| ... | @@ -0,0 +1,253 @@ |
| 1 | const builtin = @import("builtin"); |
| 2 | const std = @import("std.zig"); |
| 3 | const math = std.math; |
| 4 | const debug = std.debug; |
| 5 | const assert = std.debug.assert; |
| 6 | const testing = std.testing; |
| 7 | |
| 8 | /// There is a trade off of how quickly to fill a bloom filter; |
| 9 | /// the number of items is: |
| 10 | /// n_items / K * ln(2) |
| 11 | /// the rate of false positives is: |
| 12 | /// (1-e^(-K*N/n_items))^K |
| 13 | /// where N is the number of items |
| 14 | pub fn BloomFilter( |
| 15 | /// Size of bloom filter in cells, must be a power of two. |
| 16 | comptime n_items: usize, |
| 17 | /// Number of cells to set per item |
| 18 | comptime K: usize, |
| 19 | /// Cell type, should be: |
| 20 | /// - `bool` for a standard bloom filter |
| 21 | /// - an unsigned integer type for a counting bloom filter |
| 22 | comptime Cell: type, |
| 23 | /// endianess of the Cell |
| 24 | comptime endian: builtin.Endian, |
| 25 | /// Hash function to use |
| 26 | comptime hash: fn (out: []u8, Ki: usize, in: []const u8) void, |
| 27 | ) type { |
| 28 | assert(n_items > 0); |
| 29 | assert(math.isPowerOfTwo(n_items)); |
| 30 | assert(K > 0); |
| 31 | const cellEmpty = if (Cell == bool) false else Cell(0); |
| 32 | const cellMax = if (Cell == bool) true else math.maxInt(Cell); |
| 33 | const n_bytes = (n_items * comptime std.meta.bitCount(Cell)) / 8; |
| 34 | assert(n_bytes > 0); |
| 35 | const Io = std.packed_int_array.PackedIntIo(Cell, endian); |
| 36 | |
| 37 | return struct { |
| 38 | const Self = @This(); |
| 39 | pub const items = n_items; |
| 40 | pub const Index = math.IntFittingRange(0, n_items - 1); |
| 41 | |
| 42 | data: [n_bytes]u8 = [_]u8{0} ** n_bytes, |
| 43 | |
| 44 | pub fn reset(self: *Self) void { |
| 45 | std.mem.set(u8, self.data[0..], 0); |
| 46 | } |
| 47 | |
| 48 | pub fn @"union"(x: Self, y: Self) Self { |
| 49 | var r = Self{ .data = undefined }; |
| 50 | inline for (x.data) |v, i| { |
| 51 | r.data[i] = v | y.data[i]; |
| 52 | } |
| 53 | return r; |
| 54 | } |
| 55 | |
| 56 | pub fn intersection(x: Self, y: Self) Self { |
| 57 | var r = Self{ .data = undefined }; |
| 58 | inline for (x.data) |v, i| { |
| 59 | r.data[i] = v & y.data[i]; |
| 60 | } |
| 61 | return r; |
| 62 | } |
| 63 | |
| 64 | pub fn getCell(self: Self, cell: Index) Cell { |
| 65 | return Io.get(self.data, cell, 0); |
| 66 | } |
| 67 | |
| 68 | pub fn incrementCell(self: *Self, cell: Index) void { |
| 69 | if (Cell == bool or Cell == u1) { |
| 70 | // skip the 'get' operation |
| 71 | Io.set(&self.data, cell, 0, cellMax); |
| 72 | } else { |
| 73 | const old = Io.get(self.data, cell, 0); |
| 74 | if (old != cellMax) { |
| 75 | Io.set(&self.data, cell, 0, old + 1); |
| 76 | } |
| 77 | } |
| 78 | } |
| 79 | |
| 80 | pub fn clearCell(self: *Self, cell: Index) void { |
| 81 | Io.set(&self.data, cell, 0, cellEmpty); |
| 82 | } |
| 83 | |
| 84 | pub fn add(self: *Self, item: []const u8) void { |
| 85 | comptime var i = 0; |
| 86 | inline while (i < K) : (i += 1) { |
| 87 | var K_th_bit: packed struct { x: Index } = undefined; |
| 88 | hash(std.mem.asBytes(&K_th_bit), i, item); |
| 89 | incrementCell(self, K_th_bit.x); |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | pub fn contains(self: Self, item: []const u8) bool { |
| 94 | comptime var i = 0; |
| 95 | inline while (i < K) : (i += 1) { |
| 96 | var K_th_bit: packed struct { x: Index } = undefined; |
| 97 | hash(std.mem.asBytes(&K_th_bit), i, item); |
| 98 | if (getCell(self, K_th_bit.x) == cellEmpty) |
| 99 | return false; |
| 100 | } |
| 101 | return true; |
| 102 | } |
| 103 | |
| 104 | pub fn resize(self: Self, comptime newsize: usize) BloomFilter(newsize, K, Cell, endian, hash) { |
| 105 | var r: BloomFilter(newsize, K, Cell, endian, hash) = undefined; |
| 106 | if (newsize < n_items) { |
| 107 | std.mem.copy(u8, r.data[0..], self.data[0..r.data.len]); |
| 108 | var copied: usize = r.data.len; |
| 109 | while (copied < self.data.len) : (copied += r.data.len) { |
| 110 | for (self.data[copied .. copied + r.data.len]) |s, i| { |
| 111 | r.data[i] |= s; |
| 112 | } |
| 113 | } |
| 114 | } else if (newsize == n_items) { |
| 115 | r = self; |
| 116 | } else if (newsize > n_items) { |
| 117 | var copied: usize = 0; |
| 118 | while (copied < r.data.len) : (copied += self.data.len) { |
| 119 | std.mem.copy(u8, r.data[copied .. copied + self.data.len], self.data); |
| 120 | } |
| 121 | } |
| 122 | return r; |
| 123 | } |
| 124 | |
| 125 | /// Returns number of non-zero cells |
| 126 | pub fn popCount(self: Self) Index { |
| 127 | var n: Index = 0; |
| 128 | if (Cell == bool or Cell == u1) { |
| 129 | for (self.data) |b, i| { |
| 130 | n += @popCount(u8, b); |
| 131 | } |
| 132 | } else { |
| 133 | var i: usize = 0; |
| 134 | while (i < n_items) : (i += 1) { |
| 135 | const cell = self.getCell(@intCast(Index, i)); |
| 136 | n += if (if (Cell == bool) cell else cell > 0) Index(1) else Index(0); |
| 137 | } |
| 138 | } |
| 139 | return n; |
| 140 | } |
| 141 | |
| 142 | pub fn estimateItems(self: Self) f64 { |
| 143 | const m = comptime @intToFloat(f64, n_items); |
| 144 | const k = comptime @intToFloat(f64, K); |
| 145 | const X = @intToFloat(f64, self.popCount()); |
| 146 | return (comptime (-m / k)) * math.log1p(X * comptime (-1 / m)); |
| 147 | } |
| 148 | }; |
| 149 | } |
| 150 | |
| 151 | fn hashFunc(out: []u8, Ki: usize, in: []const u8) void { |
| 152 | var st = std.crypto.gimli.Hash.init(); |
| 153 | st.update(std.mem.asBytes(&Ki)); |
| 154 | st.update(in); |
| 155 | st.final(out); |
| 156 | } |
| 157 | |
| 158 | test "std.BloomFilter" { |
| 159 | inline for ([_]type{ bool, u1, u2, u3, u4 }) |Cell| { |
| 160 | const emptyCell = if (Cell == bool) false else Cell(0); |
| 161 | const BF = BloomFilter(128 * 8, 8, Cell, builtin.endian, hashFunc); |
| 162 | var bf = BF{}; |
| 163 | var i: usize = undefined; |
| 164 | // confirm that it is initialised to the empty filter |
| 165 | i = 0; |
| 166 | while (i < BF.items) : (i += 1) { |
| 167 | testing.expectEqual(emptyCell, bf.getCell(@intCast(BF.Index, i))); |
| 168 | } |
| 169 | testing.expectEqual(BF.Index(0), bf.popCount()); |
| 170 | testing.expectEqual(f64(0), bf.estimateItems()); |
| 171 | // fill in a few items |
| 172 | bf.incrementCell(42); |
| 173 | bf.incrementCell(255); |
| 174 | bf.incrementCell(256); |
| 175 | bf.incrementCell(257); |
| 176 | // check that they were set |
| 177 | testing.expectEqual(true, bf.getCell(42) != emptyCell); |
| 178 | testing.expectEqual(true, bf.getCell(255) != emptyCell); |
| 179 | testing.expectEqual(true, bf.getCell(256) != emptyCell); |
| 180 | testing.expectEqual(true, bf.getCell(257) != emptyCell); |
| 181 | // clear just one of them; make sure the rest are still set |
| 182 | bf.clearCell(256); |
| 183 | testing.expectEqual(true, bf.getCell(42) != emptyCell); |
| 184 | testing.expectEqual(true, bf.getCell(255) != emptyCell); |
| 185 | testing.expectEqual(false, bf.getCell(256) != emptyCell); |
| 186 | testing.expectEqual(true, bf.getCell(257) != emptyCell); |
| 187 | // reset any of the ones we've set and confirm we're back to the empty filter |
| 188 | bf.clearCell(42); |
| 189 | bf.clearCell(255); |
| 190 | bf.clearCell(257); |
| 191 | i = 0; |
| 192 | while (i < BF.items) : (i += 1) { |
| 193 | testing.expectEqual(emptyCell, bf.getCell(@intCast(BF.Index, i))); |
| 194 | } |
| 195 | testing.expectEqual(BF.Index(0), bf.popCount()); |
| 196 | testing.expectEqual(f64(0), bf.estimateItems()); |
| 197 | |
| 198 | // Lets add a string |
| 199 | bf.add("foo"); |
| 200 | testing.expectEqual(true, bf.contains("foo")); |
| 201 | { |
| 202 | // try adding same string again. make sure popcount is the same |
| 203 | const old_popcount = bf.popCount(); |
| 204 | testing.expect(old_popcount > 0); |
| 205 | bf.add("foo"); |
| 206 | testing.expectEqual(true, bf.contains("foo")); |
| 207 | testing.expectEqual(old_popcount, bf.popCount()); |
| 208 | } |
| 209 | |
| 210 | // Get back to empty filter via .reset |
| 211 | bf.reset(); |
| 212 | // Double check that .reset worked |
| 213 | i = 0; |
| 214 | while (i < BF.items) : (i += 1) { |
| 215 | testing.expectEqual(emptyCell, bf.getCell(@intCast(BF.Index, i))); |
| 216 | } |
| 217 | testing.expectEqual(BF.Index(0), bf.popCount()); |
| 218 | testing.expectEqual(f64(0), bf.estimateItems()); |
| 219 | |
| 220 | comptime var teststrings = [_][]const u8{ |
| 221 | "foo", |
| 222 | "bar", |
| 223 | "a longer string", |
| 224 | "some more", |
| 225 | "the quick brown fox", |
| 226 | "unique string", |
| 227 | }; |
| 228 | inline for (teststrings) |str| { |
| 229 | bf.add(str); |
| 230 | } |
| 231 | inline for (teststrings) |str| { |
| 232 | testing.expectEqual(true, bf.contains(str)); |
| 233 | } |
| 234 | |
| 235 | { // estimate should be close for low packing |
| 236 | const est = bf.estimateItems(); |
| 237 | testing.expect(est > @intToFloat(f64, teststrings.len) - 1); |
| 238 | testing.expect(est < @intToFloat(f64, teststrings.len) + 1); |
| 239 | } |
| 240 | |
| 241 | const larger_bf = bf.resize(4096); |
| 242 | inline for (teststrings) |str| { |
| 243 | testing.expectEqual(true, larger_bf.contains(str)); |
| 244 | } |
| 245 | testing.expectEqual(u12(bf.popCount()) * (4096 / 1024), larger_bf.popCount()); |
| 246 | |
| 247 | const smaller_bf = bf.resize(64); |
| 248 | inline for (teststrings) |str| { |
| 249 | testing.expectEqual(true, smaller_bf.contains(str)); |
| 250 | } |
| 251 | testing.expect(bf.popCount() <= u10(smaller_bf.popCount()) * (1024 / 64)); |
| 252 | } |
| 253 | } |