1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00

Refactor ujit_examples.h generator. Remove dwarfdump dependency

This commit is contained in:
Alan Wu 2020-09-23 03:02:01 -04:00
parent 2eaf55fcf1
commit 3d87eadf16
9 changed files with 223 additions and 180 deletions

View file

@ -1104,13 +1104,11 @@ incs: $(INSNS) {$(VPATH)}node_name.inc {$(VPATH)}known_errors.inc \
{$(VPATH)}vm_call_iseq_optimized.inc $(srcdir)/revision.h \
$(REVISION_H) \
$(UNICODE_DATA_HEADERS) $(ENC_HEADERS) \
{$(VPATH)}id.h {$(VPATH)}probes.dmyh {$(VPATH)}ujit_examples.h
{$(VPATH)}id.h {$(VPATH)}probes.dmyh
insns: $(INSNS)
ujit_examples.h: gen_ujit_examples.rb vm.$(OBJEXT)
$(ECHO) generating $@
$(Q) $(BASERUBY) $(srcdir)/gen_ujit_examples.rb
ujit_examples.inc: vm.$(OBJEXT)
id.h: $(tooldir)/generic_erb.rb $(srcdir)/template/id.h.tmpl $(srcdir)/defs/id.def
$(ECHO) generating $@
@ -15040,7 +15038,7 @@ transient_heap.$(OBJEXT): {$(VPATH)}vm_core.h
transient_heap.$(OBJEXT): {$(VPATH)}vm_debug.h
transient_heap.$(OBJEXT): {$(VPATH)}vm_opts.h
transient_heap.$(OBJEXT): {$(VPATH)}vm_sync.h
ujit_compile.$(OBJEXT): {$(VPATH)}ujit_examples.h
ujit_compile.$(OBJEXT): {$(VPATH)}ujit_examples.inc
util.$(OBJEXT): $(hdrdir)/ruby.h
util.$(OBJEXT): $(hdrdir)/ruby/ruby.h
util.$(OBJEXT): $(top_srcdir)/internal/compilers.h

View file

@ -1,110 +0,0 @@
def get_example_instruction_id
# TODO we could get this from the script that generates vm.inc instead of doing this song and dance
`dwarfdump --name='YARVINSN_ujit_call_example' vm.o`.each_line do |line|
if (id = line[/DW_AT_const_value\s\((\d+\))/, 1])
p [__method__, line] if $DEBUG
return id.to_i
end
end
raise
end
def get_fileoff
# use the load command to figure out the offset to the start of the content of vm.o
`otool -l vm.o`.each_line do |line|
if (fileoff = line[/fileoff (\d+)/, 1])
p [__method__, line] if $DEBUG
return fileoff.to_i
end
end
raise
end
def get_symbol_offset(symbol)
`nm vm.o`.each_line do |line|
if (offset = line[Regexp.compile('(\h+).+' + Regexp.escape(symbol) + '\Z'), 1])
p [__method__, line] if $DEBUG
return Integer(offset, 16)
end
end
raise
end
def readint8b(offset)
bytes = IO.binread('vm.o', 8, offset)
bytes.unpack('q').first # this is native endian but we want little endian. it's fine if the host moachine is x86
end
def disassemble(offset)
command = "objdump --x86-asm-syntax=intel --start-address=#{offset} --stop-address=#{offset+50} -d vm.o"
puts "Running: #{command}"
puts "feel free to verify with --reloc"
disassembly = `#{command}`
instructions = []
puts disassembly if $DEBUG
disassembly.each_line do |line|
line = line.strip
match_data = /\h+: ((?:\h\h\s?)+)\s+(\w+)/.match(line)
if match_data
bytes = match_data[1]
mnemonic = match_data[2]
instructions << [bytes, mnemonic, line]
break if mnemonic == 'jmp'
elsif !instructions.empty?
p line
raise "expected a continuous sequence of disassembly lines"
end
end
jmp_idx = instructions.find_index { |_, mnemonic, _| mnemonic == 'jmp' }
raise 'failed to find jmp' unless jmp_idx
raise 'generated code for example too long' unless jmp_idx < 10
handler_instructions = instructions[(0..jmp_idx)]
puts "Disassembly for the example handler:"
puts handler_instructions.map {|_, _, line| line}
raise 'rip reference in example makes copying unsafe' if handler_instructions.any? { |_, _, full_line| full_line.downcase.include?('rip') }
acceptable_mnemonics = %w(mov jmp lea call)
unrecognized = nil
handler_instructions.each { |i| unrecognized = i unless acceptable_mnemonics.include?(i[1]) }
raise "found an unrecognized \"#{unrecognized[1]}\" instruction in the example. List of recognized instructions: #{acceptable_mnemonics.join(', ')}" if unrecognized
raise 'found multiple jmp instructions' if handler_instructions.count { |_, mnemonic, _| mnemonic == 'jmp' } > 1
raise "the jmp instruction seems to be relative which isn't copiable" if instructions[jmp_idx][0].split.size > 4
raise 'found multiple call instructions' if handler_instructions.count { |_, mnemonic, _| mnemonic == 'call' } > 1
call_idx = handler_instructions.find_index { |_, mnemonic, _| mnemonic == 'call' }
pre_call_bytes = []
post_call_bytes = []
handler_instructions.take(call_idx).each do |bytes, mnemonic, _|
pre_call_bytes += bytes.split
end
handler_instructions[call_idx + 1, handler_instructions.size].each do |bytes, _, _|
post_call_bytes += bytes.split
end
File.write("ujit_examples.h", <<-EOF)
static const uint8_t ujit_pre_call_bytes[] = { #{pre_call_bytes.map{ |byte| '0x'+byte}.join(', ')} };
static const uint8_t ujit_post_call_bytes[] = { #{post_call_bytes.map{ |byte| '0x'+byte}.join(', ')} };
EOF
if $DEBUG
puts "file:"
puts File.binread("ujit_examples.h")
end
end
instruction_id = get_example_instruction_id
fileoff = get_fileoff
tc_table_offset = get_symbol_offset('vm_exec_core.insns_address_table')
vm_exec_core_offset = get_symbol_offset('vm_exec_core')
p instruction_id if $DEBUG
p fileoff if $DEBUG
p tc_table_offset.to_s(16) if $DEBUG
offset_to_insn_in_tc_table = fileoff + tc_table_offset + 8 * instruction_id
p offset_to_insn_in_tc_table if $DEBUG
offset_to_handler_code_from_vm_exec_core = readint8b(offset_to_insn_in_tc_table)
p offset_to_handler_code_from_vm_exec_core if $DEBUG
disassemble(vm_exec_core_offset + offset_to_handler_code_from_vm_exec_core)

View file

@ -590,7 +590,7 @@ update-known-errors:
$(IFCHANGE) $(srcdir)/defs/known_errors.def -
INSNS = opt_sc.inc optinsn.inc optunifs.inc insns.inc insns_info.inc \
vmtc.inc vm.inc mjit_compile.inc
vmtc.inc vm.inc mjit_compile.inc ujit_examples.inc
$(INSNS): $(srcdir)/insns.def vm_opts.h \
$(srcdir)/defs/opt_operand.def $(srcdir)/defs/opt_insn_unif.def \
@ -610,6 +610,8 @@ $(INSNS): $(srcdir)/insns.def vm_opts.h \
$(tooldir)/ruby_vm/models/instructions_unifications.rb \
$(tooldir)/ruby_vm/models/operands_unifications.rb \
$(tooldir)/ruby_vm/models/trace_instructions.rb \
$(tooldir)/ruby_vm/models/micro_jit.rb \
$(tooldir)/ruby_vm/models/micro_jit/example_instructions.rb \
$(tooldir)/ruby_vm/models/typemap.rb \
$(tooldir)/ruby_vm/scripts/converter.rb \
$(tooldir)/ruby_vm/scripts/insns2vm.rb \

View file

@ -13,71 +13,12 @@
require_relative 'bare_instructions'
require_relative 'operands_unifications'
require_relative 'instructions_unifications'
class RubyVM::UJITExampleInstructions
include RubyVM::CEscape
attr_reader :name
def initialize name
@name = name
end
def pretty_name
return sprintf "%s(...)(...)(...)", @name
end
def jump_destination
return @orig.name
end
def bin
return sprintf "BIN(%s)", @name
end
def width
1
end
def operands_info
""
end
def rets
return ['...']
end
def pops
return ['...']
end
def attributes
return []
end
def has_attribute? *;
return false
end
def handles_sp?
false
end
def always_leaf?
false
end
@all_examples = [new('ujit_call_example')]
def self.to_a
@all_examples
end
end
require_relative 'micro_jit'
RubyVM::Instructions = RubyVM::BareInstructions.to_a + \
RubyVM::OperandsUnifications.to_a + \
RubyVM::InstructionsUnifications.to_a + \
RubyVM::UJITExampleInstructions.to_a
RubyVM::MicroJIT::ExampleInstructions.to_a

View file

@ -0,0 +1,125 @@
#! /your/favourite/path/to/ruby
# -*- Ruby -*-
# -*- frozen_string_literal: true; -*-
# -*- warn_indent: true; -*-
#
# Copyright (c) 2020 Wu, Alan. 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.
module RubyVM::MicroJIT
class << self
def get_fileoff
# use the load command to figure out the offset to the start of the content of vm.o
`otool -l vm.o`.each_line do |line|
if (fileoff = line[/fileoff (\d+)/, 1])
p [__method__, line] if $DEBUG
return fileoff.to_i
end
end
raise
end
def get_symbol_offset(symbol)
`nm vm.o`.each_line do |line|
if (offset = line[Regexp.compile('(\h+).+' + Regexp.escape(symbol) + '\Z'), 1])
p [__method__, line] if $DEBUG
return Integer(offset, 16)
end
end
raise
end
def readint8b(offset)
bytes = IO.binread('vm.o', 8, offset)
bytes.unpack('q').first # this is native endian but we want little endian. it's fine if the host moachine is x86
end
def disassemble(offset)
command = "objdump --x86-asm-syntax=intel --start-address=#{offset} --stop-address=#{offset+50} -d vm.o"
puts "Running: #{command}"
puts "feel free to verify with --reloc"
disassembly = `#{command}`
instructions = []
puts disassembly if $DEBUG
disassembly.each_line do |line|
line = line.strip
match_data = /\h+: ((?:\h\h\s?)+)\s+(\w+)/.match(line)
if match_data
bytes = match_data[1]
mnemonic = match_data[2]
instructions << [bytes, mnemonic, line]
break if mnemonic == 'jmp'
elsif !instructions.empty?
p line
raise "expected a continuous sequence of disassembly lines"
end
end
jmp_idx = instructions.find_index { |_, mnemonic, _| mnemonic == 'jmp' }
raise 'failed to find jmp' unless jmp_idx
raise 'generated code for example too long' unless jmp_idx < 10
handler_instructions = instructions[(0..jmp_idx)]
puts "Disassembly for the example handler:"
puts handler_instructions.map {|_, _, line| line}
raise 'rip reference in example makes copying unsafe' if handler_instructions.any? { |_, _, full_line| full_line.downcase.include?('rip') }
acceptable_mnemonics = %w(mov jmp lea call)
unrecognized = nil
handler_instructions.each { |i| unrecognized = i unless acceptable_mnemonics.include?(i[1]) }
raise "found an unrecognized \"#{unrecognized[1]}\" instruction in the example. List of recognized instructions: #{acceptable_mnemonics.join(', ')}" if unrecognized
raise 'found multiple jmp instructions' if handler_instructions.count { |_, mnemonic, _| mnemonic == 'jmp' } > 1
raise "the jmp instruction seems to be relative which isn't copiable" if instructions[jmp_idx][0].split.size > 4
raise 'found multiple call instructions' if handler_instructions.count { |_, mnemonic, _| mnemonic == 'call' } > 1
call_idx = handler_instructions.find_index { |_, mnemonic, _| mnemonic == 'call' }
@pre_call_bytes = []
@post_call_bytes = []
handler_instructions.take(call_idx).each do |bytes, mnemonic, _|
@pre_call_bytes += bytes.split
end
handler_instructions[call_idx + 1, handler_instructions.size].each do |bytes, _, _|
@post_call_bytes += bytes.split
end
end
def scrape
instruction_id = RubyVM::Instructions.find_index { |insn| insn.name == 'ujit_call_example' }
fileoff = get_fileoff
tc_table_offset = get_symbol_offset('vm_exec_core.insns_address_table')
vm_exec_core_offset = get_symbol_offset('vm_exec_core')
p instruction_id if $DEBUG
p fileoff if $DEBUG
p tc_table_offset.to_s(16) if $DEBUG
offset_to_insn_in_tc_table = fileoff + tc_table_offset + 8 * instruction_id
p offset_to_insn_in_tc_table if $DEBUG
offset_to_handler_code_from_vm_exec_core = readint8b(offset_to_insn_in_tc_table)
p offset_to_handler_code_from_vm_exec_core if $DEBUG
disassemble(vm_exec_core_offset + offset_to_handler_code_from_vm_exec_core)
end
def comma_separated_hex_string(nums)
nums.map{ |byte| '0x'+byte}.join(', ')
end
def pre_call_bytes
scrape unless @pre_call_bytes
comma_separated_hex_string(@pre_call_bytes)
end
def post_call_bytes
scrape unless @post_call_bytes
comma_separated_hex_string(@post_call_bytes)
end
end
end
require_relative 'micro_jit/example_instructions'

View file

@ -0,0 +1,71 @@
#! /your/favourite/path/to/ruby
# -*- Ruby -*-
# -*- frozen_string_literal: true; -*-
# -*- warn_indent: true; -*-
#
# Copyright (c) 2020 Wu, Alan. 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.
class RubyVM::MicroJIT::ExampleInstructions
include RubyVM::CEscape
attr_reader :name
def initialize name
@name = name
end
def pretty_name
return sprintf "%s(...)(...)(...)", @name
end
def jump_destination
return @orig.name
end
def bin
return sprintf "BIN(%s)", @name
end
def width
1
end
def operands_info
""
end
def rets
return ['...']
end
def pops
return ['...']
end
def attributes
return []
end
def has_attribute? *;
return false
end
def handles_sp?
false
end
def always_leaf?
false
end
@all_examples = [new('ujit_call_example')]
def self.to_a
@all_examples
end
end

View file

@ -0,0 +1,16 @@
/* -*- C -*- */
%# Copyright (c) 2020 Wu, Alan. 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.
<%= render 'copyright' %>
<%= render 'notice', locals: {
this_file: 'contains raw instruction bytes that helps MicroJIT generate code',
edit: __FILE__,
} -%>
static const uint8_t ujit_pre_call_bytes[] = { <%= RubyVM::MicroJIT.pre_call_bytes %> };
static const uint8_t ujit_post_call_bytes[] = { <%= RubyVM::MicroJIT.post_call_bytes %> };

View file

@ -28,7 +28,7 @@
% RubyVM::TraceInstructions.to_a.each do |insn|
<%= render 'trace_instruction', locals: { insn: insn } -%>
% end
% RubyVM::UJITExampleInstructions.to_a.each do |insn|
% RubyVM::MicroJIT::ExampleInstructions.to_a.each do |insn|
INSN_ENTRY(<%= insn.name %>)
{
START_OF_ORIGINAL_INSN(<%= insn.name %>);

View file

@ -9,9 +9,9 @@
#include "ujit_asm.h"
#include "ujit_utils.h"
// TODO: give ujit_examples.h some more meaningful file name
// TODO: give ujit_examples.inc some more meaningful file name
// eg ujit_hook.h
#include "ujit_examples.h"
#include "ujit_examples.inc"
// Code generation context
typedef struct ctx_struct