Update for latt parameter; claude's predeliction for using shell variables rather than "grid" style command line args is getting annoying

This commit is contained in:
Peter Boyle
2026-08-28 10:04:36 -04:00
parent 576cebd56a
commit 9b4c388585
+16 -8
View File
@@ -56,24 +56,32 @@ module load libfabric
BIN=$root/examples/Example_pvdagm_v2_3level_DenseCoarseMatrix BIN=$root/examples/Example_pvdagm_v2_3level_DenseCoarseMatrix
OPTS1="--accelerator-threads 8 --shm 4096 --shm-mpi 1" OPTS1="--accelerator-threads 8 --shm 4096 --shm-mpi 1"
vol=16.16.16.32 # LOCAL VOLUME MATCHED to production: 48.48.48.96 on 3.6.4.4 is local 16.8.12.24; on 8 GCDs
# (1.1.2.4) that is global 16.8.24.96 with the production blockings 2.2.3.3 / 4.4.2.4:
# coarse local 8.4.4.8 and coarse-coarse local 2.1.2.2 exactly as at 288 ranks. The global
# lattice is oddly shaped and the dense bottom is tiny (N = 2.1.4.8 x 60 = 3840, one block),
# but every solver-side buffer has the production shape. (First submission, 2026-08-28,
# used 16.16.16.32 with 2^4 blocks: local 16.16.8.8 -- NOT matched.)
vol=16.8.24.96
MPI_GEOM=1.1.2.4 MPI_GEOM=1.1.2.4
# The example sizes its lattice from LATT/LS (env), NOT from --grid (first submission
# asserted heap_bytes<heap_size trying to build 48^3x96 on one node for exactly this reason).
export LATT=$vol
export LS=24
# Small-lattice setup: 16^3x32 on 8 GCDs, blocks 2^4 twice -> coarse 8.8.8.16, coarse-coarse
# 4.4.4.8 (N = 512 x 60 = 30720, 8 blocks of 3840 on a 2x4 grid; SCHUR2D_LEAF_SPAN=2 -> W=7680 leaves).
# The subspace is generated on first use and saved to SUBSPACE_FILE (a few minutes). # The subspace is generated on first use and saved to SUBSPACE_FILE (a few minutes).
export SUBSPACE_FILE=$PWD/subspace_16x32_nb64.scidac export SUBSPACE_FILE=$PWD/subspace_16x8x24x96_nb64.scidac
unset SLAB_FILE unset SLAB_FILE
export HOT_START=1 export HOT_START=1
export MASS=0.00078 export MASS=0.00078
export BLOCK=2.2.2.2 export BLOCK=2.2.3.3
export BLOCK2=2.2.2.2 export BLOCK2=4.4.2.4
export DENSE_SCHUR=1 export DENSE_SCHUR=1
export DENSE_SCHUR2D=1 export DENSE_SCHUR2D=1
export DENSE_CC=1 export DENSE_CC=1
export DENSE_SPLITK=8 export DENSE_SPLITK=1
export DENSE_DEVICE_SUM=4 export DENSE_DEVICE_SUM=4
export SCHUR2D_LEAF_SPAN=2 export SCHUR2D_LEAF_SPAN=1
export SCHUR2D_LEAF_LU=1 export SCHUR2D_LEAF_LU=1
export SCHUR2D_PROBE=0 export SCHUR2D_PROBE=0
export GRID_ALLOC_NCACHE_LARGE=64 export GRID_ALLOC_NCACHE_LARGE=64