Re: [PATCH 15/17] arm: Add string-fza.h

[Richard Earnshaw <Richard.Earnshaw@foss.arm.com>]

On 02/09/2022 21:39, Adhemerval Zanella via Libc-alpha wrote:
> From: Richard Henderson <rth@twiddle.net>
> 
> While arm has the more important string functions in assembly,
> there are still a few generic routines used.
> 
> Use the UQSUB8 insn for testing of zeros.

UQSUB8 requires ARMv6 or above.  While that's pretty likely these days, 
you might want to consider a fall-back for Armv5 or earlier if you still 
want to support those.

R.

> 
> Checked on armv7-linux-gnueabihf
> ---
>   sysdeps/arm/armv6t2/string-fza.h | 70 ++++++++++++++++++++++++++++++++
>   1 file changed, 70 insertions(+)
>   create mode 100644 sysdeps/arm/armv6t2/string-fza.h
> 
> diff --git a/sysdeps/arm/armv6t2/string-fza.h b/sysdeps/arm/armv6t2/string-fza.h
> new file mode 100644
> index 0000000000..4fe2e8383f
> --- /dev/null
> +++ b/sysdeps/arm/armv6t2/string-fza.h
> @@ -0,0 +1,70 @@
> +/* Zero byte detection; basics.  ARM version.
> +   Copyright (C) 2022 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
> +   <http://www.gnu.org/licenses/>.  */
> +
> +#ifndef _STRING_FZA_H
> +#define _STRING_FZA_H 1
> +
> +#include <string-optype.h>
> +#include <string-maskoff.h>
> +
> +/* This function returns at least one bit set within every byte
> +   of X that is zero.  */
> +
> +static inline op_t
> +find_zero_all (op_t x)
> +{
> +  /* Use unsigned saturated subtraction from 1 in each byte.
> +     That leaves 1 for every byte that was zero.  */
> +  op_t ret, ones = repeat_bytes (0x01);
> +  asm ("uqsub8 %0,%1,%2" : "=r"(ret) : "r"(ones), "r"(x));
> +  return ret;
> +}
> +
> +/* Identify bytes that are equal between X1 and X2.  */
> +
> +static inline op_t
> +find_eq_all (op_t x1, op_t x2)
> +{
> +  return find_zero_all (x1 ^ x2);
> +}
> +
> +/* Identify zero bytes in X1 or equality between X1 and X2.  */
> +
> +static inline op_t
> +find_zero_eq_all (op_t x1, op_t x2)
> +{
> +  return find_zero_all (x1) | find_zero_all (x1 ^ x2);
> +}
> +
> +/* Identify zero bytes in X1 or inequality between X1 and X2.  */
> +
> +static inline op_t
> +find_zero_ne_all (op_t x1, op_t x2)
> +{
> +  /* Make use of the fact that we'll already have ONES in a register.  */
> +  op_t ones = repeat_bytes (0x01);
> +  return find_zero_all (x1) | (find_zero_all (x1 ^ x2) ^ ones);
> +}
> +
> +/* Define the "inexact" versions in terms of the exact versions.  */
> +#define find_zero_low		find_zero_all
> +#define find_eq_low		find_eq_all
> +#define find_zero_eq_low	find_zero_eq_all
> +#define find_zero_ne_low	find_zero_ne_all
> +
> +#endif /* _STRING_FZA_H */