Merge branch 'era' of github.com:jthornber/thin-provisioning-tools into era

persistent-data/data-structures/array_block.h

@@ -19,7 +19,7 @@
 #ifndef ARRAY_BLOCK_H
 #define ARRAY_BLOCK_H
 
-#include "persistent-data/endian_utils.h"
+#include "base/endian_utils.h"
 
 //----------------------------------------------------------------
 

persistent-data/data-structures/bitset.cc

@@ -46,6 +46,10 @@ namespace persistent_data {
 				return array_.get_root();
 			}
 
+			unsigned get_nr_bits() const {
+				return nr_bits_;
+			}
+
 			void grow(unsigned new_nr_bits, bool default_value) {
 				pad_last_block(default_value);
 				resize_array(new_nr_bits, default_value);
@@ -184,7 +188,7 @@ namespace persistent_data {
 				if (n >= nr_bits_) {
 					std::ostringstream str;
 					str << "bitset index out of bounds ("
-					    << n << " >= " << nr_bits_ << endl;
+					    << n << " >= " << nr_bits_ << ")";
 					throw runtime_error(str.str());
 				}
 			}
@@ -214,6 +218,12 @@ persistent_data::bitset::get_root() const
 	return impl_->get_root();
 }
 
+unsigned
+bitset::get_nr_bits() const
+{
+	return impl_->get_nr_bits();
+}
+
 void
 persistent_data::bitset::grow(unsigned new_nr_bits, bool default_value)
 {

persistent-data/data-structures/bitset.h

@@ -16,8 +16,8 @@
 // with thin-provisioning-tools.  If not, see
 // <http://www.gnu.org/licenses/>.
 
-#ifndef BITSET_H
-#define BITSET_H
+#ifndef PERSISTENT_DATA_DATA_STRUCTURES_BITSET_H
+#define PERSISTENT_DATA_DATA_STRUCTURES_BITSET_H
 
 #include "persistent-data/run.h"
 
@@ -54,6 +54,7 @@ namespace persistent_data {
 		bitset(tm_ptr tm);
 		bitset(tm_ptr tm, block_address root, unsigned nr_bits);
 		block_address get_root() const;
+		unsigned get_nr_bits() const;
 		void grow(unsigned new_nr_bits, bool default_value);
 		void destroy();
 

persistent-data/data-structures/bloom_filter.cc

@@ -0,0 +1,146 @@
+#include "persistent-data/data-structures/bloom_filter.h"
+
+#include <stdexcept>
+
+using namespace persistent_data;
+
+//----------------------------------------------------------------
+
+namespace {
+	static const uint64_t m1 = 0x9e37fffffffc0001UL;
+	static const unsigned bits = 18;
+
+	static uint32_t hash1(block_address const &b) {
+		return (b * m1) >> bits;
+	}
+
+	static uint32_t hash2(block_address const &b) {
+		uint32_t n = b;
+
+		n = n ^ (n >> 16);
+		n = n * 0x85ebca6bu;
+		n = n ^ (n >> 13);
+		n = n * 0xc2b2ae35u;
+		n = n ^ (n >> 16);
+
+		return n;
+	}
+
+	void check_power_of_two(unsigned nr_bits) {
+		if (nr_bits & (nr_bits - 1))
+			throw std::runtime_error("bloom filter needs a power of two nr_bits");
+	}
+}
+
+//----------------------------------------------------------------
+
+bloom_filter::bloom_filter(tm_ptr tm,
+			   unsigned nr_bits, unsigned nr_probes)
+	: tm_(tm),
+	  bits_(tm),
+	  nr_probes_(nr_probes),
+	  mask_(nr_bits - 1)
+{
+	check_power_of_two(nr_bits);
+	bits_.grow(nr_bits, false);
+}
+
+bloom_filter::bloom_filter(tm_ptr tm, block_address root,
+			   unsigned nr_bits, unsigned nr_probes)
+	: tm_(tm),
+	  bits_(tm, root, nr_bits),
+	  nr_probes_(nr_probes),
+	  mask_(nr_bits - 1)
+{
+	check_power_of_two(nr_bits);
+}
+
+block_address
+bloom_filter::get_root() const
+{
+	return bits_.get_root();
+}
+
+bool
+bloom_filter::test(uint64_t b)
+{
+	vector<unsigned> probes(nr_probes_);
+	fill_probes(b, probes);
+
+	for (unsigned p = 0; p < nr_probes_; p++)
+		if (!bits_.get(probes[p]))
+			return false;
+
+	return true;
+}
+
+void
+bloom_filter::set(uint64_t b)
+{
+	vector<unsigned> probes(nr_probes_);
+	fill_probes(b, probes);
+
+	for (unsigned p = 0; p < nr_probes_; p++)
+		bits_.set(probes[p], true);
+}
+
+void
+bloom_filter::flush()
+{
+	bits_.flush();
+}
+
+void
+bloom_filter::fill_probes(block_address b, vector<unsigned> &probes) const
+{
+	uint32_t h1 = hash1(b) & mask_;
+	uint32_t h2 = hash2(b) & mask_;
+
+	probes[0] = h1;
+	for (unsigned p = 1; p < nr_probes_; p++) {
+		h1 = (h1 + h2) & mask_;
+		h2 = (h2 + p) & mask_;
+		probes[p] = h1;
+	}
+}
+
+void
+bloom_filter::print_debug(ostream &out)
+{
+	print_residency(out);
+
+	map<unsigned, unsigned> runs;
+
+	for (unsigned i = 0; i < bits_.get_nr_bits();) {
+		bool v = bits_.get(i);
+		unsigned run_length = 1;
+
+		while (++i < bits_.get_nr_bits() && bits_.get(i) == v)
+			run_length++;
+
+		map<unsigned, unsigned>::iterator it = runs.find(run_length);
+		if (it != runs.end())
+			it->second++;
+		else
+			runs.insert(make_pair(run_length, 1));
+	}
+
+	{
+		map<unsigned, unsigned>::const_iterator it;
+		for (it = runs.begin(); it != runs.end(); ++it)
+			out << it->first << ": " << it->second << endl;
+	}
+}
+
+void
+bloom_filter::print_residency(ostream &out)
+{
+	unsigned count = 0;
+	for (unsigned i = 0; i < bits_.get_nr_bits(); i++)
+		if (bits_.get(i))
+			count++;
+
+	out << "residency: " << count << "/" << bits_.get_nr_bits() << endl;
+}
+
+//----------------------------------------------------------------

persistent-data/data-structures/bloom_filter.h

@@ -0,0 +1,47 @@
+#ifndef PERSISTENT_DATA_DATA_STRUCTURES_BLOOM_FILTER_H
+#define PERSISTENT_DATA_DATA_STRUCTURES_BLOOM_FILTER_H
+
+#include "persistent-data/transaction_manager.h"
+#include "persistent-data/data-structures/bitset.h"
+
+#include <boost/shared_ptr.hpp>
+
+//----------------------------------------------------------------
+
+namespace persistent_data {
+	class bloom_filter {
+	public:
+		typedef boost::shared_ptr<bloom_filter> ptr;
+		typedef typename persistent_data::transaction_manager::ptr tm_ptr;
+
+		// nr_bits must be a power of two
+		bloom_filter(tm_ptr tm,
+			     unsigned nr_bits, unsigned nr_probes);
+
+		bloom_filter(tm_ptr tm, block_address root,
+			     unsigned nr_bits_power, unsigned nr_probes);
+
+		block_address get_root() const;
+
+		bool test(uint64_t b); // not const due to caching effects in bitset
+		void set(uint64_t b);
+		void flush();
+
+		void print_debug(ostream &out);
+
+	private:
+		void print_residency(ostream &out);
+
+		void fill_probes(block_address b, vector<unsigned> &probes) const;
+
+		tm_ptr tm_;
+		unsigned nr_bits_;
+		persistent_data::bitset bits_;
+		unsigned nr_probes_;
+		uint64_t mask_;
+	};
+}
+
+//----------------------------------------------------------------
+
+#endif

persistent-data/data-structures/btree.h

@@ -19,7 +19,7 @@
 #ifndef BTREE_H
 #define BTREE_H
 
-#include "persistent-data/endian_utils.h"
+#include "base/endian_utils.h"
 #include "persistent-data/transaction_manager.h"
 #include "persistent-data/data-structures/ref_counter.h"