTED]
;; [snip]----------------------------------------
In other words, the code became:
struct header_t a;
a.id = 0x4350494c;
a.rid = handle->rid; /* a 32bit random number */
a.st_size = sizeof( a );
a.size = 1024;
my_bzero( &a, sizeof( a ) );
lipc_send( s, &a, sizeof( a ), MSG_NOSIGNAL, -1 );
I did not read every bug in the database, but I think this is new (or not :)).
I'm sorry I can't give you the whole code, but I will try to help in any other
way.
Regards,
Mihai
--
Summary: gcc changes code order
Product: gcc
Version: 4.1.1
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: mihai dot dontu at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30855
