// Performs a union of two sets, does not allow duplicate values
struct set_t *set_union(struct set_t *set1, struct set_t *set2) {
struct set_t *temp_set_head = NULL;
struct set_t *temp_set_it = NULL;
struct set_t *set_it = set1;
// Add each element of set1
while(set_it != NULL) {
if(temp_set_head == NULL) {
temp_set_head = new_set(set_it->value);
temp_set_it = temp_set_head;
} else {
temp_set_it->next = set_add(temp_set_head, set_it->value);
if(temp_set_it->next != NULL)
temp_set_it = temp_set_it->next;
}
set_it = set_it->next;
}
// Add each element of set2
set_it = set2;
while(set_it != NULL) {
if(temp_set_head == NULL) {
temp_set_head = new_set(set_it->value);
temp_set_it = temp_set_head;
} else {
temp_set_it->next = set_add(temp_set_head, set_it->value);
if(temp_set_it->next != NULL)
temp_set_it = temp_set_it->next;
}
set_it = set_it->next;
}
return temp_set_head;
}
void test_set_operations(){
set_t *even1 = new_set(10);
set_t *even2 = new_set(10);
set_t *odd = new_set(10);
int i;
for (i=0; i < 10; i++){
set_put(even1, 2*i);
set_put(even2, 2*i);
set_put(odd, 2*i+1);
}
set_union(even1, odd);
assert(set_size(even1) == 20);
set_difference(even2, odd);
assert(set_size(even2) == 10);
set_intersection(even2, odd);
assert(set_size(even2) == 0);
set_print(even1); printf("\n");
set_optimize(even1);
set_print(even1); printf("\n");
set_print(even2); printf("\n");
set_print(odd); printf("\n");
delete_set(even1);
delete_set(even2);
delete_set(odd);
}
/*********************************************************
* construct LR(1) items. *
* This algorithm merges canonical LR(1) items together *
* based on thier common cores. *
*-------------------------------------------------------*
* This is the main LR algorithm used in this program. *
* Effectivly this is the core of clrgen and these item *
* sets are used to generate our parse table below. *
* NOTE: *
* this algorithms efficiency can be greatly improved by *
* unfolding bnf_goto into the main loop and by changing *
* the way we compute transitions by preventing the *
* function from looping through every grammar symbol *
*********************************************************/
cset* bnf_construct_items(bnf_grammar* bnf)
{
int i=0;
slist_elem* isle=0;
slist_elem* sle=0;
int iind=0;
int iadded=1;
cset* items=new_set(compare_bnf_item_sets);
cset* current_set=new_set(compare_bnf_index);
bnf_index* cbi=new_bnf_index(0,1,0,0);
if(!bnf)
return 0;
/*********************************************************
* construct the first item/start state of the item set *
*********************************************************/
cset_add(cbi->lookaheads,(void*)bnf->start_of_actions);
cset_add(cbi->lookaheads,(void*)(bnf->start_of_actions+1));
cset_add(current_set,(void*)cbi);
bnf_closure(bnf,cbi,current_set);
cset_add(items,(void*)current_set);
isle=items->members->_head;
for(;isle;isle=isle->_next,++iind)
/***********************************************************
* cycle through all symbols and compute thier transitions *
* for this current item by taking a goto closure and *
* trying to add it to the core item set. *
* if we can't add then we merge lookaheads with a *
* matching state if any because we do not want to discard *
* generated lookaheads. *
***********************************************************/
for(i=0;i<bnf->start_of_actions;++i)
{
bnf_goto_closure(bnf,(cset*)isle->_data,current_set,i);
if(current_set&¤t_set->members->_size>0)
{
if(!cset_add(items,(void*)current_set))
bnf_merge_item_lookaheads((cset*)cset_get_first_match(items,current_set),current_set);
/*
re-allocate a new set structure.
NOTE: i want to take this out and use a
cache-able version because reallocations
are unneccessary and affect performance.
*/
current_set=new_set(compare_bnf_index);
}
}
while(propagate_lookaheads(bnf,items));
return items;
}
static void init_vertex(vertex_t* v, size_t index, size_t nsymbol, size_t max_succ)
{
int i;
v->index = index;
v->succ = calloc(max_succ, sizeof(vertex_t*));
if (v->succ == NULL)
error("out of memory");
for (i = 0; i < NSETS; i += 1)
v->set[i] = new_set(nsymbol);
v->prev = new_set(nsymbol);
}
/*************************************************
* initialize the global bnf first items array. *
* This array mirrors the bnf->lexicon array and *
* and each entry here represents a first items *
* entry for the symbol of the same index in the *
* lexicon array. *
*************************************************/
void initialize_first_array(bnf_grammar* bnf)
{
first_items=(cset**)calloc(bnf->start_of_actions+1,sizeof(cset*));
int i=0;
for(;i<=bnf->start_of_actions;++i)
first_items[i]=new_set(compare_bnf_lookaheads);
}
int *intersect_sets(int *l1, int *l2, int type) /* makes the intersection of two sets */
{
int i, *l = new_set(type);
for(i = 0; i < set_size(type); i++)
l[i] = l1[i] & l2[i];
return l;
}
int *dup_set(int *l, int type) /* duplicates a set */
{
int i, *m = new_set(type);
for(i = 0; i < set_size(type); i++)
m[i] = l[i];
return m;
}
int *make_set(int n, int type) /* creates the set {n}, or the empty set if n = -1 */
{
int *l = clear_set(new_set(type), type);
if(n == -1) return l;
l[n/mod] = 1 << (n%mod);
return l;
}
// Relative complement of set1 in set2. Returns a set of elements that exist in set2 but not in set1
struct set_t *set_difference(struct set_t *set1, struct set_t *set2) {
struct set_t *temp_set_head = NULL;
struct set_t *temp_set_it = NULL;
struct set_t *set_it = set2;
// Loop through every element in set2
while(set_it != NULL) {
// Element must not exist in set1
if(!set_contains(set1, set_it->value)) {
if(temp_set_head == NULL) {
temp_set_head = new_set(set_it->value);
temp_set_it = temp_set_head;
} else {
temp_set_it->next = set_add(temp_set_head, set_it->value);
if(temp_set_it->next != NULL)
temp_set_it = temp_set_it->next;
}
}
set_it = set_it->next;
}
return temp_set_head;
}
void GameInstSet::reallocate_internal_data() {
unit_capacity *= 2;
std::vector<InstanceState> new_set(unit_capacity);
tset_add_all<GameInstSetFunctions>(&unit_set[0], unit_set.size(),
&new_set[0], new_set.size());
unit_set.swap(new_set);
//Fix pointers for grid
for (int i = 0; i < unit_capacity; i++) {
if (unit_set[i].inst) {
update_statepointer_for_reallocate_(&unit_set[i].prev_in_grid);
update_statepointer_for_reallocate_(&unit_set[i].next_in_grid);
update_statepointer_for_reallocate_(&unit_set[i].prev_same_depth);
update_statepointer_for_reallocate_(&unit_set[i].next_same_depth);
}
}
DepthMap::iterator it = depthlist_map.begin();
for (; it != depthlist_map.end(); it++) {
update_depthlist_for_reallocate_(it->second);
}
for (int i = 0; i < grid_w * grid_h; i++) {
update_depthlist_for_reallocate_(unit_grid[i]);
}
}
// Performs an intersection of two sets
struct set_t *set_intersection(struct set_t *set1, struct set_t *set2) {
if(set1 == EMPTY || set2 == EMPTY)
return EMPTY;
struct set_t *temp_set_head = NULL;
struct set_t *temp_set_it = NULL;
struct set_t *set_it = set1;
// Go through every element of set 1
set_it = set1;
while(set_it != NULL) {
// See if the elements value is also in set2
if(set_contains(set2, set_it->value)) {
if(temp_set_head == NULL) {
temp_set_head = new_set(set_it->value);
temp_set_it = temp_set_head;
} else {
// set_add will resolve any attempts to add duplicate elements
temp_set_it->next = set_add(temp_set_head, set_it->value);
if(temp_set_it->next != NULL)
temp_set_it = temp_set_it->next;
}
}
set_it = set_it->next;
}
return temp_set_head;
}
Set sub_set(Set s1,Set s2)
{
int i;
Set result = new_set();
for(i=0;i < SET_SIZE / unit_size;i++)
result[i] = s1[i] & (~s2[i]);
return result;
}
Set intersect_set(Set s1,Set s2)
{
int i;
Set result = new_set();
for(i=0;i < SET_SIZE / unit_size;i++)
result[i] = s1[i] & s2[i];
return result;
}
void test_basic(){
set_t *set = new_set(10);
int i, j;
for (i=0; i < 10; i++){
for (j=0; j < 10; j++){
set_put(set, i);
}
}
assert(set_size(set) == 10);
delete_set(set);
}
/************************************************************
* simple constructor code for bnf_index. *
************************************************************/
bnf_index* new_bnf_index(int i,
int p,
int pp,
cset* lookaheads)
{
bnf_index* bi=(bnf_index*)malloc(sizeof(bnf_index));
bi->item=i;
bi->product=p;
bi->product_part=pp;
lookaheads?
bi->lookaheads=cset_copy(lookaheads):
bi->lookaheads=new_set(compare_bnf_lookaheads);
return bi;
}
void build_node_metasets_aux(nodemetaset_t node_metasets[], uint n_nodes,
uint index)
{
// since it is pointless to consider sets of size 0, index 0 is used for sets
// of size 1 and so on.
uint size = index + 1;
// base case -- stop after reaching the full size
if(size > n_nodes)
return;
// bootstrap -- unitary sets must be built by hand
else if(size == 1)
{
// size 1 is initialised with a unitary set containing the starting node
nodeset_t new_set;
new_set.insert(0);
node_metasets[index].insert(new_set);
}
// recursion -- will out each size based on previous sizes
else
{
nodemetaset_t& previous = node_metasets[index-1];
// we'll be adding each node identifier to what already exists
for(uint node_i = 0; node_i < n_nodes; node_i++)
{
// for each set of sets that is 1 smaller than the one we're currently
// trying to build
for(nodemetaset_it nms_i = previous.begin(); nms_i != previous.end();
nms_i++)
{
// copy each set and add a different identifier to the copy each time
nodeset_t new_set((*nms_i));
new_set.insert(node_i);
// add the result to the node-set set we're currently building, but only
// if the size is correct, that is we didn't add something that was
// already there
if(new_set.size() == size)
node_metasets[index].insert(new_set);
}
}
}
// move on to the next size - remember that we incremented size at the start:
// it will be incremented again at the beginning of the next step !
build_node_metasets_aux(node_metasets, n_nodes, index+1);
}
void test_removing(){
set_t *set = new_set(0);
int i;
for (i=0; i < 1000; i++){ set_put(set, i); }
assert(set_size(set) == 1000);
set_optimize(set);
for (i=0; i < 1000; i += 2){ set_remove(set, i); }
assert(set_size(set) == 500);
for (i=0; i < 1000; i += 2){ set_put(set, i); }
assert(set_size(set) == 1000);
delete_set(set);
}
/******************************************
* returns all bnf_index's within a set *
* who's positions are greater than zero. *
* these are known as the kernels of an *
* item set. *
******************************************/
cset* bnf_get_item_kernels(cset* item)
{
cset* kernels=0;
slist_elem* sle=0;
if(!item||!item->members)
return 0;
kernels=new_set(compare_bnf_index);
sle=item->members->_head;
for(;sle;sle=sle->_next)
if(((bnf_index*)sle->_data)->product_part>0)/*any product not at 0 is a kernel*/
cset_add(kernels,sle->_data);
else if(((bnf_index*)sle->_data)->item==0&&/*recognise a start symbol aswell*/
((bnf_index*)sle->_data)->product_part==0)
cset_add(kernels,sle->_data);
return kernels;
}
// Add an element (val) to a non-empty set if it doesn't already exist in set
// Returns the previously added set_t object or NULL if nothing was added
struct set_t *set_add(struct set_t *set, const char *val) {
if(set == NULL || val == NULL)
return NULL;
// Find the end of the list
struct set_t *set_it = set;
struct set_t *set_last = set;
while(set_it != NULL) {
// Val already exists, return NULL
if(strcmp(set_it->value, val) == 0)
return NULL;
set_last = set_it;
set_it = set_it->next;
}
// Create a new set_t object. Add to the end of the list and copy the value
set_last->next = new_set(val);
return set_last->next;
}
void test_picking(){
int n = 10000;
set_t *set = new_set(0);
int i;
for (i=0; i < n; i++){
set_put(set, i);
}
assert(set_size(set) == n);
set_entry_t *p = set_head(set);
for (i=0; i < n; i++){
assert(p->key == i);
p = p->next;
}
for (i=0; i < 3*n; i++){
int v = set_get_random(set);
assert(v < n);
}
delete_set(set);
}
lc_arg_env_t *lc_arg_new_env(void)
{
lc_arg_env_t *env = XMALLOCZ(lc_arg_env_t);
env->args = new_set(lc_arg_cmp, 16);
return env;
}
static struct rr* nsec3_parse(char *name, long ttl, int type, char *s)
{
struct rr_nsec3 *rr = getmem(sizeof(*rr));
struct rr *ret_rr;
struct binary_data bitmap;
int i;
int opt_out = 0;
char *str_type = NULL;
int ltype;
i = extract_integer(&s, "hash algorithm");
if (i < 0)
return NULL;
if (i > 255)
return bitch("bad hash algorithm value");
if (i != 1)
return bitch("unrecognized or unsupported hash algorithm");
rr->hash_algorithm = i;
i = extract_integer(&s, "flags");
if (i < 0)
return NULL;
if (i > 255)
return bitch("bad flags value");
if (!(i == 0 || i == 1))
return bitch("unsupported flags value");
if (i == 1)
opt_out = 1;
rr->flags = i;
i = extract_integer(&s, "iterations");
if (i < 0)
return NULL;
if (i > 2500)
return bitch("bad iterations value");
rr->iterations = i;
/* TODO validate iteration count according to key size,
* as per http://tools.ietf.org/html/rfc5155#section-10.3 */
if (*s == '-') {
rr->salt.length = 0;
rr->salt.data = NULL;
s++;
if (*s && !isspace(*s) && *s != ';' && *s != ')')
return bitch("salt is not valid");
s = skip_white_space(s);
} else {
rr->salt = extract_hex_binary_data(&s, "salt", EXTRACT_DONT_EAT_WHITESPACE);
if (rr->salt.length <= 0)
return NULL;
if (rr->salt.length > 255)
return bitch("salt is too long");
}
rr->next_hashed_owner = extract_base32hex_binary_data(&s, "next hashed owner");
bitmap = new_set();
while (s && *s) {
str_type = extract_label(&s, "type list", "temporary");
if (!str_type) return NULL;
ltype = str2rdtype(str_type);
add_bit_to_set(&bitmap, ltype);
}
if (!s)
return NULL;
rr->type_bitmap = compressed_set(&bitmap);
ret_rr = store_record(type, name, ttl, rr);
if (ret_rr) {
G.nsec3_present = 1;
if (opt_out)
G.nsec3_opt_out_present = 1;
}
return ret_rr;
}
static void *magma_ssytrd_sb2st_parallel_section(void *arg)
{
magma_int_t my_core_id = ((magma_sbulge_id_data*)arg) -> id;
magma_sbulge_data* data = ((magma_sbulge_id_data*)arg) -> data;
magma_int_t allcores_num = data -> threads_num;
magma_int_t n = data -> n;
magma_int_t nb = data -> nb;
magma_int_t nbtiles = data -> nbtiles;
magma_int_t grsiz = data -> grsiz;
magma_int_t Vblksiz = data -> Vblksiz;
magma_int_t compT = data -> compT;
float *A = data -> A;
magma_int_t lda = data -> lda;
float *V = data -> V;
magma_int_t ldv = data -> ldv;
float *TAU = data -> TAU;
float *T = data -> T;
magma_int_t ldt = data -> ldt;
volatile magma_int_t* prog = data -> prog;
pthread_barrier_t* barrier = &(data -> barrier);
//magma_int_t sys_corenbr = 1;
#ifdef ENABLE_TIMER
real_Double_t timeB=0.0, timeT=0.0;
#endif
// with MKL and when using omp_set_num_threads instead of mkl_set_num_threads
// it need that all threads setting it to 1.
magma_setlapack_numthreads(1);
#ifdef MAGMA_SETAFFINITY
//#define PRINTAFFINITY
#ifdef PRINTAFFINITY
affinity_set print_set;
print_set.print_affinity(my_core_id, "starting affinity");
#endif
affinity_set original_set;
affinity_set new_set(my_core_id);
int check = 0;
int check2 = 0;
// bind threads
check = original_set.get_affinity();
if (check == 0) {
check2 = new_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (single cpu)\n");
}
else
{
printf("Error in sched_getaffinity\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "set affinity");
#endif
#endif
if(compT==1)
{
/* compute the Q1 overlapped with the bulge chasing+T.
* if all_cores_num=1 it call Q1 on GPU and then bulgechasing.
* otherwise the first thread run Q1 on GPU and
* the other threads run the bulgechasing.
* */
if(allcores_num==1)
{
//=========================
// bulge chasing
//=========================
#ifdef ENABLE_TIMER
timeB = magma_wtime();
#endif
magma_stile_bulge_parallel(0, 1, A, lda, V, ldv, TAU, n, nb, nbtiles, grsiz, Vblksiz, prog);
#ifdef ENABLE_TIMER
timeB = magma_wtime()-timeB;
printf(" Finish BULGE timing= %f \n" ,timeB);
#endif
//=========================
// compute the T's to be used when applying Q2
//=========================
#ifdef ENABLE_TIMER
timeT = magma_wtime();
#endif
magma_stile_bulge_computeT_parallel(0, 1, V, ldv, TAU, T, ldt, n, nb, Vblksiz);
#ifdef ENABLE_TIMER
timeT = magma_wtime()-timeT;
printf(" Finish T's timing= %f \n" ,timeT);
#endif
}else{ // allcore_num > 1
magma_int_t id = my_core_id;
magma_int_t tot = allcores_num;
//=========================
// bulge chasing
//=========================
#ifdef ENABLE_TIMER
if(id == 0)
timeB = magma_wtime();
#endif
magma_stile_bulge_parallel(id, tot, A, lda, V, ldv, TAU, n, nb, nbtiles, grsiz, Vblksiz, prog);
pthread_barrier_wait(barrier);
#ifdef ENABLE_TIMER
if(id == 0){
timeB = magma_wtime()-timeB;
printf(" Finish BULGE timing= %f \n" ,timeB);
}
#endif
//=========================
// compute the T's to be used when applying Q2
//=========================
#ifdef ENABLE_TIMER
if(id == 0)
timeT = magma_wtime();
#endif
magma_stile_bulge_computeT_parallel(id, tot, V, ldv, TAU, T, ldt, n, nb, Vblksiz);
pthread_barrier_wait(barrier);
#ifdef ENABLE_TIMER
if (id == 0){
timeT = magma_wtime()-timeT;
printf(" Finish T's timing= %f \n" ,timeT);
}
#endif
} // allcore == 1
}else{ // WANTZ = 0
//=========================
// bulge chasing
//=========================
#ifdef ENABLE_TIMER
if(my_core_id == 0)
timeB = magma_wtime();
#endif
magma_stile_bulge_parallel(my_core_id, allcores_num, A, lda, V, ldv, TAU, n, nb, nbtiles, grsiz, Vblksiz, prog);
pthread_barrier_wait(barrier);
#ifdef ENABLE_TIMER
if(my_core_id == 0){
timeB = magma_wtime()-timeB;
printf(" Finish BULGE timing= %f \n" ,timeB);
}
#endif
} // WANTZ > 0
#ifdef MAGMA_SETAFFINITY
// unbind threads
if (check == 0){
check2 = original_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (restore cpu list)\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "restored_affinity");
#endif
#endif
return 0;
}
|<product>T(ACTION){printf(\"body:<product>T(ACTION)\\n\");}\
|<body>T(PIPE)<body>{printf(\"body:<body>T(PIPE)<body>\\n\");}\
;\
product:<part>{printf(\"product:<part>\\n\");}\
|<product><part>{printf(\"product:<product><part>\\n\");}\
;\
part:T(TOKEN){printf(\"part:T(TOKEN)\\n\");}\
|T(PRODUCT){printf(\"part:T(PRODUCT)\\n\");}\
;\
*/
bool success=false;
bnf_parse_item* item=0;
static slist* product=new_slist();
static char* temp_val=0;
cset* products=new_set(comp_str);
cset* tokens=new_set(comp_str);
cset* actions=new_set(comp_str);
//beginning of stack data structure used by this parser
struct pstack_elem{void *_data;pstack_elem *_next;};
struct pcstack{size_t size;pstack_elem* first;};
pcstack* new_pcstack()
{
pcstack* pcs((pcstack*)malloc(sizeof(pcstack)));
memset((void*)pcs,0,sizeof(pcstack));
pcs->size=0;
pcs->first=0;
return pcs;
}
void* peek_stack(pcstack* stack)
void init_ident(void)
{
/* it's ok to use memcmp here, we check only strings */
id_set = new_set(memcmp, 128);
obstack_init(&id_obst);
}
//##################################################################################################
static void *magma_zapplyQ_m_parallel_section(void *arg)
{
magma_int_t my_core_id = ((magma_zapplyQ_m_id_data*)arg) -> id;
magma_zapplyQ_m_data* data = ((magma_zapplyQ_m_id_data*)arg) -> data;
magma_int_t ngpu = data -> ngpu;
magma_int_t allcores_num = data -> threads_num;
magma_int_t n = data -> n;
magma_int_t ne = data -> ne;
magma_int_t n_gpu = data -> n_gpu;
magma_int_t nb = data -> nb;
magma_int_t Vblksiz = data -> Vblksiz;
magmaDoubleComplex *E = data -> E;
magma_int_t lde = data -> lde;
magmaDoubleComplex *V = data -> V;
magma_int_t ldv = data -> ldv;
magmaDoubleComplex *TAU = data -> TAU;
magmaDoubleComplex *T = data -> T;
magma_int_t ldt = data -> ldt;
pthread_barrier_t* barrier = &(data -> barrier);
magma_int_t info;
#ifdef ENABLE_TIMER
real_Double_t timeQcpu=0.0, timeQgpu=0.0;
#endif
magma_int_t n_cpu = ne - n_gpu;
// with MKL and when using omp_set_num_threads instead of mkl_set_num_threads
// it need that all threads setting it to 1.
magma_set_lapack_numthreads(1);
#ifdef MAGMA_SETAFFINITY
//#define PRINTAFFINITY
#ifdef PRINTAFFINITY
affinity_set print_set;
print_set.print_affinity(my_core_id, "starting affinity");
#endif
affinity_set original_set;
affinity_set new_set(my_core_id);
int check = 0;
int check2 = 0;
// bind threads
check = original_set.get_affinity();
if (check == 0) {
check2 = new_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (single cpu)\n");
}
else {
printf("Error in sched_getaffinity\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "set affinity");
#endif
#endif
if (my_core_id == 0) {
//=============================================
// on GPU on thread 0:
// - apply V2*Z(:,1:N_GPU)
//=============================================
#ifdef ENABLE_TIMER
timeQgpu = magma_wtime();
#endif
magma_zbulge_applyQ_v2_m(ngpu, MagmaLeft, n_gpu, n, nb, Vblksiz, E, lde, V, ldv, T, ldt, &info);
magma_device_sync();
#ifdef ENABLE_TIMER
timeQgpu = magma_wtime()-timeQgpu;
printf(" Finish Q2_GPU GGG timing= %f\n", timeQgpu);
#endif
} else {
//=============================================
// on CPU on threads 1:allcores_num-1:
// - apply V2*Z(:,N_GPU+1:NE)
//=============================================
#ifdef ENABLE_TIMER
if (my_core_id == 1)
timeQcpu = magma_wtime();
#endif
magma_int_t n_loc = magma_ceildiv(n_cpu, allcores_num-1);
magmaDoubleComplex* E_loc = E + (n_gpu+ n_loc * (my_core_id-1))*lde;
n_loc = min(n_loc,n_cpu - n_loc * (my_core_id-1));
magma_ztile_bulge_applyQ(my_core_id, MagmaLeft, n_loc, n, nb, Vblksiz, E_loc, lde, V, ldv, TAU, T, ldt);
pthread_barrier_wait(barrier);
#ifdef ENABLE_TIMER
if (my_core_id == 1) {
timeQcpu = magma_wtime()-timeQcpu;
printf(" Finish Q2_CPU CCC timing= %f\n", timeQcpu);
}
#endif
} // END if my_core_id
#ifdef MAGMA_SETAFFINITY
// unbind threads
if (check == 0) {
check2 = original_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (restore cpu list)\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "restored_affinity");
#endif
#endif
return 0;
}
void Control::ComputeRecompilationSet(TypeDependenceChecker& dependence_checker)
{
SymbolSet type_trash_set;
//
// Find out if any source files has been touched since the last
// compilation and add all such files to recompilation_file_set.
//
FindMoreRecentInputFiles(dependence_checker.file_set);
//
// Before messing with the files, compute a list of all the types that
// have just been compiled. We need to do this here as we will be
// "Resetting" and "reScanning" some files in the loop below, which in
// effect removes the set of types to which they were associated in the
// previous compilation.
//
int length_estimate = input_java_file_set.Size(); // problem size estimate
Tuple<TypeSymbol*> input_types(length_estimate * 2);
FileSymbol* file_symbol;
for (file_symbol = (FileSymbol*) input_java_file_set.FirstElement();
file_symbol;
file_symbol = (FileSymbol*) input_java_file_set.NextElement())
{
for (unsigned i = 0; i < file_symbol -> types.Length(); i++)
input_types.Next() = file_symbol -> types[i];
}
//
// Declare the closure set, and initialize it with the Union over the
// closure of the types in the trash_bin. Essentially, we want to catch
// all "compiled" types in the compilation that has a dependence on these
// bad types.
//
SymbolSet dependents_closure(length_estimate);
for (unsigned i = 0; i < type_trash_bin.Length(); i++)
{
TypeSymbol* type = type_trash_bin[i];
if (! dependents_closure.IsElement(type))
{
if (type -> dependents_closure)
dependents_closure.Union(*type -> dependents_closure);
else dependents_closure.AddElement(type);
}
}
//
// Compute the set of types from the recompilation set that needs to be
// recompiled and update the recompilation file set.
//
SymbolSet new_set(length_estimate),
file_seen(length_estimate);
new_set = recompilation_file_set;
new_set.Union(expired_file_set);
file_seen = new_set;
// How much space do we need for a package declaration? estimate 64 tokens.
StoragePool* ast_pool = new StoragePool(64);
//
// As long as there is a new_set of files to process,...
//
do
{
//
// For each file in new_set, compute the reflexive transitive closure
// of all types contained in that file. Next, reset and rescan the
// file. If the scan was successful, iterate over the new list of
// types to see if any of them had already been introduced in the
// previous compilation via a class file. If so, add all such types to
// the dependents closure.
//
for (FileSymbol* file_symbol = (FileSymbol*) new_set.FirstElement();
file_symbol;
file_symbol = (FileSymbol*) new_set.NextElement())
{
for (unsigned i = 0; i < file_symbol -> types.Length(); i++)
{
TypeSymbol* type = file_symbol -> types[i];
if (! dependents_closure.IsElement(type))
{
if (type -> dependents_closure)
dependents_closure.Union(*type -> dependents_closure);
else dependents_closure.AddElement(type);
}
}
if (! expired_file_set.IsElement(file_symbol))
{
file_symbol -> Reset();
file_symbol -> SetJava();
scanner -> Scan(file_symbol);
LexStream* lex_stream = file_symbol -> lex_stream;
if (lex_stream) // did we have a successful scan!
{
AstPackageDeclaration* package_declaration
= parser -> PackageHeaderParse(lex_stream, ast_pool);
PackageSymbol* package
= (package_declaration
? FindOrInsertPackage(lex_stream,
package_declaration -> name)
: unnamed_package);
ast_pool -> Reset();
//
// If the file contained more than one type, only the main
// one would have been deleted. We now delete the others if
// any...
//
for (unsigned k = 0; k < lex_stream -> NumTypes(); k++)
{
TokenIndex identifier_token
= lex_stream -> Next(lex_stream -> Type(k));
NameSymbol* name_symbol =
lex_stream -> NameSymbol(identifier_token);
if (name_symbol)
{
TypeSymbol* type
= package -> FindTypeSymbol(name_symbol);
if (type && (! dependents_closure.IsElement(type)))
{
if (type -> dependents_closure)
dependents_closure.Union(*type -> dependents_closure);
else dependents_closure.AddElement(type);
}
}
}
}
}
}
//
// Iterate over the dependents_closure set. For each type T, add it to
// the trash pile. If the file with which it is associated had not yet
// been processed, mark it as having been "seen" and add it to the
// new_set to be considered later. If the file had already been
// processed but not yet added to the recompilation set, add it to the
// recompilation set, read it in and if it contains types other than
// the main one (that had previously been read in via class files) add
// those new types to the trash pile.
//
new_set.SetEmpty();
TypeSymbol* type;
for (type = (TypeSymbol*) dependents_closure.FirstElement();
type;
type = (TypeSymbol*) dependents_closure.NextElement())
{
type_trash_set.AddElement(type);
FileSymbol* file_symbol = type -> file_symbol;
if (file_symbol && (! file_seen.IsElement(file_symbol)))
{
file_seen.AddElement(file_symbol);
new_set.AddElement(file_symbol);
file_symbol -> mtime = 0; // to force a reread of the file.
}
}
//
// Check that the files in new_set exist, and if so, add them to the
// recompilation_file_set. Note that if they exist, they will be added
// because before a file is added to new_set its time stamp is reset
// to 0. See loop above...
//
FindMoreRecentInputFiles(new_set);
//
// Empty out the dependents_closure set for the next round.
//
dependents_closure.SetEmpty();
} while (! new_set.IsEmpty());
delete ast_pool;
//
// Clean up the types that were compiled in the previous compilation pass.
//
for (unsigned j = 0; j < input_types.Length(); j++)
input_types[j] -> RemoveCompilationReferences();
//
// Reset the closure sets in all the types that were considered in the
// dependence checker.
//
Tuple<TypeSymbol*>& type_list = dependence_checker.TypeList();
for (unsigned k = 0; k < type_list.Length(); k++)
{
TypeSymbol* type = type_list[k];
type -> index = TypeCycleChecker::OMEGA;
type -> unit_index = TypeCycleChecker::OMEGA;
type -> incremental_index = TypeCycleChecker::OMEGA;
delete type -> dependents_closure;
type -> dependents_closure = NULL;
}
//
// Remove all dependence edges that are no longer valid.
//
RemoveTrashedTypes(type_trash_set);
}
static void *magma_dsytrd_sb2st_parallel_section(void *arg)
{
magma_int_t my_core_id = ((magma_dbulge_id_data*)arg) -> id;
magma_dbulge_data* data = ((magma_dbulge_id_data*)arg) -> data;
magma_int_t allcores_num = data -> threads_num;
magma_int_t n = data -> n;
magma_int_t nb = data -> nb;
magma_int_t nbtiles = data -> nbtiles;
magma_int_t grsiz = data -> grsiz;
magma_int_t Vblksiz = data -> Vblksiz;
magma_int_t wantz = data -> wantz;
double *A = data -> A;
magma_int_t lda = data -> lda;
double *V = data -> V;
magma_int_t ldv = data -> ldv;
double *TAU = data -> TAU;
double *T = data -> T;
magma_int_t ldt = data -> ldt;
volatile magma_int_t* prog = data -> prog;
pthread_barrier_t* myptbarrier = &(data -> myptbarrier);
//magma_int_t sys_corenbr = 1;
#ifdef ENABLE_TIMER
real_Double_t timeB=0.0, timeT=0.0;
#endif
// with MKL and when using omp_set_num_threads instead of mkl_set_num_threads
// it need that all threads setting it to 1.
//magma_set_omp_numthreads(1);
magma_set_lapack_numthreads(1);
magma_set_omp_numthreads(1);
/*
#ifndef MAGMA_NOAFFINITY
// bind threads
cpu_set_t set;
// bind threads
CPU_ZERO( &set );
CPU_SET( my_core_id, &set );
sched_setaffinity( 0, sizeof(set), &set);
#endif
magma_set_lapack_numthreads(1);
magma_set_omp_numthreads(1);
*/
#ifndef MAGMA_NOAFFINITY
//#define PRINTAFFINITY
#ifdef PRINTAFFINITY
affinity_set print_set;
print_set.print_affinity(my_core_id, "starting affinity");
#endif
affinity_set original_set;
affinity_set new_set(my_core_id);
magma_int_t check = 0;
magma_int_t check2 = 0;
// bind threads
check = original_set.get_affinity();
if (check == 0) {
check2 = new_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (single cpu)\n");
}
else {
printf("Error in sched_getaffinity\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "set affinity");
#endif
#endif
/* compute the Q1 overlapped with the bulge chasing+T.
* if all_cores_num=1 it call Q1 on GPU and then bulgechasing.
* otherwise the first thread run Q1 on GPU and
* the other threads run the bulgechasing.
* */
//=========================
// bulge chasing
//=========================
#ifdef ENABLE_TIMER
if (my_core_id == 0)
timeB = magma_wtime();
#endif
magma_dtile_bulge_parallel(my_core_id, allcores_num, A, lda, V, ldv, TAU, n, nb, nbtiles, grsiz, Vblksiz, wantz, prog, myptbarrier);
if (allcores_num > 1) pthread_barrier_wait(myptbarrier);
#ifdef ENABLE_TIMER
if (my_core_id == 0) {
timeB = magma_wtime()-timeB;
printf(" Finish BULGE timing= %f\n", timeB);
}
#endif
//=========================
// compute the T's to be used when applying Q2
//=========================
if ( wantz > 0 ) {
#ifdef ENABLE_TIMER
if (my_core_id == 0)
timeT = magma_wtime();
#endif
magma_dtile_bulge_computeT_parallel(my_core_id, allcores_num, V, ldv, TAU, T, ldt, n, nb, Vblksiz);
if (allcores_num > 1) pthread_barrier_wait(myptbarrier);
#ifdef ENABLE_TIMER
if (my_core_id == 0) {
timeT = magma_wtime()-timeT;
printf(" Finish T's timing= %f\n", timeT);
}
#endif
}
#ifndef MAGMA_NOAFFINITY
// unbind threads
if (check == 0) {
check2 = original_set.set_affinity();
if (check2 != 0)
printf("Error in sched_setaffinity (restore cpu list)\n");
}
#ifdef PRINTAFFINITY
print_set.print_affinity(my_core_id, "restored_affinity");
#endif
#endif
return 0;
}
static struct EvalError *check_stdmacro(
enum StandardMacro stdmacro, struct Expression *args, size_t n) {
size_t length;
struct Expression expr;
struct Set *set;
switch (stdmacro) {
case F_DEFINE:
case F_SET:
if (args[0].type != E_SYMBOL) {
return new_eval_error_expr(ERR_TYPE_VAR, args[0]);
}
break;
case F_LAMBDA:
expr = args[0];
if (expr.type != E_NULL
&& expr.type != E_PAIR
&& expr.type != E_SYMBOL) {
return new_syntax_error(expr);
}
set = new_set();
while (expr.type != E_NULL) {
InternId symbol_id;
if (expr.type == E_PAIR) {
if (expr.box->car.type != E_SYMBOL) {
free_set(set);
return new_eval_error_expr(ERR_TYPE_VAR, expr.box->car);
}
symbol_id = expr.box->car.symbol_id;
} else if (expr.type == E_SYMBOL) {
symbol_id = expr.symbol_id;
} else {
free_set(set);
return new_eval_error_expr(ERR_TYPE_VAR, expr);
}
if (!add_to_set(set, symbol_id)) {
free_set(set);
return attach_code(
new_eval_error_symbol(ERR_DUP_PARAM, symbol_id),
args[0]);
}
if (expr.type == E_SYMBOL) {
break;
}
expr = expr.box->cdr;
}
free_set(set);
break;
case F_UNQUOTE:
case F_UNQUOTE_SPLICING:
return new_eval_error(ERR_UNQUOTE);
case F_COND:
for (size_t i = 0; i < n; i++) {
if (!count_list(&length, args[i]) || length < 2) {
return new_syntax_error(args[i]);
}
}
break;
case F_LET:
case F_LET_STAR:
expr = args[0];
set = new_set();
while (expr.type != E_NULL) {
if (expr.type == E_PAIR) {
if (!count_list(&length, expr.box->car) || length != 2) {
free_set(set);
return new_syntax_error(expr.box->car);
}
if (expr.box->car.box->car.type != E_SYMBOL) {
free_set(set);
return new_eval_error_expr(
ERR_TYPE_VAR, expr.box->car.box->car);
}
} else {
free_set(set);
return new_syntax_error(args[0]);
}
InternId symbol_id = expr.box->car.box->car.symbol_id;
if (!add_to_set(set, symbol_id)) {
free_set(set);
return attach_code(
new_eval_error_symbol(ERR_DUP_PARAM, symbol_id),
args[0]);
}
expr = expr.box->cdr;
}
free_set(set);
break;
default:
break;
}
return NULL;
}