[edk2-devel] MOV64 and CLANGPDB

[edk2-devel] MOV64 and CLANGPDB

[Leif Lindholm via groups.io]

I would very much like to have a MOV64 that's universally usable.
That's sort of the point.

However, I'm wondering if the error is in fact triggered by the first
stanza (movz), which doesn't mask its result, and will have 48 sign
extended bits above the ones we care about for the instruction.

Annoyingly I cannot reproduce the build error in my setup, so I can't
verify if that is the problem.

Michael, could you possibly do a test and change
  movz      Reg, (Val) >> 48, lsl #48             ; \
to
  movz      Reg, ((Val) >> 48) & 0xffff, lsl #48             ; \

on line 51 in MdePkg/Include/AArch64/AsmMacroLib.h ?

And then rebuild the MOV64 version of ArmStandaloneMmCoreEntryPoint?

/
    Leif


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#121119): https://edk2.groups.io/g/devel/message/121119
Mute This Topic: https://groups.io/mt/111172486/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [edk2-devel] MOV64 and CLANGPDB

[Michael Kubacki via groups.io]

That worked. Let me know what you'd like to do with 
https://github.com/tianocore/edk2/pull/10760.

---

Before (original code)

...ModuleEntryPoint.iiii:58:12: error: immediate must be an integer in 
range [0, 65535].

movz x9, (-1) >> 48, lsl #48 ; movk x9, ((-1) >> 32) & 0xffff, lsl #32 ; 
movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9, (-1) & 0xffff

---

After (new code)

   svc #0

   movz x9, ((-1) >> 48) & 0xffff, lsl #48 ; movk x9, ((-1) >> 32) & 
0xffff, lsl #32 ; movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9, 
(-1) & 0xffff
   cmp x0, x9
   cset x0, ne
   mov x9, xzr
   ret

---

On 2/13/2025 5:33 PM, Leif Lindholm via groups.io wrote:
> I would very much like to have a MOV64 that's universally usable.
> That's sort of the point.
> 
> However, I'm wondering if the error is in fact triggered by the first
> stanza (movz), which doesn't mask its result, and will have 48 sign
> extended bits above the ones we care about for the instruction.
> 
> Annoyingly I cannot reproduce the build error in my setup, so I can't
> verify if that is the problem.
> 
> Michael, could you possibly do a test and change
>    movz      Reg, (Val) >> 48, lsl #48             ; \
> to
>    movz      Reg, ((Val) >> 48) & 0xffff, lsl #48             ; \
> 
> on line 51 in MdePkg/Include/AArch64/AsmMacroLib.h ?
> 
> And then rebuild the MOV64 version of ArmStandaloneMmCoreEntryPoint?
> 
> /
>      Leif
> 
> 
> 
> 



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#121120): https://edk2.groups.io/g/devel/message/121120
Mute This Topic: https://groups.io/mt/111172486/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [edk2-devel] MOV64 and CLANGPDB

[Leif Lindholm via groups.io]

On Fri, 14 Feb 2025 at 01:54, Michael Kubacki
<mikuback@linux.microsoft.com> wrote:
> That worked.

Splendid! Thanks.

Ard figured out how to reproduce the issue as well - the behaviour
changes between
-target aarch64-windows-none and -target aarch64-linux-gnu, so more LLP64 fun.

> Let me know what you'd like to do with
> https://github.com/tianocore/edk2/pull/10760.

I would prefer to fix the MOV64 macro and leave the invocation alone for now.
If the optimization ot doing the direct mov is important enough, that
can be done
separately at a later point.

I pushed a fix to
https://github.com/leiflindholm/edk2/commit/a7ff8aebcad1a6621ac52007b6dc530972b26125
that you could cherry-pick into the PR.

Are you looking for this PR to hit the stable tag?
If not, the MOVx4 fix can wait too.

/
    Leif

> ---
>
> Before (original code)
>
> ...ModuleEntryPoint.iiii:58:12: error: immediate must be an integer in
> range [0, 65535].
>
> movz x9, (-1) >> 48, lsl #48 ; movk x9, ((-1) >> 32) & 0xffff, lsl #32 ;
> movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9, (-1) & 0xffff
>
> ---
>
> After (new code)
>
>    svc #0
>
>    movz x9, ((-1) >> 48) & 0xffff, lsl #48 ; movk x9, ((-1) >> 32) &
> 0xffff, lsl #32 ; movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9,
> (-1) & 0xffff
>    cmp x0, x9
>    cset x0, ne
>    mov x9, xzr
>    ret
>
> ---
>
> On 2/13/2025 5:33 PM, Leif Lindholm via groups.io wrote:
> > I would very much like to have a MOV64 that's universally usable.
> > That's sort of the point.
> >
> > However, I'm wondering if the error is in fact triggered by the first
> > stanza (movz), which doesn't mask its result, and will have 48 sign
> > extended bits above the ones we care about for the instruction.
> >
> > Annoyingly I cannot reproduce the build error in my setup, so I can't
> > verify if that is the problem.
> >
> > Michael, could you possibly do a test and change
> >    movz      Reg, (Val) >> 48, lsl #48             ; \
> > to
> >    movz      Reg, ((Val) >> 48) & 0xffff, lsl #48             ; \
> >
> > on line 51 in MdePkg/Include/AArch64/AsmMacroLib.h ?
> >
> > And then rebuild the MOV64 version of ArmStandaloneMmCoreEntryPoint?
> >
> > /
> >      Leif
> >
> >
> > 
> >
>


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#121121): https://edk2.groups.io/g/devel/message/121121
Mute This Topic: https://groups.io/mt/111172486/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [edk2-devel] MOV64 and CLANGPDB

[Michael Kubacki via groups.io]

Thanks https://github.com/tianocore/edk2/pull/10760 is updated.

I don't have a strong need for it to make the stable tag. I have a local 
solution that can be used in the meantime.

On 2/14/2025 5:48 AM, Leif Lindholm wrote:
> On Fri, 14 Feb 2025 at 01:54, Michael Kubacki
> <mikuback@linux.microsoft.com> wrote:
>> That worked.
> 
> Splendid! Thanks.
> 
> Ard figured out how to reproduce the issue as well - the behaviour
> changes between
> -target aarch64-windows-none and -target aarch64-linux-gnu, so more LLP64 fun.
> 
>> Let me know what you'd like to do with
>> https://github.com/tianocore/edk2/pull/10760.
> 
> I would prefer to fix the MOV64 macro and leave the invocation alone for now.
> If the optimization ot doing the direct mov is important enough, that
> can be done
> separately at a later point.
> 
> I pushed a fix to
> https://github.com/leiflindholm/edk2/commit/a7ff8aebcad1a6621ac52007b6dc530972b26125
> that you could cherry-pick into the PR.
> 
> Are you looking for this PR to hit the stable tag?
> If not, the MOVx4 fix can wait too.
> 
> /
>      Leif
> 
>> ---
>>
>> Before (original code)
>>
>> ...ModuleEntryPoint.iiii:58:12: error: immediate must be an integer in
>> range [0, 65535].
>>
>> movz x9, (-1) >> 48, lsl #48 ; movk x9, ((-1) >> 32) & 0xffff, lsl #32 ;
>> movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9, (-1) & 0xffff
>>
>> ---
>>
>> After (new code)
>>
>>     svc #0
>>
>>     movz x9, ((-1) >> 48) & 0xffff, lsl #48 ; movk x9, ((-1) >> 32) &
>> 0xffff, lsl #32 ; movk x9, ((-1) >> 16) & 0xffff, lsl #16 ; movk x9,
>> (-1) & 0xffff
>>     cmp x0, x9
>>     cset x0, ne
>>     mov x9, xzr
>>     ret
>>
>> ---
>>
>> On 2/13/2025 5:33 PM, Leif Lindholm via groups.io wrote:
>>> I would very much like to have a MOV64 that's universally usable.
>>> That's sort of the point.
>>>
>>> However, I'm wondering if the error is in fact triggered by the first
>>> stanza (movz), which doesn't mask its result, and will have 48 sign
>>> extended bits above the ones we care about for the instruction.
>>>
>>> Annoyingly I cannot reproduce the build error in my setup, so I can't
>>> verify if that is the problem.
>>>
>>> Michael, could you possibly do a test and change
>>>     movz      Reg, (Val) >> 48, lsl #48             ; \
>>> to
>>>     movz      Reg, ((Val) >> 48) & 0xffff, lsl #48             ; \
>>>
>>> on line 51 in MdePkg/Include/AArch64/AsmMacroLib.h ?
>>>
>>> And then rebuild the MOV64 version of ArmStandaloneMmCoreEntryPoint?
>>>
>>> /
>>>       Leif
>>>
>>>
>>> 
>>>
>>



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#121122): https://edk2.groups.io/g/devel/message/121122
Mute This Topic: https://groups.io/mt/111172486/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-