[PATCH v4 4/4] add riscv vector support for memcpy

[Adhemerval Zanella Netto]

On 21/01/25 23:04, daichengrong@iscas.ac.cn wrote:
> From: daichengrong <daichengrong@iscas.ac.cn>
> 
> Changes in v4:
>   update rvv memcpy support by compiler 
>   check whether rvv enabled by dl_hwcap
> 
> Changes in v3:
>   Remove unnecessary whitespace
>   
> Changes in v2:
>   delete size-0 branch

As before, usually we provide optimization backed up with some performance
numbers on real hardware.  

> 
> ---
>  sysdeps/riscv/multiarch/memcpy_vector.S       | 35 +++++++++++++++++++
>  .../unix/sysv/linux/riscv/multiarch/Makefile  |  8 +++++
>  .../unix/sysv/linux/riscv/multiarch/memcpy.c  |  7 ++++
>  3 files changed, 50 insertions(+)
>  create mode 100644 sysdeps/riscv/multiarch/memcpy_vector.S
> 
> diff --git a/sysdeps/riscv/multiarch/memcpy_vector.S b/sysdeps/riscv/multiarch/memcpy_vector.S
> new file mode 100644
> index 0000000000..8fddab8432
> --- /dev/null
> +++ b/sysdeps/riscv/multiarch/memcpy_vector.S
> @@ -0,0 +1,35 @@
> +/* memcpy for RISC-V Vector.
> +   Copyright (C) 2024-2025 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +
> +
> +#include <sysdep.h>
> +#include <sys/asm.h>
> +
> +ENTRY (__memcpy_vector)
> +    mv	    a6, a0
> +L(loop):
> +    vsetvli a3,a2,e8,m8,ta,mu
> +    vle8.v  v8,(a1)
> +    vse8.v  v8,(a6)
> +    add     a1,a1,a3
> +    sub     a2,a2,a3
> +    add     a6,a6,a3
> +    bnez    a2,L(loop)
> +L(ret):
> +    ret
> +END (__memcpy_vector)
> diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> index fcef5659d4..394033e077 100644
> --- a/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> +++ b/sysdeps/unix/sysv/linux/riscv/multiarch/Makefile
> @@ -5,5 +5,13 @@ sysdep_routines += \
>    memcpy_noalignment \
>    # sysdep_routines
>  
> +ifeq ($(have-gcc-riscv-rvv),yes)
> +sysdep_routines += \
> +  memcpy_vector \
> +  # rvv sysdep_routines
> +  
> +ASFLAGS-memcpy_vector.S += -march=rv64gcv
> +endif
> +
>  CFLAGS-memcpy_noalignment.c += -mno-strict-align
>  endif
> diff --git a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> index 8544f5402a..4d1fa2073d 100644
> --- a/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> +++ b/sysdeps/unix/sysv/linux/riscv/multiarch/memcpy.c
> @@ -32,11 +32,18 @@ extern __typeof (__redirect_memcpy) __libc_memcpy;
>  
>  extern __typeof (__redirect_memcpy) __memcpy_generic attribute_hidden;
>  extern __typeof (__redirect_memcpy) __memcpy_noalignment attribute_hidden;
> +extern __typeof (__redirect_memcpy) __memcpy_vector attribute_hidden;
>  
>  static inline __typeof (__redirect_memcpy) *
>  select_memcpy_ifunc (uint64_t dl_hwcap, __riscv_hwprobe_t hwprobe_func)
>  {
>    unsigned long long int v;
> +
> +#if defined(HAVE_GCC_RISCV_VECTOR) 
> +	if (dl_hwcap & COMPAT_HWCAP_ISA_V) 
> +    return __memcpy_vector;
> +#endif

Indentation seems off here.

> +
>    if (__riscv_hwprobe_one (hwprobe_func, RISCV_HWPROBE_KEY_CPUPERF_0, &v) == 0
>        && (v & RISCV_HWPROBE_MISALIGNED_MASK) == RISCV_HWPROBE_MISALIGNED_FAST)
>      return __memcpy_noalignment;