Dave Korn wrote:
Yes, I think so; aggregation is the right word for it. Or maybe
scalarization. If you wrap all these chars in a struct, can SRA handle it?
I tried
#######################################################################
/* file testmanychar.c */
extern void g (int, char *, char *, char *);
void
f (void)
{
struct {
char x0, x1, x2, x3, x4, x5, x6, x7;
} s = {0,0,0,0,0,0,0};
/* assuming s.x0 is word aligned on a x86_64, and variables are
bytes in memory, we could clear all the variables in one machine
instruction */
g (10, &s.x0, &s.x1, &s.x2);
g (20, &s.x2, &s.x3, &s.x4);
g (30, &s.x4, &s.x5, &s.x6);
g (40, &s.x6, &s.x7, &s.x0);
}
#######################################################################
I compiled it with
gcc-trunk -S -std=gnu99 -O3 -fverbose-asm \
-march=core2 -mtune=core2 testmanychar.c
and I got:
#######################################################################
.globl f
.type f, @function
f:
.LFB0:
.cfi_startproc
movq %rbx, -24(%rsp) #,
movq %rbp, -16(%rsp) #,
movq %r12, -8(%rsp) #,
movl $10, %edi #,
subq $40, %rsp #,
.cfi_def_cfa_offset 48
leaq 2(%rsp), %rbp #, tmp59
.cfi_offset 12, -16
.cfi_offset 6, -24
.cfi_offset 3, -32
leaq 4(%rsp), %r12 #, tmp64
leaq 1(%rsp), %rdx #, tmp61
movq %rbp, %rcx # tmp59,
movq %rsp, %rsi #,
movq $0, (%rsp) #, s
call g #
leaq 3(%rsp), %rdx #, tmp66
movq %r12, %rcx # tmp64,
movq %rbp, %rsi # tmp59,
movl $20, %edi #,
leaq 6(%rsp), %rbp #, tmp70
call g #
leaq 5(%rsp), %rdx #, tmp72
movq %rbp, %rcx # tmp70,
movq %r12, %rsi # tmp64,
movl $30, %edi #,
call g #
leaq 7(%rsp), %rdx #, tmp77
movq %rsp, %rcx #,
movq %rbp, %rsi # tmp70,
movl $40, %edi #,
call g #
movq 16(%rsp), %rbx #,
movq 24(%rsp), %rbp #,
movq 32(%rsp), %r12 #,
addq $40, %rsp #,
.cfi_def_cfa_offset 8
ret
.cfi_endproc
.LFE0:
.size f, .-f
.ident "GCC: (GNU) 4.5.0 20100309 (experimental) [trunk
revision 157303]"
#######################################################################
So it seems indeed that a structure can be cleared efficiently. However,
having a structure has probably also some shortcommings: I would guess
that it is much harder for GCC to keep my flags in registers only.
Regards.
--
Basile STARYNKEVITCH http://starynkevitch.net/Basile/
email: basile<at>starynkevitch<dot>net mobile: +33 6 8501 2359
8, rue de la Faiencerie, 92340 Bourg La Reine, France
*** opinions {are only mines, sont seulement les miennes} ***
