[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20161209100130.GS3174@twins.programming.kicks-ass.net>
Date: Fri, 9 Dec 2016 11:01:30 +0100
From: Peter Zijlstra <peterz@...radead.org>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: LKML <linux-kernel@...r.kernel.org>,
John Stultz <john.stultz@...aro.org>,
Ingo Molnar <mingo@...nel.org>,
David Gibson <david@...son.dropbear.id.au>,
Liav Rehana <liavr@...lanox.com>,
Chris Metcalf <cmetcalf@...lanox.com>,
Richard Cochran <richardcochran@...il.com>,
Parit Bhargava <prarit@...hat.com>,
Laurent Vivier <lvivier@...hat.com>,
"Christopher S. Hall" <christopher.s.hall@...el.com>
Subject: Re: [patch 5/6] [RFD] timekeeping: Provide optional 128bit math
On Fri, Dec 09, 2016 at 09:30:11AM +0100, Peter Zijlstra wrote:
> +static inline u64 mul_u32_u32(u32 a, u32 b)
> +{
> + u64 ret;
> +
> + asm ("mull %[b]" : "=A" (ret) : [a] "a" (a), [b] "g" (b) );
> +
> + return ret;
> +}
ARGH, that's broken on x86_64, it needs to be:
u32 high, low;
asm ("mull %[b]" : "=a" (low), "=d" (high)
: [a] "a" (a), [b] "g" (b) );
return low | ((u64)high) << 32;
The 'A' constraint doesn't work right.
And with that all the benchmark results are borken too.
root@...-ep:~/spinlocks# for i in -m64 -m32 -mx32 ; do echo $i; gcc -O3 $i -o mult mult.c -lm; ./mult; done
-m64
cond: avg: 7.474872 +- 0.008302
uncond: avg: 9.116401 +- 0.008468
128: avg: 0.826584 +- 0.005514
-m32
cond: avg: 16.604030 +- 0.009808
uncond: avg: 13.115470 +- 0.004452
-mx32
cond: avg: 6.168156 +- 0.006650
uncond: avg: 7.202092 +- 0.006813
128: avg: 0.081809 +- 0.008440
Powered by blists - more mailing lists