Closed nicolin closed 1 week ago
Mmm, that cannot work. That line is assigning double array to a scalar
what branch is this?
devel
PS: I am a RSE at UCL and we are working with partners at NVIDIA and bench-marking QUDA for a UK institute. So in the middle testing NVSHEM builds this popped up ;-) .
QUDA has merged the multiple-right-hand size thing. You may want to try the new branch in chroma feature/mrhs-mg
Hey there! QUDA’s MRHS interface had to break the way residuals are returned. I have fixed in this in the feature/mrhs-mg branch of Chroma, which is waiting to be merged as a PR (hint hint JLab peeps!). With the MRHS interface they now return an array of residua rather than a single residuum.
Best, B
From: nicolin @.> Date: Monday, October 14, 2024 at 11:38 AM To: JeffersonLab/chroma @.> Cc: Balint Joo @.>, Mention @.> Subject: Re: [JeffersonLab/chroma] Error compiling Chroma ( worked until 3 days ago) (Issue #73)
devel
PS: I am a RSE at UCL and we are working with partners at NVIDIA and bench-marking QUDA for a UK institute. So in them middle testing NVSHEM builds they popped up ;-) .
Attaching the build script I created that can maybe usefull also. ChromaBuildQuda.txthttps://github.com/user-attachments/files/17366292/ChromaBuildQuda.txt
— Reply to this email directly, view it on GitHubhttps://github.com/JeffersonLab/chroma/issues/73#issuecomment-2411623283, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AAEPL2ELLSOZISLE4I2IZITZ3PQPFAVCNFSM6AAAAABP5IE5GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMJRGYZDGMRYGM. You are receiving this because you were mentioned.Message ID: @.***>
Thanks, I will test on the that branch. The performance we get with NVSHEM is slower that QMP and MPI ( Fastest) on a single node with 4 GPUs so going to have to see what is going on and this fix will help ;-)
Hi @nicolin, are you on the UCL Center for Computational Science Slack? You can reach me there directly if you like (. If you are going to be working with QUDA and Chroma a lot we can also invite you to the QUDA slack, Best, B
From: nicolin @.> Date: Monday, October 14, 2024 at 11:46 AM To: JeffersonLab/chroma @.> Cc: Balint Joo @.>, Mention @.> Subject: Re: [JeffersonLab/chroma] Error compiling Chroma ( worked until 3 days ago) (Issue #73)
Thanks, I will test on the that branch. The performance we get with NVSHEM is slower that QMP and MPI ( Fastest) on a single node with 4 GPUs so going to have to see what is going on and this fix will help ;-)
— Reply to this email directly, view it on GitHubhttps://github.com/JeffersonLab/chroma/issues/73#issuecomment-2411642271, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AAEPL2FPGJO5JJDZFLZSKPDZ3PROXAVCNFSM6AAAAABP5IE5GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMJRGY2DEMRXGE. You are receiving this because you were mentioned.Message ID: @.***>
@bjoo thanks nice to e-meet you, yes please invite me to the QUDA slack as I will be working with Kate and Mathius also along with other nvidia contacts. I am at UCL ARC but have asked about the Center for Computational Science Slack now.
So feature/mrhs-mg fixed it .
Hi, Please send me a non-github email to bjoo AT nvidia.com so I have your email address, and I can invite you to the CUDA slack. Best wishes, Balint
From: nicolin @.> Date: Monday, October 14, 2024 at 11:59 AM To: JeffersonLab/chroma @.> Cc: Balint Joo @.>, Mention @.> Subject: Re: [JeffersonLab/chroma] Error compiling Chroma ( worked until 3 days ago) (Issue #73)
@bjoohttps://github.com/bjoo thanks nice to e-meet you, yes please invite me to the QUDA slack as I will be working with Kate and Mathius also along with other nvidia contacts. I am at UCL ARC but have asked about the Center for Computational Science Slack now.
— Reply to this email directly, view it on GitHubhttps://github.com/JeffersonLab/chroma/issues/73#issuecomment-2411667493, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AAEPL2BAEDLM7RVU7QDV2ITZ3PS6XAVCNFSM6AAAAABP5IE5GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMJRGY3DONBZGM. You are receiving this because you were mentioned.Message ID: @.***>
@bjoo /home/dp006/dp006/dc-gove1/QudaBench/FASTSUM/QCDSolvers/chroma/lib/actions/ferm/invert/quda_solvers/syssolver_linop_clover_quda_multigrid_w.cc: In member function ‘Chroma::SystemSolverResults_t Chroma::LinOpSysSolverQUDAMULTIGRIDClover::qudaInvert(Chroma::CloverTermT<QDP::OLattice<QDP::PSpinVector<QDP::PColorVector<QDP::RComplex<QDP::Word >, 3>, 4> >, QDP::OLattice<QDP::PScalar<QDP::PColorMatrix<QDP::RComplex<QDP::Word >, 3> > > >&, Chroma::CloverTermT<QDP::OLattice<QDP::PSpinVector<QDP::PColorVector<QDP::RComplex<QDP::Word >, 3>, 4> >, QDP::OLattice<QDP::PScalar<QDP::PColorMatrix<QDP::RComplex<QDP::Word >, 3> > > >&, const T&, Chroma::LinOpSysSolverQUDAMULTIGRIDClover::T&) const’:
/home/dp006/dp006/dc-gove1/QudaBench/FASTSUM/QCDSolvers/chroma/lib/actions/ferm/invert/quda_solvers/syssolver_linop_clover_quda_multigrid_w.cc:114:32: error: no match for ‘operator=’ (operand types are ‘QDP::Real’ {aka ‘QDP::OScalar<QDP::PScalar<QDP::PScalar<QDP::RScalar<QDP::Word > > > >’} and ‘const double [128]’)
114 | ret.resid = quda_inv_param.true_res;