| ... | ... | @@ -37,78 +37,83 @@ pub const XxHash3 = xxhash.XxHash3; |
| 37 | 37 | pub const XxHash64 = xxhash.XxHash64; |
| 38 | 38 | pub const XxHash32 = xxhash.XxHash32; |
| 39 | 39 | |
| 40 | | /// Deprecated in favor of `int`. |
| 41 | | pub fn uint32(input: u32) u32 { |
| 42 | | return int(input); |
| 43 | | } |
| 44 | | |
| 45 | | /// Applies a bit-mangling transformation to an unsigned integer type `T`. |
| 46 | | /// Optimized per type: for `u16` and `u32`, Skeeto's xorshift-multiply; for `u64`, Maiga's mx3. |
| 47 | | /// Falls back on an avalanche pattern for other integer types, ensuring high entropy. |
| 40 | /// Easy & fast hash function for integer types |
| 48 | 41 | pub fn int(input: anytype) @TypeOf(input) { |
| 42 | // This function is only intended for integer types |
| 49 | 43 | const info = @typeInfo(@TypeOf(input)).int; |
| 50 | | if (info.signedness == .signed) { |
| 51 | | const Unsigned = @Type(.{ .int = .{ .signedness = .unsigned, .bits = info.bits } }); |
| 52 | | const casted: Unsigned = @bitCast(input); |
| 53 | | return @bitCast(int(casted)); |
| 54 | | } else if (info.bits < 4) { |
| 55 | | return @truncate(int(@as(u4, input))); |
| 56 | | } |
| 57 | | var x = input; |
| 58 | | switch (info.bits) { |
| 59 | | 16 => { |
| 60 | | // https://github.com/skeeto/hash-prospector |
| 61 | | // 3-round xorshift-multiply (-Xn3) |
| 62 | | // bias = 0.0045976709018820602 |
| 63 | | x = (x ^ (x >> 7)) *% 0x2993; |
| 64 | | x = (x ^ (x >> 5)) *% 0xe877; |
| 65 | | x = (x ^ (x >> 9)) *% 0x0235; |
| 66 | | x = x ^ (x >> 10); |
| 67 | | }, |
| 68 | | 32 => { |
| 69 | | // https://github.com/skeeto/hash-prospector |
| 70 | | x = (x ^ (x >> 17)) *% 0xed5ad4bb; |
| 71 | | x = (x ^ (x >> 11)) *% 0xac4c1b51; |
| 72 | | x = (x ^ (x >> 15)) *% 0x31848bab; |
| 73 | | x = x ^ (x >> 14); |
| 74 | | }, |
| 75 | | 64 => { |
| 76 | | // https://github.com/jonmaiga/mx3 |
| 77 | | // https://github.com/jonmaiga/mx3/blob/48924ee743d724aea2cafd2b4249ef8df57fa8b9/mx3.h#L17 |
| 78 | | const c = 0xbea225f9eb34556d; |
| 79 | | x = (x ^ (x >> 32)) *% c; |
| 80 | | x = (x ^ (x >> 29)) *% c; |
| 81 | | x = (x ^ (x >> 32)) *% c; |
| 82 | | x = x ^ (x >> 29); |
| 83 | | }, |
| 84 | | else => { |
| 85 | | // This construction provides robust avalanche properties, but it is not optimal for any given size. |
| 86 | | const hsize = info.bits >> 1; |
| 87 | | const c = comptime blk: { |
| 88 | | const max = (1 << info.bits) - 1; |
| 89 | | var mul = 1; |
| 90 | | while (mul * 3 < max) mul *= 3; |
| 91 | | break :blk ((mul ^ (mul >> hsize)) | 1); |
| 92 | | }; |
| 93 | | inline for (0..2) |_| { |
| 94 | | x = (x ^ (x >> hsize + 1)) *% c; |
| 95 | | x = (x ^ (x >> hsize - 1)) *% c; |
| 44 | const bits = info.bits; |
| 45 | // Convert input to unsigned integer (easier to deal with) |
| 46 | const Uint = @Type(.{ .int = .{ .bits = bits, .signedness = .unsigned } }); |
| 47 | const u_input: Uint = @bitCast(input); |
| 48 | if (bits > 256) @compileError("bit widths > 256 are unsupported, use std.hash.autoHash functionality."); |
| 49 | // For bit widths that don't have a dedicated function, use a heuristic |
| 50 | // construction with a multiplier suited to diffusion - |
| 51 | // a mod 2^bits where a^2 - 46 * a + 1 = 0 mod 2^(bits + 4), |
| 52 | // on Mathematica: bits = 256; BaseForm[Solve[1 - 46 a + a^2 == 0, a, Modulus -> 2^(bits + 4)][[-1]][[1]][[2]], 16] |
| 53 | const mult: Uint = @truncate(0xfac2e27ed2036860a062b5f264d80a512b00aa459b448bf1eca24d41c96f59e5b); |
| 54 | // The bit width of the input integer determines how to hash it |
| 55 | const output = switch (bits) { |
| 56 | 0...2 => u_input *% mult, |
| 57 | 16 => uint16(u_input), |
| 58 | 32 => uint32(u_input), |
| 59 | 64 => uint64(u_input), |
| 60 | else => blk: { |
| 61 | var x: Uint = u_input; |
| 62 | inline for (0..4) |_| { |
| 63 | x ^= x >> (bits / 2); |
| 64 | x *%= mult; |
| 96 | 65 | } |
| 97 | | x ^= (x >> hsize); |
| 66 | break :blk x; |
| 98 | 67 | }, |
| 99 | | } |
| 68 | }; |
| 69 | return @bitCast(output); |
| 70 | } |
| 71 | |
| 72 | /// Source: https://github.com/skeeto/hash-prospector |
| 73 | fn uint16(input: u16) u16 { |
| 74 | var x: u16 = input; |
| 75 | x = (x ^ (x >> 7)) *% 0x2993; |
| 76 | x = (x ^ (x >> 5)) *% 0xe877; |
| 77 | x = (x ^ (x >> 9)) *% 0x0235; |
| 78 | x = x ^ (x >> 10); |
| 79 | return x; |
| 80 | } |
| 81 | |
| 82 | /// DEPRECATED: use std.hash.int() |
| 83 | /// Source: https://github.com/skeeto/hash-prospector |
| 84 | pub fn uint32(input: u32) u32 { |
| 85 | var x: u32 = input; |
| 86 | x = (x ^ (x >> 17)) *% 0xed5ad4bb; |
| 87 | x = (x ^ (x >> 11)) *% 0xac4c1b51; |
| 88 | x = (x ^ (x >> 15)) *% 0x31848bab; |
| 89 | x = x ^ (x >> 14); |
| 90 | return x; |
| 91 | } |
| 92 | |
| 93 | /// Source: https://github.com/jonmaiga/mx3 |
| 94 | fn uint64(input: u64) u64 { |
| 95 | var x: u64 = input; |
| 96 | const c = 0xbea225f9eb34556d; |
| 97 | x = (x ^ (x >> 32)) *% c; |
| 98 | x = (x ^ (x >> 29)) *% c; |
| 99 | x = (x ^ (x >> 32)) *% c; |
| 100 | x = x ^ (x >> 29); |
| 100 | 101 | return x; |
| 101 | 102 | } |
| 102 | 103 | |
| 103 | 104 | test int { |
| 104 | 105 | const expectEqual = @import("std").testing.expectEqual; |
| 105 | | try expectEqual(0xC, int(@as(u4, 1))); |
| 106 | | try expectEqual(0x4F, int(@as(u8, 1))); |
| 107 | | try expectEqual(0x4F, int(@as(i8, 1))); |
| 106 | try expectEqual(0x1, int(@as(u1, 1))); |
| 107 | try expectEqual(0x3, int(@as(u2, 1))); |
| 108 | try expectEqual(0x4, int(@as(u3, 1))); |
| 109 | try expectEqual(0xD6, int(@as(u8, 1))); |
| 108 | 110 | try expectEqual(0x2880, int(@as(u16, 1))); |
| 111 | try expectEqual(0x2880, int(@as(i16, 1))); |
| 112 | try expectEqual(0x838380, int(@as(u24, 1))); |
| 109 | 113 | try expectEqual(0x42741D6, int(@as(u32, 1))); |
| 114 | try expectEqual(0x42741D6, int(@as(i32, 1))); |
| 110 | 115 | try expectEqual(0x71894DE00D9981F, int(@as(u64, 1))); |
| 111 | | try expectEqual(0x50BC2BB18910C3DE0BAA2CE0D0C5B83E, int(@as(u128, 1))); |
| 116 | try expectEqual(0x71894DE00D9981F, int(@as(i64, 1))); |
| 112 | 117 | } |
| 113 | 118 | |
| 114 | 119 | test { |