InternPool: add a dump function

So we can see stats
This commit is contained in:
Andrew Kelley 2023-05-05 18:07:25 -07:00
parent 9ec0017f46
commit 0471638734
2 changed files with 148 additions and 12 deletions

View File

@ -2026,6 +2026,11 @@ pub fn update(comp: *Compilation, main_progress_node: *std.Progress.Node) !void
try comp.performAllTheWork(main_progress_node);
if (comp.bin_file.options.module) |module| {
std.debug.print("intern pool stats for '{s}':\n", .{
comp.bin_file.options.root_name,
});
module.intern_pool.dump();
if (comp.bin_file.options.is_test and comp.totalErrorCount() == 0) {
// The `test_functions` decl has been intentionally postponed until now,
// at which point we must populate it with the list of test functions that

View File

@ -722,10 +722,10 @@ pub const Tag = enum(u8) {
/// data is float value bitcasted to u32.
float_f32,
/// An f64 value.
/// data is payload index to Float64.
/// data is extra index to Float64.
float_f64,
/// An f128 value.
/// data is payload index to Float128.
/// data is extra index to Float128.
float_f128,
/// An extern function.
extern_func,
@ -877,6 +877,33 @@ pub const Int = struct {
limbs_len: u32,
};
/// A f64 value, broken up into 2 u32 parts.
pub const Float64 = struct {
piece0: u32,
piece1: u32,
pub fn get(self: Float64) f64 {
const int_bits = @as(u64, self.piece0) | (@as(u64, self.piece1) << 32);
return @bitCast(u64, int_bits);
}
};
/// A f128 value, broken up into 4 u32 parts.
pub const Float128 = struct {
piece0: u32,
piece1: u32,
piece2: u32,
piece3: u32,
pub fn get(self: Float128) f128 {
const int_bits = @as(u128, self.piece0) |
(@as(u128, self.piece1) << 32) |
(@as(u128, self.piece2) << 64) |
(@as(u128, self.piece3) << 96);
return @bitCast(f128, int_bits);
}
};
pub fn init(ip: *InternPool, gpa: Allocator) !void {
assert(ip.items.len == 0);
@ -1293,20 +1320,42 @@ fn addLimbsAssumeCapacity(ip: *InternPool, limbs: []const usize) void {
}
fn extraData(ip: InternPool, comptime T: type, index: usize) T {
const fields = std.meta.fields(T);
var i: usize = index;
var result: T = undefined;
inline for (fields) |field| {
inline for (@typeInfo(T).Struct.fields, 0..) |field, i| {
const int32 = ip.extra.items[i + index];
@field(result, field.name) = switch (field.type) {
u32 => ip.extra.items[i],
Index => @intToEnum(Index, ip.extra.items[i]),
i32 => @bitCast(i32, ip.extra.items[i]),
Pointer.Flags => @bitCast(Pointer.Flags, ip.extra.items[i]),
Pointer.PackedOffset => @bitCast(Pointer.PackedOffset, ip.extra.items[i]),
Pointer.VectorIndex => @intToEnum(Pointer.VectorIndex, ip.extra.items[i]),
u32 => int32,
Index => @intToEnum(Index, int32),
i32 => @bitCast(i32, int32),
Pointer.Flags => @bitCast(Pointer.Flags, int32),
Pointer.PackedOffset => @bitCast(Pointer.PackedOffset, int32),
Pointer.VectorIndex => @intToEnum(Pointer.VectorIndex, int32),
else => @compileError("bad field type: " ++ @typeName(field.type)),
};
}
return result;
}
/// Asserts the struct has 32-bit fields and the number of fields is evenly divisible by 2.
fn limbData(ip: InternPool, comptime T: type, index: usize) T {
switch (@sizeOf(usize)) {
@sizeOf(u32) => return extraData(ip, T, index),
@sizeOf(u64) => {},
else => @compileError("unsupported host"),
}
var result: T = undefined;
inline for (@typeInfo(T).Struct.fields, 0..) |field, i| {
const host_int = ip.limbs.items[index + i / 2];
const int32 = if (i % 2 == 0)
@truncate(u32, host_int)
else
@truncate(u32, host_int >> 32);
@field(result, field.name) = switch (field.type) {
u32 => int32,
Index => @intToEnum(Index, int32),
else => @compileError("bad field type: " ++ @typeName(field.type)),
};
i += 1;
}
return result;
}
@ -1350,3 +1399,85 @@ pub fn childType(ip: InternPool, i: Index) Index {
else => unreachable,
};
}
pub fn dump(ip: InternPool) void {
dumpFallible(ip, std.heap.page_allocator) catch return;
}
fn dumpFallible(ip: InternPool, arena: Allocator) anyerror!void {
const items_size = (1 + 4) * ip.items.len;
const extra_size = 4 * ip.extra.items.len;
const limbs_size = 8 * ip.limbs.items.len;
// TODO: map overhead size is not taken into account
const total_size = @sizeOf(InternPool) + items_size + extra_size + limbs_size;
std.debug.print(
\\InternPool size: {d} bytes
\\ items: {d} bytes
\\ extra: {d} bytes
\\ limbs: {d} bytes
\\
, .{ total_size, items_size, extra_size, limbs_size });
const tags = ip.items.items(.tag);
const datas = ip.items.items(.data);
const TagStats = struct {
count: usize = 0,
bytes: usize = 0,
};
var counts = std.AutoArrayHashMap(Tag, TagStats).init(arena);
for (tags, datas) |tag, data| {
const gop = try counts.getOrPut(tag);
if (!gop.found_existing) gop.value_ptr.* = .{};
gop.value_ptr.count += 1;
gop.value_ptr.bytes += 1 + 4 + @as(usize, switch (tag) {
.type_int_signed => 0,
.type_int_unsigned => 0,
.type_array => @sizeOf(Vector),
.type_vector => @sizeOf(Vector),
.type_pointer => @sizeOf(Pointer),
.type_optional => 0,
.type_error_union => @sizeOf(ErrorUnion),
.type_enum_simple => @sizeOf(EnumSimple),
.simple_type => 0,
.simple_value => 0,
.simple_internal => 0,
.int_u32 => 0,
.int_i32 => 0,
.int_usize => 0,
.int_comptime_int_u32 => 0,
.int_comptime_int_i32 => 0,
.int_positive,
.int_negative,
.enum_tag_positive,
.enum_tag_negative,
=> b: {
const int = ip.limbData(Int, data);
break :b @sizeOf(Int) + int.limbs_len * 8;
},
.float_f32 => 0,
.float_f64 => @sizeOf(Float64),
.float_f128 => @sizeOf(Float128),
.extern_func => @panic("TODO"),
.func => @panic("TODO"),
});
}
const SortContext = struct {
map: *std.AutoArrayHashMap(Tag, TagStats),
pub fn lessThan(ctx: @This(), a_index: usize, b_index: usize) bool {
const values = ctx.map.values();
return values[a_index].bytes > values[b_index].bytes;
}
};
counts.sort(SortContext{ .map = &counts });
const len = @min(50, tags.len);
std.debug.print("top 50 tags:\n", .{});
for (counts.keys()[0..len], counts.values()[0..len]) |tag, stats| {
std.debug.print(" {s}: {d} occurrences, {d} total bytes\n", .{
@tagName(tag), stats.count, stats.bytes,
});
}
}