Fixed first_edge initialization bug.

This commit is contained in:
davi 2010-11-08 22:02:18 -02:00
parent cde9f72c9e
commit 676d34073c
8 changed files with 84 additions and 57 deletions

View File

@ -110,8 +110,10 @@ CMPH_METHOD_DECL(void_type, rehash)() {
<< slack_.size() << " keys in slack " << slack_.size() << " keys in slack "
<< values_.size() << " key in total" << std::endl; << values_.size() << " key in total" << std::endl;
slack_type().swap(slack_); slack_type().swap(slack_);
table_.Reset(make_iterator_first(values_.begin()), bool success = table_.Reset(
make_iterator_first(values_.end())); make_iterator_first(values_.begin()),
make_iterator_first(values_.end()));
assert(success);
std::vector<value_type> new_values(values_.size()); std::vector<value_type> new_values(values_.size());
for (const_iterator it = values_.begin(), end = values_.end(); for (const_iterator it = values_.begin(), end = values_.end();
it != end; ++it) { it != end; ++it) {

View File

@ -69,6 +69,7 @@ bool MPHTable::GenerateQueue(
// At this point queue head is the number of edges touching at least one // At this point queue head is the number of edges touching at least one
// vertex of degree 1. // vertex of degree 1.
cerr << "Queue head " << queue_head << " Queue tail " << queue_tail << endl; cerr << "Queue head " << queue_head << " Queue tail " << queue_tail << endl;
graph->DebugGraph();
while (queue_tail != queue_head) { while (queue_tail != queue_head) {
cmph_uint32 current_edge = queue[queue_tail++]; cmph_uint32 current_edge = queue[queue_tail++];
graph->RemoveEdge(current_edge); graph->RemoveEdge(current_edge);

View File

@ -3,6 +3,7 @@
// Minimal perfect hash abstraction implementing the BDZ algorithm // Minimal perfect hash abstraction implementing the BDZ algorithm
#include <cassert>
#include <cmath> #include <cmath>
#include <unordered_map> // for std::hash #include <unordered_map> // for std::hash
#include <vector> #include <vector>
@ -129,6 +130,11 @@ cmph_uint32 MPHTable::index(const Key& key) const {
h[0] = h[0] % r_; h[0] = h[0] % r_;
h[1] = h[1] % r_ + r_; h[1] = h[1] % r_ + r_;
h[2] = h[2] % r_ + (r_ << 1); h[2] = h[2] % r_ + (r_ << 1);
assert(g_.size());
cerr << "g_.size() " << g_.size() << " h0 >> 2 " << (h[0] >> 2) << endl;
assert((h[0] >> 2) <g_.size());
assert((h[1] >> 2) <g_.size());
assert((h[2] >> 2) <g_.size());
cmph_uint32 vertex = h[(get_2bit_value(g_, h[0]) + get_2bit_value(g_, h[1]) + get_2bit_value(g_, h[2])) % 3]; cmph_uint32 vertex = h[(get_2bit_value(g_, h[0]) + get_2bit_value(g_, h[1]) + get_2bit_value(g_, h[2])) % 3];
cerr << "Search found vertex " << vertex << endl; cerr << "Search found vertex " << vertex << endl;
return Rank(vertex); return Rank(vertex);

View File

@ -9,7 +9,7 @@ using std::endl;
using std::vector; using std::vector;
namespace { namespace {
static const cmph_uint8 kInvalidEdge = std::numeric_limits<cmph_uint8>::max(); static const cmph_uint32 kInvalidEdge = std::numeric_limits<cmph_uint32>::max();
} }
namespace cxxmph { namespace cxxmph {
@ -65,5 +65,17 @@ void TriGraph::RemoveEdge(cmph_uint32 current_edge) {
--vertex_degree_[vertex]; --vertex_degree_[vertex];
} }
} }
void TriGraph::DebugGraph() const {
int i;
for(i = 0; i < edges_.size(); i++){
cerr << i << " " << edges_[i][0] << " " << edges_[i][1] << " " << edges_[i][2]
<< " nexts " << next_edge_[i][0] << " " << next_edge_[i][1] << " " << next_edge_[i][2] << endl;
}
for(i = 0; i < first_edge_.size();i++){
cerr << "first for vertice " <<i << " " << first_edge_[i] << endl;
}
}
} // namespace cxxmph } // namespace cxxmph

View File

@ -30,6 +30,7 @@ class TriGraph {
void AddEdge(const Edge& edge); void AddEdge(const Edge& edge);
void RemoveEdge(cmph_uint32 edge_id); void RemoveEdge(cmph_uint32 edge_id);
void ExtractEdgesAndClear(std::vector<Edge>* edges); void ExtractEdgesAndClear(std::vector<Edge>* edges);
void DebugGraph() const;
const std::vector<Edge>& edges() const { return edges_; } const std::vector<Edge>& edges() const { return edges_; }
const std::vector<cmph_uint8>& vertex_degree() const { return vertex_degree_; } const std::vector<cmph_uint8>& vertex_degree() const { return vertex_degree_; }

View File

@ -2,7 +2,7 @@ bin_PROGRAMS = cmph
lib_LTLIBRARIES = libcmph.la lib_LTLIBRARIES = libcmph.la
include_HEADERS = cmph.h cmph_types.h cmph_time.h chd_ph.h include_HEADERS = cmph.h cmph_types.h cmph_time.h chd_ph.h
libcmph_la_SOURCES = hash.h hash.c \ libcmph_la_SOURCES = hash.h hash.c \
jenkins_hash.h jenkins_hash.c\ jenkins_hash.h jenkins_hash.c MurmurHash2.h\
hash_state.h debug.h \ hash_state.h debug.h \
vstack.h vstack.c vqueue.h vqueue.c\ vstack.h vstack.c vqueue.h vqueue.c\
graph.h graph.c bitbool.h \ graph.h graph.c bitbool.h \

View File

@ -178,6 +178,7 @@ static int bdz_generate_queue(cmph_uint32 nedges, cmph_uint32 nvertices, bdz_que
}; };
}; };
DEBUGP("Queue head %d Queue tail %d\n", queue_head, queue_tail); DEBUGP("Queue head %d Queue tail %d\n", queue_head, queue_tail);
bdz_dump_graph(graph3,graph3->nedges,graph3->nedges+graph3->nedges/4);
while(queue_tail!=queue_head){ while(queue_tail!=queue_head){
curr_edge=queue[queue_tail++]; curr_edge=queue[queue_tail++];
bdz_remove_edge(graph3,curr_edge); bdz_remove_edge(graph3,curr_edge);

View File

@ -9,6 +9,7 @@
#define DEBUG #define DEBUG
#include "debug.h" #include "debug.h"
#include "MurmurHash2.h"
#define hashsize(n) ((cmph_uint32)1<<(n)) #define hashsize(n) ((cmph_uint32)1<<(n))
#define hashmask(n) (hashsize(n)-1) #define hashmask(n) (hashsize(n)-1)
@ -99,63 +100,67 @@ void jenkins_state_destroy(jenkins_state_t *state)
inline void __jenkins_hash_vector(cmph_uint32 seed, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes) inline void __jenkins_hash_vector(cmph_uint32 seed, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes)
{ {
register cmph_uint32 len, length; int i;
for (i = 0; i < 3; ++i) {
/* Set up the internal state */ hashes[i] = MurmurHash2(k, keylen, seed + i);
length = keylen;
len = length;
hashes[0] = hashes[1] = 0x9e3779b9; /* the golden ratio; an arbitrary value */
hashes[2] = seed; /* the previous hash value - seed in our case */
/*---------------------------------------- handle most of the key */
while (len >= 12)
{
hashes[0] += ((cmph_uint32)k[0] +((cmph_uint32)k[1]<<8) +((cmph_uint32)k[2]<<16) +((cmph_uint32)k[3]<<24));
hashes[1] += ((cmph_uint32)k[4] +((cmph_uint32)k[5]<<8) +((cmph_uint32)k[6]<<16) +((cmph_uint32)k[7]<<24));
hashes[2] += ((cmph_uint32)k[8] +((cmph_uint32)k[9]<<8) +((cmph_uint32)k[10]<<16)+((cmph_uint32)k[11]<<24));
mix(hashes[0],hashes[1],hashes[2]);
k += 12; len -= 12;
} }
// register cmph_uint32 len, length;
/*------------------------------------- handle the last 11 bytes */ //
hashes[2] += length; // /* Set up the internal state */
switch(len) /* all the case statements fall through */ // length = keylen;
{ // len = length;
case 11: // hashes[0] = hashes[1] = 0x9e3779b9; /* the golden ratio; an arbitrary value */
hashes[2] +=((cmph_uint32)k[10]<<24); // hashes[2] = seed; /* the previous hash value - seed in our case */
case 10: //
hashes[2] +=((cmph_uint32)k[9]<<16); // /*---------------------------------------- handle most of the key */
case 9 : // while (len >= 12)
hashes[2] +=((cmph_uint32)k[8]<<8); // {
/* the first byte of hashes[2] is reserved for the length */ // hashes[0] += ((cmph_uint32)k[0] +((cmph_uint32)k[1]<<8) +((cmph_uint32)k[2]<<16) +((cmph_uint32)k[3]<<24));
case 8 : // hashes[1] += ((cmph_uint32)k[4] +((cmph_uint32)k[5]<<8) +((cmph_uint32)k[6]<<16) +((cmph_uint32)k[7]<<24));
hashes[1] +=((cmph_uint32)k[7]<<24); // hashes[2] += ((cmph_uint32)k[8] +((cmph_uint32)k[9]<<8) +((cmph_uint32)k[10]<<16)+((cmph_uint32)k[11]<<24));
case 7 : // mix(hashes[0],hashes[1],hashes[2]);
hashes[1] +=((cmph_uint32)k[6]<<16); // k += 12; len -= 12;
case 6 : // }
hashes[1] +=((cmph_uint32)k[5]<<8); //
case 5 : // /*------------------------------------- handle the last 11 bytes */
hashes[1] +=(cmph_uint8) k[4]; // hashes[2] += length;
case 4 : // switch(len) /* all the case statements fall through */
hashes[0] +=((cmph_uint32)k[3]<<24); // {
case 3 : // case 11:
hashes[0] +=((cmph_uint32)k[2]<<16); // hashes[2] +=((cmph_uint32)k[10]<<24);
case 2 : // case 10:
hashes[0] +=((cmph_uint32)k[1]<<8); // hashes[2] +=((cmph_uint32)k[9]<<16);
case 1 : // case 9 :
hashes[0] +=(cmph_uint8)k[0]; // hashes[2] +=((cmph_uint32)k[8]<<8);
/* case 0: nothing left to add */ // /* the first byte of hashes[2] is reserved for the length */
} // case 8 :
// hashes[1] +=((cmph_uint32)k[7]<<24);
mix(hashes[0],hashes[1],hashes[2]); // case 7 :
// hashes[1] +=((cmph_uint32)k[6]<<16);
// case 6 :
// hashes[1] +=((cmph_uint32)k[5]<<8);
// case 5 :
// hashes[1] +=(cmph_uint8) k[4];
// case 4 :
// hashes[0] +=((cmph_uint32)k[3]<<24);
// case 3 :
// hashes[0] +=((cmph_uint32)k[2]<<16);
// case 2 :
// hashes[0] +=((cmph_uint32)k[1]<<8);
// case 1 :
// hashes[0] +=(cmph_uint8)k[0];
// /* case 0: nothing left to add */
// }
//
// mix(hashes[0],hashes[1],hashes[2]);
} }
cmph_uint32 jenkins_hash(jenkins_state_t *state, const char *k, cmph_uint32 keylen) cmph_uint32 jenkins_hash(jenkins_state_t *state, const char *k, cmph_uint32 keylen)
{ {
cmph_uint32 hashes[3]; // cmph_uint32 hashes[3];
__jenkins_hash_vector(state->seed, k, keylen, hashes); // __jenkins_hash_vector(state->seed, k, keylen, hashes);
return hashes[2]; // return hashes[2];
/* cmph_uint32 a, b, c; cmph_uint32 a, b, c;
cmph_uint32 len, length; cmph_uint32 len, length;
// Set up the internal state // Set up the internal state
@ -209,7 +214,6 @@ cmph_uint32 jenkins_hash(jenkins_state_t *state, const char *k, cmph_uint32 keyl
/// report the result /// report the result
return c; return c;
*/
} }
void jenkins_hash_vector_(jenkins_state_t *state, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes) void jenkins_hash_vector_(jenkins_state_t *state, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes)