On Friday, 22 January 2016 at 14:06:42 UTC, Adam D. Ruppe wrote:
On Friday, 22 January 2016 at 12:18:53 UTC, anonymous wrote:
I don't know much about these things, but it seems to be the
`ret;`.
Right. This is an ordinary D function so the compiler generates
code to set up a stack for local variables. It looks like:
push ebp;
mov ebp, esp;
sub EBP, some_size;
/* sometimes a few other register saves */
/*
your code here
*/
/* sometimes a few other register restores */
leave;
ret;
`leave` btw is the same as `mov esp,ebp; pop ebp;` - it undoes
the result of those first three instructions.
All this setup stuff is about creating a stack frame for the
function's local variables. If you ret without restoring the
frame, all local variables (and return addresses!) from there
on are going to be out of sync and will lead to memory access
violations. That's what happened to you.
If you want to write a whole function in assembly without the
compiler inserting any additional code, start it off with `asm
{ naked; }` inside so dmd knows what you are trying to do. Then
you are in complete control.
Otherwise, remember to clear the frame correctly, or better
yet, just return using the ordinary D statement instead of the
asm instruction.
naked version:
int pmovmskb2(byte16 v)
{
asm
{
naked;
push RBP;
mov RBP, RSP;
sub RSP, 0x20;
movdqa dword ptr[RBP-0x10], XMM0;
mov dword ptr[RBP-0x18], 0;
movdqa XMM0, dword ptr[RBP-0x10];
pmovmskb EAX, XMM0;
mov RSP, RBP;
pop RBP;
ret;
}
}
Note that there is maybe a DMD codegen bug because the asm
generated for the non naked version copy the result to the stack
and then the stack to result but after pmovmskb it's already
setup in EAX.
000000000044C580h push rbp
000000000044C581h mov rbp, rsp
000000000044C584h sub rsp, 20h
000000000044C588h movdqa dqword ptr [rbp-10h], xmm0
000000000044C58Dh mov dword ptr [rbp-18h], 00000000h
000000000044C594h movdqa xmm0, dqword ptr [rbp-10h]
000000000044C599h pmovmskb eax, xmm0 ; already in result
000000000044C59Dh mov dword ptr [rbp-18h], eax ; what?
000000000044C5A0h mov eax, dword ptr [rbp-18h] ; what?
000000000044C5A3h mov rsp, rbp
000000000044C5A6h pop rbp
000000000044C5A7h ret
