Is it faster/better to use std.mem.concat when using only strings in a std.fmt.allocPrint? Or does it mostly depend on whats more readable?
#std.mem.concat vs std.fmt.allocPrint?
honest hill
lyric gazelle
concat is probably faster
because it doesn't us the format logic
and it doesn't require to use std.fmt.count on it
honest hill
yea that's what i was suspecting, thanks for confirming
wasnt sure if there was some sort of optimization to force it to use concat
dawn shale
just out of curiosity i did a couple tiny tests, its using fixedbufferallocator so probably not very representative but at the same time probably as optimal as the asm will get
main.main:
sub $0xa8,%rsp
vmovaps -0xa51f(%rip),%xmm0 ; 0x200240
vmovaps %xmm0,-0x60(%rsp)
movabs $0x68746e6576656c65,%rax
mov %rax,-0x50(%rsp)
int3
lea -0x60(%rsp),%rax
mov %rax,-0x70(%rsp)
movq $0x18,-0x68(%rsp)
lea -0x70(%rsp),%rax
mov %rax,-0x78(%rsp)
add $0xa8,%rsp
ret
main.main:
push %rbx
sub $0x130,%rsp
lea 0x30(%rsp),%rax
mov %rax,0x10(%rsp)
movq $0x18,0x18(%rsp)
movq $0x0,0x20(%rsp)
lea 0x10(%rsp),%rbx
mov %rbx,0x8(%rsp)
lea 0x8(%rsp),%rcx
mov $0x208c01,%edi
mov $0x5,%esi
mov $0x207e70,%edx
call 0x20ae90 ; <fmt.formatBuf__anon_4124>
test %ax,%ax
jne 0x20a7a6 ; <main.main+166>
lea 0x8(%rsp),%rcx
mov $0x2092a8,%edi
mov $0x6,%esi
mov $0x207e70,%edx
call 0x20ae90 ; <fmt.formatBuf__anon_4124>
test %ax,%ax
jne 0x20a7a6 ; <main.main+166>
lea 0x8(%rsp),%rcx
mov $0x209295,%edi
mov $0x5,%esi
mov $0x207e70,%edx
call 0x20ae90 ; <fmt.formatBuf__anon_4124>
test %ax,%ax
jne 0x20a7a6 ; <main.main+166>
lea 0x8(%rsp),%rcx
mov $0x208f5e,%edi
mov $0x8,%esi
mov $0x207e70,%edx
call 0x20ae90 ; <fmt.formatBuf__anon_4124>
test %ax,%ax
je 0x20a7bf ; <main.main+191>
int3
mov %rax,0x10(%rsp)
mov %rcx,0x18(%rsp)
mov %rbx,0x28(%rsp)
add $0x130,%rsp
pop %rbx
ret
mov 0x10(%rsp),%rax
mov 0x20(%rsp),%rcx
jmp 0x20a7a6 ; <main.main+166>
first is concat, second is allocPrint