Tsung Chun Lin <[email protected]> writes:
> From ddb7852c92dc0222af9eeec1deafce753b3a7067 Mon Sep 17 00:00:00 2001
> From: Jim Tsung-Chun Lin <[email protected]>
> Date: Mon, 30 Dec 2024 15:32:12 +0800
> Subject: [PATCH] Prefer scalar_int_mode if the size - 1 is equal to
> UNITS_PER_WORD.
>
> Don't use the QI vector if its size is equal to UNITS_PER_WORD for
> better code generation.
That makes sense, but...
> Before patch:
>
> vsetivli zero,4,e8,mf4,ta,ma
> vmv.v.i v1,0
> addi a4,sp,12
> vse8.v v1,0(a4)
>
> After patch:
>
> sw zero,12(sp)
> ---
> gcc/expr.cc | 2 +-
> gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c | 1 -
> 2 files changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/gcc/expr.cc b/gcc/expr.cc
> index babf00f34dc..e2abef1f544 100644
> --- a/gcc/expr.cc
> +++ b/gcc/expr.cc
> @@ -1062,7 +1062,7 @@ widest_fixed_size_mode_for_size (unsigned int size,
> by_pieces_operation op)
> gcc_checking_assert (size > 1);
>
> /* Use QI vector only if size is wider than a WORD. */
> - if (can_use_qi_vectors (op) && size > UNITS_PER_WORD)
> + if (can_use_qi_vectors (op) && (size - 1) > UNITS_PER_WORD)
...it looks like this will still use UNITS_PER_WORD QI vectors for
size == UNITS_PER_WORD + 2.
How about instead removing the size > UNITS_PER_WORD check above and
instead checking:
if (is_a<fixed_size_mode> (mode, &candidate)
&& GET_MODE_SIZE (candidate) > UNITS_PER_WORD
&& GET_MODE_INNER (candidate) == QImode)
in the loop (with the second line being new)?
Thanks,
Richard
> {
> machine_mode mode;
> fixed_size_mode candidate;
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c
> b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c
> index 52e2580c53e..6549ae61c67 100644
> --- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c
> @@ -51,5 +51,4 @@ void p(int buf, __builtin_va_list ab, int q) {
> } while (k);
> }
>
> -/* { dg-final { scan-assembler-times
> {vsetivli\tzero,\s*4,\s*e8,\s*mf4,\s*t[au],\s*m[au]} 1 } } */
> /* { dg-final { scan-assembler-times
> {vsetivli\tzero,\s*8,\s*e8,\s*mf2,\s*t[au],\s*m[au]} 1 } } */
