Re: [PATCH] sparc/vdso: Add helper function for 64-bit right shift on 32-bit target
From: Thomas Gleixner
Date: Sun Aug 04 2024 - 11:44:56 EST
On Sun, Aug 04 2024 at 10:39, Koakuma via wrote:
> From: Koakuma <koachan@xxxxxxxxxxxxxx>
>
> Add helper function for 64-bit right shift on 32-bit target so that
> clang does not emit a runtime library call.
>
> Signed-off-by: Koakuma <koachan@xxxxxxxxxxxxxx>
> ---
> Hi~
>
> This adds a small function to do 64-bit right shifts for use in vDSO
> code, needed so that clang does not emit a call to runtime library.
> ---
> arch/sparc/vdso/vclock_gettime.c | 8 ++++----
> include/vdso/math64.h | 28 ++++++++++++++++++++++++++++
> --- a/include/vdso/math64.h
> +++ b/include/vdso/math64.h
> @@ -21,6 +21,34 @@ __iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
> return ret;
> }
>
> +#if BITS_PER_LONG == 32
> +/* This is to prevent the compiler from emitting a call to __lshrdi3. */
> +static __always_inline u64
> +__shr64(u64 val, int amt)
> +{
> + u32 mask = (1U << amt) - 1;
> + u32 lo = val;
> + u32 hi = val >> 32;
> + u32 mi;
> +
> + if (amt >= 32)
> + return hi >> (amt - 32);
> +
> +
> + mi = (hi & mask) << (32 - amt);
> + hi >>= amt;
> + lo = (lo >> amt) | mi;
> +
> + return ((u64) hi) << 32 | lo;
> +}
> +#else
> +static __always_inline u64
> +__shr64(u64 val, int amt)
> +{
> + return val >> amt;
> +}
Why does this sparc'ism need to be in generic code?
Thanks,
tglx