Boost logo

Boost :

Subject: Re: [boost] Mathias Gaunard / NumScale relationship
From: Mathias Gaunard (mathias.gaunard_at_[hidden])
Date: 2017-04-08 21:15:33

On 8 April 2017 at 16:00, Joel FALCOU via Boost <boost_at_[hidden]>

> Hi Niall, hi Andrey,
> I'm taking on to send you an email in private w/r to your questions on the
> list regarding Boost.SIMD, Mathias Gaunard and its relationship with
> NumScale. I'm adding Michael in CC: as he knows about the whole ordeal
> already and will corroborate my claims.
> Mathias has been one of the engineer I hired and then associate with to
> fund MetaScale, the company that will later become NumScale. He let the
> company after being ousted by one of our early investor for basically
> running the company into the wall and generally not knowing how to do
> business. He left for Bloomberg and started actually speaking about
> Boost.SIMD *as if* the internal information he gathered from within
> NumScale could be released to the public without our consent inclding
> internal benchmark report and road-map information. I imagine you
> understand why the cease&desist letter was then send.
> We have a bunch of other people using Boost.SIMD and they're doing fine
> (including other company and a bunch of OS project). I dont see why we'll
> be suing people using our open source software, which is different from
> using piece of the software he should not have.
> Since being laid off, Mathias kept trying to make our company and effort
> within open source fails by playing the victim and basically retelling
> history as he see fit. Also note that Mathias verbally threaten me that he
> will make this review a failure for us.
> This is the factual story of the whole thing:
> - 2003-2006 I worked in my PhD on something called E.V.E which was a EDSL
> in C++ able to supp0rot SIMD operations on PPC Altivec.
> - 2008 I'm now an assistant professor and work on a project called NT2
> (that has been spoken about on the list a bunch of time) in collaboration
> with Jean-Thierry Lapresté which works on a huge amount of libm
> vectorisation code. Inside NT2, a remnant of E.V.E lingers and get worked
> on to support x86.
> - 2010 or some such, We start isolating the SIMD layer from NT2. At the
> same time, the university authorizes the release and open exploitation of
> NT2 as a whole in open source. NT2 and its internal soon-to-be Boost.SIMD
> is released under a Boost Licence.
> - around 2011, *I hire* Mathias to work on some part of NT2 and related
> project that culminates in 2012 at the creation of MetaScale.
> - We start speaking of NT2 and Boost.SIMD separately and it's probably
> around this time I raise the idea to make an actual submission to Boost.
> - 2014 for reasons that has no value being exposed here, Mathias leaves
> MetaScale and NumScale is funded to replace it.
> - 2015 The amount of people actively working on NT2 plummets and
> NumScale's team decide to reuse parts of Boost.SIMD into what bSIMD will
> become.
> - 2016 the rewrite is complete and the lack of support on NT2 make me
> pull the plug on the project. NumScale decide to open source back some
> parts of bSIMD, leaving this new Boost.SIMD made out of bSIMD be the sole
> remnant of this.
> And as your concerns goes Andrey about NumScale blocking PR or w/e coming
> into Boost.SIMD, The key point here is that Mathias is no longer affiliated
> with NumScale and has no information nor actual reason to speak on our
> behalf. NumScale has no plan to at short term release ITS OWN support for
> some architectures. This does not preclude anybody trying to undertake the
> effort to support an other architecture as long as yes, it is not Mathias
> pulling up the files we own he scavenged before leaving. We had had someone
> from ETH Zurich starting a port on QPX, something we also started but never
> finished due to other constraints. If this ever completes then, yeah it'll
> probably go inside the Open Source version if it's of enough good quality.
> We have various plan and discussion going on with different vendor and
> open sourcing more is planned at later date and depending on things I can
> not legally speak of because NDA.
> I'm planning to write or have Michael answer those issues publicly in a
> way I don't spread too much information that has nothing to do in public. I
> would appreciate if this line of discussion stay private for now until we
> address this properly in public.
> Meanwhile, I'm open to answer your other question.

Most of those claims are inaccurate and blatant defamation, for which I
could trigger legal action.
They can be proved false trivially, but let's not dwell on that.

I have no agenda in bringing harm to NumScale, Boost.SIMD, and the people
associated with it, I'm just making people aware of the situation about
using that technology. I think the technology is somewhat interesting, if
over-engineered for the little it does, but that the IP situation is too
messy and badly managed to risk it.

That Bloomberg LP was sent a cease-and-desist letter where NumScale claimed
ownership on Boost.SIMD and asked for destruction of all related material,
despite that material being covered by the Boost Software License and being
published in academic journals, is a fact. (I can provide justification
upon request)
When I asked Joel what this was about, he told me it was just the new board
being over-zealous to protect their IP, and they only had suspicion of
copyright infringement and no proof, which proved to be wrong. No apology
was ever issued.

Of course there are other issues of dubious legal nature concerning that
company, and I could go and correct the history of how things actually
transpired between me and NumScale, but that seems irrelevant to the
question at hand, which is whether Boost.SIMD is a good fit to be
distributed as part of the Boost C++ libraries.

Boost list run by bdawes at, gregod at, cpdaniel at, john at