plessl-llvmruby 0.0.3

data/ext/llvmruby.h

@@ -0,0 +1,73 @@
+#define __STDC_LIMIT_MACROS
+
+#include "llvm/Module.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Constants.h"
+#include "llvm/Instructions.h"
+#include "llvm/ModuleProvider.h"
+#include "llvm/PassManager.h"
+#include "llvm/LinkAllPasses.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Transforms/Scalar.h"
+#include "llvm/Analysis/Verifier.h"
+#include "llvm/ExecutionEngine/JIT.h"
+#include "llvm/ExecutionEngine/Interpreter.h"
+#include "llvm/ExecutionEngine/GenericValue.h"
+#include "llvm/Support/IRBuilder.h"
+#include <iostream>
+using namespace llvm;
+
+#include "ruby.h"
+
+extern VALUE cLLVMRuby;
+extern VALUE cLLVMValue;
+extern VALUE cLLVMUser;
+extern VALUE cLLVMUse;
+extern VALUE cLLVMModule;
+extern VALUE cLLVMFunction;
+extern VALUE cLLVMBasicBlock;
+extern VALUE cLLVMBuilder;
+extern VALUE cLLVMType;
+extern VALUE cLLVMPointerType;
+extern VALUE cLLVMStructType;
+extern VALUE cLLVMArrayType;
+extern VALUE cLLVMVectorType;
+extern VALUE cLLVMFunctionType;
+extern VALUE cLLVMInstruction;
+extern VALUE cLLVMBinaryOperator;
+extern VALUE cLLVMAllocationInst;
+
+#define HANDLE_TERM_INST(Num, Opcode, Klass) extern VALUE cLLVM##Klass;
+#define HANDLE_MEMORY_INST(Num, Opcode, Klass) extern VALUE cLLVM##Klass;
+#define HANDLE_OTHER_INST(Num, Opcode, Klass) extern VALUE cLLVM##Klass;
+#include "llvm/Instruction.def"
+
+extern VALUE cLLVMBinaryOps;
+extern VALUE cLLVMPhi;
+extern VALUE cLLVMPassManager;
+
+#define LLVM_VAL(obj) ((Value*)DATA_PTR(obj))
+#define LLVM_USER(obj) ((User*)DATA_PTR(obj))
+#define LLVM_USE(obj) ((Use*)DATA_PTR(obj))
+#define LLVM_TYPE(obj) ((Type*)DATA_PTR(obj))
+#define LLVM_FUNC_TYPE(obj) ((FunctionType*)DATA_PTR(obj))
+#define LLVM_MODULE(obj) ((Module*)DATA_PTR(obj))
+#define LLVM_FUNCTION(obj) ((Function*)DATA_PTR(obj))
+#define LLVM_BASIC_BLOCK(obj) ((BasicBlock*)DATA_PTR(obj))
+#define LLVM_INSTRUCTION(obj) ((Instruction*)DATA_PTR(obj))
+#define LLVM_PHI(obj) ((PHINode*)DATA_PTR(obj))
+
+#define CHECK_TYPE(val, klass)\
+  if(!rb_obj_is_kind_of(val, klass)) {\
+    rb_raise(rb_eTypeError, "wrong argument type: %s given, expected %s", rb_obj_classname(val), rb_class2name(klass));\
+  }
+
+extern "C" {
+VALUE llvm_value_wrap(Value*);
+VALUE llvm_user_wrap(User*);
+VALUE llvm_use_wrap(Use*);
+VALUE llvm_function_wrap(Function*);
+VALUE llvm_basic_block_wrap(BasicBlock*);
+VALUE llvm_function_create_block(VALUE);
+VALUE llvm_instruction_wrap(Instruction*);
+}

data/lib/llvm.rb

@@ -0,0 +1,146 @@
+require 'llvmruby'
+
+class Fixnum
+  def llvm(type = LLVM::MACHINE_WORD)
+    LLVM::Value.get_constant(type, self)
+  end
+end
+
+class Float
+  def llvm
+    LLVM::Value.get_float_constant(self)
+  end
+end
+
+class Object
+  def immediate
+    LLVM::Value.get_immediate_constant(self)
+  end
+end
+
+class LLVM::Value
+  def llvm
+    self
+  end
+
+  def immediate
+    self
+  end
+end
+
+module LLVM
+  # enum llvm::Type::TypeID
+  VoidTyID, FloatTyID, DoubleTyID, X86_FP80TyID, FP128TyID, PPC_FP128TyID, LabelTyID, IntegerTyID, 
+  FunctionTyID, StructTyID, ArrayTyID, PointerTyID, OpaqueTyID, VectorTyID = (0..13).to_a
+  
+  class Builder
+    def self.add_bin_op(op)
+      define_method(op.downcase) do |x, y|
+        bin_op(Instruction.const_get(op), x, y)
+      end
+    end
+    
+    bin_ops = ['Add', 'Sub', 'Mul', 'UDiv', 'SDiv', 'FDiv', 'URem', 'SRem', 'FRem']
+    bin_ops += ['Shl', 'LShr', 'AShr', 'And', 'Or', 'Xor']
+    bin_ops.each {|op| add_bin_op(op)}
+
+    Instruction.constants.grep(/^ICMP_/) do |pred|
+      define_method(pred.downcase) do |x, y|
+        icmp(Instruction.const_get(pred), x, y)
+      end
+    end
+
+    Instruction.constants.grep(/^FCMP_/) do |pred|
+      define_method(pred.downcase) do |x, y|
+        fcmp(Instruction.const_get(pred), x, y)
+      end
+    end
+
+    def self.define_cast(name, inst)
+      define_method(name) do |val, dest_type|
+        cast(inst, val, dest_type)
+      end
+    end
+
+    define_cast(:trunc,      Instruction::Trunc)
+    define_cast(:zext,       Instruction::ZExt)
+    define_cast(:sext,       Instruction::SExt)
+    define_cast(:fp_to_si,   Instruction::FPToSI)
+    define_cast(:fp_to_ui,   Instruction::FPToUI)
+    define_cast(:ui_to_fp,   Instruction::UIToFP)
+    define_cast(:si_to_fp,   Instruction::SIToFP)
+    define_cast(:fp_trunc,   Instruction::FPTrunc)
+    define_cast(:fp_ext,     Instruction::FPExt)
+    define_cast(:ptr_to_int, Instruction::PtrToInt)
+    define_cast(:int_to_ptr, Instruction::IntToPtr)
+    define_cast(:bit_cast,   Instruction::BitCast)
+ 
+    def write(&b)
+      instance_eval(&b)
+    end
+  end
+
+  # describe structures used by the ruby 1.8/1.9 interpreters
+  module RubyInternals
+    FIXNUM_FLAG = 0x1.llvm
+    CHAR = Type::Int8Ty
+    P_CHAR = Type::pointer(CHAR)
+    LONG = MACHINE_WORD 
+    INT = Type::Int32Ty
+    VALUE = MACHINE_WORD
+    P_VALUE = Type::pointer(VALUE)
+    ID = MACHINE_WORD
+    RBASIC = Type::struct([VALUE, VALUE])
+    RARRAY = Type::struct([RBASIC, LONG, LONG, P_VALUE])
+    P_RARRAY = Type::pointer(RARRAY)
+    RSTRING = Type::struct([RBASIC, LONG, P_CHAR, VALUE])
+    P_RSTRING = Type::pointer(RSTRING)
+  end
+
+  # include this into the builder to get methods for manipulating ruby values
+  module RubyHelpers
+    include RubyInternals
+
+    def fixnum?(val)
+      self.and(FIXNUM_FLAG, val)
+    end
+
+    def num2fix(val)
+      shifted = shl(val, 1.llvm)
+      xor(FIXNUM_FLAG, shifted)
+    end
+
+    def fix2int(val)
+      x = xor(FIXNUM_FLAG, val)
+      lshr(val, 1.llvm)
+    end
+
+    def slen(str)
+      val_ptr = int_to_ptr(str, P_RSTRING)
+      len_ptr = struct_gep(val_ptr, 1)
+      load(len_ptr)
+    end
+
+    def alen(ary)
+      val_ptr = int_to_ptr(ary, P_RARRAY)
+      len_ptr = struct_gep(val_ptr, 1)
+      load(len_ptr)
+    end
+
+    def aref(ary, idx)
+      val_ptr = int_to_ptr(ary, P_RARRAY)
+      data_ptr = struct_gep(val_ptr, 3)
+      data_ptr = load(data_ptr)
+      slot_n = gep(data_ptr, idx.llvm)
+      load(slot_n)
+    end
+
+    def aset(ary, idx, set)
+      val_ptr = int_to_ptr(ary, P_RARRAY)
+      data_ptr = struct_gep(val_ptr, 3)
+      data_ptr = load(data_ptr)
+      slot_n = gep(data_ptr, idx.llvm)
+      store(set, slot_n)
+    end
+  end
+end

data/lib/ruby_vm.rb

@@ -0,0 +1,352 @@
+require 'llvm'
+include LLVM
+include RubyInternals
+
+class Symbol
+  # turn a symbol object_id into a VALUE
+  # from gc.c, symbols object_id's are calculated like this:
+  # SYM2ID(x) = RSHIFT((unsigned long)x,8)
+  # object_id = (SYM2ID(obj) * sizeof(RVALUE) + (4 << 2)) | FIXNUM_FLAG;
+  def llvm
+    (((object_id/20) << 8) | 0xe).llvm(MACHINE_WORD)
+  end
+end
+
+class Object
+  def llvm
+    LLVM::Value.get_immediate_constant(self)
+  end
+
+  def llvm_send(f)
+    # for now, pass the receiver as the first argument
+    ExecutionEngine.runFunction(f, self)
+  end
+end
+
+class Builder
+  include RubyHelpers
+
+  def self.set_globals(b)
+    @@stack = b.alloca(VALUE, 100)
+    @@stack_ptr = b.alloca(P_VALUE, 0)
+    b.store(@@stack, @@stack_ptr)
+    @@locals = b.alloca(VALUE, 100)
+  end
+
+  def stack
+    @@stack
+  end
+
+  def stack_ptr
+    @@stack_ptr
+  end
+
+  def push(val)
+    sp = load(stack_ptr)
+    store(val, sp)
+    new_sp = gep(sp, 1.llvm)
+    store(new_sp, stack_ptr)
+  end
+
+  def pop
+    sp = load(stack_ptr)
+    new_sp = gep(sp, -1.llvm)
+    store(new_sp, stack_ptr)
+    load(new_sp)
+  end
+
+  def peek(n = 1)
+    sp = load(stack_ptr)
+    peek_sp = gep(sp, (-n).llvm)
+    load(peek_sp)
+  end
+
+  def topn(n)
+    sp = load(stack_ptr)
+    idx = sub((-1).llvm(Type::Int32Ty), n)
+    top_sp = gep(sp, idx)
+    load(top_sp)
+  end
+
+  def locals
+    @@locals
+  end
+end
+
+class OpCodeBuilder
+  def initialize(mod, func, rb_funcs)
+    @mod = mod
+    @func = func
+    @rb_funcs = rb_funcs
+  end
+
+  def get_self
+    @func.arguments.first
+  end
+
+  def nop(b, oprnd)
+  end
+
+  def putnil(b, oprnd)
+    b.push(nil.immediate)
+  end
+
+  def putself(b, oprnd)
+    b.push(get_self)
+  end
+
+  def putobject(b, oprnd)
+    b.push(oprnd.llvm)
+  end
+
+  def pop(b, oprnd)
+    b.pop
+  end
+
+  def dup(b, oprnd)
+    b.push(b.peek)
+  end
+
+  def swap(b, oprn) 
+    v1 = b.pop
+    v2 = b.pop
+    b.push(v1)
+    b.push(v2) 
+  end
+
+  def setlocal(b, oprnd)
+    v = b.pop
+    local_slot = b.gep(b.locals, oprnd.llvm)
+    b.store(v, local_slot)
+  end
+
+  def getlocal(b, oprnd)
+    local_slot = b.gep(b.locals, oprnd.llvm)
+    val = b.load(local_slot)
+    b.push(val)
+  end
+
+  def opt_plus(b, oprnd)
+    v1 = b.fix2int(b.pop)
+    v2 = b.fix2int(b.pop)
+    sum = b.add(v1, v2)
+    b.push(b.num2fix(sum))
+  end
+
+  def opt_minus(b, oprnd)
+    v1 = b.fix2int(b.pop)
+    v2 = b.fix2int(b.pop)
+    sum = b.sub(v2, v1)
+    b.push(b.num2fix(sum))
+  end
+
+  def opt_mult(b, oprnd)
+    v1 = b.fix2int(b.pop)
+    v2 = b.fix2int(b.pop)
+    mul = b.mul(v1, v2)
+    b.push(b.num2fix(mul))
+  end
+
+  def opt_aref(b, oprnd)
+    idx = b.fix2int(b.pop)
+    ary = b.pop
+    out = b.aref(ary, idx)
+    b.push(out)
+  end
+  
+  def opt_aset(b, oprnd)
+    set = b.pop
+    idx = b.fix2int(b.pop)
+    ary = b.pop
+    b.call(@rb_funcs[:rb_ary_store], ary, idx, set)
+    b.push(set)
+  end
+
+  def opt_length(b, oprnd)
+    recv  = b.pop
+    len = b.alen(recv)
+    len = b.num2fix(len)
+    b.push(len)
+  end
+  
+  def opt_lt(b, oprnd)
+    obj = b.pop
+    recv = b.pop
+    x = b.fix2int(recv)
+    y = b.fix2int(obj)
+    val = b.icmp_slt(x, y)
+    val = b.int_cast(val, LONG, false)
+    val = b.mul(val, 2.llvm)
+    b.push(val)
+  end
+  
+  def opt_gt(b, oprnd)
+    obj = b.pop
+    recv = b.pop
+    x = b.fix2int(recv)
+    y = b.fix2int(obj)
+    val = b.icmp_sgt(x, y)
+    val = b.int_cast(val, LONG, false)
+    val = b.mul(val, 2.llvm)
+    b.push(val)
+  end
+
+  def opt_ge(b, oprnd)
+    obj = b.pop
+    recv = b.pop
+    x = b.fix2int(recv)
+    y = b.fix2int(obj)
+    val = b.icmp_sge(x, y)
+    val = b.int_cast(val, LONG, false)
+    val = b.mul(val, 2.llvm)
+    b.push(val)
+  end
+
+  def getinstancevariable(b, oprnd)
+    id = b.call(@rb_funcs[:rb_to_id], oprnd.llvm)
+    v = b.call(@rb_funcs[:rb_ivar_get], get_self, id)
+    b.push(v)
+  end
+ 
+  def setinstancevariable(b, oprnd)
+    new_val = b.peek
+    id = b.call(@rb_funcs[:rb_to_id], oprnd.llvm)
+    b.call(@rb_funcs[:rb_ivar_set], get_self, id, new_val)
+  end
+
+  def newarray(b, oprnd)
+    ary = b.call(@rb_funcs[:rb_ary_new])
+    b.push(ary)
+  end
+ 
+  def newhash(b, oprnd)
+    hash = b.call(@rb_funcs[:rb_hash_new])
+    i = oprnd.llvm(Type::Int32Ty) # This is an integer not a fixnum
+
+    entry_block = @func.create_block
+    loop_block = @func.create_block
+    exit_block = @func.create_block
+
+    b.br(entry_block)
+    b.set_insert_point(entry_block)
+    cmp = b.icmp_sgt(i, 0.llvm(Type::Int32Ty)) 
+    b.cond_br(cmp, loop_block, exit_block)
+
+    b.set_insert_point(loop_block)
+    idx = b.phi(Type::Int32Ty)
+    idx.add_incoming(i, entry_block)
+    next_idx = b.sub(idx, 2.llvm(Type::Int32Ty))
+    idx.add_incoming(next_idx, loop_block)
+
+    n_1 = b.sub(idx, 1.llvm(Type::Int32Ty))
+    n_2 = b.sub(idx, 2.llvm(Type::Int32Ty))
+    b.call(@rb_funcs[:rb_hash_aset], hash, b.topn(n_1), b.topn(n_2))
+    
+    cmp = b.icmp_sgt(next_idx, 0.llvm(Type::Int32Ty))
+    b.cond_br(cmp, loop_block, exit_block)
+
+    b.set_insert_point(exit_block)
+    b.push(hash)
+  end
+
+  def send(b, oprnd)
+    recv = nil.immediate
+    id = b.call(@rb_funcs[:rb_to_id], :inspect.immediate)
+    argc = 0.llvm(Type::Int32Ty)
+    val = b.call(@rb_funcs[:rb_funcall2], recv, id, argc, b.stack)
+    b.push(val)
+  end
+end
+
+class RubyVM
+  def self.start
+    @module = LLVM::Module.new('ruby_vm')
+    ExecutionEngine.get(@module)
+
+    @rb_ary_new = @module.external_function('rb_ary_new', ftype(VALUE, []))
+    @rb_hash_new = @module.external_function('rb_hash_new', ftype(VALUE, []))
+    @rb_hash_aset = @module.external_function('rb_hash_aset', ftype(VALUE, [VALUE, VALUE, VALUE]))
+    @rb_ary_store = @module.external_function('rb_ary_store', ftype(VALUE, [VALUE, LONG, VALUE]))
+    @rb_to_id = @module.external_function('rb_to_id', ftype(VALUE, [VALUE]))
+    @rb_ivar_get = @module.external_function('rb_ivar_get', ftype(VALUE, [VALUE, ID]))
+    @rb_ivar_set = @module.external_function('rb_ivar_set', ftype(VALUE, [VALUE, ID, VALUE]))
+    @rb_funcall2 = @module.external_function('rb_funcall2', ftype(VALUE, [VALUE, ID, INT, P_VALUE]))
+
+    @rb_funcs = {
+      :rb_ary_new => @rb_ary_new,
+      :rb_hash_new => @rb_hash_new,
+      :rb_hash_aset => @rb_hash_aset,
+      :rb_ary_store => @rb_ary_store,
+      :rb_to_id => @rb_to_id,
+      :rb_ivar_get => @rb_ivar_get,
+      :rb_ivar_set => @rb_ivar_set,
+      :rb_funcall2 => @rb_funcall2
+    }
+
+    @func_n = 0
+  end
+
+  def self.ftype(ret, args)
+    Type.function(ret, args)
+  end
+
+  def self.call_bytecode(bytecode, farg)
+    f = compile_bytecode(bytecode)
+    ExecutionEngine.run_function(f, nil, farg)
+  end
+
+  def self.method_send(recv, compiled_method, farg = nil)
+    ExecutionEngine.run_function(compiled_method, recv, farg)
+  end
+
+  def self.compile_bytecode(bytecode) 
+    f = @module.get_or_insert_function("vm_func#{@func_n}", Type.function(VALUE, [VALUE, VALUE]))
+    @func_n += 1
+
+    get_self = f.arguments[0]
+
+    entry_block = f.create_block
+    b = entry_block.builder
+    Builder.set_globals(b)
+    b.push(f.arguments[1])
+
+    blocks = bytecode.map { f.create_block } 
+    exit_block = f.create_block
+    blocks << exit_block
+    b.br(blocks.first)
+
+    op_builder = OpCodeBuilder.new(@module, f, @rb_funcs)
+
+    bytecode.each_with_index do |opcode, i|
+      op, arg = opcode
+
+      block = blocks[i] 
+      b.set_insert_point(block)
+
+      case op
+      when :jump
+        b.br(blocks[arg])
+      when :branchif
+        v = b.pop
+        cmp = b.icmp_eq(v, 0.llvm)
+        b.cond_br(cmp, blocks[i+1], blocks[arg])
+      when :branchunless
+        v = b.pop
+        cmp = b.icmp_eq(v, 0.llvm)
+        b.cond_br(cmp, blocks[arg], blocks[i+1])
+      else
+        op_builder.__send__(op, b, arg)
+      end
+
+      if op != :jump && op != :branchif && op != :branchunless
+        b.br(blocks[i+1])
+      end
+    end
+
+    b = exit_block.builder
+    ret_val = b.pop
+    b.return(ret_val)
+
+    f
+  end
+end