Re: MPS prstack

[Gerd Möllmann]

Helmut Eller <eller.helmut@gmail.com> writes:

> On Mon, May 27 2024, Gerd Möllmann wrote:
>
>>>   prstack.stack[prstack.sp++] = e;
>>
>> Hm. No idea when the increment is sequenced. The side effect of postfix
>> ++ could be applied anywhere before the next sequence point. But where
>> is that sequence point? There is one at the end of the whole expression
>> I think.
>
> The generated code with O0 looks like this:
>
>    ...
>    0x000055555587fcb1 <+28>:    call   0x55555587fbd0 <grow_print_stack>
>    0x000055555587fcb6 <+33>:    mov    0x8845c3(%rip),%rcx        # 
> 0x555556104280 <prstack>
>    0x000055555587fcbd <+40>:    mov    0x8845cc(%rip),%rax        # 
> 0x555556104290 <prstack+16>
>    0x000055555587fcc4 <+47>:    lea    0x1(%rax),%rdx
>    0x000055555587fcc8 <+51>:    mov    %rdx,0x8845c1(%rip)        # 
> 0x555556104290 <prstack+16>
>    0x000055555587fccf <+58>:    mov    %rax,%rdx
>    0x000055555587fcd2 <+61>:    mov    %rdx,%rax
>    0x000055555587fcd5 <+64>:    shl    $0x3,%rax
>    0x000055555587fcd9 <+68>:    sub    %rdx,%rax
>    0x000055555587fcdc <+71>:    shl    $0x3,%rax
>    0x000055555587fce0 <+75>:    add    %rcx,%rax
>    0x000055555587fce3 <+78>:    mov    0x10(%rbp),%rcx
>    0x000055555587fce7 <+82>:    mov    0x18(%rbp),%rbx
>    ...
>
> So it looks like prstack.sp is updated with the move at +51, i.e. before
> the stack slot is initialized.
>
> We could write
>
>   prstack.stack[prstack.sp] = e;
>   prstack.sp += 1;
>
> Not sure how we then guarantee that e still in a register or the control
> stack at the second line.
>
> But my question was more whether MPS actually does call scan_prstack at
> "any moment" or if this is just a theoretical possibility for some
> possible future version of MPS.

I think it is "at any moment". At least I can't find anything in the
docs hinting at something else.

Which probably means that we can't wholy trust stack points. They may be
off.