Re: rex.W prefix in jumps on x86_64

[Marc Nieper-Wißkirchen]

PS:

The following patch corrects at least my issue. (The _callr function
already did the correct thing.) All tests pass (on an Intel Core
i7-9700).

diff --git a/lib/jit_x86-cpu.c b/lib/jit_x86-cpu.c
index 8150fa1..1db522a 100644
--- a/lib/jit_x86-cpu.c
+++ b/lib/jit_x86-cpu.c
@@ -3457,7 +3457,7 @@ _calli_p(jit_state_t *_jit, jit_word_t i0)
 static void
 _jmpr(jit_state_t *_jit, jit_int32_t r0)
 {
-    rex(0, WIDE, _NOREG, _NOREG, r0);
+    rex(0, 0, _NOREG, _NOREG, r0);
     ic(0xff);
     mrm(0x03, 0x04, r7(r0));
 }


Am Mo., 15. Juni 2020 um 16:44 Uhr schrieb Marc Nieper-Wißkirchen
<marc.nieper+gnu@gmail.com>:
>
> Hi Paul,
>
> > >>  look at the disassembly, I see things like
> > >>
> > >>  rex.W jmpq *%rax
> > >>
> > >>  when an instruction like `jmpr R0' is jitted. I am wondering about
> > >> the
> > >>  REX prefix, which doesn't seem to be necessary. In 64 bit-mode,
> > >> where
> > >>  the operand size is fixed for jumps and calls.
> > >>
> >
> > Was that introduced by commit 8dc376f?. This one introduced a bug but I
> > can't exactly put my finger on what's going on.
>
> I am using the git head version, so that commit is included in my test
> code. But that commit has something to do with jmpi's, not jmpr's.
>
> Which (other) bug has been introduced by that commit?
>
> Marc