Home Original page

[PATCH 3/6] x86: improve operand reversal

Jan Beulich JBeulich@suse.com
Tue Aug 7 13:13:00 GMT 2018 
>>> On 07.08.18 at 14:06, <hjl.tools@gmail.com> wrote:
> On Tue, Aug 7, 2018 at 12:37 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>> On 06.08.18 at 18:25, <hjl.tools@gmail.com> wrote:
>>> On Mon, Aug 6, 2018 at 8:59 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>>>> On 06.08.18 at 17:09, <hjl.tools@gmail.com> wrote:
>>>>> But there is no indication at all in your patch to show that it does
>>>>> anything remotely to {load} nor {store}.   All your testcase changes
>>>>> are for the ".s" suffix.
>>>>
>>>> There's a whole lot of stuff getting added to *pseudos.s.
>>>
>>> That will make sure that {load} and {store} are handler properly
>>> by actually testing them, instead of relying on the deprecated
>>> ".s" suffix.
>>
>> I'm afraid I'm now confused by the "will" in your response: This
>> makes it sound as if you assume something is yet to be added. But
>> all those tests are already there. Bottom line - I'm still hanging in
>> the air as to which way to proceed (see the earlier enumerated
>> three options).
> 
> Your change is't needed unless you can show that it improves {load}
> or {store}.

H.J., please. What is the purpose of me adding a whole lot of stuff
to *pseudos.s? Try assembling this without this patch in place. Then
comment out all the lines that cause errors, and look at the resulting
mess: {load} and {store} don't have their supposed meaning in
quite a few cases without this patch (in fact often the opposite
results). Hence the need to disconnect {store} from the (as you say
deprecated) .s suffix, so that .s can (mostly) remain as bogus as it
always was (just not producing as many errors anymore, which -
as said - is mostly a side effect of fixing the pseudo prefixes), while
{store} can be made act as its name says.

Jan
More information about the Binutils mailing list