adrien/zig-dimal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

Compare commits

base: adrien:cc39c1a6f46eb3db4447b0379b7d0dceaa34f244
Branches Tags
adrien:main
adrien:0.1.1
adrien:0.1.0
..
compare: adrien:fb922e3d5ceee23258b5c4c36cbcce5a3e3e1b9f
Branches Tags
adrien:main
adrien:0.1.1
adrien:0.1.0

No commits in common. "cc39c1a6f46eb3db4447b0379b7d0dceaa34f244" and "fb922e3d5ceee23258b5c4c36cbcce5a3e3e1b9f" have entirely different histories.
cc39c1a6f4 ... fb922e3d5c

1 changed files with 151 additions and 117 deletions
Show Stats Expand all files Collapse all files

268
src/benchmark.zig
View File

@ -6,9 +6,6 @@ const Vector = @import("Vector.zig").Vector;
var io: Io = undefined; var io: Io = undefined;
pub fn main(init: std.process.Init) !void { pub fn main(init: std.process.Init) !void {
const zone = tracy.ZoneN(@src(), "Main Loop");
defer zone.End();
var stdout_buf: [4096]u8 = undefined; var stdout_buf: [4096]u8 = undefined;
var stdout_writer: std.Io.File.Writer = .init(.stdout(), init.io, &stdout_buf); var stdout_writer: std.Io.File.Writer = .init(.stdout(), init.io, &stdout_buf);
try stdout_writer.interface.print("Starting Benchmarks...", .{}); try stdout_writer.interface.print("Starting Benchmarks...", .{});
@ -23,8 +20,6 @@ pub fn main(init: std.process.Init) !void {
try stdout_writer.flush(); try stdout_writer.flush();
try bench_Vector(&stdout_writer.interface); try bench_Vector(&stdout_writer.interface);
try stdout_writer.flush(); try stdout_writer.flush();
tracy.FrameMark();
} }
fn getTime() Io.Timestamp { fn getTime() Io.Timestamp {
@ -42,6 +37,8 @@ fn bench_Scalar(writer: *std.Io.Writer) !void {
const ITERS: usize = 100_000; const ITERS: usize = 100_000;
const SAMPLES: usize = 10; const SAMPLES: usize = 10;
var gsink: f64 = 0;
const getVal = struct { const getVal = struct {
fn f(comptime TT: type, i: usize, comptime mask: u7) TT { fn f(comptime TT: type, i: usize, comptime mask: u7) TT {
const v: u8 = @as(u8, @truncate(i & @as(usize, mask))) + 1; const v: u8 = @as(u8, @truncate(i & @as(usize, mask))) + 1;
@ -100,29 +97,29 @@ fn bench_Scalar(writer: *std.Io.Writer) !void {
var samples: [SAMPLES]f64 = undefined; var samples: [SAMPLES]f64 = undefined;
for (0..SAMPLES) |s_idx| { for (0..SAMPLES) |s_idx| {
var sink: T = 0;
const t_start = getTime(); const t_start = getTime();
for (0..ITERS) |i| { for (0..ITERS) |i| {
std.mem.doNotOptimizeAway( const r = if (comptime std.mem.eql(u8, op_name, "add"))
{ (M{ .value = getVal(T, i, 63) }).add(M{ .value = getVal(T, i +% 7, 63) })
_ = if (comptime std.mem.eql(u8, op_name, "add")) else if (comptime std.mem.eql(u8, op_name, "sub"))
(M{ .value = getVal(T, i, 63) }).add(M{ .value = getVal(T, i +% 7, 63) }) (M{ .value = getVal(T, i +% 10, 63) }).sub(M{ .value = getVal(T, i, 63) })
else if (comptime std.mem.eql(u8, op_name, "sub")) else if (comptime std.mem.eql(u8, op_name, "mulBy"))
(M{ .value = getVal(T, i +% 10, 63) }).sub(M{ .value = getVal(T, i, 63) }) (M{ .value = getVal(T, i, 63) }).mulBy(M{ .value = getVal(T, i +% 1, 63) })
else if (comptime std.mem.eql(u8, op_name, "mulBy")) else if (comptime std.mem.eql(u8, op_name, "divBy"))
(M{ .value = getVal(T, i, 63) }).mulBy(M{ .value = getVal(T, i +% 1, 63) }) (M{ .value = getVal(T, i +% 10, 63) }).divBy(S{ .value = getVal(T, i, 63) })
else if (comptime std.mem.eql(u8, op_name, "divBy")) else if (comptime std.mem.eql(u8, op_name, "scale"))
(M{ .value = getVal(T, i +% 10, 63) }).divBy(S{ .value = getVal(T, i, 63) }) (M{ .value = getVal(T, i, 63) }).scale(getVal(T, i +% 2, 63))
else if (comptime std.mem.eql(u8, op_name, "scale")) else
(M{ .value = getVal(T, i, 63) }).scale(getVal(T, i +% 2, 63)) (KM{ .value = getVal(T, i, 15) }).to(M);
else
(KM{ .value = getVal(T, i, 15) }).to(M); if (comptime @typeInfo(T) == .float) sink += r.value else sink ^= r.value;
},
);
} }
const t_end = getTime(); const t_end = getTime();
samples[s_idx] = @as(f64, @floatFromInt(t_start.durationTo(t_end).toNanoseconds())); samples[s_idx] = @as(f64, @floatFromInt(t_start.durationTo(t_end).toNanoseconds()));
fold(T, &gsink, sink);
} }
const stats = computeStats(&samples, ITERS); const stats = computeStats(&samples, ITERS);
@ -155,6 +152,9 @@ fn bench_Scalar(writer: *std.Io.Writer) !void {
} }
try writer.print("└──────────────┴───────┴───────┴───────┴───────┴───────┴───────┴───────┴───────┘\n", .{}); try writer.print("└──────────────┴───────┴───────┴───────┴───────┴───────┴───────┴───────┴───────┘\n", .{});
try writer.print("\nAnti-optimisation sink: {d:.4}\n", .{gsink});
try std.testing.expect(gsink != 0);
} }
fn bench_vsNative(writer: *std.Io.Writer) !void { fn bench_vsNative(writer: *std.Io.Writer) !void {
@ -173,6 +173,8 @@ fn bench_vsNative(writer: *std.Io.Writer) !void {
const TNames = .{ "i32", "i64", "i128", "f32", "f64" }; const TNames = .{ "i32", "i64", "i128", "f32", "f64" };
const Ops = .{ "add", "mulBy", "divBy" }; const Ops = .{ "add", "mulBy", "divBy" };
var gsink: f64 = 0;
try writer.print( try writer.print(
\\ \\
\\ Scalar vs Native Overhead Analysis \\ Scalar vs Native Overhead Analysis
@ -191,39 +193,45 @@ fn bench_vsNative(writer: *std.Io.Writer) !void {
const M = Scalar(T, .init(.{ .L = 1 }), .init(.{})); const M = Scalar(T, .init(.{ .L = 1 }), .init(.{}));
const S = Scalar(T, .init(.{ .T = 1 }), .init(.{})); const S = Scalar(T, .init(.{ .T = 1 }), .init(.{}));
std.mem.doNotOptimizeAway({ for (0..SAMPLES) |_| {
for (0..SAMPLES) |_| { // --- 1. Benchmark Native ---
// --- 1. Benchmark Native --- var n_sink: T = 0;
const n_start = getTime(); const n_start = getTime();
for (0..ITERS) |i| { for (0..ITERS) |i| {
const a = getValT(T, i); const a = getValT(T, i);
const b = getValT(T, 2); const b = getValT(T, 2);
_ = if (comptime std.mem.eql(u8, op_name, "add")) const r = if (comptime std.mem.eql(u8, op_name, "add"))
a + b a + b
else if (comptime std.mem.eql(u8, op_name, "mulBy")) else if (comptime std.mem.eql(u8, op_name, "mulBy"))
a * b a * b
else if (comptime @typeInfo(T) == .int) @divTrunc(a, b) else a / b; else if (comptime @typeInfo(T) == .int) @divTrunc(a, b) else a / b;
}
const n_end = getTime();
native_total_ns += @as(f64, @floatFromInt(n_start.durationTo(n_end).toNanoseconds()));
// --- 2. Benchmark Scalar --- if (comptime @typeInfo(T) == .float) n_sink += r else n_sink ^= r;
const q_start = getTime();
for (0..ITERS) |i| {
const qa = M{ .value = getValT(T, i) };
const qb = if (comptime std.mem.eql(u8, op_name, "divBy")) S{ .value = getValT(T, 2) } else M{ .value = getValT(T, 2) };
_ = if (comptime std.mem.eql(u8, op_name, "add"))
qa.add(qb)
else if (comptime std.mem.eql(u8, op_name, "mulBy"))
qa.mulBy(qb)
else
qa.divBy(qb);
}
const q_end = getTime();
quantity_total_ns += @as(f64, @floatFromInt(q_start.durationTo(q_end).toNanoseconds()));
} }
}); const n_end = getTime();
native_total_ns += @as(f64, @floatFromInt(n_start.durationTo(n_end).toNanoseconds()));
fold(T, &gsink, n_sink);
// --- 2. Benchmark Scalar ---
var q_sink: T = 0;
const q_start = getTime();
for (0..ITERS) |i| {
const qa = M{ .value = getValT(T, i) };
const qb = if (comptime std.mem.eql(u8, op_name, "divBy")) S{ .value = getValT(T, 2) } else M{ .value = getValT(T, 2) };
const r = if (comptime std.mem.eql(u8, op_name, "add"))
qa.add(qb)
else if (comptime std.mem.eql(u8, op_name, "mulBy"))
qa.mulBy(qb)
else
qa.divBy(qb);
if (comptime @typeInfo(T) == .float) q_sink += r.value else q_sink ^= r.value;
}
const q_end = getTime();
quantity_total_ns += @as(f64, @floatFromInt(q_start.durationTo(q_end).toNanoseconds()));
fold(T, &gsink, q_sink);
}
const avg_n = (native_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS)); const avg_n = (native_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS));
const avg_q = (quantity_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS)); const avg_q = (quantity_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS));
@ -237,6 +245,8 @@ fn bench_vsNative(writer: *std.Io.Writer) !void {
} }
try writer.print("└───────────┴──────┴───────────┴───────────┴───────────┘\n", .{}); try writer.print("└───────────┴──────┴───────────┴───────────┴───────────┘\n", .{});
try writer.print("\nAnti-optimisation sink: {d:.4}\n", .{gsink});
try std.testing.expect(gsink != 0);
} }
fn bench_crossTypeVsNative(writer: *std.Io.Writer) !void { fn bench_crossTypeVsNative(writer: *std.Io.Writer) !void {
@ -270,6 +280,8 @@ fn bench_crossTypeVsNative(writer: *std.Io.Writer) !void {
const TNames = .{ "i16", "i64", "i128", "f32", "f64" }; const TNames = .{ "i16", "i64", "i128", "f32", "f64" };
const Ops = .{ "add", "mulBy", "divBy" }; const Ops = .{ "add", "mulBy", "divBy" };
var gsink: f64 = 0;
try writer.print( try writer.print(
\\ \\
\\ Cross-Type Overhead Analysis: Scalar vs Native \\ Cross-Type Overhead Analysis: Scalar vs Native
@ -290,54 +302,60 @@ fn bench_crossTypeVsNative(writer: *std.Io.Writer) !void {
const M2 = Scalar(T2, .init(.{ .L = 1 }), .init(.{})); const M2 = Scalar(T2, .init(.{ .L = 1 }), .init(.{}));
const S2 = Scalar(T2, .init(.{ .T = 1 }), .init(.{})); const S2 = Scalar(T2, .init(.{ .T = 1 }), .init(.{}));
std.mem.doNotOptimizeAway({ for (0..SAMPLES) |_| {
for (0..SAMPLES) |_| { // --- 1. Benchmark Native (Cast T2 to T1, then math) ---
// --- 1. Benchmark Native (Cast T2 to T1, then math) --- var n_sink: T1 = 0;
const n_start = getTime(); const n_start = getTime();
for (0..ITERS) |i| { for (0..ITERS) |i| {
const a = getValT(T1, i); const a = getValT(T1, i);
const b_raw = getValT(T2, 2); const b_raw = getValT(T2, 2);
const b = castTo(T1, T2, b_raw); const b = castTo(T1, T2, b_raw);
_ = if (comptime std.mem.eql(u8, op_name, "add")) const r = if (comptime std.mem.eql(u8, op_name, "add"))
a + b a + b
else if (comptime std.mem.eql(u8, op_name, "mulBy")) else if (comptime std.mem.eql(u8, op_name, "mulBy"))
a * b a * b
else if (comptime @typeInfo(T1) == .int) else if (comptime @typeInfo(T1) == .int)
@divTrunc(a, b) @divTrunc(a, b)
else else
a / b; a / b;
}
const n_end = getTime();
native_total_ns += @as(f64, @floatFromInt(n_start.durationTo(n_end).toNanoseconds()));
// --- 2. Benchmark Scalar --- if (comptime @typeInfo(T1) == .float) n_sink += r else n_sink ^= r;
const q_start = getTime();
for (0..ITERS) |i| {
const qa = M1{ .value = getValT(T1, i) };
const qb = if (comptime std.mem.eql(u8, op_name, "divBy"))
S2{ .value = getValT(T2, 2) }
else
M2{ .value = getValT(T2, 2) };
_ = if (comptime std.mem.eql(u8, op_name, "add"))
qa.add(qb)
else if (comptime std.mem.eql(u8, op_name, "mulBy"))
qa.mulBy(qb)
else
qa.divBy(qb);
}
const q_end = getTime();
quantity_total_ns += @as(f64, @floatFromInt(q_start.durationTo(q_end).toNanoseconds()));
} }
const n_end = getTime();
native_total_ns += @as(f64, @floatFromInt(n_start.durationTo(n_end).toNanoseconds()));
fold(T1, &gsink, n_sink);
const avg_n = (native_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS)); // --- 2. Benchmark Scalar ---
const avg_q = (quantity_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS)); var q_sink: T1 = 0;
const slowdown = avg_q / avg_n; const q_start = getTime();
for (0..ITERS) |i| {
const qa = M1{ .value = getValT(T1, i) };
const qb = if (comptime std.mem.eql(u8, op_name, "divBy"))
S2{ .value = getValT(T2, 2) }
else
M2{ .value = getValT(T2, 2) };
try writer.print("│ {s:<7} │ {s:<4} │ {s:<4} │ {d:>7.2}ns │ {d:>7.2}ns │ {d:>8.2}x │\n", .{ const r = if (comptime std.mem.eql(u8, op_name, "add"))
op_name, TNames[t1_idx], TNames[t2_idx], avg_n, avg_q, slowdown, qa.add(qb)
}); else if (comptime std.mem.eql(u8, op_name, "mulBy"))
qa.mulBy(qb)
else
qa.divBy(qb);
if (comptime @typeInfo(T1) == .float) q_sink += r.value else q_sink ^= r.value;
}
const q_end = getTime();
quantity_total_ns += @as(f64, @floatFromInt(q_start.durationTo(q_end).toNanoseconds()));
fold(T1, &gsink, q_sink);
}
const avg_n = (native_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS));
const avg_q = (quantity_total_ns / SAMPLES) / @as(f64, @floatFromInt(ITERS));
const slowdown = avg_q / avg_n;
try writer.print("│ {s:<7} │ {s:<4} │ {s:<4} │ {d:>7.2}ns │ {d:>7.2}ns │ {d:>8.2}x │\n", .{
op_name, TNames[t1_idx], TNames[t2_idx], avg_n, avg_q, slowdown,
}); });
} }
} }
@ -347,11 +365,14 @@ fn bench_crossTypeVsNative(writer: *std.Io.Writer) !void {
} }
try writer.print("└─────────┴──────┴──────┴───────────┴───────────┴───────────┘\n", .{}); try writer.print("└─────────┴──────┴──────┴───────────┴───────────┴───────────┘\n", .{});
try writer.print("\nAnti-optimisation sink: {d:.4}\n", .{gsink});
try std.testing.expect(gsink != 0);
} }
fn bench_Vector(writer: *std.Io.Writer) !void { fn bench_Vector(writer: *std.Io.Writer) !void {
const ITERS: usize = 10_000; const ITERS: usize = 10_000;
const SAMPLES: usize = 10; const SAMPLES: usize = 10;
var gsink: f64 = 0;
const getVal = struct { const getVal = struct {
fn f(comptime TT: type, i: usize, comptime mask: u7) TT { fn f(comptime TT: type, i: usize, comptime mask: u7) TT {
@ -399,33 +420,44 @@ fn bench_Vector(writer: *std.Io.Writer) !void {
var samples: [SAMPLES]f64 = undefined; var samples: [SAMPLES]f64 = undefined;
std.mem.doNotOptimizeAway({ for (0..SAMPLES) |s_idx| {
for (0..SAMPLES) |s_idx| { var sink: T = 0;
const t_start = getTime();
for (0..ITERS) |i| {
const v1 = V.initDefault(getVal(T, i, 63));
if (comptime std.mem.eql(u8, op_name, "add")) { const t_start = getTime();
const v2 = V.initDefault(getVal(T, i +% 7, 63)); for (0..ITERS) |i| {
_ = v1.add(v2); const v1 = V.initDefault(getVal(T, i, 63));
} else if (comptime std.mem.eql(u8, op_name, "scale")) {
const sc = getVal(T, i +% 2, 63); if (comptime std.mem.eql(u8, op_name, "add")) {
_ = v1.scale(sc); const v2 = V.initDefault(getVal(T, i +% 7, 63));
} else if (comptime std.mem.eql(u8, op_name, "mulByScalar")) { const res = v1.add(v2);
const s_val = Q_time{ .value = getVal(T, i +% 2, 63) }; for (res.data) |val| {
_ = v1.mulByScalar(s_val); if (comptime @typeInfo(T) == .float) sink += val else sink ^= val;
} else if (comptime std.mem.eql(u8, op_name, "length")) {
_ = v1.length();
} }
} else if (comptime std.mem.eql(u8, op_name, "scale")) {
const sc = getVal(T, i +% 2, 63);
const res = v1.scale(sc);
for (res.data) |val| {
if (comptime @typeInfo(T) == .float) sink += val else sink ^= val;
}
} else if (comptime std.mem.eql(u8, op_name, "mulByScalar")) {
const s_val = Q_time{ .value = getVal(T, i +% 2, 63) };
const res = v1.mulByScalar(s_val);
for (res.data) |val| {
if (comptime @typeInfo(T) == .float) sink += val else sink ^= val;
}
} else if (comptime std.mem.eql(u8, op_name, "length")) {
const r = v1.length();
if (comptime @typeInfo(T) == .float) sink += r else sink ^= r;
} }
const t_end = getTime();
samples[s_idx] = @as(f64, @floatFromInt(t_start.durationTo(t_end).toNanoseconds()));
} }
const t_end = getTime();
const median_ns_per_op = computeStats(&samples, ITERS); samples[s_idx] = @as(f64, @floatFromInt(t_start.durationTo(t_end).toNanoseconds()));
try writer.print(" {d:>7.1} │", .{median_ns_per_op}); fold(T, &gsink, sink);
}); }
const median_ns_per_op = computeStats(&samples, ITERS);
try writer.print(" {d:>7.1} │", .{median_ns_per_op});
} }
try writer.print("\n", .{}); try writer.print("\n", .{});
} }
@ -435,4 +467,6 @@ fn bench_Vector(writer: *std.Io.Writer) !void {
} }
} }
try writer.print("└─────────────┴──────┴─────────┴─────────┴─────────┘\n", .{}); try writer.print("└─────────────┴──────┴─────────┴─────────┴─────────┘\n", .{});
try writer.print("\nAnti-optimisation sink: {d:.4}\n", .{gsink});
try std.testing.expect(gsink != 0);
} }