Since C++ supports (most) of C's functions, let's test the new and improved IO utils performance!
Most notably cout and, the classic, printf
cout.cc
#include "iostream"
int main() {
std::cout << "Hello World!\n";
}
printf.cc
#include "iostream"
int main() {
printf("Hello World!\n");
}
Running g++ -fno-asynchronous-unwind-tables -fno-exceptions -fno-rtti -fverbose-asm -Wall -Wextra printf.cc -O3 -masm=intel -S -o printf
.section __TEXT,__text,regular,pure_instructions
.build_version macos, 12, 0 sdk_version 12, 3
.intel_syntax noprefix
.globl _main ## -- Begin function main
.p2align 4, 0x90
_main: ## @main
## %bb.0:
push rbp
mov rbp, rsp
lea rdi, [rip + L_str]
call _puts
xor eax, eax
pop rbp
ret
## -- End function
.section __TEXT,__cstring,cstring_literals
L_str: ## @str
.asciz "Hello World!"
.subsections_via_symbols
And using objdump -d on the compiled binary from standard g++ printf.cc:
printf: file format mach-o 64-bit x86-64
Disassembly of section __TEXT,__text:
0000000100003f70 <_main>:
100003f70: 55 pushq %rbp
100003f71: 48 89 e5 movq %rsp, %rbp
100003f74: 48 8d 3d 2b 00 00 00 leaq 43(%rip), %rdi ## 0x100003fa6 <dyld_stub_binder+0x100003fa6>
100003f7b: b0 00 movb $0, %al
100003f7d: e8 04 00 00 00 callq 0x100003f86 <dyld_stub_binder+0x100003f86>
100003f82: 31 c0 xorl %eax, %eax
100003f84: 5d popq %rbp
100003f85: c3 retq
Disassembly of section __TEXT,__stubs:
0000000100003f86 <__stubs>:
100003f86: ff 25 74 40 00 00 jmpq *16500(%rip) ## 0x100008000 <dyld_stub_binder+0x100008000>
Disassembly of section __TEXT,__stub_helper:
0000000100003f8c <__stub_helper>:
100003f8c: 4c 8d 1d 75 40 00 00 leaq 16501(%rip), %r11 ## 0x100008008 <__dyld_private>
100003f93: 41 53 pushq %r11
100003f95: ff 25 65 00 00 00 jmpq *101(%rip) ## 0x100004000 <dyld_stub_binder+0x100004000>
100003f9b: 90 nop
100003f9c: 68 00 00 00 00 pushq $0
100003fa1: e9 e6 ff ff ff jmp 0x100003f8c <__stub_helper>
Compare this to the cout implementation:
.section __TEXT,__text,regular,pure_instructions
.build_version macos, 12, 0 sdk_version 12, 3
.intel_syntax noprefix
.globl _main ## -- Begin function main
.p2align 4, 0x90
_main: ## @main
## %bb.0:
push rbp
mov rbp, rsp
mov rdi, qword ptr [rip + __ZNSt3__14coutE@GOTPCREL]
lea rsi, [rip + L_.str]
mov edx, 13
call __ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m
xor eax, eax
pop rbp
ret
## -- End function
.globl __ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m ## -- Begin function _ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m
.weak_def_can_be_hidden __ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m
.p2align 4, 0x90
__ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m: ## @_ZNSt3__124__put_character_sequenceIcNS_11char_traitsIcEEEERNS_13basic_ostreamIT_T0_EES7_PKS4_m
## %bb.0:
push rbp
mov rbp, rsp
push r15
push r14
push r13
push r12
push rbx
sub rsp, 40
mov r14, rdx
mov r13, rsi
mov rbx, rdi
lea rdi, [rbp - 80]
mov rsi, rbx
call __ZNSt3__113basic_ostreamIcNS_11char_traitsIcEEE6sentryC1ERS3_
cmp byte ptr [rbp - 80], 0
je LBB1_5
## %bb.1:
mov rax, qword ptr [rbx]
mov rax, qword ptr [rax - 24]
mov rdi, qword ptr [rbx + rax + 40]
lea r12, [rbx + rax]
mov ecx, 176
and ecx, dword ptr [rbx + rax + 8]
add r14, r13
cmp ecx, 32
mov r15, r13
cmove r15, r14
mov eax, dword ptr [rbx + rax + 144]
cmp eax, -1
jne LBB1_3
## %bb.2:
lea rax, [rbp - 56]
mov qword ptr [rbp - 64], rdi ## 8-byte Spill
mov rdi, rax
mov rsi, r12
call __ZNKSt3__18ios_base6getlocEv
mov rsi, qword ptr [rip + __ZNSt3__15ctypeIcE2idE@GOTPCREL]
lea rdi, [rbp - 56]
call __ZNKSt3__16locale9use_facetERNS0_2idE
mov rcx, qword ptr [rax]
mov rdi, rax
mov esi, 32
call qword ptr [rcx + 56]
mov byte ptr [rbp - 41], al ## 1-byte Spill
lea rdi, [rbp - 56]
call __ZNSt3__16localeD1Ev
mov rdi, qword ptr [rbp - 64] ## 8-byte Reload
movsx eax, byte ptr [rbp - 41] ## 1-byte Folded Reload
mov dword ptr [r12 + 144], eax
LBB1_3:
movsx r9d, al
mov rsi, r13
mov rdx, r15
mov rcx, r14
mov r8, r12
call __ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_
test rax, rax
jne LBB1_5
## %bb.4:
mov rax, qword ptr [rbx]
mov rax, qword ptr [rax - 24]
lea rdi, [rbx + rax]
mov esi, dword ptr [rbx + rax + 32]
or esi, 5
call __ZNSt3__18ios_base5clearEj
LBB1_5:
lea rdi, [rbp - 80]
call __ZNSt3__113basic_ostreamIcNS_11char_traitsIcEEE6sentryD1Ev
mov rax, rbx
add rsp, 40
pop rbx
pop r12
pop r13
pop r14
pop r15
pop rbp
ret
## -- End function
.private_extern __ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_ ## -- Begin function _ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_
.globl __ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_
.weak_def_can_be_hidden __ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_
.p2align 4, 0x90
__ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_: ## @_ZNSt3__116__pad_and_outputIcNS_11char_traitsIcEEEENS_19ostreambuf_iteratorIT_T0_EES6_PKS4_S8_S8_RNS_8ios_baseES4_
## %bb.0:
push rbp
mov rbp, rsp
push r15
push r14
push r13
push r12
push rbx
sub rsp, 56
test rdi, rdi
je LBB2_19
## %bb.1:
mov r14, r8
mov r15, rcx
mov r13, rdi
mov dword ptr [rbp - 44], r9d ## 4-byte Spill
mov rax, rcx
sub rax, rsi
mov rcx, qword ptr [r8 + 24]
xor r12d, r12d
sub rcx, rax
cmovg r12, rcx
mov qword ptr [rbp - 88], rdx ## 8-byte Spill
mov rbx, rdx
sub rbx, rsi
test rbx, rbx
jle LBB2_3
## %bb.2:
mov rax, qword ptr [r13]
mov rdi, r13
mov rdx, rbx
call qword ptr [rax + 96]
cmp rax, rbx
jne LBB2_19
LBB2_3:
test r12, r12
jle LBB2_15
## %bb.4:
cmp r12, 23
mov qword ptr [rbp - 80], r14 ## 8-byte Spill
jae LBB2_8
## %bb.5:
lea eax, [r12 + r12]
mov byte ptr [rbp - 72], al
lea r14, [rbp - 71]
jmp LBB2_9
LBB2_8:
lea rbx, [r12 + 16]
and rbx, -16
mov rdi, rbx
call __Znwm
mov r14, rax
mov qword ptr [rbp - 56], rax
or rbx, 1
mov qword ptr [rbp - 72], rbx
mov qword ptr [rbp - 64], r12
LBB2_9:
mov eax, dword ptr [rbp - 44] ## 4-byte Reload
movzx esi, al
mov rdi, r14
mov rdx, r12
call _memset
mov byte ptr [r14 + r12], 0
test byte ptr [rbp - 72], 1
je LBB2_11
## %bb.10:
mov rsi, qword ptr [rbp - 56]
jmp LBB2_12
LBB2_11:
lea rsi, [rbp - 71]
LBB2_12:
mov r14, qword ptr [rbp - 80] ## 8-byte Reload
mov rax, qword ptr [r13]
mov rdi, r13
mov rdx, r12
call qword ptr [rax + 96]
mov rbx, rax
test byte ptr [rbp - 72], 1
je LBB2_14
## %bb.13:
mov rdi, qword ptr [rbp - 56]
call __ZdlPv
LBB2_14:
cmp rbx, r12
jne LBB2_19
LBB2_15:
mov rsi, qword ptr [rbp - 88] ## 8-byte Reload
sub r15, rsi
test r15, r15
jle LBB2_17
## %bb.16:
mov rax, qword ptr [r13]
mov rdi, r13
mov rdx, r15
call qword ptr [rax + 96]
cmp rax, r15
jne LBB2_19
LBB2_17:
mov qword ptr [r14 + 24], 0
jmp LBB2_20
LBB2_19:
xor r13d, r13d
LBB2_20:
mov rax, r13
add rsp, 56
pop rbx
pop r12
pop r13
pop r14
pop r15
pop rbp
ret
## -- End function
.section __TEXT,__cstring,cstring_literals
L_.str: ## @.str
.asciz "Hello World!\n"
.subsections_via_symbols
And a disassembled dump that is too big to put here!
results
| process | printf | cout |
|---|---|---|
g++ -> asm (with cleaning) |
20 | 223 |
g++ -> asm (no cleaning) |
25 | 1585 |
objdump on binary |
29 | 1028 |
It's weird that the disassembled binary has fewer SLOC than the standard assembly output - but printf is still undeniably faster!
So why use cout?