Re: [PATCH] x86/hweight: Don't clobber %rdi

From: Borislav Petkov <bp@alien8.de>
To: Jiri Kosina <jikos@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Ville Syrj??l?? <ville.syrjala@linux.intel.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	the arch/x86 maintainers <x86@kernel.org>,
	Borislav Petkov <bp@suse.de>, "H . Peter Anvin" <hpa@zytor.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Brian Gerst <brgerst@gmail.com>,
	Denys Vlasenko <dvlasenk@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@kernel.org>
Subject: Re: [PATCH] x86/hweight: Don't clobber %rdi
Date: Mon, 8 Aug 2016 21:17:01 +0200	[thread overview]
Message-ID: <20160808191701.GH511@nazgul.tnic> (raw)
In-Reply-To: <alpine.LRH.2.00.1608082050110.19185@gjva.wvxbf.pm>

On Mon, Aug 08, 2016 at 08:53:16PM +0200, Jiri Kosina wrote:
> Ville's patch is the proper thing to do; I still think the comments 
> probably should be removed as they might provide more confusion than help, 
> but that's a separate discussion.

Right, so while trying to debug this with a pen and paper, I found that
adding a subsript to the w variable makes it more understandable. I.e.,
something like this:

	w_1 -= (w >> 1) & 0x55555555;
	w_2 =  (w_1 & 0x33333333) + ((w_1 >> 2) & 0x33333333);

and then you'd have

	movl %eax, %edx				# w
	shrl %edx				# w >>= 1
	andl $0x55555555, %edx			# (w >> 1) & 0x55555555
	subl %edx, %eax				# w_1 = w - ...

	movl %eax, %edx				# w_1
	shrl $2, %eax				# w_1 >>= 2
	andl $0x33333333, %edx			# (w_1 >> 2) & 0x33333333
	andl $0x33333333, %eax			# w_1 &= 0x33333333
	addl %edx, %eax				# w_2 = ... + ...

...

Anyway, something like that. I find that more readable TBH.

Thanks and thanks for confirming Ville's patch fixes it on your box.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--