vmovdqu issue

[Biterider]

Hi
I'm having some problems with the vmovdqu instruction. It seems that the memory argument is not being encoded correctly or I'm doing something wrong.
The test code is very simple:

Code: [Select]

.data
MyYmm YMMWORD 123456789012345678901234567890

.code
start proc
  vmovdqu ymm0, MyYmm
  ret
start endp

The debug output looks like:

Code: [Select]

00007FF615751000 C5 FE 6F 85 F8 1F 00 00 vmovdqu     ymm0,ymmword ptr [rbp+1FF8h] 
00007FF615751008 C3                   ret 
triggering a GPF.

Regards, Biterider

[jj2007]

Same with 32-bit code. Where does the [?bp... creep in?

ML 14.0 assembles it as follows (but complains about bad data initialiser etc):
0040658A                ³.  C5FE:6F05 00704000      vmovdqu ymm0, [407000]

[Biterider]

Hi
Playing a bit with this instruction, I've noticed that writing the ymm register back to memory is encoded correctly.


Biterider

[habran]

Hi Biterider,
good find, this is definitely a bug
It works fine with a new codegen.c and it will be easy fixed in next release:

Code: [Select]

000000013FF35A3C C5 FE 6F 05 7F 5C 00 00 vmovdqu     ymm0,ymmword ptr [MyYmm (013FF3B6C3h)Thanks

[Biterider]

Hi habran
Thank you for addressing this issue. Do you have an idea when do you will get a fix?
In the meantime, I'm going to replace vmovdqu with vlddqu, which does a similar job.

Biterider

[habran]

Hi Biterider,
I don't know if John is available to upload the fix so I attached InstrTablev2.h for you
replace the one in H folder with this one and build uasm source
that will fix the problem.

[Biterider]

Thank you Habran!

[johnsa]

This fix is now also on 2.50 branch on Git.

[Biterider]