[PATCH] erofs-utils: avoid redundant memcpy and sha256() for dedupe

[zhaoyifan (H)]

Hi Zijie,

It would be quite appreciated if you could help us polish the 
multithreading -Ededupe implementation. I will try to rebase the 
existing code to the latest codebase ASAP.

You could find the design decision in multithreading -Ededupe in this paper:

https://dl.acm.org/doi/pdf/10.1145/3671016.3671395


Thanks,

Yifan

On 2025/8/15 17:54, wangzijie wrote:
>> Hi Zijie,
>>
>> On 2025/8/15 16:44, wangzijie wrote:
>>> We have already use xxh64() for filtering first for dedupe, when we
>>> need to skip the same xxh64 hash, no need to do memcpy and sha256(),
>>> relocate the code to avoid it.
>>>
>>> Signed-off-by: wangzijie <wangzijie1 at honor.com>
>> Thanks for the patch, it makes sense to me since we only keep one
>> record according to xxh64 (instead of sha256) for now:
>>
>> Reviewed-by: Gao Xiang <hsiangkao at linux.alibaba.com>
>>
>> Although I think multi-threaded deduplication is more useful, see:
>> https://github.com/erofs/erofs-utils/issues/25
>> but I'm not sure if you're interested in it... ;-)
> Hi Xiang,
> Thank you for providing this information, I want to optimize mkfs time with
> dedupe option and send this patch. I will find time to research Yifan's demo
> of multi-threaded deduplication and try to provide some help.
>
>> Thanks,
>> Gao Xiang
>
>