On Sun, Jun 22, 2025 at 1:32 PM Jan Hubicka <hubi...@ucw.cz> wrote:
>
> > Since there is
> >
> > /* X86_TUNE_SPLIT_LONG_MOVES: Avoid instructions moving immediates
> > directly to memory. */
> > DEF_TUNE (X86_TUNE_SPLIT_LONG_MOVES, "split_long_moves", m_PPRO)
>
> If I recall correctly, this tune was added for PentiumPro which had
> problem decoding moves with long immediate and is a performance
> optimization rather than code size one.
>
> As you discuss in the PR, i686 preffers
> xorl %eax, %eax
> movl %eax, aligned_heap_area
> over
> movl $0, aligned_heap_area
> which has too long encoding.
> >
> > to avoid long immediate store instructions, like
> >
> > c7 02 00 00 00 00 movl $0x0,(%rdx)
> > c7 02 ff ff ff ff movl $0xffffffff,(%rdx)
> >
> > add TARGET_USE_AND0_ORM1_STORE and enable *mov<mode>_(and|or) for
> > TARGET_USE_AND0_ORM1_STORE, which is true for TARGET_SPLIT_LONG_MOVES or
> > -Oz, to also generate:
> >
> > 83 22 00 andl $0x0,(%rdx)
> > 83 0a ff orl $0xffffffff,(%rdx)
>
> I think this will not work well on PPro hardware since it will do it as
> read-modify-write, while
Since read-modify-write is enabled for PentiumPro:
/* X86_TUNE_READ_MODIFY_WRITE: Enable use of read modify write instructions
such as "add $1, mem". */
DEF_TUNE (X86_TUNE_READ_MODIFY_WRITE, "read_modify_write",
~(m_PENT | m_LAKEMONT))
should this
/* Generate "and $0,mem" and "or $-1,mem", instead of "mov $0,mem" and
"mov $-1,mem" with shorter encoding for TARGET_SPLIT_LONG_MOVES with
TARGET_READ_MODIFY_WRITE or -Oz. */
#define TARGET_USE_AND0_ORM1_STORE \
((TARGET_SPLIT_LONG_MOVES && TARGET_READ_MODIFY_WRITE) \
|| (optimize_insn_for_size_p () && optimize_size > 1))
work?
> xorl %eax, %eax
> movl %eax, (%rdx)
> is executed as store saving the useless load.
>
> Honza
--
H.J.
