#!/bin/bash -l
#SBATCH --job-name=smoother-modes
#SBATCH --nodes=36
#SBATCH --ntasks-per-node=8
#SBATCH --cpus-per-task=7
#SBATCH --gpus-per-node=8
#SBATCH --time=1:00:00
#SBATCH --account=phy157_dwf
#SBATCH --gpu-bind=none
#SBATCH --exclusive
#SBATCH --mem=0
#SBATCH -S 0

##############################################################################
# The 1402.2585 p.13 comparison on this machine: adaptive GCR smoothers vs
# the same polynomials frozen (GCRReplaySmoother: recorded for the first
# PolyRecordIters outer steps, then replayed with no inner products), vs a
# Chebyshev 1/x fit on the measured interval.  Same banked point otherwise.
#
#   M1 gcr    / gcr      reference
#   M2 replay / replay   both levels frozen after PolyRecordIters steps
#   M3 replay / gcr      fine frozen only (the fine smoother is the reduction-
#                        heavy one at Nrhs=1)
#   M4 cheb   / gcr      fine Chebyshev [FineChebLo,FineChebHi] order Fso
#   M5 cheb   / cheb
#
# Laptop 8^4 findings (hot config, Ls=4, NBASIS=8): replay/replay converges
# (28 vs 23 outer); cheb on the FINE level diverges there while cheb on the
# coarse level is fine -- the tiny operator has modes the fixed polynomial
# amplifies (left of / off the axis) that GCR handles adaptively.  Production
# spectrum is near-normal with edge 130.5 (shift 0.1) so M4/M5 may behave
# differently; they are cheap to include and cheap to discard.
#
# Readouts per mode: Fouter count, s/RHS, and the FINAL exact-halo residual.
##############################################################################

cat << EOF > select_gpu
#!/bin/bash
export GPU_MAP=(0 1 2 3 7 6 5 4)
export NUMA_MAP=(3 3 1 1 2 2 0 0)
export GPU=\${GPU_MAP[\$SLURM_LOCALID]}
export NUMA=\${NUMA_MAP[\$SLURM_LOCALID]}
export HIP_VISIBLE_DEVICES=\$GPU
unset ROCR_VISIBLE_DEVICES
if [ \$SLURM_PROCID = "0" ]; then echo \$*; fi
exec numactl -m \$NUMA -N \$NUMA \$*
EOF
chmod +x ./select_gpu

root=$HOME/ParallelIO/systems/Frontier
source $root/sourceme-rocm7.2.sh

export OMP_NUM_THREADS=7
export MPICH_GPU_SUPPORT_ENABLED=1
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export MPICH_OFI_NIC_POLICY=GPU
module load libfabric
unset DENSE_GATHER DENSE_GATHER_FORCE DENSE_GATHER_DEBUG DENSE_GATHER_MIN_BYTES

OPTS1="--accelerator-threads 8 --shm 4096 --shm-mpi 1 --device-mem 32000"
vol=48.48.48.96
MPI_GEOM=3.6.4.4

# banked point (2026-08-26 sweep): Css 2.0, Nstep 2, Fso 6, mmax 4, svm 8
export FineSmootherShift=0.1
export FineSmootherOrder=6
export FineSmootherMmax=4
export CoarseSmootherShift=2.0
export CoarseSmootherNstep=2
export CoarseSmootherMmax=2
export CoarseSolverTol=0.05
export CoarseSolverOrder=200
export CoarseSolverMmax=8
export OuterTol=1e-8
export OuterMmax=6
export OuterNstep=12
export SUBSPACE_FILE=/lustre/orion/phy157/proj-shared/phy157_dwf/paboyle/subspace_nb64.scidac
unset  SLAB_FILE
export DENSE_SCHUR=1
export DENSE_SCHUR2D=1
export MASS=0.00078
export BLOCK=2.2.3.3
export BLOCK2=4.4.2.4
export L3_TOL=3.0e-1
export L3_MAXIT=2
export L3_NSTEP=50
export DENSE_CC=1
export DENSE_APPLY_PROFILE=1
unset  DENSE_CC_CHECK
export DENSE_SPLITK=128
export DENSE_DEVICE_SUM=2          # cartesian P2P ring: no 8 MB device-allreduce cliff (NRHS=12 abort)
export GRID_ALLOC_NCACHE_LARGE=64
export NRHS=4
export PowerIterations=0
export SmootherCoeffLog=0

# frozen-polynomial controls
export PolyRecordIters=4           # outer steps recorded before the switch
export FineChebLo=3.0              # harvested |R|<0.1 edge / PowerIteration edge x1.05
export FineChebHi=137.0
export CoarseChebLo=8.0
export CoarseChebHi=47.0           # shift 2.0: edge 43.3 x1.08

run_mode () {
  name=$1; export FineSmootherMode=$2; export CoarseSmootherMode=$3
  echo "----- $name : FineSmootherMode=$FineSmootherMode CoarseSmootherMode=$CoarseSmootherMode -----"
  fname=log.modes.$name
  srun -N36 -n288 ./select_gpu $root/examples/Example_pvdagm_v2_3level_DenseCoarseMatrix \
       --mpi ${MPI_GEOM} --grid $vol $OPTS1 --comms-overlap > $fname 2>&1
  echo "  $(grep -h 'V2 3-level solve Nrhs' $fname | tr '\n' ' ')"
  echo "  $(grep -h 'Fouter MrhsPGCR: Converged' $fname | sed 's/.*Converged/Converged/' | tr '\n' ' ')"
  echo "  $(grep -h 'FINAL Nrhs .: worst' $fname | tr '\n' ' ')"
  grep -h "SwitchableSmoother\|GCRCoefficients .*calls" $fname | head -6
}

run_mode M1_gcr_gcr        gcr    gcr
run_mode M2_replay_replay  replay replay
run_mode M3_replay_gcr     replay gcr
run_mode M4_cheb_gcr       cheb   gcr
run_mode M5_cheb_cheb      cheb   cheb

echo "========================================================="
echo "summary"
grep -h "V2 3-level solve Nrhs 1" log.modes.* | sed 's/.*V2/V2/'
echo "========================================================="