On 08/06/2014 15:50, Bill Hart wrote: > The mpfr-3.1.2 test suite fails against MPIR-2.7.0-alpha. Specifically > atan returns the wrong value. > > With some help from Fredrik Johansson we've tracked it down to a bug in > mpn_divrem. > > Specifically the quotient a/b is computed incorrectly with: > > a=+0*2^0+0*2^64+0*2^128+0*2^192+0*2^256+0*2^320+0*2^384+0*2^448+9223372036854775808*2^512 > > b=+5441491000443887104*2^0+16890151831357120748*2^64+2531194513324792887*2^128+15023743366172631521*2^192 > > MPIR 2.6.0 returns the correct values with: > > q=+2813613562306264048*2^0+18446744073709551575*2^64+18446744073709551615*2^128+18446744073709551615*2^192+11324819608111655951*2^256 > > r=+0*2^0+17436446882452594688*2^64+10431979954583400456*2^128+818554020547913635*2^192+217622286998165104*2^256 > > MPIR-2.7.0-alpha returns the incorrect values: > > q=+2813613562306263966*2^0+18446744073709551575*2^64+18446744073709551615*2^128+4202895142513760287*2^192+11324819608111655953*2^256 > > r=+0*2^0+2470107076112546816*2^64+11918624597650930617*2^128+5462319302375862668*2^192+4800654167612485773*2^256 > > I will hopefully have some time later today to begin trcking down the > bug, which surely lies in the division basecase, probably the > precomputed inverse code.
Is this a failure in the standard MPFR tests? I ask because this bug doesn't show up on Windows x64. Brian -- You received this message because you are subscribed to the Google Groups "mpir-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to mpir-devel+unsubscr...@googlegroups.com. To post to this group, send email to mpir-devel@googlegroups.com. Visit this group at http://groups.google.com/group/mpir-devel. For more options, visit https://groups.google.com/d/optout.