/*
* Straightforward Sieve of Eratosthenes, but skipping 3.
*
* Uses 1 bit per odd number.
*
* Time for Pi(10^10) = 48.5s
*/
static WTYPE* sieve_base23(WTYPE end)
{
WTYPE* mem;
size_t n, s;
size_t last = (end+1)/2;
mem = (WTYPE*) calloc( NWORDS(last), sizeof(WTYPE) );
assert(mem != 0);
SET_ARRAY_BIT(mem, 1/2); /* 1 is composite */
/* Mark all multiples of 3. Could skip if callers know this. */
for (n = 3*3; n <= end; n += 2*3) SET_ARRAY_BIT(mem,n/2);
n = 5;
while ( (n*n) <= end ) {
if (!IS_SET_ARRAY_BIT(mem,n/2)) {
for (s = n*n; s <= end; s += 2*n) {
SET_ARRAY_BIT(mem,s/2);
}
}
n += 2;
if ( ((n*n) <= end) && (!IS_SET_ARRAY_BIT(mem,n/2)) ) {
for (s = n*n; s <= end; s += 2*n) {
SET_ARRAY_BIT(mem,s/2);
}
}
n += 4;
}
return mem;
}
static tracer_cmdresp_t _event_op_set_all(tracer_event_state_t state)
{
int ii;
int saveBit;
saveBit = GET_ARRAY_BIT(event_op_ctrl_t,
tracer_config.event_op_ctrl,
TRACER_EVENT_RESERVE_0);
if (TRACER_EVENT_ON == state)
{
if ((TRACER_EVENT_ID_MAX - 1) !=
tracer_config.event_in_use_count)
{
memset (tracer_config.event_op_ctrl, 0xFF,
TRACER_EVENT_ID_MAX >> 3);
for (ii = (TRACER_EVENT_ID_MAX & (~(uint32)0x07));
ii < TRACER_EVENT_ID_MAX; ii++)
{
SET_ARRAY_BIT(event_op_ctrl_t, tracer_config.event_op_ctrl,
ii);
}
if (0 == saveBit)
{
CLR_ARRAY_BIT(event_op_ctrl_t,
tracer_config.event_op_ctrl,
TRACER_EVENT_RESERVE_0);
}
tracer_config.event_in_use_count =
TRACER_EVENT_ID_MAX - 1; // less RESERVE0 event
return TRACER_CMDRESP_SUCCESS;
}
}
// Dependency: bTracerInitialized is TRUE.
static tracer_cmdresp_t _entity_op_set(tracer_ost_entity_id_enum_t eid,
tracer_entity_state_t state)
{
if (TRACER_ENTITY_SWEVT == eid)
{
return TRACER_CMDRESP_F_INVALID;
}
if (((0 == state) ? TRACER_ENTITY_OFF : TRACER_ENTITY_ON) !=
GET_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl, eid))
{
if (TRACER_ENTITY_ON == state)
{
SET_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl, eid);
tracer_config.entity_in_use_count++;
}
else
{
CLR_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl, eid);
tracer_config.entity_in_use_count--;
}
return TRACER_CMDRESP_SUCCESS;
}
return TRACER_CMDRESP_S_UNCHANGED;
}
// Dependency: bTracerInitialized is TRUE.
static tracer_cmdresp_t _entity_op_set_all(tracer_entity_state_t state)
{
tracer_cmdresp_t ret_val;
int ii;
ret_val = TRACER_CMDRESP_S_UNCHANGED;
for (ii = 0; ii < TRACER_NUM_USER_ENTITIES; ii++)
{
if (TRACER_ENTITY_SWEVT != tracer_entity_vals[ii])
{
if (((0 == state) ? TRACER_ENTITY_OFF : TRACER_ENTITY_ON) !=
GET_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl,
tracer_entity_vals[ii]))
{
if (TRACER_ENTITY_ON == state)
{
SET_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl,
tracer_entity_vals[ii]);
tracer_config.entity_in_use_count++;
}
else
{
CLR_ARRAY_BIT(entity_op_ctrl_t, tracer_config.entity_op_ctrl,
tracer_entity_vals[ii]);
tracer_config.entity_in_use_count--;
}
ret_val = TRACER_CMDRESP_SUCCESS;
}
}
}
return ret_val;
}
/*
* Naive Wheel factoring based on algorithm Ek from Sorenson 1991
*
* Uses 1 bit per odd number.
*
* Note we're including initialization code that marks all 3,5 multiples.
* If the caller didn't look at these, this could be skipped.
*
* Time for Pi(10^10) = 46.4s
*/
static WTYPE* sieve_eratek(WTYPE end)
{
WTYPE* mem;
size_t p, f, x;
size_t last = (end+1)/2;
static const WTYPE wheel[] = {1, 7, 11, 13, 17, 19, 23, 29};
static const WTYPE W[] = {0,6,0,0,0,0,0,4,0,0,0,2,0,4,0,0,0,2,0,4,0,0,0,6,0,0,0,0,0,2,0};
mem = (WTYPE*) calloc( NWORDS(last), sizeof(WTYPE) );
assert(mem != 0);
/* Mark all multiples of 3 and 5 as composite. */
//for (p = 3*3; p <= end; p += 2*3) SET_ARRAY_BIT(mem,p/2);
//for (p = 5*5; p <= end; p += 2*5) SET_ARRAY_BIT(mem,p/2);
p = 9;
while (p <= end) {
SET_ARRAY_BIT(mem,p/2); p += 6; if (p > end) break; // mark 9, p = 15
SET_ARRAY_BIT(mem,p/2); p += 6; if (p > end) break; // mark 15, p = 21
SET_ARRAY_BIT(mem,p/2); p += 4; if (p > end) break; // mark 21, p = 25
SET_ARRAY_BIT(mem,p/2); p += 2; if (p > end) break; // mark 25, p = 27
SET_ARRAY_BIT(mem,p/2); p += 6; if (p > end) break; // mark 27, p = 33
SET_ARRAY_BIT(mem,p/2); p += 2; if (p > end) break; // mark 33, p = 35
SET_ARRAY_BIT(mem,p/2); p += 4; // mark 35, p = 39
}
p = 7;
while ((p*p) <= end) {
{
size_t fidx = p%30;
f = p;
/* Here's the problem -- for each prime, we're walking the array from
* start to finish 8 times. The operation count is the same as the
* faster wheel-30 sieves, but this is just horrible for the cache. */
while (f < p+30) {
for (x = p*f; x <= end; x += p*30)
SET_ARRAY_BIT(mem,x/2);
size_t move = W[fidx];
f += move; fidx += move;
if (fidx > 30) fidx -= 30;
}
}
//p = next_prime(p);
do { p += 2; } while (IS_SET_ARRAY_BIT(mem,p/2));
}
SET_ARRAY_BIT(mem, 1/2); /* 1 is composite */
CLR_ARRAY_BIT(mem, 3/2); /* 3 is prime */
CLR_ARRAY_BIT(mem, 5/2); /* 5 is prime */
return mem;
}
/*
* Better Sieve of Atkin.
*
* Uses 1 bit per odd number.
*
* Just some simple optimizations to make it a little better. Still not good.
*
* Time for Pi(10^10) = 97.2s
*/
static WTYPE* sieve_atkin_2(WTYPE end)
{
WTYPE* mem;
size_t x, y, n, sqlimit;
size_t last = (end+1+1)/2;
long loopend, y_limit, dn;
end++;
mem = (WTYPE*) malloc( NWORDS(last) * sizeof(WTYPE) );
assert(mem != 0);
/* mark everything as a composite */
memset(mem, 0xFF, NBYTES(last));
sqlimit = sqrtf(end);
for (x = 1; x <= sqlimit; x++) {
{
size_t xx4 = 4*x*x;
y = 1;
for (n = xx4+1; n <= end; n = xx4+y*y) {
size_t nmod12 = n%12;
if ( (nmod12 == 1) || (nmod12 == 5) )
XOR_ARRAY_BIT(mem,n/2);
y++;
}
}
{
size_t xx3 = 3*x*x;
y = 1;
for (n = xx3+1; n <= end; n = xx3+y*y) {
size_t nmod12 = n%12;
if (nmod12 == 7)
XOR_ARRAY_BIT(mem,n/2);
y++;
}
y = x-1;
while ( y*y >= xx3 )
y--;
for (n = xx3-y*y; y >= 1 && n <= end; n = xx3-y*y) {
size_t nmod12 = n%12;
if (nmod12 == 11)
XOR_ARRAY_BIT(mem,n/2);
y--;
}
}
}
/* Mark all squares of primes as composite */
for (n = 5; n <= sqlimit; n += 2)
if (!IS_SET_ARRAY_BIT(mem,n/2))
for (y = n*n; y <= end; y += 2*n*n)
SET_ARRAY_BIT(mem,y/2);
CLR_ARRAY_BIT(mem, 3/2); /* 3 is prime */
return mem;
}
/*
* Straightforward Sieve of Eratosthenes.
*
* Uses 1 bit per odd number.
*
* Time for Pi(10^10) = 54.6s
*/
static WTYPE* sieve_erat(WTYPE end)
{
WTYPE* mem;
size_t n, s;
size_t last = (end+1)/2;
mem = (WTYPE*) calloc( NWORDS(last), sizeof(WTYPE) );
assert(mem != 0);
// Tight:
// for (n = 3; (n*n) <= end; n = next_prime(n))
// for (s = n*n; s <= end; s += 2*n)
// SET_ARRAY_BIT(mem,s/2);
n = 3;
while ( (n*n) <= end) {
for (s = n*n; s <= end; s += 2*n)
SET_ARRAY_BIT(mem,s/2);
// Could do: n = next_prime(n)
do { n += 2; } while (IS_SET_ARRAY_BIT(mem,n/2));
}
SET_ARRAY_BIT(mem, 1/2); /* 1 is composite */
return mem;
}
/*
* Naive Sieve of Atkin.
*
* Uses 1 bit per odd number.
*
* This is really slow. Just keeping it here as a reference.
*
* Time for Pi(10^10) = 123.5s
*/
static WTYPE* sieve_atkin_naive(WTYPE end)
{
WTYPE* mem;
size_t x, y, n, sqlimit;
size_t last = (end+1+1)/2;
long loopend, y_limit, dn;
end++;
mem = (WTYPE*) malloc( NWORDS(last) * sizeof(WTYPE) );
assert(mem != 0);
/* mark everything as a composite */
memset(mem, 0xFF, NBYTES(last));
sqlimit = sqrt(end);
for (x = 1; x <= sqlimit; x++) {
for (y = 1; y <= sqlimit; y++) {
n = 4*x*x + y*y;
if ( (n <= end) && (n % 12 == 1 || n % 12 == 5) )
XOR_ARRAY_BIT(mem,n/2);
n = 3*x*x + y*y;
if ( (n <= end) && (n % 12 == 7) )
XOR_ARRAY_BIT(mem,n/2);
n = 3*x*x - y*y;
if ( (n <= end) && (x > y) && (n % 12 == 11) )
XOR_ARRAY_BIT(mem,n/2);
}
}
/* Mark all squares of primes as composite */
for (n = 5; n <= sqlimit; n += 2)
if (!IS_SET_ARRAY_BIT(mem,n/2))
for (y = n*n; y <= end; y += 2*n*n)
SET_ARRAY_BIT(mem,y/2);
CLR_ARRAY_BIT(mem, 3/2); /* 3 is prime */
return mem;
}