On Tuesday 21 June 2005 16:08, [email protected] wrote:
> Memory is memory. Pushed from the stack or as a 4 byte immediate value you
> still have to get those 4 bytes somewhere although with the pointer you DO
> actually stand a chance GCC might enregister the pointer variable.
>
> Sure you don't think instruction bytes fetching is free ;->
I think that instruction fetch is deeply prefetched, thus
push immediate
do not need to wait for operand, no AGU to allocate, nothing,
while
push N(%ebp)
depends on %ebp value, on (%ebp+N) value, will need AGU.
> BTW, I don't give a shit about the size advantage. Put the 3 byte EBP ref
> and the 5 byte push imm32 in a loop and measure them - I know what the
> answer will be.
You know wrong.
# cat t.c
#define rdtscl(low) asm volatile("rdtsc" : "=a" (low) : : "edx")
#define NL "\n"
#include <stdio.h>
int main() {
int i,k,start,end;
int v = 1234;
for(k=0; k<6; k++) {
rdtscl(start);
for(i=0; i<1000; i++) {
asm(NL
" push %0" NL
" pop %%eax" NL
" push %0" NL
" pop %%eax" NL
" push %0" NL
" pop %%eax" NL
" push %0" NL
" pop %%eax" NL
: /* outputs */
: "m" (v) /* inputs */
: "ax" /* clobbers */
);
}
rdtscl(end);
printf("Took %u CPU cycles ", end-start);
rdtscl(start);
for(i=0; i<1000; i++) {
asm(NL
" push %%ebx" NL
" pop %%eax" NL
" push %%ebx" NL
" pop %%eax" NL
" push %%ebx" NL
" pop %%eax" NL
" push %%ebx" NL
" pop %%eax" NL
: /* outputs */
: /* inputs */
: "ax" /* clobbers */
);
}
rdtscl(end);
printf("Took %u CPU cycles\n", end-start);
}
return 0;
}
# gcc -O2 -falign-loops=64 t.c
# ./a.out
Took 9574 CPU cycles Took 8068 CPU cycles
Took 9575 CPU cycles Took 8058 CPU cycles
Took 9134 CPU cycles Took 8061 CPU cycles
Took 9134 CPU cycles Took 8043 CPU cycles
Took 9134 CPU cycles Took 8043 CPU cycles
Took 9134 CPU cycles Took 8043 CPU cycles
# cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 11
model name : Intel(R) Celeron(TM) CPU 1200MHz
stepping : 1
cpu MHz : 1196.207
cache size : 256 KB
physical id : 0
siblings : 1
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 2
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse
bogomips : 2359.29
# gcc -O2 -falign-loops=64 -c t.c
# objdump -d -r t.o
t.o: file format elf32-i386
Disassembly of section .text:
00000000 <main>:
0: 55 push %ebp
1: 89 e5 mov %esp,%ebp
3: 53 push %ebx
4: 50 push %eax
5: 83 e4 f0 and $0xfffffff0,%esp
8: 83 ec 10 sub $0x10,%esp
b: c7 45 f8 d2 04 00 00 movl $0x4d2,0xfffffff8(%ebp)
12: 31 db xor %ebx,%ebx
14: 0f 31 rdtsc
16: ba e7 03 00 00 mov $0x3e7,%edx
1b: 89 c1 mov %eax,%ecx
1d: 90 nop
1e: 90 nop
1f: 90 nop
20: 90 nop
21: 90 nop
22: 90 nop
23: 90 nop
24: 90 nop
25: 90 nop
26: 90 nop
27: 90 nop
28: 90 nop
29: 90 nop
2a: 90 nop
2b: 90 nop
2c: 90 nop
2d: 90 nop
2e: 90 nop
2f: 90 nop
30: 90 nop
31: 90 nop
32: 90 nop
33: 90 nop
34: 90 nop
35: 90 nop
36: 90 nop
37: 90 nop
38: 90 nop
39: 90 nop
3a: 90 nop
3b: 90 nop
3c: 90 nop
3d: 90 nop
3e: 90 nop
3f: 90 nop
40: ff 75 f8 pushl 0xfffffff8(%ebp)
43: 58 pop %eax
44: ff 75 f8 pushl 0xfffffff8(%ebp)
47: 58 pop %eax
48: ff 75 f8 pushl 0xfffffff8(%ebp)
4b: 58 pop %eax
4c: ff 75 f8 pushl 0xfffffff8(%ebp)
4f: 58 pop %eax
50: 4a dec %edx
51: 79 ed jns 40 <main+0x40>
53: 0f 31 rdtsc
55: 83 ec 08 sub $0x8,%esp
58: 29 c8 sub %ecx,%eax
5a: 50 push %eax
5b: 68 00 00 00 00 push $0x0
5c: R_386_32 .rodata.str1.1
60: e8 fc ff ff ff call 61 <main+0x61>
61: R_386_PC32 printf
65: 0f 31 rdtsc
67: ba e7 03 00 00 mov $0x3e7,%edx
6c: 89 c1 mov %eax,%ecx
6e: 83 c4 10 add $0x10,%esp
71: eb 0d jmp 80 <main+0x80>
73: 90 nop
74: 90 nop
75: 90 nop
76: 90 nop
77: 90 nop
78: 90 nop
79: 90 nop
7a: 90 nop
7b: 90 nop
7c: 90 nop
7d: 90 nop
7e: 90 nop
7f: 90 nop
80: 53 push %ebx
81: 58 pop %eax
82: 53 push %ebx
83: 58 pop %eax
84: 53 push %ebx
85: 58 pop %eax
86: 53 push %ebx
87: 58 pop %eax
88: 4a dec %edx
89: 79 f5 jns 80 <main+0x80>
8b: 0f 31 rdtsc
8d: 83 ec 08 sub $0x8,%esp
90: 29 c8 sub %ecx,%eax
92: 50 push %eax
93: 68 14 00 00 00 push $0x14
94: R_386_32 .rodata.str1.1
98: 43 inc %ebx
99: e8 fc ff ff ff call 9a <main+0x9a>
9a: R_386_PC32 printf
9e: 83 c4 10 add $0x10,%esp
a1: 83 fb 05 cmp $0x5,%ebx
a4: 0f 8e 6a ff ff ff jle 14 <main+0x14>
aa: 31 c0 xor %eax,%eax
ac: 8b 5d fc mov 0xfffffffc(%ebp),%ebx
af: c9 leave
b0: c3 ret
--
vda
-
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/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
