Re: [patch 13/17] Immediate Values - x86 Optimization

[Mathieu Desnoyers]

* H. Peter Anvin (hpa@xxxxxxxxx) wrote:
> Mathieu Desnoyers wrote:
>> * H. Peter Anvin (hpa@xxxxxxxxx) wrote:
>>> Mathieu Desnoyers wrote:
>>>> Ok, so the most flexible solution that I see, that should fit for both
>>>> x86 and x86_64 would be :
>>>> 1 byte  :       "=q" : "a", "b", "c", or "d" register for the i386.  For
>>>>                        x86-64 it is equivalent to "r" class (for 8-bit
>>>>                        instructions that do not use upper halves).
>>>> 2, 4, 8 bytes : "=r" : A register operand is allowed provided that it is 
>>>> in a
>>>>                        general register.
>>> Any reason to keep carrying this completely misleading comment chunk 
>>> still?
>>>
>>> 	-hpa
>> This comment explains why I use the =q constraint for the 1 bytes
>> immediate value. It makes sure we use an instruction with 1-byte opcode,
>> without REX.R prefix, on x86_64.
>
> No, it doesn't.  That would be "=Q".
>
> 	-hpa

Ok. Sorry, it's been a few months since we looked at this. So the =q
opcode lets the compiler choose instructions with or without REX prefix.
We can allow this because 

- We don't need the opcode length in the stop_machine_run() version
- we support variable length opcode in the nmi-safe version

Am I remembering correctly now ?

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/