On Sun, 05 Apr 2015, Joe Perches wrote:
> On Mon, 2015-04-06 at 06:26 +0200, Nicholas Mc Guire wrote:
> > On Sun, 05 Apr 2015, Joe Perches wrote:
> > > Try it and look at the generated .lst files with and
> > > without the patch I sent.
> []
> > from all that I understood it should
> > be doable both as macro and inline.
>
> I think it _should_ be doable too but I also think
> the only reason gcc doesn't optimize the inline
> is because gcc's optimizer isn't good enough yet.
>
"unfortunately" I can't blame it on gcc - here is the initial toy-case
- test.c and either testi.h or testm.h included
- m = TIMEOUT or m = atoi(argv[1]);
both in the inline and the macro case gcc reduced the code to a single
load mediate or register instruction for the constant - so the optimizer
is doing its job.
test.c:
#include <stdio.h>
#define HZ 100
#define MSECS_PER_SEC 1000
#define TIMEOUT 100
#include "testi.h" /* inline msecs_to_jiffies */
//#include "testm.h" /* macro versions */
int main(int argc, char **argv) {
//int m = atoi(argv[0]); /* non-const */
int m = TIMEOUT; /* const */
printf("%lu\n",msecs_to_jiffies(m));
return 0;
}
testm.h:
#define msecs_to_jiffies(m) \
(__builtin_constant_p (m) \
? ((m) * HZ / MSECS_PER_SEC ) : __msecs_to_jiffies(m))
unsigned long __msecs_to_jiffies(int m)
{
return m * HZ / MSECS_PER_SEC ;
}
first case with a non-const
main:
.LFB12:
.cfi_startproc
subq $8, %rsp #,
.cfi_def_cfa_offset 16
movq 8(%rsi), %rdi # MEM[(char * *)argv_2(D) + 8B], MEM[(char *
*)argv_2(D) + 8B]
xorl %eax, %eax #
call atoi #
movl $1717986919, %edx #, tmp69
movl %eax, %ecx #, m
movl $.LC0, %edi #,
imull %edx # tmp69
sarl $31, %ecx #, tmp71
xorl %eax, %eax #
sarl $2, %edx #, tmp67
subl %ecx, %edx # tmp71, tmp67
movslq %edx, %rsi # tmp67, tmp72
call printf #
o
second with a constant:
main:
.LFB12:
.cfi_startproc
subq $8, %rsp #,
.cfi_def_cfa_offset 16
movl $10, %esi #,
movl $.LC0, %edi #,
xorl %eax, %eax #
call printf #
inline:
-------
testi.h:
static inline unsigned long __msecs_to_jiffies(int m)
{
return m * HZ / MSECS_PER_SEC;
}
static inline unsigned long msecs_to_jiffies(int m)
{
return __builtin_constant_p (m) ?
(m) * HZ / MSECS_PER_SEC : __msecs_to_jiffies(m);
}
first case with a non-const
main:
.LFB13:
.cfi_startproc
subq $8, %rsp #,
.cfi_def_cfa_offset 16
movq (%rsi), %rdi # *argv_1(D),
xorl %eax, %eax #
call atoi #
movl $1717986919, %edx #, tmp68
movl %eax, %ecx #, m
movl $.LC0, %edi #,
imull %edx # tmp68
sarl $31, %ecx #, tmp70
xorl %eax, %eax #
sarl $2, %edx #, tmp66
subl %ecx, %edx # tmp70, tmp66
movslq %edx, %rsi # tmp66, tmp71
call printf #
second with a constant:
main:
.LFB13:
.cfi_startproc
subq $8, %rsp #,
.cfi_def_cfa_offset 16
xorl %esi, %esi #
movl $.LC0, %edi #,
xorl %eax, %eax #
call printf #
giving it another run from scratch somewhere I simply screwed up or
overlooked some detail.
thx!
hofrat
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
