Hacker News Reader

Comment by Flow | Hacker News Reader

Exactly what purpose does the surrounding instructions serve in this and similar simple cases? Is it compiler dogma or a missed uncommon optimization?

        push    rbp
        mov     rbp, rsp
        popcnt  eax, edi
        pop     rbp
        ret

View on HN

It's a function invocation following x86 C calling convention in assembly. push ebp pushes the base stack pointer of the caller of count() onto the stack, and mov ebp, esp sets the base stack pointer for the callee count(). pop ebp pops the base stack pointer and ret pops the return address into the program counter. count_ones() is replaced with its assembly counterpart, which stores the result in return register eax. This is as optimized as count() as a user-created function can be in assembly given the calling convention. https://en.m.wikipedia.org/wiki/X86_calling_conventions

Correction: rsp and rbp are 64-bit registers storing 64-bit stack memory addresses, while popcount eax, edi is taking the 32-bit parameter as input and returning the 32-bit return value. The calling convention is x64 C.

GCC, for example, allows a "-fomit-frame-pointer" [1] optimization option which would get rid of this. I'm not sure why this isn't done by default in optimized builds. Maybe it has something to do with stack unwinding: if the functions panics for some reason, or triggers a CPU fault, there's no way to get the correct backtrace if you don't have the frame pointers on the stack.

[1]: https://gcc.gnu.org/onlinedocs/gcc-3.4.4/gcc/Optimize-Option...

rustc will omit the frame pointer if you add "-C debuginfo=0".

  example::count:
        popcnt  eax, edi
        ret

IIRC it's needed on x86 but not on x86_64

A function is being called, so it gets a stack frame at the start of the call and that stack frame goes away when it returns.

rbp points to the base of the current stack frame (register base pointer) and rsp points to the top of the current stack frame (register stack pointer).

What the function actually does is just the single popcnt instruction.

Yes, I get the reserve space for local variables and so on.

But this is a leaf-function and one that has no register trashing or temporary variables. Could this be inlined later on via link-time optimizations? Will the linker then remove that function prologue and epilogue?

Yeah, it can certainly be inlined or optimized away with various compiler options. However, as you might surmise, debugging without stack frames becomes rather difficult as functions corresponding to source code disappear into a goo of assembly.

The example with slightly different options (as mentioned in another comment) gives what you're looking for: https://godbolt.org/g/GlkEQK

gcc has the -fomit-leaf-frame-pointer option at least.

Those are the function prologue and function epilogue. https://en.wikipedia.org/wiki/Function_prologue

Reply on news.ycombinator.com