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;
From:   David Laight <David.Laight@ACULAB.COM>
To:     'Andrew Morton' <akpm@linux-foundation.org>,
        Nathan Chancellor <nathan@kernel.org>
CC:     Nick Desaulniers <ndesaulniers@google.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "clang-built-linux@googlegroups.com" 
        <clang-built-linux@googlegroups.com>,
        "terrelln@fb.com" <terrelln@fb.com>
Subject: RE: [PATCH] lib/zstd: Fix bitwise vs logical operators
Thread-Topic: [PATCH] lib/zstd: Fix bitwise vs logical operators
Thread-Index: AQHXkwP9hKCn1U30S0yWYygV6tj6Nat7Wbtg
Date:   Thu, 19 Aug 2021 21:22:54 +0000
Message-ID: <defb9e5133234835950c21511d776fb9@AcuMS.aculab.com>
References: <20210815004154.1781834-1-nathan@kernel.org>
 <20210816180532.6e85c93f84d62dc720d21178@linux-foundation.org>
In-Reply-To: <20210816180532.6e85c93f84d62dc720d21178@linux-foundation.org>
Accept-Language: en-GB, en-US
MIME-Version: 1.0
Content-Language: en-US
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8BIT
Precedence: bulk

From: Andrew Morton
> Sent: 17 August 2021 02:06
> 
> On Sat, 14 Aug 2021 17:41:54 -0700 Nathan Chancellor <nathan@kernel.org> wrote:
> 
> > clang warns several times along the lines of:
> >
> > lib/zstd/compress.c:1043:7: warning: bitwise and of boolean expressions; did you mean logical and?
> [-Wbool-operation-and]
> >                 if ((offset_1 > 0) & (ZSTD_read32(ip + 1 - offset_1) == ZSTD_read32(ip + 1))) {
> >                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >                                    &&
> >
> > Bitwise ANDs do not short circuit, meaning that the ZSTD_read32 calls
> > will be evaluated even if the first condition is not true. This is not
> > always a problem but it is not a standard way to do conditionals so
> > replace the bitwise ANDs with logical ones to fix the warning and make
> > the code clearer.
> >
> > ...
> >
> > --- a/lib/zstd/compress.c
> > +++ b/lib/zstd/compress.c
> > @@ -1040,7 +1040,7 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx *cctx, const void *src, size_t
> sr
> >  		const BYTE *match = base + matchIndex;
> >  		hashTable[h] = curr; /* update hash table */
> >
> > -		if ((offset_1 > 0) & (ZSTD_read32(ip + 1 - offset_1) == ZSTD_read32(ip + 1))) {
> > +		if ((offset_1 > 0) && (ZSTD_read32(ip + 1 - offset_1) == ZSTD_read32(ip + 1))) {
> 
> yeah, this is a late night party trick which is sometimes used to
> attempt to speed things up by avoiding a branch.  It is perhaps
> beneficial if the LHS is almost always true.  I guess.
> 
> I'd prefer that the code not do this - it's silly, looks wrong and I
> bet it's unmeasurable.
> 
> But I think this code is supposed to be kept in sync with an
> out-of-tree upstream version so this change might be problematic.

Except that in this case you want the short-circuit.
The RH condition is slow and always true when offset_1 is zero.
OTOH subtracting offset_1 makes the lines look odd (out of context).

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)