/*\ blocking vector get
\*/
void armcill_getv(int proc, int bytes, int count, void* src[], void* dst[])
{
int _j, i, batch, issued=0;
void *ps, *pd;
#if 0
printf("%d: getv %d\n", armci_me, count); fflush(stdout);
#endif
for (_j = 0; _j < count; ){
/* how big a batch of requests can we issue */
batch = (count - _j )<max_pending ? count - _j : max_pending;
_j += batch;
for(i=0; i< batch; i++){
if(get_dscr[cur_get])elan_wait(get_dscr[cur_get],100);
else pending_get++;
ps = src[issued];
pd = dst[issued];
get_dscr[cur_get]= elan_get(elan_base->state,ps, pd,(size_t)bytes,proc);
issued++;
cur_get++;
if(cur_get>=max_pending)cur_get=0;
}
}
if(issued != count)
armci_die2("armci-elan getv:mismatch %d %d \n", count,issued);
for(i=0; i<max_pending; i++) if(get_dscr[i]){
elan_wait(get_dscr[i],100);
get_dscr[i]=(ELAN_EVENT*)0;
}
}
int elan_long_swap(long *target, long value, int vp)
{
long result;
#ifdef _LP64
elan_wait(elan_atomic64(a, ELAN_ATOMIC_SWAP, target, value, 0, vp, &result), elan_base->waitType);
#else
elan_wait(elan_atomic32(a, ELAN_ATOMIC_SWAP, target, value, 0, vp, &result), elan_base->waitType);
#endif
return(result);
}
int elan_long_fadd(long *target, long inc, int vp)
{
long result;
#ifdef _LP64
elan_wait(elan_atomic64(a, ELAN_ATOMIC_ADD, target, inc, 0, vp, &result), elan_base->waitType);
#else
elan_wait(elan_atomic32(a, ELAN_ATOMIC_ADD, target, inc, 0, vp, &result), elan_base->waitType);
#endif
return(result);
}
static
int run_pp_c(int msize, int loops)
{
int cnt;
assert(msize <= MAX_MSIZE);
ELAN_EVENT *event = NULL;
for (cnt = 0; cnt < loops; cnt++) {
unsigned len = msize;
s_buf->len = len;
s_buf->mark = 1;
r_buf->mark = 0;
event = elan_put(base->state,
s_buf->buf - len,
remote_rbuf->buf - len,
len + TAIL_SIZE,
RANK_SERVER);
elan_wait(event, base->waitType);
while (r_buf->mark != 1) idle();
}
return 0;
}
static
void run_pp_server(void)
{
ELAN_EVENT *event = NULL;
while (1) {
while (r_buf->mark != 1) idle();
unsigned len = r_buf->len;
//memcpy(s_buf->buf - len, r_buf->buf - len, len);
s_buf->len = len;
s_buf->mark = 1;
r_buf->mark = 0;
event = elan_put(base->state,
s_buf->buf - len,
remote_rbuf->buf - len,
len + TAIL_SIZE,
RANK_CLIENT);
elan_wait(event, base->waitType);
}
}
int elan_int_swap(int *target, int value, int vp)
{
int result;
elan_wait(elan_atomic32(a, ELAN_ATOMIC_SWAP, target, value, 0, vp, &result), elan_base->waitType);
return(result);
}
int elan_int_fadd(int *target, int inc, int vp)
{
int result;
elan_wait(elan_atomic32(a, ELAN_ATOMIC_ADD, target, inc, 0, vp, &result), elan_base->waitType);
return(result);
}
void armcill_get2D(int proc, int bytes, int count, void* src_ptr,int src_stride,
void* dst_ptr,int dst_stride)
{
u_int acount[2];
acount[0]=bytes;
acount[1]=count;
elan_wait(elan_getss(_pgsstate,src_ptr,dst_ptr, &src_stride, &dst_stride, acount,1,proc),elan_base->waitType);
}
void armcill_getv(int proc, int bytes, int count, void* src[], void* dst[])
{
int _j, issued=0;
#if 0
printf("%d: getv %d\n", armci_me, count); fflush(stdout);
#endif
for (_j = 0; _j < count; _j++ ){
_src[issued] = src[_j];
_dst[issued] = dst[_j];
issued++;
if(issued == MAX_VECS){
elan_wait(elan_getv(_pgctrl,_src,_dst,bytes,issued,proc),elan_base->waitType);
issued=0;
}
}
if(issued)elan_wait(elan_getv(_pgctrl,_src,_dst,bytes,issued,proc),elan_base->waitType);
}
void armcill_wait_put()
{
int i;
if(!pending_put)return;
else pending_put=0;
for(i=0; i<max_pending; i++) if(put_dscr[i]){
elan_wait(put_dscr[i],100);
put_dscr[i]=(ELAN_EVENT*)0;
}
}
void armcill_get2D(int proc, int bytes, int count, void* src_ptr,int src_stride,
void* dst_ptr,int dst_stride)
{
int _j, issued=0;
char *ps=src_ptr, *pd=dst_ptr;
#if 0
printf("%d: getv %d\n", armci_me, count); fflush(stdout);
#endif
for (_j = 0; _j < count; _j++ ){
_src[issued] = ps;
_dst[issued] = pd;
ps += src_stride;
pd += dst_stride;
issued++;
if(issued == MAX_VECS){
elan_wait(elan_getv(_pgctrl,_src,_dst,bytes,issued,proc),elan_base->waitType);
issued=0;
}
}
if(issued)elan_wait(elan_getv(_pgctrl,_src,_dst,bytes,issued,proc),elan_base->waitType);
}
/*\ strided get, nonblocking
\*/
void armcill_get2D(int proc, int bytes, int count, void* src_ptr,int src_stride,
void* dst_ptr,int dst_stride)
{
int _j, i, batch, issued=0;
char *ps=src_ptr, *pd=dst_ptr;
#if 1
for (_j = 0; _j < count; ){
/* how big a batch of requests can we issue */
batch = (count - _j )<max_pending ? count - _j : max_pending;
_j += batch;
for(i=0; i< batch; i++){
#if 1
if(get_dscr[cur_get])elan_wait(get_dscr[cur_get],100);
else pending_get++;
get_dscr[cur_get]=elan_get(elan_base->state,ps,pd, (size_t)bytes, proc);
#else
elan_wait(elan_get(elan_base->state, ps, pd, (size_t)bytes, proc),elan_base->waitType);
#endif
issued++;
ps += src_stride;
pd += dst_stride;
cur_get++;
if(cur_get>=max_pending)cur_get=0;
}
}
if(issued != count)
armci_die2("armci-elan get:mismatch %d %d \n", count,issued);
#else
for (_j = 0; _j < count; _j++){
elan_wait(elan_get(elan_base->state, ps, pd, (size_t)bytes, proc),elan_base->waitType);
ps += src_stride;
pd += dst_stride;
}
#endif
}
void armcill_putS(int proc, void* src_ptr, int src_stride_arr[], void* dst_ptr,
int dst_stride_arr[], int count[], int stride_levels)
{
elan_wait(elan_putss(_pgsstate,src_ptr,dst_ptr, src_stride_arr,
dst_stride_arr, count, stride_levels, proc),elan_base->waitType);
}
