-
-
Notifications
You must be signed in to change notification settings - Fork 2.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
optimized memcpy #18912
Merged
Merged
optimized memcpy #18912
Changes from all commits
Commits
Show all changes
3 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -1,149 +1,213 @@ | ||
const std = @import("std"); | ||
const assert = std.debug.assert; | ||
const common = @import("./common.zig"); | ||
const builtin = @import("builtin"); | ||
|
||
comptime { | ||
if (builtin.object_format != .c) { | ||
@export(&memcpy, .{ .name = "memcpy", .linkage = common.linkage, .visibility = common.visibility }); | ||
@export(&memmove, .{ .name = "memmove", .linkage = common.linkage, .visibility = common.visibility }); | ||
const export_options: std.builtin.ExportOptions = .{ | ||
.name = "memcpy", | ||
.linkage = common.linkage, | ||
.visibility = common.visibility, | ||
}; | ||
|
||
if (builtin.mode == .ReleaseSmall) | ||
@export(&memcpySmall, export_options) | ||
else | ||
@export(&memcpyFast, export_options); | ||
} | ||
} | ||
|
||
fn memcpy(noalias opt_dest: ?[*]u8, noalias opt_src: ?[*]const u8, len: usize) callconv(.C) ?[*]u8 { | ||
return memmove(opt_dest, opt_src, len); | ||
const Element = if (std.simd.suggestVectorLength(u8)) |vec_size| | ||
@Type(.{ .vector = .{ | ||
.child = u8, | ||
.len = vec_size, | ||
} }) | ||
else | ||
usize; | ||
|
||
comptime { | ||
assert(@sizeOf(Element) >= @alignOf(Element)); | ||
assert(std.math.isPowerOfTwo(@sizeOf(Element))); | ||
Comment on lines
+30
to
+31
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. All types have a size that is a multiple of its alignment, therefore the first condition isn't checking for anything. If you need a power of two, just use |
||
} | ||
|
||
fn memmove(opt_dest: ?[*]u8, opt_src: ?[*]const u8, len: usize) callconv(.C) ?[*]u8 { | ||
// a port of https://github.com/facebook/folly/blob/1c8bc50e88804e2a7361a57cd9b551dd10f6c5fd/folly/memcpy.S | ||
if (len == 0) { | ||
@branchHint(.unlikely); | ||
return opt_dest; | ||
} | ||
fn memcpySmall(noalias dest: ?[*]u8, noalias src: ?[*]const u8, len: usize) callconv(.C) ?[*]u8 { | ||
@setRuntimeSafety(builtin.is_test); | ||
|
||
const dest = opt_dest.?; | ||
const src = opt_src.?; | ||
|
||
if (len < 8) { | ||
@branchHint(.unlikely); | ||
if (len == 1) { | ||
@branchHint(.unlikely); | ||
dest[0] = src[0]; | ||
} else if (len >= 4) { | ||
@branchHint(.unlikely); | ||
blockCopy(dest, src, 4, len); | ||
} else { | ||
blockCopy(dest, src, 2, len); | ||
} | ||
return dest; | ||
for (0..len) |i| { | ||
dest.?[i] = src.?[i]; | ||
} | ||
|
||
if (len > 32) { | ||
@branchHint(.unlikely); | ||
if (len > 256) { | ||
@branchHint(.unlikely); | ||
copyMove(dest, src, len); | ||
return dest; | ||
} | ||
copyLong(dest, src, len); | ||
return dest; | ||
} | ||
return dest; | ||
} | ||
|
||
if (len > 16) { | ||
@branchHint(.unlikely); | ||
blockCopy(dest, src, 16, len); | ||
return dest; | ||
} | ||
fn memcpyFast(noalias dest: ?[*]u8, noalias src: ?[*]const u8, len: usize) callconv(.C) ?[*]u8 { | ||
@setRuntimeSafety(builtin.is_test); | ||
|
||
const small_limit = 2 * @sizeOf(Element); | ||
|
||
blockCopy(dest, src, 8, len); | ||
if (copySmallLength(small_limit, dest.?, src.?, len)) return dest; | ||
|
||
copyForwards(dest.?, src.?, len); | ||
|
||
return dest; | ||
} | ||
|
||
inline fn blockCopy(dest: [*]u8, src: [*]const u8, block_size: comptime_int, len: usize) void { | ||
const first = @as(*align(1) const @Vector(block_size, u8), src[0..block_size]).*; | ||
const second = @as(*align(1) const @Vector(block_size, u8), src[len - block_size ..][0..block_size]).*; | ||
dest[0..block_size].* = first; | ||
dest[len - block_size ..][0..block_size].* = second; | ||
} | ||
inline fn copySmallLength( | ||
comptime small_limit: comptime_int, | ||
dest: [*]u8, | ||
src: [*]const u8, | ||
len: usize, | ||
) bool { | ||
if (len < 16) { | ||
copyLessThan16(dest, src, len); | ||
return true; | ||
} | ||
|
||
inline fn copyLong(dest: [*]u8, src: [*]const u8, len: usize) void { | ||
var array: [8]@Vector(32, u8) = undefined; | ||
if (comptime 2 < (std.math.log2(small_limit) + 1) / 2) { | ||
if (copy16ToSmallLimit(small_limit, dest, src, len)) return true; | ||
} | ||
|
||
inline for (.{ 64, 128, 192, 256 }, 0..) |N, i| { | ||
array[i * 2] = src[(N / 2) - 32 ..][0..32].*; | ||
array[(i * 2) + 1] = src[len - N / 2 ..][0..32].*; | ||
return false; | ||
} | ||
|
||
if (len <= N) { | ||
@branchHint(.unlikely); | ||
for (0..i + 1) |j| { | ||
dest[j * 32 ..][0..32].* = array[j * 2]; | ||
dest[len - ((j * 32) + 32) ..][0..32].* = array[(j * 2) + 1]; | ||
} | ||
return; | ||
} | ||
inline fn copyLessThan16( | ||
dest: [*]u8, | ||
src: [*]const u8, | ||
len: usize, | ||
) void { | ||
@setRuntimeSafety(builtin.is_test); | ||
if (len < 4) { | ||
if (len == 0) return; | ||
dest[0] = src[0]; | ||
dest[len / 2] = src[len / 2]; | ||
dest[len - 1] = src[len - 1]; | ||
return; | ||
} | ||
copyRange4(4, dest, src, len); | ||
} | ||
|
||
inline fn copyMove(dest: [*]u8, src: [*]const u8, len: usize) void { | ||
if (@intFromPtr(src) >= @intFromPtr(dest)) { | ||
@branchHint(.unlikely); | ||
copyForward(dest, src, len); | ||
} else if (@intFromPtr(src) + len > @intFromPtr(dest)) { | ||
@branchHint(.unlikely); | ||
overlapBwd(dest, src, len); | ||
} else { | ||
copyForward(dest, src, len); | ||
inline fn copy16ToSmallLimit( | ||
comptime small_limit: comptime_int, | ||
dest: [*]u8, | ||
src: [*]const u8, | ||
len: usize, | ||
) bool { | ||
@setRuntimeSafety(builtin.is_test); | ||
inline for (2..(std.math.log2(small_limit) + 1) / 2 + 1) |p| { | ||
const limit = 1 << (2 * p); | ||
if (len < limit) { | ||
copyRange4(limit / 4, dest, src, len); | ||
return true; | ||
} | ||
} | ||
return false; | ||
} | ||
|
||
inline fn copyForward(dest: [*]u8, src: [*]const u8, len: usize) void { | ||
const tail: @Vector(32, u8) = src[len - 32 ..][0..32].*; | ||
inline fn copyForwards( | ||
noalias dest: [*]u8, | ||
noalias src: [*]const u8, | ||
len: usize, | ||
) void { | ||
@setRuntimeSafety(builtin.is_test); | ||
assert(len >= 2 * @sizeOf(Element)); | ||
|
||
dest[0..@sizeOf(Element)].* = src[0..@sizeOf(Element)].*; | ||
const alignment_offset = @alignOf(Element) - @intFromPtr(src) % @alignOf(Element); | ||
const n = len - alignment_offset; | ||
const d = dest + alignment_offset; | ||
const s = src + alignment_offset; | ||
|
||
copyBlocksAlignedSource(@ptrCast(d), @alignCast(@ptrCast(s)), n); | ||
|
||
// copy last `@sizeOf(Element)` bytes unconditionally, since block copy | ||
// methods only copy a multiple of `@sizeOf(Element)` bytes. | ||
const offset = len - @sizeOf(Element); | ||
dest[offset..][0..@sizeOf(Element)].* = src[offset..][0..@sizeOf(Element)].*; | ||
} | ||
|
||
const N: usize = len & ~@as(usize, 127); | ||
var i: usize = 0; | ||
inline fn copyBlocksAlignedSource( | ||
noalias dest: [*]align(1) Element, | ||
noalias src: [*]const Element, | ||
max_bytes: usize, | ||
) void { | ||
copyBlocks(dest, src, max_bytes); | ||
} | ||
|
||
while (i < N) : (i += 128) { | ||
dest[i..][0..32].* = src[i..][0..32].*; | ||
dest[i + 32 ..][0..32].* = src[i + 32 ..][0..32].*; | ||
dest[i + 64 ..][0..32].* = src[i + 64 ..][0..32].*; | ||
dest[i + 96 ..][0..32].* = src[i + 96 ..][0..32].*; | ||
} | ||
/// Copies the largest multiple of `@sizeOf(T)` bytes from `src` to `dest`, | ||
/// that is less than `max_bytes` where `T` is the child type of `src` and | ||
/// `dest`. | ||
inline fn copyBlocks( | ||
noalias dest: anytype, | ||
noalias src: anytype, | ||
max_bytes: usize, | ||
) void { | ||
@setRuntimeSafety(builtin.is_test); | ||
|
||
const T = @typeInfo(@TypeOf(dest)).pointer.child; | ||
comptime assert(T == @typeInfo(@TypeOf(src)).pointer.child); | ||
|
||
if (len - i <= 32) { | ||
@branchHint(.unlikely); | ||
dest[len - 32 ..][0..32].* = tail; | ||
} else { | ||
copyLong(dest[i..], src[i..], len - i); | ||
const loop_count = max_bytes / @sizeOf(T); | ||
|
||
for (dest[0..loop_count], src[0..loop_count]) |*d, s| { | ||
d.* = s; | ||
dweiller marked this conversation as resolved.
Show resolved
Hide resolved
|
||
} | ||
} | ||
|
||
inline fn overlapBwd(dest: [*]u8, src: [*]const u8, len: usize) void { | ||
var array: [5]@Vector(32, u8) = undefined; | ||
array[0] = src[len - 32 ..][0..32].*; | ||
inline for (1..5) |i| array[i] = src[(i - 1) << 5 ..][0..32].*; | ||
|
||
const end: usize = (@intFromPtr(dest) + len - 32) & 31; | ||
const range = len - end; | ||
var s = src + range; | ||
var d = dest + range; | ||
|
||
while (@intFromPtr(s) > @intFromPtr(src + 128)) { | ||
// zig fmt: off | ||
const first = @as(*align(1) const @Vector(32, u8), @ptrCast(s - 32)).*; | ||
const second = @as(*align(1) const @Vector(32, u8), @ptrCast(s - 64)).*; | ||
const third = @as(*align(1) const @Vector(32, u8), @ptrCast(s - 96)).*; | ||
const fourth = @as(*align(1) const @Vector(32, u8), @ptrCast(s - 128)).*; | ||
|
||
@as(*align(32) @Vector(32, u8), @alignCast(@ptrCast(d - 32))).* = first; | ||
@as(*align(32) @Vector(32, u8), @alignCast(@ptrCast(d - 64))).* = second; | ||
@as(*align(32) @Vector(32, u8), @alignCast(@ptrCast(d - 96))).* = third; | ||
@as(*align(32) @Vector(32, u8), @alignCast(@ptrCast(d - 128))).* = fourth; | ||
// zig fmt: on | ||
|
||
s -= 128; | ||
d -= 128; | ||
} | ||
/// copy `len` bytes from `src` to `dest`; `len` must be in the range | ||
/// `[copy_len, 4 * copy_len)`. | ||
inline fn copyRange4( | ||
comptime copy_len: comptime_int, | ||
noalias dest: [*]u8, | ||
noalias src: [*]const u8, | ||
len: usize, | ||
) void { | ||
@setRuntimeSafety(builtin.is_test); | ||
comptime assert(std.math.isPowerOfTwo(copy_len)); | ||
assert(len >= copy_len); | ||
assert(len < 4 * copy_len); | ||
|
||
const a = len & (copy_len * 2); | ||
const b = a / 2; | ||
|
||
const last = len - copy_len; | ||
const pen = last - b; | ||
|
||
dest[0..copy_len].* = src[0..copy_len].*; | ||
dest[b..][0..copy_len].* = src[b..][0..copy_len].*; | ||
dest[pen..][0..copy_len].* = src[pen..][0..copy_len].*; | ||
dest[last..][0..copy_len].* = src[last..][0..copy_len].*; | ||
} | ||
|
||
test { | ||
const S = struct { | ||
fn testFunc(comptime copy_func: anytype) !void { | ||
const max_len = 1024; | ||
var buffer: [max_len + @alignOf(Element) - 1]u8 align(@alignOf(Element)) = undefined; | ||
for (&buffer, 0..) |*b, i| { | ||
b.* = @intCast(i % 97); | ||
} | ||
var dest: [max_len + @alignOf(Element) - 1]u8 align(@alignOf(Element)) = undefined; | ||
|
||
for (0..max_len) |copy_len| { | ||
for (0..@alignOf(Element)) |s_offset| { | ||
for (0..@alignOf(Element)) |d_offset| { | ||
@memset(&dest, 0xff); | ||
const s = buffer[s_offset..][0..copy_len]; | ||
const d = dest[d_offset..][0..copy_len]; | ||
_ = copy_func(@ptrCast(d.ptr), @ptrCast(s.ptr), s.len); | ||
std.testing.expectEqualSlices(u8, s, d) catch |e| { | ||
std.debug.print("error encountered for length={d}, s_offset={d}, d_offset={d}\n", .{ | ||
copy_len, s_offset, d_offset, | ||
}); | ||
return e; | ||
}; | ||
} | ||
} | ||
} | ||
} | ||
}; | ||
|
||
inline for (array[1..], 0..) |vec, i| dest[i * 32 ..][0..32].* = vec; | ||
dest[len - 32 ..][0..32].* = array[0]; | ||
try S.testFunc(memcpySmall); | ||
try S.testFunc(memcpyFast); | ||
} |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.