Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 17 Feb 2021 12:46:36 -0800
From:      Ryan Libby <rlibby@freebsd.org>
To:        Alex Richardson <arichardson@freebsd.org>
Cc:        src-committers <src-committers@freebsd.org>, dev-commits-src-all@freebsd.org, dev-commits-src-main@freebsd.org
Subject:   Re: git: 8fa6abb6f4f6 - main - Expose clang's alignment builtins and use them for roundup2/rounddown2
Message-ID:  <CAHgpiFyAcK5%2BZh0bdOdMw12Yj96RHo08g9qaNh5C50k1jwP1bA@mail.gmail.com>
In-Reply-To: <202102031604.113G4SQq019037@gitrepo.freebsd.org>
References:  <202102031604.113G4SQq019037@gitrepo.freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Feb 3, 2021 at 8:04 AM Alex Richardson <arichardson@freebsd.org> wrote:
>
> The branch main has been updated by arichardson:
>
> URL: https://cgit.FreeBSD.org/src/commit/?id=8fa6abb6f4f64f4f23e2920e2aea7996566851a4
>
> commit 8fa6abb6f4f64f4f23e2920e2aea7996566851a4
> Author:     Alex Richardson <arichardson@FreeBSD.org>
> AuthorDate: 2021-02-03 15:27:17 +0000
> Commit:     Alex Richardson <arichardson@FreeBSD.org>
> CommitDate: 2021-02-03 16:02:54 +0000
>
>     Expose clang's alignment builtins and use them for roundup2/rounddown2
>
>     This makes roundup2/rounddown2 type- and const-preserving and allows
>     using it on pointer types without casting to uintptr_t first. Not
>     performing pointer-to-integer conversions also helps the compiler's
>     optimization passes and can therefore result in better code generation.
>     When using it with integer values there should be no change other than
>     the compiler checking that the alignment value is a valid power-of-two.
>
>     I originally implemented these builtins for CHERI a few years ago and
>     they have been very useful for CheriBSD. However, they are also useful
>     for non-CHERI code so I was able to upstream them for Clang 10.0.
>
>     Rationale from the clang documentation:
>     Clang provides builtins to support checking and adjusting alignment
>     of pointers and integers. These builtins can be used to avoid relying
>     on implementation-defined behavior of arithmetic on integers derived
>     from pointers. Additionally, these builtins retain type information
>     and, unlike bitwise arithmetic, they can perform semantic checking on
>     the alignment value.
>
>     There is also a feature request for GCC, so GCC may also support it in
>     the future: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98641
>
>     Reviewed By:    brooks, jhb, imp
>     Differential Revision: https://reviews.freebsd.org/D28332
> ---
>  sys/sys/cdefs.h | 19 +++++++++++++++++++
>  sys/sys/param.h |  4 ++--
>  2 files changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/sys/sys/cdefs.h b/sys/sys/cdefs.h
> index 75bedd4b8128..72ef942084f2 100644
> --- a/sys/sys/cdefs.h
> +++ b/sys/sys/cdefs.h
> @@ -884,4 +884,23 @@
>  #define        __guarded_by(x)         __lock_annotate(guarded_by(x))
>  #define        __pt_guarded_by(x)      __lock_annotate(pt_guarded_by(x))
>
> +/* Alignment builtins for better type checking and improved code generation. */
> +/* Provide fallback versions for other compilers (GCC/Clang < 10): */
> +#if !__has_builtin(__builtin_is_aligned)
> +#define __builtin_is_aligned(x, align) \
> +       (((__uintptr_t)x & ((align) - 1)) == 0)
> +#endif
> +#if !__has_builtin(__builtin_align_up)
> +#define __builtin_align_up(x, align)   \
> +       ((__typeof__(x))(((__uintptr_t)(x)+((align)-1))&(~((align)-1))))
> +#endif
> +#if !__has_builtin(__builtin_align_down)
> +#define __builtin_align_down(x, align) \
> +       ((__typeof__(x))((x)&(~((align)-1))))
> +#endif
> +
> +#define __align_up(x, y) __builtin_align_up(x, y)
> +#define __align_down(x, y) __builtin_align_down(x, y)
> +#define __is_aligned(x, y) __builtin_is_aligned(x, y)
> +

Since these are only valid for powers of 2, I think it would be good to
indicate that in the names (__align_up2() etc).

>  #endif /* !_SYS_CDEFS_H_ */
> diff --git a/sys/sys/param.h b/sys/sys/param.h
> index 079357a19d47..d6f1eb21dcd2 100644
> --- a/sys/sys/param.h
> +++ b/sys/sys/param.h
> @@ -305,9 +305,9 @@
>  #endif
>  #define        nitems(x)       (sizeof((x)) / sizeof((x)[0]))
>  #define        rounddown(x, y) (((x)/(y))*(y))
> -#define        rounddown2(x, y) ((x)&(~((y)-1)))          /* if y is power of two */
> +#define        rounddown2(x, y) __align_down(x, y) /* if y is power of two */
>  #define        roundup(x, y)   ((((x)+((y)-1))/(y))*(y))  /* to any y */
> -#define        roundup2(x, y)  (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */
> +#define        roundup2(x, y)  __align_up(x, y) /* if y is powers of two */
>  #define powerof2(x)    ((((x)-1)&(x))==0)
>
>  /* Macros for min/max. */

This broke the gcc build:
https://ci.freebsd.org/job/FreeBSD-main-amd64-gcc6_build/3200/console
00:40:30 --- all_subdir_firewire ---
00:40:30 In file included from /workspace/src/sys/sys/types.h:43:0,
00:40:30                  from /workspace/src/sys/sys/param.h:99,
00:40:30                  from /workspace/src/sys/dev/firewire/fwohci.c:40:
00:40:30 /workspace/src/sys/dev/firewire/fwohci.c: In function
'fwohci_get_plen':
00:40:30 /workspace/src/sys/dev/firewire/fwohci.c:2699:17: error:
'typeof' applied to a bit-field
00:40:30    r += roundup2(fp->mode.wreqb.len, sizeof(uint32_t));

We could:
 - Drop the cast for the fallback.
 - Cast with (__typeof__(+(x))) which unfortunately promotes e.g. char
   to int but otherwise I think behaves okay, and the promotion was
   previously happening for roundup2/rounddown2 anyway.
 - Punt the casting burden to callers of roundup2/rounddown2.

Ryan



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAHgpiFyAcK5%2BZh0bdOdMw12Yj96RHo08g9qaNh5C50k1jwP1bA>