On Fri, Jan 31, 2025 at 6:37 PM Aleksandar Rakic
<aleksandar.ra...@htecgroup.com> wrote:
>
> From: Robert Suchanek <robert.sucha...@imgtec.com>
>
> Cherry-picked 4194c529fade9b3106d118cac63b71bc8b13f7be
> from https://github.com/MIPS/gcc
This (and some other of the patches) lack the appropriate ChangeLog
entry embedded in the commit message.
> Signed-off-by: Robert Suchanek <robert.sucha...@imgtec.com>
> Signed-off-by: Faraz Shahbazker <fshahbaz...@wavecomp.com>
> Signed-off-by: Aleksandar Rakic <aleksandar.ra...@htecgroup.com>
> ---
> gcc/config/mips/mips.cc | 8 +++++++-
> gcc/config/mips/mips.h | 5 +++++
> 2 files changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/config/mips/mips.cc b/gcc/config/mips/mips.cc
> index 3185fa9633e..cd4bce71ae8 100644
> --- a/gcc/config/mips/mips.cc
> +++ b/gcc/config/mips/mips.cc
> @@ -9631,7 +9631,13 @@ mips_expand_block_move (rtx dest, rtx src, rtx length,
> rtx alignment)
> {
> if (ISA_HAS_COPY)
> return mips16_expand_copy (dest, src, length, alignment);
> - else if (INTVAL (length) <= MIPS_MAX_MOVE_BYTES_PER_LOOP_ITER)
> + else if (INTVAL (length) <= MIPS_MAX_MOVE_BYTES_PER_LOOP_ITER
> + /* We increase slightly the maximum number of bytes in
> + a straight-line block if the source and destination
> + are aligned to the register width. */
> + || (!optimize_size
> + && INTVAL (alignment) == UNITS_PER_WORD
> + && INTVAL (length) <= MIPS_MAX_MOVE_MEM_STRAIGHT))
> {
> mips_block_move_straight (dest, src, INTVAL (length),
> INTVAL (alignment));
> diff --git a/gcc/config/mips/mips.h b/gcc/config/mips/mips.h
> index efd23a262f9..0245287f9bf 100644
> --- a/gcc/config/mips/mips.h
> +++ b/gcc/config/mips/mips.h
> @@ -3171,6 +3171,11 @@ while (0)
> #define MIPS_MAX_MOVE_BYTES_STRAIGHT \
> (MIPS_MAX_MOVE_BYTES_PER_LOOP_ITER * 2)
>
> +/* The maximum number of bytes that can be copied by any expanded block move;
> + see mips_expand_block_move. */
> +#define MIPS_MAX_MOVE_MEM_STRAIGHT \
> + (MIPS_MAX_MOVE_BYTES_PER_LOOP_ITER * 3)
> +
> /* The base cost of a memcpy call, for MOVE_RATIO and friends. These
> values were determined experimentally by benchmarking with CSiBE.
> In theory, the call overhead is higher for TARGET_ABICALLS (especially
> --
> 2.34.1
