static void multroutine(const void *const arg)
{
const idargument *const ia = (idargument *)arg;
const workset *const ws = ia->tp->extra;
const runconfig *const rc = ia->tp->rc;
const unsigned id = ia->id;
const unsigned sz = rc->size;
const unsigned l = sz;
const unsigned m = sz;
const unsigned n = sz;
const unsigned tr = tilerows;
const unsigned tc = tilecols;
const joblayout al = definejob(rc, id, l, m, tr, tc);
const unsigned adiff = (al.baseoffset - al.mapoffset) / sizeof(eltype);
runconfig brc = *rc;
brc.nworkers = 1;
const joblayout bl = definejob(&brc, 0, m, n, tc, tr);
const joblayout rl = definejob(rc, id, l, n, tr, tr);
const unsigned rdiff = (rl.baseoffset - rl.mapoffset) / sizeof(eltype);
const eltype *const a
= (const eltype *)peekmap(rc, ws->fda,
al.mapoffset, al.maplength, pmshared)
+ adiff;
const eltype *const b
= (const eltype *)peekmap(rc, ws->fdb,
bl.mapoffset, bl.maplength, pmshared);
eltype *const r
= (eltype *)peekmap(rc, ws->fdr,
rl.mapoffset, rl.maplength, pmwrite | pmshared)
+ rdiff;
matmul(a, b, al.baserow, al.nrows, m, n, r);
printf("mult %03u with a:%u rows is done on core %d\n",
id, al.nrows, sched_getcpu());
}
static void randroutine(const void *const arg)
{
const idargument *const ia = (idargument *)arg;
const workset *const ws = ia->tp->extra;
const runconfig *const rc = ia->tp->rc;
const unsigned id = ia->id;
const unsigned sz = rc->size;
const unsigned l = sz;
const unsigned m = sz;
const unsigned n = sz;
const unsigned tr = tilerows;
const unsigned tc = tilecols;
const joblayout al = definejob(rc, id, l, m, tr, tc);
const joblayout bl = definejob(rc, id, m, n, tc, tr);
const unsigned adiff = (al.baseoffset - al.mapoffset) / sizeof(eltype);
eltype *const a
= (eltype *)peekmap(rc, ws->fda, al.mapoffset, al.maplength,
pmwrite | pmshared)
+ adiff;
const unsigned bdiff = (bl.baseoffset - bl.mapoffset) / sizeof(eltype);
eltype *const b
= (eltype *)peekmap(rc, ws->fdb, bl.mapoffset, bl.maplength,
pmwrite | pmshared)
+ bdiff;
matfill(id, al.absolutebaserow, a, al.baserow, al.nrows, m, tc,
elrand);
matfill(id * 5, bl.absolutebaserow, b, bl.baserow, bl.nrows, n, tr,
elrand);
printf("rand %03u with a:%u b:%u rows is done on core %d\n",
id, al.nrows, bl.nrows, sched_getcpu());
}
int main(const int argc, const char *const argv[])
{
const runconfig *const rc = formconfig(argc, argv, 64, 1024);
const unsigned sz = rc->size;
printf("\tmatrix size: %fMiB\n",
(double)sz * sz * sizeof(eltype) / (double)(1 << 20));
const workset ws = {
.fda = makeshm(rc, sz * sz * sizeof(eltype)),
.fdb = makeshm(rc, sz * sz * sizeof(eltype)),
.fdr = makeshm(rc, sz * sz * sizeof(eltype)) };
treeplugin tp = {
.makeargument = makeidargument,
.dropargument = dropidargument,
.rc = rc,
.extra = (void *)&ws };
printf("randomization\n");
fflush(stdout);
tp.treeroutine = randroutine;
treespawn(&tp);
printf("multiplication\n");
fflush(stdout);
tp.treeroutine = multroutine;
treespawn(&tp);
dropshm(ws.fda);
dropshm(ws.fdb);
printf("some values\n");
const unsigned tr = tilerows;
const unsigned m = sz;
const eltype *const r
= (const eltype *)peekmap(rc, ws.fdr,
0, m * m * sizeof(eltype), 0);
matdump(r, m, m, tr, tr, 127, 237, 8, 8);
dropshm(ws.fdr);
return 0;
}
static gint
grady (const guchar *image,
gint x,
gint y)
{
gint val = 0;
val = val + peekmap (image, x-1, y-1);
val = val + 2 * peekmap (image, x, y-1);
val = val + peekmap (image, x+1, y-1);
val = val - peekmap (image, x-1, y+1);
val = val - 2 * peekmap (image, x, y+1);
val = val - peekmap (image, x+1, y+1);
return val;
}
