[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CANn89i+u9JJf=DqPyOjwe639bdFYTNm0ny_POH2gmoaX4djLcA@mail.gmail.com>
Date: Mon, 19 Sep 2016 16:33:53 -0700
From: Eric Dumazet <edumazet@...gle.com>
To: Stephen Hemminger <stephen@...workplumber.org>
Cc: Neal Cardwell <ncardwell@...gle.com>,
David Miller <davem@...emloft.net>,
netdev <netdev@...r.kernel.org>, Van Jacobson <vanj@...gle.com>,
Yuchung Cheng <ycheng@...gle.com>,
Nandita Dukkipati <nanditad@...gle.com>,
Soheil Hassas Yeganeh <soheil@...gle.com>
Subject: Re: [PATCH v3 net-next 16/16] tcp_bbr: add BBR congestion control
>
> It generates some slightly smaller code.
> if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts)
> - 3e7: 0f b6 c0 movzbl %al,%eax
> - 3ea: 83 f8 03 cmp $0x3,%eax
> - 3ed: 0f 86 d4 00 00 00 jbe 4c7 <bbr_lt_bw_sampling.isra.6+0x157>
> + 3e7: 3c 03 cmp $0x3,%al
> + 3e9: 0f 86 d1 00 00 00 jbe 4c0 <bbr_lt_bw_sampling.isra.6+0x150>
>
> And different code for abs
> /* Is new bw close to the lt_bw from the previous interval? */
> diff = abs(bw - bbr->lt_bw);
> - 47a: 44 89 e2 mov %r12d,%edx
> - 47d: 29 c2 sub %eax,%edx
> - 47f: 89 d1 mov %edx,%ecx
> - 481: 89 d3 mov %edx,%ebx
> + 475: 44 89 e3 mov %r12d,%ebx
> + 478: 29 c3 sub %eax,%ebx
> + 47a: 89 da mov %ebx,%edx
> + 47c: c1 fa 1f sar $0x1f,%edx
> + 47f: 31 d3 xor %edx,%ebx
> + 481: 29 d3 sub %edx,%ebx
>
> The biggest change is getting rid of the always true conditional.
>
...
>
> Overall, it really makes little difference. Actual file sizes come out the same.
> The idea is more to document what is variable
> and what is immutable in the algorithm.
SGTM, thanks Stephen !
Powered by blists - more mailing lists