Cat program in x64 FASM

Question

I wrote cat program in x64 FASM and I would like to know if there is anything that can be improved.

This implementation of cat supports:

• Reading from stdin.
• Multiple arguments.

I checked few cases and program behaves exactly like original cat command.

Is there anything I could do to improve performance of this program? I compared it to original cat and my program is around 200ms slower on 512MB file.

Code:

format ELF64 executable 3
entry start
BUFSIZE equ 1048576
segment readable writeable
 fd dq 0
 buf rb BUFSIZE
segment readable executable
start:
 mov rbx, 1
 cmp [rsp], rbx
 je read
open:
 xor rsi, rsi ;fd=open(pathname,flags)
 mov rdi, [rsp+rbx*8+8]
 mov rax, 2
 syscall
 mov [fd], rax
read:
 mov rdx, BUFSIZE ;bytes_read=read(fd,buf,BUFSIZE)
 mov rsi, buf
 mov rdi, [fd]
 mov rax, 0
 syscall
write:
 mov rdx, rax ;write(STDOUT_FILENO,buf,bytes_read)
 mov rsi, buf
 mov rdi, 1
 mov rax, 1
 syscall
 test rdx, rdx ;if(bytes_read!=0) goto read
 jnz read
close:
 mov rdi, [fd] ;close(fd)
 mov rax, 3
 syscall
 inc rbx
 cmp rbx, [rsp] ;if(rbx<argc)
 jb open
exit:
 xor rdi, rdi
 mov rax, 60
 syscall

• Martin R: I have rolled back your last edit. Please don't change or add to the code in your question after you have received answers. See What should I do when someone answers my question? Thank you.

Answer 1

Linux has the sendfile system call, which copies data between file descriptors instead of copying the data to userspace and back to kernel space. That may be more efficient.

Your code looks very clean and organized. If you had used named constants instead of magic numbers, you might not even need some of the comments:

 mov rdx, tax
 mov rsi, but
 mov rdi, STDOUT_FILENO
 mov rax, SYS_write
 syscall

But I think even with these constants, the comments are still helpful, so I'd probably keep them anyway.

You should add error handling for stdout write errors, such as ENOSPACE.

• DeBos: I wasn't aware of sendfile to be honest. I will try that and let you know if I see any difference in performance. My code looks very clean and organized thanks to other user. I wrote same program in x86 NASM before, but even back then my code wasn't that bad I think. These are not magic numbers as there are comments, but do you think its good idea to kinda rewrite needed constants that are defined in C and remove comments as everything will be clearly visible without them?
• DeBos: In relation to error handling, I'm firstly making program work, then optimizing it and adding error handling at the end.
• Roland Illig: I intentionally wrote "you might not need comments". I'd say just give it a try and see which variant looks nicer. :)
• DeBos: I added improved code to the question, so you can tell me which variant looks nicer/more readable/cleaner to you. Personally code without comments looks cleaner to me, but thats my code, so it will always be clean to me (until I will leave it for week and then come back later of course).
• DeBos: EDIT: you can find improved code here: pastebin.com/pS2ku0xz