Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
IronPort-SDR: 2eV/3JIdk4JRqiUVtPIylTC+qN6yb9u9NVc/MWImnbFSX9XyhCgdDQFY0Bp/ce/ByS7VElN27c
 KH/SuxGQ6YfQ==
IronPort-SDR: 6Ck2M2xBKPYBeWY4Di7+IhB7A6KYcAiZ8Lol/Kpwit0nDqg7wfTGRZ5/kpPaL8Ddn/6XnYHM2n
 grfjSpDl/8sQ==
Date:   Fri, 24 Jul 2020 07:45:03 -0700
From:   Andi Kleen <ak@linux.intel.com>
To:     Ian Rogers <irogers@google.com>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@redhat.com>,
        Namhyung Kim <namhyung@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        linux-kernel@vger.kernel.org, Stephane Eranian <eranian@google.com>
Subject: Re: [PATCH] perf bench: Add benchmark of find_next_bit
Message-ID: <20200724144503.GD1180481@tassilo.jf.intel.com>
References: <20200724071959.3110510-1-irogers@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200724071959.3110510-1-irogers@google.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jul 24, 2020 at 12:19:59AM -0700, Ian Rogers wrote:
> for_each_set_bit, or similar functions like for_each_cpu, may be hot
> within the kernel. If many bits were set then one could imagine on
> Intel a "bt" instruction with every bit may be faster than the function
> call and word length find_next_bit logic. Add a benchmark to measure
> this.
 
> This benchmark on AMD rome and Intel skylakex shows "bt" is not a good
> option except for very small bitmaps.

Small bitmaps is a common case in the kernel (e.g. cpu bitmaps) 

But the current code isn't that great for small bitmaps. It always looks horrific
when I look at PT traces or brstackinsn, especially since it was optimized
purely for code size at some point.

Probably would be better to have different implementations for
different sizes.

-Andi