Peter Boyle
Peter Boyle
Hi Isaaku, can you identify where in the HMC its is running slow? it prints continual timestamps and should be easy to compare the logs Peter
OK, suspect thrashing in between host and GPU in the Cshift which I haven't optimised. Will try and optimise that and see if it helps.
Cshift confirmed indeed slow, have been working on optimising. Have a much faster version in most cases, but not in develop yet. Will get back to you.
Hi Isaaku, can you try the branch "sycl" - I have speed up the Cshift 12x on a V100. Not sure if this will be enough, but you are also...
I have sped up the Shift 12x (on 16^3 volume) on V100. Since the drop was 50x this is perhaps not a big enough factor yet, but your solver is...
Sorry - I broke the code, hold off on that
looks ok, I was doing something wrong. This branch configures with modified commands ../configure --enable-accelerator=cuda --enable-comms=mpi --enable-simd=GPU CXX=nvcc CXXFLAGS="-ccbin mpicxx -gencode arch=compute_70,code=sm_70 -std=c++11"
Hmm... Summit, V100. SYCL branch. ./Test_contfrac_force Grid : Message : ================================================ Grid : Message : MPI is initialised and logging filters activated Grid : Message : ================================================ Grid : Message...
Test_hmc_IwasakiGauge Grid : Message : 182.640130 s : Total H after trajectory = 118697.576176948 dH = -0.262491766450694 Grid : Message : 182.640230 s : -------------------------------------------------- Grid : Message : 182.640270...
For me the hmc_EODWFRatio is working... puzzled will look at your logs now