APX ZU indication

APX ZU indication

[Jan Beulich]

Lili,

before you or your colleagues even get to coding this up, I'd like
to bring up this aspect, now that there is that "Assembly Syntax
Recommendations" do as well (without, sadly, offering any way for
providing feedback): For SETcc, embedding the indicator in the
mnemonic (or alternatively as some kind of mnemonic extension, along
the lines of {sae} or {nf}), simply expressing this via operands
would seem more tidy / readable:

	setnz	%eax

instead of, as that doc has it,

	setunz	%al

making clear at the first glance what register is updated. (Even
if said doc wouldn't be updated, I'd like us to consider supporting
such as an extension, right from the start.)

The situation is, admittedly, less clear for IMUL, as

	imul	$3, %dx, %ecx
	imulw	$3, (%rdx), %ecx

aren't entirely unambiguous - they could also indicate
multiplications storing the full 32-bit result in the destination.
Hence I don't mean to extend the above suggestion to IMUL.

Jan

RE: APX ZU indication

[Cui, Lili]

> Lili,
> 
> before you or your colleagues even get to coding this up, I'd like to bring up
> this aspect, now that there is that "Assembly Syntax Recommendations" do
> as well (without, sadly, offering any way for providing feedback): For SETcc,
> embedding the indicator in the mnemonic (or alternatively as some kind of
> mnemonic extension, along the lines of {sae} or {nf}), simply expressing this
> via operands would seem more tidy / readable:
> 
> 	setnz	%eax
> 
> instead of, as that doc has it,
> 
> 	setunz	%al
> 

You meant setzunz, right?

> making clear at the first glance what register is updated. (Even if said doc
> wouldn't be updated, I'd like us to consider supporting such as an extension,
> right from the start.)
> 
> The situation is, admittedly, less clear for IMUL, as
> 
> 	imul	$3, %dx, %ecx
> 	imulw	$3, (%rdx), %ecx
> 
> aren't entirely unambiguous - they could also indicate multiplications storing
> the full 32-bit result in the destination.
> Hence I don't mean to extend the above suggestion to IMUL.
> 

Regarding how to express ZU of SETcc, we initially had a version similar to your suggestion, which is to change the size of reg to distinguish them, instead of the current distinction by adding suffix zu. The reason for the modification was to make the expression of ZU of SETcc and IMUL consistent. I think Binutils could support "setnz %eax", but gcc doesn't use it that way, this would make it seem a bit redundant.

Regards,
Lili.

Re: APX ZU indication

[Jan Beulich]

On 28.04.2024 09:43, Cui, Lili wrote:
>> Lili,
>>
>> before you or your colleagues even get to coding this up, I'd like to bring up
>> this aspect, now that there is that "Assembly Syntax Recommendations" do
>> as well (without, sadly, offering any way for providing feedback): For SETcc,
>> embedding the indicator in the mnemonic (or alternatively as some kind of
>> mnemonic extension, along the lines of {sae} or {nf}), simply expressing this
>> via operands would seem more tidy / readable:
>>
>> 	setnz	%eax
>>
>> instead of, as that doc has it,
>>
>> 	setunz	%al
>>
> 
> You meant setzunz, right?

Indeed; too many 'z'. Part of why I dislike that "zu" infix, especially
when followed by a condition code (there's far less room for confusion with
IMUL).

>> making clear at the first glance what register is updated. (Even if said doc
>> wouldn't be updated, I'd like us to consider supporting such as an extension,
>> right from the start.)
>>
>> The situation is, admittedly, less clear for IMUL, as
>>
>> 	imul	$3, %dx, %ecx
>> 	imulw	$3, (%rdx), %ecx
>>
>> aren't entirely unambiguous - they could also indicate multiplications storing
>> the full 32-bit result in the destination.
>> Hence I don't mean to extend the above suggestion to IMUL.
>>
> 
> Regarding how to express ZU of SETcc, we initially had a version similar to your suggestion, which is to change the size of reg to distinguish them, instead of the current distinction by adding suffix zu. The reason for the modification was to make the expression of ZU of SETcc and IMUL consistent. I think Binutils could support "setnz %eax", but gcc doesn't use it that way, this would make it seem a bit redundant.

IMUL is an outlier anyway, so consistency there is of questionable value,
imo. If you/they insist on "zu" for SETcc, then may I ask that the spec at
least formally permit both variants? That'll kind of fit with permitting
variations on other APX aspects as well. (Yet better would be for the insn
page to explicitly name both forms.)

Jan