> _______________________________________
> From: James Hogan
> Sent: Monday, July 24, 2017 3:24 AM
> To: Aleksandar Markovic
> Cc: linux-mips@xxxxxxxxxxxxxx; Aleksandar Markovic; Miodrag Dinic; Goran Ferenc; Douglas Leung; linux-> kernel@xxxxxxxxxxxxxxx; Paul Burton; Petar Jovanovic; Raghu Gandham; Ralf Baechle
> Subject: Re: [PATCH v3 11/16] MIPS: math-emu: <MADDF|MSUBF>.<D|S>: Fix NaN propagation
>
> On Fri, Jul 21, 2017 at 04:09:09PM +0200, Aleksandar Markovic wrote:
> > From: Aleksandar Markovic <aleksandar.markovic@xxxxxxxxxx>
> >
> > Fix the cases of <MADDF|MSUBF>.<D|S> when any of three inputs is any
> > NaN. Correct behavior of <MADDF|MSUBF>.<D|S> fd, fs, ft is following:
> >
> > - if any of inputs is sNaN, return a sNaN using following rules: if
> > only one input is sNaN, return that one; if more than one input is
> > sNaN, order of precedence for return value is fd, fs, ft
> > - if no input is sNaN, but at least one of inputs is qNaN, return a
> > qNaN using following rules: if only one input is qNaN, return that
> > one; if more than one input is qNaN, order of precedence for
> > return value is fd, fs, ft
> >
> > The previous code contained handling of some above cases, but not all.
> > Also, such handling was scattered into various cases of
> > "switch (CLPAIR(xc, yc))" statement and elsewhere. With this patch,
> > this logic is placed in one place, and "switch (CLPAIR(xc, yc))" is
> > significantly simplified.
> >
> > The relevant example:
> >
> > MADDF.S fd,fs,ft:
> > If fs contains qNaN1, ft contains qNaN2, and fd contains qNaN3, fd
> > is going to contain qNaN3 (without this patch, it used to contain
> > qNaN1).
> >
>
> Fixes: e24c3bec3e8e ("MIPS: math-emu: Add support for the MIPS R6 MADDF FPU instruction")
> Fixes: 83d43305a1df ("MIPS: math-emu: Add support for the MIPS R6 MSUBF FPU instruction")
>
In v4, I will add these lines to commit messages of all MADDF/MSUBF patches from this series.
> > Signed-off-by: Miodrag Dinic <miodrag.dinic@xxxxxxxxxx>
> > Signed-off-by: Goran Ferenc <goran.ferenc@xxxxxxxxxx>
> > Signed-off-by: Aleksandar Markovic <aleksandar.markovic@xxxxxxxxxx>
>
> > If backported, I suspect commits:
> > 6162051e87f6 ("MIPS: math-emu: Unify ieee754sp_m{add,sub}f")
> > and
> > d728f6709bcc ("MIPS: math-emu: Unify ieee754dp_m{add,sub}f")
> > in 4.7 will require manual backporting between 4.3 and 4.7 (due to
> > separation of maddf/msubf before that point), so I suppose tagging
> > stable 4.7+ and backporting is best (assuming you consider this fix
> > worth backporting).
I am going to tag all MADDF/MSUBF patches "stable 4.7+" and all MIN/MAX/MINA/MAXA patches "stable 4.3+" in v4.
> > ---
> > arch/mips/math-emu/dp_maddf.c | 71 ++++++++++++++-----------------------------
> > arch/mips/math-emu/sp_maddf.c | 69 ++++++++++++++---------------------------
> > 2 files changed, 46 insertions(+), 94 deletions(-)
> ...
> > + /* handle the cases when at least one of x, y or z is a NaN */
> > + if (((xc == IEEE754_CLASS_SNAN) || (xc == IEEE754_CLASS_QNAN)) ||
> > + ((yc == IEEE754_CLASS_SNAN) || (yc == IEEE754_CLASS_QNAN)) ||
> > + ((zc == IEEE754_CLASS_SNAN) || (zc == IEEE754_CLASS_QNAN))) {
>
> This condition basically covers all of the cases below. Any particular
> reason not to skip it ...
> > + /* order of precedence is z, x, y */
> > + if (zc == IEEE754_CLASS_SNAN)
> > + return ieee754dp_nanxcpt(z);
> > + if (xc == IEEE754_CLASS_SNAN)
> > + return ieee754dp_nanxcpt(x);
> > + if (yc == IEEE754_CLASS_SNAN)
> > + return ieee754dp_nanxcpt(y);
> > + if (zc == IEEE754_CLASS_QNAN)
> > + return z;
> > + if (xc == IEEE754_CLASS_QNAN)
> > + return x;
> > return y;
>
> ... and make this return conditional on (yc == IEEE754_CLASS_QNAN)?
You are right. I am going to reorganize the code as you suggested in v4.
Regards,
Aleksandar
