Environment Variable Configuration
The following environment variables are used in the applied math libraries.
Environment Variable |
Meaning |
Valid Value |
Default Value |
|---|---|---|---|
HGEMM_F16F16F32_MAX_KB |
Maximum number of KBs for hgemm |
Integer |
1'000'000'000'000L |
KML_SCADSS_VARIANT |
SCADSS scheduling policy |
Integer (0 or 1). The default value is 0. |
0 |
KSCALAPACK_REDIST_TO_LOCAL_APPLICABLE |
Local memory adaptation for ScaLAPACK |
O\D\F |
D |
KML_NUM_THREADS |
Number of local threads |
Integer |
0 |
KML_ALLOCATE_POLICY |
alloc rule |
Integer |
0 |
KML_ALLOCATE_HUGEPAGE_SIZE |
alloc hugepage size |
Integer |
512*1024*1024 |
KML_PXGETRF_FORCE_CALU |
CALU algorithm for the pgetrf function |
Integer |
0 |
KML_FAST_EIGENSOLVER |
High-performance eigenvalue solver |
Integer |
1 |
KML_TSQR_CHECK |
Result check for tall and skinny QR factorization |
Integer |
0 |
KML_MPI_USE_SHARED_MEMORY |
Shared memory used by MPI |
Integer |
1 |
OMP_PLACES |
OMP built-in parameter |
cores |
- |
OMP_PROC_BIND |
OMP built-in parameter |
TRUE |
- |
KML_VERBOSE |
Print level |
0/1/2/3 |
0 |
Environment Variable |
Meaning |
Valid Value |
Default Value |
|---|---|---|---|
KML_DSS_PPN |
Number of processes on each node after autotune is enabled. |
Positive integer |
2 |
KML_DSS_SCHE_MODE |
DSS scheduling mode. |
STATIC\DYNAMIC\KUPL\LAST |
LAST |
KML_DSS_SCALE_ALGO |
Scaling is performed based on the maximum values of the rows and columns. Enabling this variable improves calculation precision when there is a significant discrepancy between the maximum row and column values in the original matrix. If ROWCOL_SCAN is selected, valid matrix values must be passed in the Analyze phase. |
NONE\ROWCOL_SCAN |
NONE |
KML_DSS_SGEMM_ISA |
Selects a proper operator optimization branch for the solver SGEMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_DGEMM_ISA |
Selects a proper operator optimization branch for the solver DGEMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_CGEMM_ISA |
Selects a proper operator optimization branch for the solver CGEMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_ZGEMM_ISA |
Selects a proper operator optimization branch for the solver ZGEMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_DSYRK_ISA |
Selects a proper operator optimization branch for the solver DSYRK. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_ZSYRK_ISA |
Selects a proper operator optimization branch for the solver ZSYRK. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_DSYRKMM_ISA |
Selects a proper operator optimization branch for the solver DSYRKMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_ZSYRKMM_ISA |
Selects a proper operator optimization branch for the solver ZSYRKMM. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_SGETRF_ISA |
Selects a proper operator optimization branch for the solver SGETRF. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_DGETRF_ISA |
Selects a proper operator optimization branch for the solver DGETRF. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_CGETRF_ISA |
Selects a proper operator optimization branch for the solver CGETRF. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_ZGETRF_ISA |
Selects a proper operator optimization branch for the solver ZGETRF. |
NEON\SVE256 |
Server type-based determination |
KML_DSS_DPOTRF_ISA |
Selects a proper operator optimization branch for the solver DPOTRF. |
NEON\SVE256 |
Server type-based determination |
Server type-based determination: The default value is NEON for TaiShan 200 1280 and TaiShan 200 2280, and SVE256 for TaiShan 200 1280 V2 and TaiShan 200 2280 V2.