CP_FM_GEMM
Section can be repeated.
Benchmark and test the cp_fm_gemm routines by multiplying C=A*B [Edit on GitHub]
Keywords
Keyword descriptions
- FORCE_BLOCKSIZE
Type: logical
Default: F
Lone keyword: T
Usage: FORCE_BLOCKSIZEDescription: Forces the blocksize, even if this implies that a few processes might have no data
- GRID_2D
Type: integer[2]
Default: 1 1
Usage: GRID_2D 64 16Description: Explicitly set the blacs 2D processor layout. If the product differs from the number of MPI ranks, it is ignored and a default nearly square layout is used.
- K
Type: integer
Default: 256
Usage: K 1024Description: Dimension 1 of C
- M
Type: integer
Default: 256
Usage: M 1024Description: Inner dimension M
- N
Type: integer
Default: 256
Usage: N 1024Description: Dimension 2 of C
- NCOL_BLOCK
Type: integer
Default: 32
Usage: NCOL_BLOCK 64Description: block_size for cols
- NROW_BLOCK
Type: integer
Default: 32
Usage: nrow_block 64Description: block_size for rows
- N_LOOP
Type: integer
Default: 10
Usage: N_LOOP 10Description: Number of cp_fm_gemm operations being timed (useful for small matrices).
- ROW_MAJOR
Type: logical
Default: T
Lone keyword: T
Usage: ROW_MAJOR .FALSE.Description: Use a row major blacs grid
- TRANSA
Type: logical
Default: F
Lone keyword: T
Usage: TRANSADescription: Transpose matrix A
- TRANSB
Type: logical
Default: F
Lone keyword: T
Usage: TRANSBDescription: Transpose matrix B