[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <85d9e45a-9631-a139-2d65-86a6753a35e6@ispras.ru>
Date: Fri, 13 Sep 2019 14:48:15 +0300
From: Denis Efremov <efremov@...ras.ru>
To: Matthew Wilcox <willy@...radead.org>
Cc: akpm@...ux-foundation.org, Akinobu Mita <akinobu.mita@...il.com>,
Jan Kara <jack@...e.cz>, linux-kernel@...r.kernel.org,
Matthew Wilcox <matthew@....cx>, dm-devel@...hat.com,
linux-fsdevel@...r.kernel.org, linux-media@...r.kernel.org,
Erdem Tumurov <erdemus@...il.com>,
Vladimir Shelekhov <vshel@....nsk.su>
Subject: Re: [PATCH v2] lib/memweight.c: open codes bitmap_weight()
Hi,
Sorry for reviving this conversation, but it looks to me like
this function could be reduced to a single bitmap_weight call:
static inline size_t memweight(const void *ptr, size_t bytes)
{
BUG_ON(bytes >= UINT_MAX / BITS_PER_BYTE);
return bitmap_weight(ptr, bytes * BITS_PER_BYTE);
}
Comparing to the current implementation
https://elixir.bootlin.com/linux/latest/source/lib/memweight.c#L11
this results in a signification simplification.
__bitmap_weight already count last bits with hweight_long as we
discussed earlier.
int __bitmap_weight(const unsigned long *bitmap, unsigned int bits)
{
...
if (bits % BITS_PER_LONG)
w += hweight_long(bitmap[k] & BITMAP_LAST_WORD_MASK(bits));
...
}
and __arch_hweight* functions use popcnt instruction.
I've briefly tested the equivalence of 2 implementations on x86_64 with
fuzzing here: https://gist.github.com/evdenis/95a8b9b8041e09368b31c3a9510491a5
What do you think making this function static inline and moving it
to include/linux/string.h? I could prepare a patch for it and add some tests for
memweight and bitmap_weight. Or maybe I miss something again?
Best regards,
Denis
Powered by blists - more mailing lists