1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/tool/ruby_vm/models/bare_instructions.rb
Alan Wu 89e7997622 Combine call info and cache to speed up method invocation
To perform a regular method call, the VM needs two structs,
`rb_call_info` and `rb_call_cache`. At the moment, we allocate these two
structures in separate buffers. In the worst case, the CPU needs to read
4 cache lines to complete a method call. Putting the two structures
together reduces the maximum number of cache line reads to 2.

Combining the structures also saves 8 bytes per call site as the current
layout uses separate two pointers for the call info and the call cache.
This saves about 2 MiB on Discourse.

This change improves the Optcarrot benchmark at least 3%. For more
details, see attached bugs.ruby-lang.org ticket.

Complications:
 - A new instruction attribute `comptime_sp_inc` is introduced to
 calculate SP increase at compile time without using call caches. At
 compile time, a `TS_CALLDATA` operand points to a call info struct, but
 at runtime, the same operand points to a call data struct. Instruction
 that explicitly define `sp_inc` also need to define `comptime_sp_inc`.
 - MJIT code for copying call cache becomes slightly more complicated.
 - This changes the bytecode format, which might break existing tools.

[Misc #16258]
2019-10-24 18:03:42 +09:00

236 lines
6.8 KiB
Ruby
Executable file

#! /your/favourite/path/to/ruby
# -*- Ruby -*-
# -*- frozen_string_literal: true; -*-
# -*- warn_indent: true; -*-
#
# Copyright (c) 2017 Urabe, Shyouhei. All rights reserved.
#
# This file is a part of the programming language Ruby. Permission is hereby
# granted, to either redistribute and/or modify this file, provided that the
# conditions mentioned in the file COPYING are met. Consult the file for
# details.
require_relative '../loaders/insns_def'
require_relative 'c_expr'
require_relative 'typemap'
require_relative 'attribute'
class RubyVM::BareInstructions
attr_reader :template, :name, :opes, :pops, :rets, :decls, :expr
def initialize opts = {}
@template = opts[:template]
@name = opts[:name]
@loc = opts[:location]
@sig = opts[:signature]
@expr = RubyVM::CExpr.new opts[:expr]
@opes = typesplit @sig[:ope]
@pops = typesplit @sig[:pop].reject {|i| i == '...' }
@rets = typesplit @sig[:ret].reject {|i| i == '...' }
@attrs = opts[:attributes].map {|i|
RubyVM::Attribute.new i.merge(:insn => self)
}.each_with_object({}) {|a, h|
h[a.key] = a
}
@attrs_orig = @attrs.dup
check_attribute_consistency
predefine_attributes
end
def pretty_name
n = @sig[:name]
o = @sig[:ope].map{|i| i[/\S+$/] }.join ', '
p = @sig[:pop].map{|i| i[/\S+$/] }.join ', '
r = @sig[:ret].map{|i| i[/\S+$/] }.join ', '
return sprintf "%s(%s)(%s)(%s)", n, o, p, r
end
def bin
return "BIN(#{name})"
end
def call_attribute name
return sprintf 'attr_%s_%s(%s)', name, @name, \
@opes.map {|i| i[:name] }.compact.join(', ')
end
def has_attribute? k
@attrs_orig.has_key? k
end
def attributes
return @attrs \
. sort_by {|k, _| k } \
. map {|_, v| v }
end
def width
return 1 + opes.size
end
def declarations
return @variables \
. values \
. group_by {|h| h[:type] } \
. sort_by {|t, v| t } \
. map {|t, v| [t, v.map {|i| i[:name] }.sort ] } \
. map {|t, v|
sprintf("MAYBE_UNUSED(%s) %s", t, v.join(', '))
}
end
def preamble
# preamble makes sense for operand unifications
return []
end
def sc?
# sc stands for stack caching.
return false
end
def cast_to_VALUE var, expr = var[:name]
RubyVM::Typemap.typecast_to_VALUE var[:type], expr
end
def cast_from_VALUE var, expr = var[:name]
RubyVM::Typemap.typecast_from_VALUE var[:type], expr
end
def operands_info
opes.map {|o|
c, _ = RubyVM::Typemap.fetch o[:type]
next c
}.join
end
def handles_sp?
/\b(false|0)\b/ !~ @attrs.fetch('handles_sp').expr.expr
end
def always_leaf?
@attrs.fetch('leaf').expr.expr == 'true;'
end
def handle_canary stmt
# Stack canary is basically a good thing that we want to add, however:
#
# - When the instruction returns variadic number of return values,
# it is not easy to tell where is the stack top. We can't but
# skip it.
#
# - When the instruction body is empty (like putobject), we can
# say for 100% sure that canary is a waste of time.
#
# So we skip canary for those cases.
return '' if @sig[:ret].any? {|i| i == '...' }
return '' if @expr.blank?
return " #{stmt};\n"
end
def inspect
sprintf "#<%s %s@%s:%d>", self.class.name, @name, @loc[0], @loc[1]
end
def has_ope? var
return @opes.any? {|i| i[:name] == var[:name] }
end
def has_pop? var
return @pops.any? {|i| i[:name] == var[:name] }
end
def use_call_data?
@use_call_data ||=
@variables.find { |_, var_info| var_info[:type] == 'CALL_DATA' }
end
private
def check_attribute_consistency
if has_attribute?('sp_inc') \
&& use_call_data? \
&& !has_attribute?('comptime_sp_inc')
# As the call cache caches information that can only be obtained at
# runtime, we do not need it when compiling from AST to bytecode. This
# attribute defines an expression that computes the stack pointer
# increase based on just the call info to avoid reserving space for the
# call cache at compile time. In the expression, all call data operands
# are mapped to their call info counterpart. Additionally, all mentions
# of `cd` in the operand name are replaced with `ci`.
raise "Please define attribute `comptime_sp_inc` for `#{@name}`"
end
end
def generate_attribute t, k, v
@attrs[k] ||= RubyVM::Attribute.new \
insn: self, \
name: k, \
type: t, \
location: [], \
expr: v.to_s + ';'
return @attrs[k] ||= attr
end
def predefine_attributes
# Beware: order matters here because some attribute depends another.
generate_attribute 'const char*', 'name', "insn_name(#{bin})"
generate_attribute 'enum ruby_vminsn_type', 'bin', bin
generate_attribute 'rb_num_t', 'open', opes.size
generate_attribute 'rb_num_t', 'popn', pops.size
generate_attribute 'rb_num_t', 'retn', rets.size
generate_attribute 'rb_num_t', 'width', width
generate_attribute 'rb_snum_t', 'sp_inc', rets.size - pops.size
generate_attribute 'bool', 'handles_sp', default_definition_of_handles_sp
generate_attribute 'bool', 'leaf', default_definition_of_leaf
end
def default_definition_of_handles_sp
# Insn with ISEQ should yield it; can handle sp.
return opes.any? {|o| o[:type] == 'ISEQ' }
end
def default_definition_of_leaf
# Insn that handles SP can never be a leaf.
if not has_attribute? 'handles_sp' then
return ! default_definition_of_handles_sp
elsif handles_sp? then
return "! #{call_attribute 'handles_sp'}"
else
return true
end
end
def typesplit a
@variables ||= {}
a.map do |decl|
md = %r'
(?<comment> /[*] [^*]* [*]+ (?: [^*/] [^*]* [*]+ )* / ){0}
(?<ws> \g<comment> | \s ){0}
(?<ident> [_a-zA-Z] [0-9_a-zA-Z]* ){0}
(?<type> (?: \g<ident> \g<ws>+ )* \g<ident> ){0}
(?<var> \g<ident> ){0}
\G \g<ws>* \g<type> \g<ws>+ \g<var>
'x.match(decl)
@variables[md['var']] ||= {
decl: decl,
type: md['type'],
name: md['var'],
}
end
end
@instances = RubyVM::InsnsDef.map {|h|
new h.merge(:template => h)
}
def self.fetch name
@instances.find do |insn|
insn.name == name
end or raise IndexError, "instruction not found: #{name}"
end
def self.to_a
@instances
end
end