lkml.org 
[lkml]   [2018]   [Jul]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2] x86-64: use 32-bit XOR to zero registers
>>> On 09.07.18 at 10:33, <pavel@ucw.cz> wrote:
> Anyway, normally assembler is the one who chooses instruction
> encoding.

There are different possible views here, and I personally think that
while it is a compiler's job to chose optimal encodings, assemblers
shouldn't by default alter what the programmer (or compiler) has
specified, unless there are multiple encodings that are _entirely_
identical in effect, but which are _truly_ different encodings (and
presence/absence of instruction prefixes to me doesn't count as
"truly different"). This view is particularly on the basis that
assembly programmers often imply a certain encoding to be chosen,
be it just to imply size, or to go as far as meaning to run-time patch
code.

> xor %rax, %rax is equivalent to xor %eax, %eax; first variant is
> slower on some CPUs, second variant will take one extra byte due to
> operand size prefix IIRC...

The second variant is actually shorter by one byte.

> Should the assembler generate right
> variant according to the CPU type?

An "optimization" mode has recently been added to gas, but in general
as well as for the sake of older gas I don't think we should rely on this
new behavior (which is also off by default) except in cases where the
impact on the source code would be undesirable (as can e.g. be the
case when macro-izing things).

Jan


\
 
 \ /
  Last update: 2018-07-09 10:47    [W:0.029 / U:3.460 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site