root / man / man3 / HPL_logsort.3
Historique | Voir | Annoter | Télécharger (2,32 ko)
| 1 | 1 | equemene | .TH HPL_logsort 3 "September 10, 2008" "HPL 2.0" "HPL Library Functions" |
|---|---|---|---|
| 2 | 1 | equemene | .SH NAME |
| 3 | 1 | equemene | HPL_logsort \- Sort the processes in logarithmic order. |
| 4 | 1 | equemene | .SH SYNOPSIS |
| 5 | 1 | equemene | \fB\&#include "hpl.h"\fR |
| 6 | 1 | equemene | |
| 7 | 1 | equemene | \fB\&void\fR |
| 8 | 1 | equemene | \fB\&HPL_logsort(\fR |
| 9 | 1 | equemene | \fB\&const int\fR |
| 10 | 1 | equemene | \fI\&NPROCS\fR, |
| 11 | 1 | equemene | \fB\&const int\fR |
| 12 | 1 | equemene | \fI\&ICURROC\fR, |
| 13 | 1 | equemene | \fB\&int *\fR |
| 14 | 1 | equemene | \fI\&IPLEN\fR, |
| 15 | 1 | equemene | \fB\&int *\fR |
| 16 | 1 | equemene | \fI\&IPMAP\fR, |
| 17 | 1 | equemene | \fB\&int *\fR |
| 18 | 1 | equemene | \fI\&IPMAPM1\fR |
| 19 | 1 | equemene | \fB\&);\fR |
| 20 | 1 | equemene | .SH DESCRIPTION |
| 21 | 1 | equemene | \fB\&HPL_logsort\fR |
| 22 | 1 | equemene | computes an array IPMAP and its inverse IPMAPM1 that |
| 23 | 1 | equemene | contain the logarithmic sorted processes id with repect to the local |
| 24 | 1 | equemene | number of rows of U that they own. This is necessary to ensure that |
| 25 | 1 | equemene | the logarithmic spreading of U is optimal in terms of number of steps |
| 26 | 1 | equemene | and communication volume as well. In other words, the larget pieces |
| 27 | 1 | equemene | of U will be sent a minimal number of times. |
| 28 | 1 | equemene | .SH ARGUMENTS |
| 29 | 1 | equemene | .TP 8 |
| 30 | 1 | equemene | NPROCS (global input) const int |
| 31 | 1 | equemene | On entry, NPROCS specifies the number of process rows in the |
| 32 | 1 | equemene | process grid. NPROCS is at least one. |
| 33 | 1 | equemene | .TP 8 |
| 34 | 1 | equemene | ICURROC (global input) const int |
| 35 | 1 | equemene | On entry, ICURROC is the source process row. |
| 36 | 1 | equemene | .TP 8 |
| 37 | 1 | equemene | IPLEN (global input/output) int * |
| 38 | 1 | equemene | On entry, IPLEN is an array of dimension NPROCS+1, such that |
| 39 | 1 | equemene | IPLEN[0] is 0, and IPLEN[i] contains the number of rows of U, |
| 40 | 1 | equemene | that process i-1 has. On exit, IPLEN[i] is the number of |
| 41 | 1 | equemene | rows of U in the processes before process IPMAP[i] after the |
| 42 | 1 | equemene | sort, with the convention that IPLEN[NPROCS] is the total |
| 43 | 1 | equemene | number of rows of the panel. In other words, IPLEN[i+1] - |
| 44 | 1 | equemene | IPLEN[i] is the number of rows of A that should be moved to |
| 45 | 1 | equemene | the process IPMAP[i]. IPLEN is such that the number of rows |
| 46 | 1 | equemene | of the source process row is IPLEN[1] - IPLEN[0], and the |
| 47 | 1 | equemene | remaining entries of this array are sorted so that the |
| 48 | 1 | equemene | quantities IPLEN[i+1]-IPLEN[i] are logarithmically sorted. |
| 49 | 1 | equemene | .TP 8 |
| 50 | 1 | equemene | IPMAP (global output) int * |
| 51 | 1 | equemene | On entry, IPMAP is an array of dimension NPROCS. On exit, |
| 52 | 1 | equemene | array contains the logarithmic mapping of the processes. In |
| 53 | 1 | equemene | other words, IPMAP[myroc] is the corresponding sorted process |
| 54 | 1 | equemene | coordinate. |
| 55 | 1 | equemene | .TP 8 |
| 56 | 1 | equemene | IPMAPM1 (global output) int * |
| 57 | 1 | equemene | On entry, IPMAPM1 is an array of dimension NPROCS. On exit, |
| 58 | 1 | equemene | this array contains the inverse of the logarithmic mapping |
| 59 | 1 | equemene | contained in IPMAP: IPMAPM1[ IPMAP[i] ] = i, for all i in |
| 60 | 1 | equemene | [0.. NPROCS) |
| 61 | 1 | equemene | .SH SEE ALSO |
| 62 | 1 | equemene | .BR HPL_plindx1 \ (3), |
| 63 | 1 | equemene | .BR HPL_plindx10 \ (3), |
| 64 | 1 | equemene | .BR HPL_pdlaswp01N \ (3), |
| 65 | 1 | equemene | .BR HPL_pdlaswp01T \ (3). |