関数の入口・出口における処理はpush/popとsub/addでどのくらい性能差が出るか

asm.s

        .text
        .intel_syntax noprefix

        .globl  pushpop
        .p2align        4, 0x90
        .type   pushpop,@function
pushpop:
        push rax
        mov  rax, qword ptr [rdi]
        lea  rcx, [rax + 1]
        mov  qword ptr [rdi], rcx
        pop  rcx
        ret

        .globl  subadd
        .p2align        4, 0x90
        .type   subadd,@function
subadd:
        sub  rsp, 8
        mov  rax, qword ptr [rdi]
        lea  rcx, [rax + 1]
        mov  qword ptr [rdi], rcx
        add  rsp, 8
        ret

        # スタックを実行不可とする（ビルド時の警告抑制）
        .section        ".note.GNU-stack","",@progbits

disas.txt

objdump -d -Mintel a.out

a.out:     file format elf64-x86-64


Disassembly of section .init:

0000000000001000 <_init>:
    1000:	f3 0f 1e fa          	endbr64
    1004:	48 83 ec 08          	sub    rsp,0x8
    1008:	48 8b 05 c1 2f 00 00 	mov    rax,QWORD PTR [rip+0x2fc1]        # 3fd0 <__gmon_start__@Base>
    100f:	48 85 c0             	test   rax,rax
    1012:	74 02                	je     1016 <_init+0x16>
    1014:	ff d0                	call   rax
    1016:	48 83 c4 08          	add    rsp,0x8
    101a:	c3                   	ret

Disassembly of section .plt:

0000000000001020 <printf@plt-0x10>:
    1020:	ff 35 ca 2f 00 00    	push   QWORD PTR [rip+0x2fca]        # 3ff0 <_GLOBAL_OFFSET_TABLE_+0x8>
    1026:	ff 25 cc 2f 00 00    	jmp    QWORD PTR [rip+0x2fcc]        # 3ff8 <_GLOBAL_OFFSET_TABLE_+0x10>
    102c:	0f 1f 40 00          	nop    DWORD PTR [rax+0x0]

0000000000001030 <printf@plt>:
    1030:	ff 25 ca 2f 00 00    	jmp    QWORD PTR [rip+0x2fca]        # 4000 <printf@GLIBC_2.2.5>
    1036:	68 00 00 00 00       	push   0x0
    103b:	e9 e0 ff ff ff       	jmp    1020 <_init+0x20>

Disassembly of section .plt.got:

0000000000001040 <__cxa_finalize@plt>:
    1040:	ff 25 9a 2f 00 00    	jmp    QWORD PTR [rip+0x2f9a]        # 3fe0 <__cxa_finalize@GLIBC_2.2.5>
    1046:	66 90                	xchg   ax,ax

Disassembly of section .text:

0000000000001050 <_start>:
    1050:	f3 0f 1e fa          	endbr64
    1054:	31 ed                	xor    ebp,ebp
    1056:	49 89 d1             	mov    r9,rdx
    1059:	5e                   	pop    rsi
    105a:	48 89 e2             	mov    rdx,rsp
    105d:	48 83 e4 f0          	and    rsp,0xfffffffffffffff0
    1061:	50                   	push   rax
    1062:	54                   	push   rsp
    1063:	45 31 c0             	xor    r8d,r8d
    1066:	31 c9                	xor    ecx,ecx
    1068:	48 8d 3d e1 00 00 00 	lea    rdi,[rip+0xe1]        # 1150 <main>
    106f:	ff 15 4b 2f 00 00    	call   QWORD PTR [rip+0x2f4b]        # 3fc0 <__libc_start_main@GLIBC_2.34>
    1075:	f4                   	hlt
    1076:	66 2e 0f 1f 84 00 00 	cs nop WORD PTR [rax+rax*1+0x0]
    107d:	00 00 00 

0000000000001080 <deregister_tm_clones>:
    1080:	48 8d 3d 91 2f 00 00 	lea    rdi,[rip+0x2f91]        # 4018 <__TMC_END__>
    1087:	48 8d 05 8a 2f 00 00 	lea    rax,[rip+0x2f8a]        # 4018 <__TMC_END__>
    108e:	48 39 f8             	cmp    rax,rdi
    1091:	74 15                	je     10a8 <deregister_tm_clones+0x28>
    1093:	48 8b 05 2e 2f 00 00 	mov    rax,QWORD PTR [rip+0x2f2e]        # 3fc8 <_ITM_deregisterTMCloneTable@Base>
    109a:	48 85 c0             	test   rax,rax
    109d:	74 09                	je     10a8 <deregister_tm_clones+0x28>
    109f:	ff e0                	jmp    rax
    10a1:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]
    10a8:	c3                   	ret
    10a9:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]

00000000000010b0 <register_tm_clones>:
    10b0:	48 8d 3d 61 2f 00 00 	lea    rdi,[rip+0x2f61]        # 4018 <__TMC_END__>
    10b7:	48 8d 35 5a 2f 00 00 	lea    rsi,[rip+0x2f5a]        # 4018 <__TMC_END__>
    10be:	48 29 fe             	sub    rsi,rdi
    10c1:	48 89 f0             	mov    rax,rsi
    10c4:	48 c1 ee 3f          	shr    rsi,0x3f
    10c8:	48 c1 f8 03          	sar    rax,0x3
    10cc:	48 01 c6             	add    rsi,rax
    10cf:	48 d1 fe             	sar    rsi,1
    10d2:	74 14                	je     10e8 <register_tm_clones+0x38>
    10d4:	48 8b 05 fd 2e 00 00 	mov    rax,QWORD PTR [rip+0x2efd]        # 3fd8 <_ITM_registerTMCloneTable@Base>
    10db:	48 85 c0             	test   rax,rax
    10de:	74 08                	je     10e8 <register_tm_clones+0x38>
    10e0:	ff e0                	jmp    rax
    10e2:	66 0f 1f 44 00 00    	nop    WORD PTR [rax+rax*1+0x0]
    10e8:	c3                   	ret
    10e9:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]

00000000000010f0 <__do_global_dtors_aux>:
    10f0:	f3 0f 1e fa          	endbr64
    10f4:	80 3d 1d 2f 00 00 00 	cmp    BYTE PTR [rip+0x2f1d],0x0        # 4018 <__TMC_END__>
    10fb:	75 2b                	jne    1128 <__do_global_dtors_aux+0x38>
    10fd:	55                   	push   rbp
    10fe:	48 83 3d da 2e 00 00 	cmp    QWORD PTR [rip+0x2eda],0x0        # 3fe0 <__cxa_finalize@GLIBC_2.2.5>
    1105:	00 
    1106:	48 89 e5             	mov    rbp,rsp
    1109:	74 0c                	je     1117 <__do_global_dtors_aux+0x27>
    110b:	48 8b 3d fe 2e 00 00 	mov    rdi,QWORD PTR [rip+0x2efe]        # 4010 <__dso_handle>
    1112:	e8 29 ff ff ff       	call   1040 <__cxa_finalize@plt>
    1117:	e8 64 ff ff ff       	call   1080 <deregister_tm_clones>
    111c:	c6 05 f5 2e 00 00 01 	mov    BYTE PTR [rip+0x2ef5],0x1        # 4018 <__TMC_END__>
    1123:	5d                   	pop    rbp
    1124:	c3                   	ret
    1125:	0f 1f 00             	nop    DWORD PTR [rax]
    1128:	c3                   	ret
    1129:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]

0000000000001130 <frame_dummy>:
    1130:	f3 0f 1e fa          	endbr64
    1134:	e9 77 ff ff ff       	jmp    10b0 <register_tm_clones>
    1139:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]

0000000000001140 <inc>:
    1140:	8b 07                	mov    eax,DWORD PTR [rdi]
    1142:	8d 48 01             	lea    ecx,[rax+0x1]
    1145:	89 0f                	mov    DWORD PTR [rdi],ecx
    1147:	c3                   	ret
    1148:	0f 1f 84 00 00 00 00 	nop    DWORD PTR [rax+rax*1+0x0]
    114f:	00 

0000000000001150 <main>:
    1150:	41 57                	push   r15
    1152:	41 56                	push   r14
    1154:	53                   	push   rbx
    1155:	48 83 ec 10          	sub    rsp,0x10
    1159:	48 c7 44 24 08 00 00 	mov    QWORD PTR [rsp+0x8],0x0
    1160:	00 00 
    1162:	0f 31                	rdtsc
    1164:	48 89 d3             	mov    rbx,rdx
    1167:	48 c1 e3 20          	shl    rbx,0x20
    116b:	48 09 c3             	or     rbx,rax
    116e:	41 bf 00 ca 9a 3b    	mov    r15d,0x3b9aca00
    1174:	4c 8d 74 24 08       	lea    r14,[rsp+0x8]
    1179:	0f 1f 80 00 00 00 00 	nop    DWORD PTR [rax+0x0]
    1180:	4c 89 f7             	mov    rdi,r14
    1183:	e8 f8 00 00 00       	call   1280 <pushpop>
    1188:	49 ff cf             	dec    r15
    118b:	75 f3                	jne    1180 <main+0x30>
    118d:	0f 31                	rdtsc
    118f:	48 c1 e2 20          	shl    rdx,0x20
    1193:	48 09 c2             	or     rdx,rax
    1196:	48 29 da             	sub    rdx,rbx
    1199:	66 48 0f 6e ca       	movq   xmm1,rdx
    119e:	66 0f 62 0d 6a 0e 00 	punpckldq xmm1,XMMWORD PTR [rip+0xe6a]        # 2010 <_IO_stdin_used+0x10>
    11a5:	00 
    11a6:	66 0f 5c 0d 72 0e 00 	subpd  xmm1,XMMWORD PTR [rip+0xe72]        # 2020 <_IO_stdin_used+0x20>
    11ad:	00 
    11ae:	66 0f 28 c1          	movapd xmm0,xmm1
    11b2:	66 0f 15 c1          	unpckhpd xmm0,xmm1
    11b6:	f2 0f 58 c1          	addsd  xmm0,xmm1
    11ba:	f2 0f 5e 05 6e 0e 00 	divsd  xmm0,QWORD PTR [rip+0xe6e]        # 2030 <_IO_stdin_used+0x30>
    11c1:	00 
    11c2:	48 8b 4c 24 08       	mov    rcx,QWORD PTR [rsp+0x8]
    11c7:	48 8d 3d 6a 0e 00 00 	lea    rdi,[rip+0xe6a]        # 2038 <_IO_stdin_used+0x38>
    11ce:	41 bf 00 ca 9a 3b    	mov    r15d,0x3b9aca00
    11d4:	be 00 ca 9a 3b       	mov    esi,0x3b9aca00
    11d9:	b0 01                	mov    al,0x1
    11db:	e8 50 fe ff ff       	call   1030 <printf@plt>
    11e0:	48 c7 44 24 08 00 00 	mov    QWORD PTR [rsp+0x8],0x0
    11e7:	00 00 
    11e9:	0f 31                	rdtsc
    11eb:	48 89 d3             	mov    rbx,rdx
    11ee:	48 c1 e3 20          	shl    rbx,0x20
    11f2:	48 09 c3             	or     rbx,rax
    11f5:	4c 8d 74 24 08       	lea    r14,[rsp+0x8]
    11fa:	66 0f 1f 44 00 00    	nop    WORD PTR [rax+rax*1+0x0]
    1200:	4c 89 f7             	mov    rdi,r14
    1203:	e8 88 00 00 00       	call   1290 <subadd>
    1208:	49 ff cf             	dec    r15
    120b:	75 f3                	jne    1200 <main+0xb0>
    120d:	0f 31                	rdtsc
    120f:	48 c1 e2 20          	shl    rdx,0x20
    1213:	48 09 c2             	or     rdx,rax
    1216:	48 29 da             	sub    rdx,rbx
    1219:	66 48 0f 6e ca       	movq   xmm1,rdx
    121e:	66 0f 62 0d ea 0d 00 	punpckldq xmm1,XMMWORD PTR [rip+0xdea]        # 2010 <_IO_stdin_used+0x10>
    1225:	00 
    1226:	66 0f 5c 0d f2 0d 00 	subpd  xmm1,XMMWORD PTR [rip+0xdf2]        # 2020 <_IO_stdin_used+0x20>
    122d:	00 
    122e:	66 0f 28 c1          	movapd xmm0,xmm1
    1232:	66 0f 15 c1          	unpckhpd xmm0,xmm1
    1236:	f2 0f 58 c1          	addsd  xmm0,xmm1
    123a:	f2 0f 5e 05 ee 0d 00 	divsd  xmm0,QWORD PTR [rip+0xdee]        # 2030 <_IO_stdin_used+0x30>
    1241:	00 
    1242:	48 8b 4c 24 08       	mov    rcx,QWORD PTR [rsp+0x8]
    1247:	48 8d 3d 24 0e 00 00 	lea    rdi,[rip+0xe24]        # 2072 <_IO_stdin_used+0x72>
    124e:	be 00 ca 9a 3b       	mov    esi,0x3b9aca00
    1253:	b0 01                	mov    al,0x1
    1255:	e8 d6 fd ff ff       	call   1030 <printf@plt>
    125a:	48 8d 3d 4a 0e 00 00 	lea    rdi,[rip+0xe4a]        # 20ab <_IO_stdin_used+0xab>
    1261:	be 02 00 00 00       	mov    esi,0x2
    1266:	ba 03 00 00 00       	mov    edx,0x3
    126b:	31 c0                	xor    eax,eax
    126d:	e8 be fd ff ff       	call   1030 <printf@plt>
    1272:	31 c0                	xor    eax,eax
    1274:	48 83 c4 10          	add    rsp,0x10
    1278:	5b                   	pop    rbx
    1279:	41 5e                	pop    r14
    127b:	41 5f                	pop    r15
    127d:	c3                   	ret
    127e:	66 90                	xchg   ax,ax

0000000000001280 <pushpop>:
    1280:	50                   	push   rax
    1281:	48 8b 07             	mov    rax,QWORD PTR [rdi]
    1284:	48 8d 48 01          	lea    rcx,[rax+0x1]
    1288:	48 89 0f             	mov    QWORD PTR [rdi],rcx
    128b:	59                   	pop    rcx
    128c:	c3                   	ret
    128d:	0f 1f 00             	nop    DWORD PTR [rax]

0000000000001290 <subadd>:
    1290:	48 83 ec 08          	sub    rsp,0x8
    1294:	48 8b 07             	mov    rax,QWORD PTR [rdi]
    1297:	48 8d 48 01          	lea    rcx,[rax+0x1]
    129b:	48 89 0f             	mov    QWORD PTR [rdi],rcx
    129e:	48 83 c4 08          	add    rsp,0x8
    12a2:	c3                   	ret

Disassembly of section .fini:

00000000000012a4 <_fini>:
    12a4:	f3 0f 1e fa          	endbr64
    12a8:	48 83 ec 08          	sub    rsp,0x8
    12ac:	48 83 c4 08          	add    rsp,0x8
    12b0:	c3                   	ret

main.c

#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

#define LOOP 1000000000L

uint64_t pushpop(uint64_t*);
uint64_t subadd(uint64_t*);

uint64_t run_bench(const char *name, uint64_t (*f)(uint64_t*), size_t loop) {
  uint64_t start = __builtin_readcyclecounter();
  uint64_t cnt = 0;
  for (size_t i = 0; i < loop; ++i) {
    f(&cnt);
  }
  uint64_t end = __builtin_readcyclecounter();
  uint64_t elapsed = end - start;
  return elapsed;
}

int main(int argc, char *argv[]) {
  size_t loop = LOOP;
  size_t num_bench = 3;
  if (argc >= 2) {
    loop = strtol(argv[1], NULL, 0);
  }
  if (argc >= 3) {
    num_bench = strtol(argv[2], NULL, 0);
  }

  printf("num_bench: %zu\n"
         "loop: %zu\n"
         "\n",
         num_bench, loop);

  printf("| run |   pp elapsed | pp average |   sa elapsed | sa average |\n"
         "|-----|--------------|------------|--------------|------------|\n");

  uint64_t sum_pp = 0, sum_sa = 0;
  for (size_t i = 0; i < num_bench; ++i) {
    uint64_t elapsed_pp = run_bench("pushpop", pushpop, loop);
    uint64_t elapsed_sa = run_bench("subadd ", subadd, loop);
    sum_pp += elapsed_pp;
    sum_sa += elapsed_sa;

    printf("| %3zu | %12lu | %#10.6f | %12lu | %#10.6f |\n",
           i + 1, elapsed_pp, (double)elapsed_pp / loop, elapsed_sa, (double)elapsed_sa / loop);
  }

  printf("| avg | %#12.1f | %#10.6f | %#12.1f | %#10.6f |\n",
         (double)sum_pp / num_bench, (double)sum_pp / loop / num_bench,
         (double)sum_sa / num_bench, (double)sum_sa / loop / num_bench);
}

Makefile

SRCS := main.c asm.s

all:
        make -r a.out

a.out: $(SRCS) Makefile
        clang -O2 -o $@ $(SRCS)

.PHONY: disas
disas:
        objdump -d -Mintel a.out

計測結果.md

$ ./a.out
pushpop: 1000000000 times => elapsed 5713678920, average 5.71368, cnt = 1000000000
subadd: 1000000000 times => elapsed 5709335488, average 5.70934, cnt = 1000000000
inc(&i) = 2, i = 3
$ ./a.out
pushpop: 1000000000 times => elapsed 5702241796, average 5.70224, cnt = 1000000000
subadd: 1000000000 times => elapsed 5710384354, average 5.71038, cnt = 1000000000
inc(&i) = 2, i = 3
$ ./a.out
pushpop: 1000000000 times => elapsed 5707916838, average 5.70792, cnt = 1000000000
subadd: 1000000000 times => elapsed 5708659500, average 5.70866, cnt = 1000000000
inc(&i) = 2, i = 3

実験回	pushpop	subadd	diff (pp - sa)
1	5.71368	5.70934	0.00434
2	5.70224	5.71038	-0.00814
3	5.70792	5.70866	-0.00074

差は無いと言って良い？

Ubuntu 24.04 on WSL2
CPU: AMD Ryzen 9 5950X 16-Core Processor 3.4GHz

$ time ./a.out
num_bench: 3
loop: 1000000000

run	pp elapsed	pp average	sa elapsed	sa average
1	5714961367	5.714961	5695248421	5.695248
2	5697037688	5.697038	5693585820	5.693586
3	5694287530	5.694288	5695270470	5.695270
avg	5702095528.3	5.702096	5694701570.3	5.694702

real 0m10.013s
user 0m10.057s
sys 0m0.001s

Ubuntu 24.04 on WSL2
CPU: Intel(R) Core(TM) Ultra 7 165U 2.688GHz

$ time ./a.out
num_bench: 3
loop: 1000000000

run	pp elapsed	pp average	sa elapsed	sa average
1	1605336124	1.605336	1691755508	1.691756
2	1647749740	1.647750	1680450364	1.680450
3	1690682586	1.690683	1666465438	1.666465
avg	1647922816.7	1.647923	1679557103.3	1.679557

real 0m3.640s
user 0m3.714s
sys 0m0.001s