From debbugs-submit-bounces@debbugs.gnu.org Wed Oct 27 11:56:43 2021 Received: (at 51198) by debbugs.gnu.org; 27 Oct 2021 15:56:43 +0000 Received: from localhost ([127.0.0.1]:50361 helo=debbugs.gnu.org) by debbugs.gnu.org with esmtp (Exim 4.84_2) (envelope-from ) id 1mflIA-0001tg-Tb for submit@debbugs.gnu.org; Wed, 27 Oct 2021 11:56:43 -0400 Received: from eggs.gnu.org ([209.51.188.92]:44926) by debbugs.gnu.org with esmtp (Exim 4.84_2) (envelope-from ) id 1mflI7-0001tK-Vm for 51198@debbugs.gnu.org; Wed, 27 Oct 2021 11:56:42 -0400 Received: from fencepost.gnu.org ([2001:470:142:3::e]:41760) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1mflI2-0006V4-Hz; Wed, 27 Oct 2021 11:56:34 -0400 DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=gnu.org; s=fencepost-gnu-org; h=MIME-Version:In-Reply-To:Date:References:Subject:To: From; bh=ewzpM+EmjEIU7YKg1iXYPpYrQxN0oR1SSZk176vpC2o=; b=Ddajol19vU1w/pQl54VE /pPvSM+8sPiuGVfeJIHSSZ4kOIymjp3klRYxaRHuviAC2lzJ71w/Z50rf9dLngAIqdiZlGg8ClwCJ HC73/qYKidHSA+NWvHaJpSs4MEFYTtiwxZNRIVArhCb3VfDyQb1n7BauTOGSCanyLlozyX/8dwZSO 57mrvR7sO2Xouojb1lGUAsC0QAz0Vpzn2Yl7KdWbpfffJS/ICFI/9SnpTNF38ra9mC1ITLbOVxhIr LX2Orv+U1mni7fCEhMLRX2FNjudSTbGuBAejJ5rY4Wnb735CSFU9CBvgr5AaQvLmkPYK5AAs0AdIe T/SG4Ahhr1SVQQ==; Received: from 91-160-117-201.subs.proxad.net ([91.160.117.201]:54244 helo=ribbon) by fencepost.gnu.org with esmtpsa (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1mflI2-0008M6-2X; Wed, 27 Oct 2021 11:56:34 -0400 From: =?utf-8?Q?Ludovic_Court=C3=A8s?= To: Leo Famulari Subject: Re: bug#51198: [PATCH] gnu: Add b2sum. References: <878ryfjudi.fsf@guix.i-did-not-set--mail-host-address--so-tickle-me> <874k92kgaq.fsf@guix.i-did-not-set--mail-host-address--so-tickle-me> Date: Wed, 27 Oct 2021 17:56:31 +0200 In-Reply-To: (Leo Famulari's message of "Wed, 27 Oct 2021 11:38:08 -0400") Message-ID: <875ytifols.fsf_-_@gnu.org> User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.2 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Spam-Score: -2.3 (--) X-Debbugs-Envelope-To: 51198 Cc: 51198@debbugs.gnu.org, =?utf-8?Q?Nicol=C3=B2?= Balzarotti X-BeenThere: debbugs-submit@debbugs.gnu.org X-Mailman-Version: 2.1.18 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: debbugs-submit-bounces@debbugs.gnu.org Sender: "Debbugs-submit" X-Spam-Score: -3.3 (---) Hi! Leo Famulari skribis: > On Wed, Oct 27, 2021 at 10:44:45AM +0200, Nicol=C3=B2 Balzarotti wrote: >> This blog post [fn:1] on guix-hpc address the "Pre-built binaries >> vs. performance" dilemma. >> >> [fn:1] https://hpc.guix.info/blog/2018/01/pre-built-binaries-vs-performa= nce/ > > It's helpful, but it doesn't look like we are using that method in Guix > anymore. It was removed in this commit: > > https://git.savannah.gnu.org/cgit/guix.git/commit/?id=3D969adb235ee34decb= 65255e1ea821ff0e221ed3d > > I guess it learned how to do runtime feature detection? BLIS has run-time/load-time feature detection now (which is the main approach the blog argues for). >> I guess the easiest way is to provide a variant (b2sum-avx or something >> like that) with avx enabled. Else, I'd just go with the unoptimized >> version as it happens for many other packages, but let's hear from >> others. > > The easiest thing is require local building, since it's an extremely > cheap build. Only 1.5 seconds on my laptop, total (not just the build > phase). With blis, one had to consider the lengthy build time. > > Because performance is critical for a tool like this, and because it's > cheap to build, I've attached a patch to require local building. This is > easier for me than creating a set of package variants that will need to > be expanded for each new generation of CPUs :) > >>From 8b862425310cf631c30c823eb0fa2bfd79d36823 Mon Sep 17 00:00:00 2001 > From: Leo Famulari > Date: Wed, 27 Oct 2021 11:29:40 -0400 > Subject: [PATCH] gnu: b2sum: Build on the local machine. > > * gnu/packages/crypto.scm (b2sum)[arguments]: Set '#:substitutable? #f'. > --- > gnu/packages/crypto.scm | 13 +++++++++++-- > 1 file changed, 11 insertions(+), 2 deletions(-) > > diff --git a/gnu/packages/crypto.scm b/gnu/packages/crypto.scm > index 3acd147f25..de8bcf1d70 100644 > --- a/gnu/packages/crypto.scm > +++ b/gnu/packages/crypto.scm > @@ -858,9 +858,18 @@ (define-public b2sum > (base32 "04z631v0vzl52g73v390ask5fnzi5wg83lcjkjhpmmymaz= 0jn152")))) > (build-system gnu-build-system) > (arguments > - `(#:make-flags (list (string-append "CC=3D" ,(cc-for-target)) > - (string-append "PREFIX=3D" (assoc-ref %outpu= ts "out"))) > + `(;; By default, b2sum uses the compiler to generate instructions > + ;; tailored to the CPU of the running machine, using "-march=3D= native". > + ;; This gives a ~1.5x speedup on a Core i5-6300U with a large d= ataset > + ;; paged in, whereas compilation of b2sum takes ~1.5 seconds. > + ;; b2sum does not support run-time feature detection: > + ;; https://github.com/BLAKE2/BLAKE2/issues/1 > + ;; For more information, see the discussion beginning here: > + ;; https://debbugs.gnu.org/cgi/bugreport.cgi?bug=3D51198#13 > + #:substitutable? #f It=E2=80=99s okay but not entirely sufficient: on a cluster setup, you typi= cally talk to a daemon that=E2=80=99s on another machine, so the CPU features it= =E2=80=99ll detect there may be different from those you=E2=80=99ll use. Likewise, as discussed on IRC, you=E2=80=99d also need #:local-build? #t. Anyway, I=E2=80=99d suggest using =E2=80=98guix hash -H blake2s-256=E2=80= =99 or similar (it uses libgcrypt, which does the right thing). I think the latest Coreutils provide a generic =E2=80=98cksum=E2=80=99, too, that probably does the righ= t thing. Ludo=E2=80=99.