Received-SPF: pass (google.com: domain of linux-kernel+bounces-4606-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Message-ID: <00ffc2e3-446b-473e-89b9-a859f36e43dd@bytedance.com>
Date: Tue, 19 Dec 2023 10:34:36 +0800
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] maple_tree: Avoid checking other gaps after getting the
 largest gap
To: "Liam R. Howlett" <Liam.Howlett@Oracle.com>
References: <20231215074632.82045-1-zhangpeng.00@bytedance.com>
 <20231218202014.mpotsekdkszasn4t@revolver>
 <20231218202845.buffbdq3vhpiv2py@revolver>
Cc: maple-tree@lists.infradead.org, akpm@linux-foundation.org,
 linux-kernel@vger.kernel.org, linux-mm@kvack.org,
 Peng Zhang <zhangpeng.00@bytedance.com>
From: Peng Zhang <zhangpeng.00@bytedance.com>
In-Reply-To: <20231218202845.buffbdq3vhpiv2py@revolver>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit


在 2023/12/19 04:28, Liam R. Howlett 写道:
> * Liam R. Howlett <Liam.Howlett@Oracle.com> [231218 15:20]:
>> * Peng Zhang <zhangpeng.00@bytedance.com> [231215 02:46]:
>>> The last range stored in maple tree is typically quite large. By
>>> checking if it exceeds the sum of the remaining ranges in that node, it
>>> is possible to avoid checking all other gaps.
>>>
>>> Running the maple tree test suite in user mode almost always results in
>>> a near 100% hit rate for this optimization.
>>
>> This should only be triggered for right-most nodes and root though,
>> correct (mas->max  == ULONG_MAX from just before this)?
Yes, only for right-most nodes and root.
>>
>> I wonder if it's worth special case checking the first gap if the node
>> min is 0 as well.  Might be worth looking at, but this patch is
>> certainly worth doing.
> 
> Actually, not just when the min is 0, we have a special case close to
> here for slot 0 so we could just check the same sort of thing there.
I think that the first slot in a node does not have any special
significance. It has a lower probability of being the largest gap,
so it may not be worth considering.
> 
>>
>>>
>>> Signed-off-by: Peng Zhang <zhangpeng.00@bytedance.com>
>>
>> Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
>>
>>> ---
>>>   lib/maple_tree.c | 3 +++
>>>   1 file changed, 3 insertions(+)
>>>
>>> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
>>> index c9a970ea20dd..6f241bb38799 100644
>>> --- a/lib/maple_tree.c
>>> +++ b/lib/maple_tree.c
>>> @@ -1518,6 +1518,9 @@ static unsigned long mas_leaf_max_gap(struct ma_state *mas)
>>>   		gap = ULONG_MAX - pivots[max_piv];
>>>   		if (gap > max_gap)
>>>   			max_gap = gap;
>>> +
>>> +		if (max_gap > pivots[max_piv] - mas->min)
>>> +			return max_gap;
>>>   	}
>>>   
>>>   	for (; i <= max_piv; i++) {
>>> -- 
>>> 2.20.1
>>>
>