improve va_arg assembly on arm targets

[folkertdev]

tracking issue: #44930

For this example

#![feature(c_variadic)]

#[unsafe(no_mangle)]
unsafe extern "C" fn variadic(a: f64, mut args: ...) -> f64 {
    let b = args.arg::<f64>();
    let c = args.arg::<f64>();

    a + b + c
}

We currently generate (via llvm):

variadic:
    sub     sp, sp, #12
    stmib   sp, {r2, r3}
    vmov    d0, r0, r1
    add     r0, sp, #4
    vldr    d1, [sp, #4]
    add     r0, r0, #15
    bic     r0, r0, #7
    vadd.f64        d0, d0, d1
    add     r1, r0, #8
    str     r1, [sp]
    vldr    d1, [r0]
    vadd.f64        d0, d0, d1
    vmov    r0, r1, d0
    add     sp, sp, #12
    bx      lr

LLVM is not doing a good job. By using our own emit_ptr_va_arg we can save 3 instructions:

variadic:
	sub sp, sp, #12
	stmib sp, {r2, r3}
	vmov d0, r0, r1
	add r2, sp, #4
	vldr d1, [sp, #4]
	add r2, r2, #16
	vldr d2, [sp, #12]
	vadd.f64 d0, d0, d1
	vadd.f64 d0, d0, d2
	vmov r0, r1, d0
	str r2, [sp], #12
	bx lr

But clang generates even fewer instructions by removing the use of r2 in the above code. The crucial missing piece is ending the lifetime of the va_list value, see https://godbolt.org/z/TK3rsdExG. That va_list is generated internally, so I'm guessing that is why the lifetime end is not emitted by default? But the va_list is a stack value so obviously the lifetime really does end before a return.

With that we generate exactly the instructions that clang generates (though the loads and adds are slightly reordered):

variadic:
    sub sp, sp, #12
    stmib sp, {r2, r3}
    vmov d0, r0, r1
    vldr d1, [sp, #4]
    vldr d2, [sp, #12]
    vadd.f64 d0, d0, d1
    vadd.f64 d0, d0, d2
    vmov r0, r1, d0
    add sp, sp, #12
    bx lr

The arguments to emit_ptr_va_arg are based on the clang implemenation.

r? @workingjubilee (I can re-roll if your queue is too full, but you do seem like the right person here)

[rustbot]

workingjubilee is currently at their maximum review capacity.
They may take a while to respond.

[rustbot]

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin