Merge pull request #3106 from ziglang/hash-tooling-changes

Hash tooling changes
This commit is contained in:
Marc Tiehuis 2019-08-21 21:54:12 -10:00 committed by GitHub
commit 0e75fef1de
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 306 additions and 176 deletions

View File

@ -1,8 +1,10 @@
// zig run benchmark.zig --release-fast --override-std-dir ..
const builtin = @import("builtin");
const std = @import("std");
const std = @import("../std.zig");
const time = std.time;
const Timer = time.Timer;
const crypto = @import("../crypto.zig");
const crypto = std.crypto;
const KiB = 1024;
const MiB = 1024 * KiB;
@ -14,7 +16,7 @@ const Crypto = struct {
name: []const u8,
};
const hashes = []Crypto{
const hashes = [_]Crypto{
Crypto{ .ty = crypto.Md5, .name = "md5" },
Crypto{ .ty = crypto.Sha1, .name = "sha1" },
Crypto{ .ty = crypto.Sha256, .name = "sha256" },
@ -45,7 +47,7 @@ pub fn benchmarkHash(comptime Hash: var, comptime bytes: comptime_int) !u64 {
return throughput;
}
const macs = []Crypto{
const macs = [_]Crypto{
Crypto{ .ty = crypto.Poly1305, .name = "poly1305" },
Crypto{ .ty = crypto.HmacMd5, .name = "hmac-md5" },
Crypto{ .ty = crypto.HmacSha1, .name = "hmac-sha1" },
@ -75,7 +77,7 @@ pub fn benchmarkMac(comptime Mac: var, comptime bytes: comptime_int) !u64 {
return throughput;
}
const exchanges = []Crypto{Crypto{ .ty = crypto.X25519, .name = "x25519" }};
const exchanges = [_]Crypto{Crypto{ .ty = crypto.X25519, .name = "x25519" }};
pub fn benchmarkKeyExchange(comptime DhKeyExchange: var, comptime exchange_count: comptime_int) !u64 {
std.debug.assert(DhKeyExchange.minimum_key_length >= DhKeyExchange.secret_length);
@ -135,13 +137,16 @@ pub fn main() !void {
var buffer: [1024]u8 = undefined;
var fixed = std.heap.FixedBufferAllocator.init(buffer[0..]);
const args = try std.os.argsAlloc(&fixed.allocator);
const args = try std.process.argsAlloc(&fixed.allocator);
var filter: ?[]u8 = "";
var i: usize = 1;
while (i < args.len) : (i += 1) {
if (std.mem.eql(u8, args[i], "--seed")) {
if (std.mem.eql(u8, args[i], "--mode")) {
try stdout.print("{}\n", builtin.mode);
return;
} else if (std.mem.eql(u8, args[i], "--seed")) {
i += 1;
if (i == args.len) {
usage();

View File

@ -269,8 +269,8 @@ pub const Blake2b512 = Blake2b(512);
fn Blake2b(comptime out_len: usize) type {
return struct {
const Self = @This();
const block_length = 128;
const digest_length = out_len / 8;
pub const block_length = 128;
pub const digest_length = out_len / 8;
const iv = [8]u64{
0x6a09e667f3bcc908,

View File

@ -420,8 +420,8 @@ pub const Sha512 = Sha2_64(Sha512Params);
fn Sha2_64(comptime params: Sha2Params64) type {
return struct {
const Self = @This();
const block_length = 128;
const digest_length = params.out_len / 8;
pub const block_length = 128;
pub const digest_length = params.out_len / 8;
s: [8]u64,
// Streaming Cache

273
std/hash/benchmark.zig Normal file
View File

@ -0,0 +1,273 @@
// zig run benchmark.zig --release-fast --override-std-dir ..
const builtin = @import("builtin");
const std = @import("std");
const time = std.time;
const Timer = time.Timer;
const hash = std.hash;
const KiB = 1024;
const MiB = 1024 * KiB;
const GiB = 1024 * MiB;
var prng = std.rand.DefaultPrng.init(0);
const Hash = struct {
ty: type,
name: []const u8,
has_iterative_api: bool = true,
init_u8s: ?[]const u8 = null,
init_u64: ?u64 = null,
};
const siphash_key = "0123456789abcdef";
const hashes = [_]Hash{
Hash{
.ty = hash.Wyhash,
.name = "wyhash",
.init_u64 = 0,
},
Hash{
.ty = hash.SipHash64(1, 3),
.name = "siphash(1,3)",
.init_u8s = siphash_key,
},
Hash{
.ty = hash.SipHash64(2, 4),
.name = "siphash(2,4)",
.init_u8s = siphash_key,
},
Hash{
.ty = hash.Fnv1a_64,
.name = "fnv1a",
},
Hash{
.ty = hash.Adler32,
.name = "adler32",
},
Hash{
.ty = hash.crc.Crc32WithPoly(.IEEE),
.name = "crc32-slicing-by-8",
},
Hash{
.ty = hash.crc.Crc32SmallWithPoly(.IEEE),
.name = "crc32-half-byte-lookup",
},
Hash{
.ty = hash.CityHash32,
.name = "cityhash-32",
.has_iterative_api = false,
},
Hash{
.ty = hash.CityHash64,
.name = "cityhash-64",
.has_iterative_api = false,
},
Hash{
.ty = hash.Murmur2_32,
.name = "murmur2-32",
.has_iterative_api = false,
},
Hash{
.ty = hash.Murmur2_64,
.name = "murmur2-64",
.has_iterative_api = false,
},
Hash{
.ty = hash.Murmur3_32,
.name = "murmur3-32",
.has_iterative_api = false,
},
};
const Result = struct {
hash: u64,
throughput: u64,
};
const block_size: usize = 8192;
pub fn benchmarkHash(comptime H: var, bytes: usize) !Result {
var h = blk: {
if (H.init_u8s) |init| {
break :blk H.ty.init(init);
}
if (H.init_u64) |init| {
break :blk H.ty.init(init);
}
break :blk H.ty.init();
};
var block: [block_size]u8 = undefined;
prng.random.bytes(block[0..]);
var offset: usize = 0;
var timer = try Timer.start();
const start = timer.lap();
while (offset < bytes) : (offset += block.len) {
h.update(block[0..]);
}
const end = timer.read();
const elapsed_s = @intToFloat(f64, end - start) / time.ns_per_s;
const throughput = @floatToInt(u64, @intToFloat(f64, bytes) / elapsed_s);
return Result{
.hash = h.final(),
.throughput = throughput,
};
}
pub fn benchmarkHashSmallKeys(comptime H: var, key_size: usize, bytes: usize) !Result {
const key_count = bytes / key_size;
var block: [block_size]u8 = undefined;
prng.random.bytes(block[0..]);
var i: usize = 0;
var timer = try Timer.start();
const start = timer.lap();
var sum: u64 = 0;
while (i < key_count) : (i += 1) {
const small_key = block[0..key_size];
sum +%= blk: {
if (H.init_u8s) |init| {
break :blk H.ty.hash(init, small_key);
}
if (H.init_u64) |init| {
break :blk H.ty.hash(init, small_key);
}
break :blk H.ty.hash(small_key);
};
}
const end = timer.read();
const elapsed_s = @intToFloat(f64, end - start) / time.ns_per_s;
const throughput = @floatToInt(u64, @intToFloat(f64, bytes) / elapsed_s);
return Result{
.hash = sum,
.throughput = throughput,
};
}
fn usage() void {
std.debug.warn(
\\throughput_test [options]
\\
\\Options:
\\ --filter [test-name]
\\ --seed [int]
\\ --count [int]
\\ --key-size [int]
\\ --iterative-only
\\ --help
\\
);
}
fn mode(comptime x: comptime_int) comptime_int {
return if (builtin.mode == builtin.Mode.Debug) x / 64 else x;
}
// TODO(#1358): Replace with builtin formatted padding when available.
fn printPad(stdout: var, s: []const u8) !void {
var i: usize = 0;
while (i < 12 - s.len) : (i += 1) {
try stdout.print(" ");
}
try stdout.print("{}", s);
}
pub fn main() !void {
var stdout_file = try std.io.getStdOut();
var stdout_out_stream = stdout_file.outStream();
const stdout = &stdout_out_stream.stream;
var buffer: [1024]u8 = undefined;
var fixed = std.heap.FixedBufferAllocator.init(buffer[0..]);
const args = try std.process.argsAlloc(&fixed.allocator);
var filter: ?[]u8 = "";
var count: usize = mode(128 * MiB);
var key_size: usize = 32;
var seed: u32 = 0;
var test_iterative_only = false;
var i: usize = 1;
while (i < args.len) : (i += 1) {
if (std.mem.eql(u8, args[i], "--mode")) {
try stdout.print("{}\n", builtin.mode);
return;
} else if (std.mem.eql(u8, args[i], "--seed")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
seed = try std.fmt.parseUnsigned(u32, args[i], 10);
// we seed later
} else if (std.mem.eql(u8, args[i], "--filter")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
filter = args[i];
} else if (std.mem.eql(u8, args[i], "--count")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
const c = try std.fmt.parseUnsigned(usize, args[i], 10);
count = c * MiB;
} else if (std.mem.eql(u8, args[i], "--key-size")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
key_size = try std.fmt.parseUnsigned(usize, args[i], 10);
if (key_size > block_size) {
try stdout.print("key_size cannot exceed block size of {}\n", block_size);
std.os.exit(1);
}
} else if (std.mem.eql(u8, args[i], "--iterative-only")) {
test_iterative_only = true;
} else if (std.mem.eql(u8, args[i], "--help")) {
usage();
return;
} else {
usage();
std.os.exit(1);
}
}
inline for (hashes) |H| {
if (filter == null or std.mem.indexOf(u8, H.name, filter.?) != null) {
if (!test_iterative_only or H.has_iterative_api) {
try stdout.print("{}\n", H.name);
// Always reseed prior to every call so we are hashing the same buffer contents.
// This allows easier comparison between different implementations.
if (H.has_iterative_api) {
prng.seed(seed);
const result = try benchmarkHash(H, count);
try stdout.print(" iterative: {:4} MiB/s [{x:0<16}]\n", result.throughput / (1 * MiB), result.hash);
}
if (!test_iterative_only) {
prng.seed(seed);
const result_small = try benchmarkHashSmallKeys(H, key_size, count);
try stdout.print(" small keys: {:4} MiB/s [{x:0<16}]\n", result_small.throughput / (1 * MiB), result_small.hash);
}
}
}
}
}

View File

@ -9,17 +9,17 @@ const std = @import("../std.zig");
const debug = std.debug;
const testing = std.testing;
pub const Polynomial = struct {
const IEEE = 0xedb88320;
const Castagnoli = 0x82f63b78;
const Koopman = 0xeb31d82e;
pub const Polynomial = enum(u32) {
IEEE = 0xedb88320,
Castagnoli = 0x82f63b78,
Koopman = 0xeb31d82e,
};
// IEEE is by far the most common CRC and so is aliased by default.
pub const Crc32 = Crc32WithPoly(Polynomial.IEEE);
pub const Crc32 = Crc32WithPoly(.IEEE);
// slicing-by-8 crc32 implementation.
pub fn Crc32WithPoly(comptime poly: u32) type {
pub fn Crc32WithPoly(comptime poly: Polynomial) type {
return struct {
const Self = @This();
const lookup_tables = comptime block: {
@ -31,7 +31,7 @@ pub fn Crc32WithPoly(comptime poly: u32) type {
var j: usize = 0;
while (j < 8) : (j += 1) {
if (crc & 1 == 1) {
crc = (crc >> 1) ^ poly;
crc = (crc >> 1) ^ @enumToInt(poly);
} else {
crc = (crc >> 1);
}
@ -100,7 +100,7 @@ pub fn Crc32WithPoly(comptime poly: u32) type {
}
test "crc32 ieee" {
const Crc32Ieee = Crc32WithPoly(Polynomial.IEEE);
const Crc32Ieee = Crc32WithPoly(.IEEE);
testing.expect(Crc32Ieee.hash("") == 0x00000000);
testing.expect(Crc32Ieee.hash("a") == 0xe8b7be43);
@ -108,7 +108,7 @@ test "crc32 ieee" {
}
test "crc32 castagnoli" {
const Crc32Castagnoli = Crc32WithPoly(Polynomial.Castagnoli);
const Crc32Castagnoli = Crc32WithPoly(.Castagnoli);
testing.expect(Crc32Castagnoli.hash("") == 0x00000000);
testing.expect(Crc32Castagnoli.hash("a") == 0xc1d04330);
@ -116,7 +116,7 @@ test "crc32 castagnoli" {
}
// half-byte lookup table implementation.
pub fn Crc32SmallWithPoly(comptime poly: u32) type {
pub fn Crc32SmallWithPoly(comptime poly: Polynomial) type {
return struct {
const Self = @This();
const lookup_table = comptime block: {
@ -127,7 +127,7 @@ pub fn Crc32SmallWithPoly(comptime poly: u32) type {
var j: usize = 0;
while (j < 8) : (j += 1) {
if (crc & 1 == 1) {
crc = (crc >> 1) ^ poly;
crc = (crc >> 1) ^ @enumToInt(poly);
} else {
crc = (crc >> 1);
}
@ -164,7 +164,7 @@ pub fn Crc32SmallWithPoly(comptime poly: u32) type {
}
test "small crc32 ieee" {
const Crc32Ieee = Crc32SmallWithPoly(Polynomial.IEEE);
const Crc32Ieee = Crc32SmallWithPoly(.IEEE);
testing.expect(Crc32Ieee.hash("") == 0x00000000);
testing.expect(Crc32Ieee.hash("a") == 0xe8b7be43);
@ -172,7 +172,7 @@ test "small crc32 ieee" {
}
test "small crc32 castagnoli" {
const Crc32Castagnoli = Crc32SmallWithPoly(Polynomial.Castagnoli);
const Crc32Castagnoli = Crc32SmallWithPoly(.Castagnoli);
testing.expect(Crc32Castagnoli.hash("") == 0x00000000);
testing.expect(Crc32Castagnoli.hash("a") == 0xc1d04330);

View File

@ -152,8 +152,8 @@ fn SipHash(comptime T: type, comptime c_rounds: usize, comptime d_rounds: usize)
pub fn hash(key: []const u8, input: []const u8) T {
var c = Self.init(key);
c.update(input);
return c.final();
@inlineCall(c.update, input);
return @inlineCall(c.final);
}
};
}

View File

@ -1,148 +0,0 @@
const builtin = @import("builtin");
const std = @import("std");
const time = std.time;
const Timer = time.Timer;
const hash = std.hash;
const KiB = 1024;
const MiB = 1024 * KiB;
const GiB = 1024 * MiB;
var prng = std.rand.DefaultPrng.init(0);
const Hash = struct {
ty: type,
name: []const u8,
init_u8s: ?[]const u8 = null,
init_u64: ?u64 = null,
};
const siphash_key = "0123456789abcdef";
const hashes = [_]Hash{
Hash{ .ty = hash.Wyhash, .name = "wyhash", .init_u64 = 0 },
Hash{ .ty = hash.SipHash64(1, 3), .name = "siphash(1,3)", .init_u8s = siphash_key },
Hash{ .ty = hash.SipHash64(2, 4), .name = "siphash(2,4)", .init_u8s = siphash_key },
Hash{ .ty = hash.Fnv1a_64, .name = "fnv1a" },
Hash{ .ty = hash.Crc32, .name = "crc32" },
};
const Result = struct {
hash: u64,
throughput: u64,
};
pub fn benchmarkHash(comptime H: var, bytes: usize) !Result {
var h = blk: {
if (H.init_u8s) |init| {
break :blk H.ty.init(init);
}
if (H.init_u64) |init| {
break :blk H.ty.init(init);
}
break :blk H.ty.init();
};
var block: [8192]u8 = undefined;
prng.random.bytes(block[0..]);
var offset: usize = 0;
var timer = try Timer.start();
const start = timer.lap();
while (offset < bytes) : (offset += block.len) {
h.update(block[0..]);
}
const end = timer.read();
const elapsed_s = @intToFloat(f64, end - start) / time.ns_per_s;
const throughput = @floatToInt(u64, @intToFloat(f64, bytes) / elapsed_s);
return Result{
.hash = h.final(),
.throughput = throughput,
};
}
fn usage() void {
std.debug.warn(
\\throughput_test [options]
\\
\\Options:
\\ --filter [test-name]
\\ --seed [int]
\\ --count [int]
\\ --help
\\
);
}
fn mode(comptime x: comptime_int) comptime_int {
return if (builtin.mode == builtin.Mode.Debug) x / 64 else x;
}
// TODO(#1358): Replace with builtin formatted padding when available.
fn printPad(stdout: var, s: []const u8) !void {
var i: usize = 0;
while (i < 12 - s.len) : (i += 1) {
try stdout.print(" ");
}
try stdout.print("{}", s);
}
pub fn main() !void {
var stdout_file = try std.io.getStdOut();
var stdout_out_stream = stdout_file.outStream();
const stdout = &stdout_out_stream.stream;
var buffer: [1024]u8 = undefined;
var fixed = std.heap.FixedBufferAllocator.init(buffer[0..]);
const args = try std.process.argsAlloc(&fixed.allocator);
var filter: ?[]u8 = "";
var count: usize = mode(128 * MiB);
var i: usize = 1;
while (i < args.len) : (i += 1) {
if (std.mem.eql(u8, args[i], "--seed")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
const seed = try std.fmt.parseUnsigned(u32, args[i], 10);
prng.seed(seed);
} else if (std.mem.eql(u8, args[i], "--filter")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
filter = args[i];
} else if (std.mem.eql(u8, args[i], "--count")) {
i += 1;
if (i == args.len) {
usage();
std.os.exit(1);
}
const c = try std.fmt.parseUnsigned(u32, args[i], 10);
count = c * MiB;
} else if (std.mem.eql(u8, args[i], "--help")) {
usage();
return;
} else {
usage();
std.os.exit(1);
}
}
inline for (hashes) |H| {
if (filter == null or std.mem.indexOf(u8, H.name, filter.?) != null) {
const result = try benchmarkHash(H, count);
try printPad(stdout, H.name);
try stdout.print(": {:4} MiB/s [{:16}]\n", result.throughput / (1 * MiB), result.hash);
}
}
}

View File

@ -116,8 +116,8 @@ pub const Wyhash = struct {
pub fn hash(seed: u64, input: []const u8) u64 {
var c = Wyhash.init(seed);
c.update(input);
return c.final();
@inlineCall(c.update, input);
return @inlineCall(c.final);
}
};