[PATCH 16/34] Add a generic and minimal Bloom filter implementation

From: SZEDER Gábor <szeder.dev@gmail.com>
Date: May 29, 2020, 08:50 UTC
Message-ID: <20200529085038.26008-17-szeder.dev@gmail.com>
In-Reply-To: <20200529085038.26008-1-szeder.dev@gmail.com>

We'll use Bloom filters to store the paths modified between each commit and one of its parents. This means many small Bloom filters, which puts some constraints on the hashing scheme, as discussed in the previous commit. However, Bloom filters might turn out to be useful in other parts of Git as well, where they might be free from those constraints. So add a Bloom filter implementation that is fairly generic, and copes only with allocation, initialization and release of the filter and with manipulating the bits in its bit array, but leaves all the hashing to the user of the filter.

Unfortunately, it's still somewhat specific, though, in how it maps
the hashes to specific bits in memory, i.e. how it maps the hashes to
array indices and how the bit array is indexed.  We'll see how that
will pan out...
---
 Makefile       |  1 +
 bloom-filter.c | 91 ++++++++++++++++++++++++++++++++++++++++++++++++++
 bloom-filter.h | 47 ++++++++++++++++++++++++++
 3 files changed, 139 insertions(+)
 create mode 100644 bloom-filter.c
 create mode 100644 bloom-filter.h

diff --git a/Makefile b/Makefile
index 09f98b777c..a9db1e8d9b 100644
--- a/Makefile
+++ b/Makefile
@@ -839,6 +839,7 @@ LIB_OBJS += base85.o
 LIB_OBJS += bisect.o
 LIB_OBJS += blame.o
 LIB_OBJS += blob.o
+LIB_OBJS += bloom-filter.o
 LIB_OBJS += branch.o
 LIB_OBJS += bulk-checkin.o
 LIB_OBJS += bundle.o
diff --git a/bloom-filter.c b/bloom-filter.c
new file mode 100644
index 0000000000..919da26462
--- /dev/null
+++ b/bloom-filter.c
@@ -0,0 +1,91 @@
+#include "bloom-filter.h"
+
+void bloom_filter_init(struct bloom_filter *bf, unsigned int nr_hashes,
+		       uint32_t nr_elements)
+{
+	/* n * k / ln(2) ≈ n * k / 0.69315 ≈ n * k * 10 / 7 */
+	uint32_t nr_bits = st_mult(st_mult(nr_elements, nr_hashes), 10) / 7;
+	uint32_t nr_bytes = st_add(nr_bits, 7) / 8;
+	/*
+	 * But we round up to fully utilize all bytes, thus lowering the
+	 * probability of false positives a bit.
+	 */
+	bf->nr_bits = nr_bytes * 8;
+	bf->bits = xcalloc(nr_bytes, sizeof(*(bf->bits)));
+}
+
+void bloom_filter_init_with_size(struct bloom_filter *bf, uint32_t nr_bits)
+{
+	uint32_t nr_bytes = st_add(nr_bits, 7) / 8;
+	bf->nr_bits = nr_bits;
+	bf->bits = xcalloc(nr_bytes, sizeof(*(bf->bits)));
+}
+
+void bloom_filter_free(struct bloom_filter *bf)
+{
+	FREE_AND_NULL(bf->bits);
+	bf->nr_bits = 0;
+}
+
+uint32_t bloom_filter_bytes(struct bloom_filter *bf)
+{
+	return (bf->nr_bits + 7) / 8;
+}
+
+void bloom_filter_clear_all_bits(struct bloom_filter *bf)
+{
+	memset(bf->bits, 0, bloom_filter_bytes(bf));
+}
+
+void bloom_filter_set_all_bits(struct bloom_filter *bf)
+{
+	memset(bf->bits, 0xff, bloom_filter_bytes(bf));
+}
+
+static inline uint32_t bit_offset(uint32_t nr_bits, uint32_t hash)
+{
+	return hash % nr_bits;
+}
+
+static inline uint32_t byte_offset(uint32_t nr_bits, uint32_t bit_offset)
+{
+	return (nr_bits - 1) / 8 - bit_offset / 8;
+}
+
+static inline uint8_t byte_mask(uint32_t bit_offset)
+{
+	return 1 << (bit_offset % 8);
+}
+
+static inline void bloom_filter_set_one_bit(struct bloom_filter *bf,
+					    uint32_t hash)
+{
+	uint32_t offset = bit_offset(bf->nr_bits, hash);
+	bf->bits[byte_offset(bf->nr_bits, offset)] |= byte_mask(offset);
+}
+
+void bloom_filter_set_bits(struct bloom_filter *bf, const uint32_t *hashes,
+			   unsigned int nr_hashes)
+{
+	unsigned int i;
+	for (i = 0; i < nr_hashes; i++)
+		bloom_filter_set_one_bit(bf, hashes[i]);
+}
+
+static inline int bloom_filter_check_one_bit(struct bloom_filter *bf,
+					     uint32_t hash)
+{
+	uint32_t offset = bit_offset(bf->nr_bits, hash);
+	return bf->bits[byte_offset(bf->nr_bits, offset)] & byte_mask(offset);
+}
+
+enum bloom_result bloom_filter_check_bits(struct bloom_filter *bf,
+					  const uint32_t *hashes,
+					  unsigned int nr_hashes)
+{
+	unsigned int i;
+	for (i = 0; i < nr_hashes; i++)
+		if (!bloom_filter_check_one_bit(bf, hashes[i]))
+			return BLOOM_DEFINITELY_NOT;
+	return BLOOM_POSSIBLY_YES;
+}
diff --git a/bloom-filter.h b/bloom-filter.h
new file mode 100644
index 0000000000..2fcd4a2c19
--- /dev/null
+++ b/bloom-filter.h
@@ -0,0 +1,47 @@
+#ifndef BLOOM_FILTER_H
+#define BLOOM_FILTER_H
+
+#include "git-compat-util.h"
+
+enum bloom_result {
+	/*
+	 * BLOOM_CANT_TELL is a value that a caller can use to report that
+	 * a Bloom filter is not available; bloom_filter_check_bits() will
+	 * never return it.
+	 */
+	BLOOM_CANT_TELL = -1,
+	BLOOM_DEFINITELY_NOT = 0,
+	BLOOM_POSSIBLY_YES = 1
+};
+
+struct bloom_filter {
+	uint32_t nr_bits;
+	uint8_t *bits;
+};
+
+/*
+ * Initialize a Bloom filter with the number of bits that is (close to)
+ * optimal to hold the given number of elements using the given number
+ * of hashes per element.
+ */
+void bloom_filter_init(struct bloom_filter *bf, uint32_t nr_hashes,
+		       uint32_t nr_elements);
+
+/* Initialize a Bloom filter with the given number of bits */
+void bloom_filter_init_with_size(struct bloom_filter *bf, uint32_t nr_bits);
+
+void bloom_filter_free(struct bloom_filter *bf);
+
+/* Return the size of the Bloom filter's bit array in bytes */
+uint32_t bloom_filter_bytes(struct bloom_filter *bf);
+
+void bloom_filter_clear_all_bits(struct bloom_filter *bf);
+void bloom_filter_set_all_bits(struct bloom_filter *bf);
+
+void bloom_filter_set_bits(struct bloom_filter *bf, const uint32_t *hashes,
+			   unsigned int nr_hashes);
+enum bloom_result bloom_filter_check_bits(struct bloom_filter *bf,
+					  const uint32_t *hashes,
+					  unsigned int nr_hashes);
+
+#endif

-- 
2.27.0.rc1.431.g5c813f95dc