inst_sel.rs 217 KB
Newer Older
1
use ast::ir::*;
2
use ast::ptr::*;
qinsoon's avatar
qinsoon committed
3
use ast::inst::*;
4
use ast::op;
qinsoon's avatar
qinsoon committed
5
use ast::op::*;
qinsoon's avatar
qinsoon committed
6
use ast::types;
qinsoon's avatar
qinsoon committed
7
use ast::types::*;
qinsoon's avatar
qinsoon committed
8
use vm::VM;
qinsoon's avatar
qinsoon committed
9
use runtime::mm;
10 11
use runtime::mm::objectmodel::OBJECT_HEADER_SIZE;
use runtime::mm::objectmodel::OBJECT_HEADER_OFFSET;
12 13 14 15
use runtime::ValueLocation;
use runtime::thread;
use runtime::entrypoints;
use runtime::entrypoints::RuntimeEntrypoint;
16 17

use compiler::CompilerPass;
18
use compiler::backend;
19
use compiler::backend::RegGroup;
qinsoon's avatar
qinsoon committed
20
use compiler::backend::PROLOGUE_BLOCK_NAME;
qinsoon's avatar
qinsoon committed
21 22 23
use compiler::backend::x86_64;
use compiler::backend::x86_64::CodeGenerator;
use compiler::backend::x86_64::ASMCodeGen;
qinsoon's avatar
qinsoon committed
24 25
use compiler::machine_code::CompiledFunction;
use compiler::frame::Frame;
26

27
use utils::math;
qinsoon's avatar
qinsoon committed
28
use utils::POINTER_SIZE;
29

30
use std::collections::HashMap;
qinsoon's avatar
qinsoon committed
31
use std::any::Any;
32

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
lazy_static! {
    pub static ref LONG_4_TYPE : P<MuType> = P(
        MuType::new(new_internal_id(), MuType_::mustruct(Mu("long_4"), vec![UINT32_TYPE.clone(); 4]))
    );

    pub static ref UITOFP_C0 : P<Value> = P(Value{
        hdr: MuEntityHeader::named(new_internal_id(), Mu("UITOFP_C0")),
        ty : LONG_4_TYPE.clone(),
        v  : Value_::Constant(Constant::List(vec![
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT32_TYPE.clone(),
                v : Value_::Constant(Constant::Int(1127219200u64))
            }),
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT32_TYPE.clone(),
                v : Value_::Constant(Constant::Int(1160773632u64))
            }),
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT32_TYPE.clone(),
                v : Value_::Constant(Constant::Int(0u64))
            }),
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT32_TYPE.clone(),
                v : Value_::Constant(Constant::Int(0u64))
            })
        ]))
    });

    pub static ref QUAD_2_TYPE : P<MuType> = P(
        MuType::new(new_internal_id(), MuType_::mustruct(Mu("quad_2"), vec![UINT64_TYPE.clone(); 2]))
    );

    pub static ref UITOFP_C1 : P<Value> = P(Value{
        hdr: MuEntityHeader::named(new_internal_id(), Mu("UITOFP_C1")),
        ty : QUAD_2_TYPE.clone(),
        v  : Value_::Constant(Constant::List(vec![
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT64_TYPE.clone(),
                v : Value_::Constant(Constant::Int(4841369599423283200u64))
            }),
            P(Value{
                hdr: MuEntityHeader::unnamed(new_internal_id()),
                ty: UINT64_TYPE.clone(),
                v : Value_::Constant(Constant::Int(4985484787499139072u64))
            })
        ]))
    });
85

qinsoon's avatar
qinsoon committed
86 87
    pub static ref FPTOUI_C_DOUBLE : P<Value> = P(Value{
        hdr: MuEntityHeader::named(new_internal_id(), Mu("FPTOUI_C_DOUBLE")),
88 89 90
        ty : UINT64_TYPE.clone(),
        v  : Value_::Constant(Constant::Int(4890909195324358656u64))
    });
qinsoon's avatar
qinsoon committed
91 92 93 94 95 96

    pub static ref FPTOUI_C_FLOAT : P<Value> = P(Value{
        hdr: MuEntityHeader::named(new_internal_id(), Mu("FPTOUI_C_FLOAT")),
        ty : UINT32_TYPE.clone(),
        v  : Value_::Constant(Constant::Int(1593835520u64))
    });
97 98
}

99 100
const INLINE_FASTPATH : bool = false;

101
pub struct InstructionSelection {
102
    name: &'static str,
103
    backend: Box<CodeGenerator>,
104 105

    current_fv_id: MuID,
qinsoon's avatar
qinsoon committed
106
    current_callsite_id: usize,
qinsoon's avatar
qinsoon committed
107 108
    current_frame: Option<Frame>,
    current_block: Option<MuName>,
qinsoon's avatar
qinsoon committed
109 110 111 112
    current_func_start: Option<ValueLocation>,
    // key: block id, val: callsite that names the block as exception block
    current_exn_callsites: HashMap<MuID, Vec<ValueLocation>>,
    // key: block id, val: block location
113 114 115 116
    current_exn_blocks: HashMap<MuID, ValueLocation>,

    current_constants: HashMap<MuID, P<Value>>,
    current_constants_locs: HashMap<MuID, P<Value>>
117 118
}

119
impl <'a> InstructionSelection {
qinsoon's avatar
qinsoon committed
120
    #[cfg(feature = "aot")]
121
    pub fn new() -> InstructionSelection {
122 123
        InstructionSelection{
            name: "Instruction Selection (x64)",
124
            backend: Box::new(ASMCodeGen::new()),
125

126
            current_fv_id: 0,
qinsoon's avatar
qinsoon committed
127
            current_callsite_id: 0,
qinsoon's avatar
qinsoon committed
128 129 130
            current_frame: None,
            current_block: None,
            current_func_start: None,
qinsoon's avatar
qinsoon committed
131 132
            // key: block id, val: callsite that names the block as exception block
            current_exn_callsites: HashMap::new(), 
133 134 135 136
            current_exn_blocks: HashMap::new(),

            current_constants: HashMap::new(),
            current_constants_locs: HashMap::new()
137 138
        }
    }
qinsoon's avatar
qinsoon committed
139 140 141 142 143

    #[cfg(feature = "jit")]
    pub fn new() -> InstructionSelection {
        unimplemented!()
    }
144 145
    
    // in this pass, we assume that
qinsoon's avatar
qinsoon committed
146
    // * we do not need to backup/restore caller-saved registers
147 148
    // if any of these assumption breaks, we will need to re-emit the code
    #[allow(unused_variables)]
149
    fn instruction_select(&mut self, node: &'a TreeNode, f_content: &FunctionContent, f_context: &mut FunctionContext, vm: &VM) {
150
        trace!("instsel on node#{} {}", node.id(), node);
qinsoon's avatar
qinsoon committed
151 152
        
        match node.v {
153 154
            TreeNode_::Instruction(ref inst) => {
                match inst.v {
qinsoon's avatar
qinsoon committed
155
                    Instruction_::Branch2{cond, ref true_dest, ref false_dest, true_prob} => {
qinsoon's avatar
qinsoon committed
156
                        trace!("instsel on BRANCH2");
157 158
                        // 'branch_if_true' == true, we emit cjmp the same as CmpOp  (je  for EQ, jne for NE)
                        // 'branch_if_true' == false, we emit opposite cjmp as CmpOp (jne for EQ, je  for NE)
159
                        let (fallthrough_dest, branch_dest, branch_if_true) = {
160 161 162 163 164
                            // get current block and next block in trace (fallthrough block)
                            let cur_block = f_content.get_block_by_name(self.current_block.as_ref().unwrap().clone());
                            let next_block_in_trace = cur_block.control_flow.get_hottest_succ().unwrap();

                            if next_block_in_trace == true_dest.target {
165 166 167
                                (true_dest, false_dest, false)
                            } else {
                                (false_dest, true_dest, true)
168
                            }
169
                        };
170
                        
171
                        let ref ops = inst.ops;
172
                        
173 174
                        self.process_dest(&ops, fallthrough_dest, f_content, f_context, vm);
                        self.process_dest(&ops, branch_dest, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
175
                        
176
                        let branch_target = f_content.get_block(branch_dest.target).name().unwrap();
177 178 179
    
                        let ref cond = ops[cond];
                        
qinsoon's avatar
qinsoon committed
180
                        if self.match_cmp_res(cond) {
181
                            trace!("emit cmp_res-branch2");
182
                            match self.emit_cmp_res(cond, f_content, f_context, vm) {
183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
                                op::CmpOp::EQ => {
                                    if branch_if_true {
                                        self.backend.emit_je(branch_target);
                                    } else {
                                        self.backend.emit_jne(branch_target);
                                    }
                                },
                                op::CmpOp::NE => {
                                    if branch_if_true {
                                        self.backend.emit_jne(branch_target);
                                    } else {
                                        self.backend.emit_je(branch_target);
                                    }
                                },
                                op::CmpOp::UGE => {
                                    if branch_if_true {
                                        self.backend.emit_jae(branch_target);
                                    } else {
                                        self.backend.emit_jb(branch_target);
                                    }
                                },
                                op::CmpOp::UGT => {
                                    if branch_if_true {
                                        self.backend.emit_ja(branch_target);
                                    } else {
                                        self.backend.emit_jbe(branch_target);
                                    }
                                },
                                op::CmpOp::ULE => {
                                    if branch_if_true {
                                        self.backend.emit_jbe(branch_target);
                                    } else {
                                        self.backend.emit_ja(branch_target);
                                    }
                                },
                                op::CmpOp::ULT => {
                                    if branch_if_true {
                                        self.backend.emit_jb(branch_target);
                                    } else {
                                        self.backend.emit_jae(branch_target);
                                    }
                                },
                                op::CmpOp::SGE => {
                                    if branch_if_true {
                                        self.backend.emit_jge(branch_target);
                                    } else {
                                        self.backend.emit_jl(branch_target);
                                    }
                                },
                                op::CmpOp::SGT => {
                                    if branch_if_true {
                                        self.backend.emit_jg(branch_target);
                                    } else {
                                        self.backend.emit_jle(branch_target);
                                    }
                                },
                                op::CmpOp::SLE => {
                                    if branch_if_true {
                                        self.backend.emit_jle(branch_target);
                                    } else {
                                        self.backend.emit_jg(branch_target);
                                    }
                                },
                                op::CmpOp::SLT => {
                                    if branch_if_true {
                                        self.backend.emit_jl(branch_target);
                                    } else {
                                        self.backend.emit_jge(branch_target);
                                    }
                                },
qinsoon's avatar
qinsoon committed
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298

                                // floating point

                                op::CmpOp::FOEQ | op::CmpOp::FUEQ => {
                                    if branch_if_true {
                                        self.backend.emit_je(branch_target);
                                    } else {
                                        self.backend.emit_jne(branch_target);
                                    }
                                },
                                op::CmpOp::FONE | op::CmpOp::FUNE => {
                                    if branch_if_true {
                                        self.backend.emit_jne(branch_target);
                                    } else {
                                        self.backend.emit_je(branch_target);
                                    }
                                },
                                op::CmpOp::FOGT | op::CmpOp::FUGT => {
                                    if branch_if_true {
                                        self.backend.emit_ja(branch_target);
                                    } else {
                                        self.backend.emit_jbe(branch_target);
                                    }
                                },
                                op::CmpOp::FOGE | op::CmpOp::FUGE => {
                                    if branch_if_true {
                                        self.backend.emit_jae(branch_target);
                                    } else {
                                        self.backend.emit_jb(branch_target);
                                    }
                                },
                                op::CmpOp::FOLT | op::CmpOp::FULT => {
                                    if branch_if_true {
                                        self.backend.emit_jb(branch_target);
                                    } else {
                                        self.backend.emit_jae(branch_target);
                                    }
                                },
                                op::CmpOp::FOLE | op::CmpOp::FULE => {
                                    if branch_if_true {
                                        self.backend.emit_jbe(branch_target);
                                    } else {
                                        self.backend.emit_ja(branch_target);
                                    }
                                },

qinsoon's avatar
qinsoon committed
299 300 301 302
                                _ => unimplemented!()
                            }
                        } else if self.match_ireg(cond) {
                            trace!("emit ireg-branch2");
303
                            
304
                            let cond_reg = self.emit_ireg(cond, f_content, f_context, vm);
305
                            
qinsoon's avatar
qinsoon committed
306
                            // emit: cmp cond_reg 1
qinsoon's avatar
qinsoon committed
307
                            self.backend.emit_cmp_imm_r(1, &cond_reg);
qinsoon's avatar
qinsoon committed
308
                            // emit: je #branch_dest
309 310 311 312 313
                            if branch_if_true {
                                self.backend.emit_je(branch_target);
                            } else {
                                self.backend.emit_jne(branch_target);
                            }
qinsoon's avatar
qinsoon committed
314 315
                        } else {
                            unimplemented!();
316
                        }
317
                    },
qinsoon's avatar
qinsoon committed
318 319

                    Instruction_::Select{cond, true_val, false_val} => {
qinsoon's avatar
qinsoon committed
320 321
                        use ast::op::CmpOp::*;

qinsoon's avatar
qinsoon committed
322
                        trace!("instsel on SELECT");
323
                        let ref ops = inst.ops;
qinsoon's avatar
qinsoon committed
324 325 326 327 328

                        let ref cond = ops[cond];
                        let ref true_val = ops[true_val];
                        let ref false_val = ops[false_val];

329 330 331 332 333 334 335 336 337 338 339 340 341
                        // generate compare
                        let cmpop = if self.match_cmp_res(cond) {
                            self.emit_cmp_res(cond, f_content, f_context, vm)
                        } else if self.match_ireg(cond) {
                            let tmp_cond = self.emit_ireg(cond, f_content, f_context, vm);
                            // emit: cmp cond_reg 1
                            self.backend.emit_cmp_imm_r(1, &tmp_cond);

                            EQ
                        } else {
                            panic!("expected cond to be ireg, found {}", cond)
                        };

qinsoon's avatar
qinsoon committed
342 343 344
                        if self.match_ireg(true_val) {
                            // moving integers/pointers
                            let tmp_res   = self.get_result_value(node);
qinsoon's avatar
qinsoon committed
345

346
                            // use cmov for 16/32/64bit integer
qinsoon's avatar
qinsoon committed
347
                            // use jcc  for 8 bit
348
                            // FIXME: could use 32bit register to implement 8bit select
qinsoon's avatar
qinsoon committed
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
                            match tmp_res.ty.get_int_length() {
                                // cmov
                                Some(len) if len > 8 => {
                                    let tmp_true  = self.emit_ireg(true_val, f_content, f_context, vm);
                                    let tmp_false = self.emit_ireg(false_val, f_content, f_context, vm);

                                    // mov tmp_false -> tmp_res
                                    self.backend.emit_mov_r_r(&tmp_res, &tmp_false);

                                    match cmpop {
                                        EQ  => self.backend.emit_cmove_r_r (&tmp_res, &tmp_true),
                                        NE  => self.backend.emit_cmovne_r_r(&tmp_res, &tmp_true),
                                        SGE => self.backend.emit_cmovge_r_r(&tmp_res, &tmp_true),
                                        SGT => self.backend.emit_cmovg_r_r (&tmp_res, &tmp_true),
                                        SLE => self.backend.emit_cmovle_r_r(&tmp_res, &tmp_true),
                                        SLT => self.backend.emit_cmovl_r_r (&tmp_res, &tmp_true),
                                        UGE => self.backend.emit_cmovae_r_r(&tmp_res, &tmp_true),
                                        UGT => self.backend.emit_cmova_r_r (&tmp_res, &tmp_true),
                                        ULE => self.backend.emit_cmovbe_r_r(&tmp_res, &tmp_true),
                                        ULT => self.backend.emit_cmovb_r_r (&tmp_res, &tmp_true),

                                        FOEQ | FUEQ => self.backend.emit_cmove_r_r (&tmp_res, &tmp_true),
                                        FONE | FUNE => self.backend.emit_cmovne_r_r(&tmp_res, &tmp_true),
                                        FOGT | FUGT => self.backend.emit_cmova_r_r (&tmp_res, &tmp_true),
                                        FOGE | FUGE => self.backend.emit_cmovae_r_r(&tmp_res, &tmp_true),
                                        FOLT | FULT => self.backend.emit_cmovb_r_r (&tmp_res, &tmp_true),
                                        FOLE | FULE => self.backend.emit_cmovbe_r_r(&tmp_res, &tmp_true),

                                        _ => unimplemented!()
                                    }
                                }
                                // jcc
                                _ => {
382 383 384
                                    let blk_true  = format!("{}_{}_select_true",  self.current_fv_id, node.id());
                                    let blk_false = format!("{}_{}_select_false", self.current_fv_id, node.id());
                                    let blk_end   = format!("{}_{}_select_end",   self.current_fv_id, node.id());
qinsoon's avatar
qinsoon committed
385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408

                                    // jump to blk_true if true
                                    match cmpop {
                                        EQ  => self.backend.emit_je (blk_true.clone()),
                                        NE  => self.backend.emit_jne(blk_true.clone()),
                                        SGE => self.backend.emit_jge(blk_true.clone()),
                                        SGT => self.backend.emit_jg (blk_true.clone()),
                                        SLE => self.backend.emit_jle(blk_true.clone()),
                                        SLT => self.backend.emit_jl (blk_true.clone()),
                                        UGE => self.backend.emit_jae(blk_true.clone()),
                                        UGT => self.backend.emit_ja (blk_true.clone()),
                                        ULE => self.backend.emit_jbe(blk_true.clone()),
                                        ULT => self.backend.emit_jb (blk_true.clone()),

                                        FOEQ | FUEQ => self.backend.emit_je (blk_true.clone()),
                                        FONE | FUNE => self.backend.emit_jne(blk_true.clone()),
                                        FOGT | FUGT => self.backend.emit_ja (blk_true.clone()),
                                        FOGE | FUGE => self.backend.emit_jae(blk_true.clone()),
                                        FOLT | FULT => self.backend.emit_jb (blk_true.clone()),
                                        FOLE | FULE => self.backend.emit_jbe(blk_true.clone()),

                                        _ => unimplemented!()
                                    }

409
                                    // finishing current block
qinsoon's avatar
qinsoon committed
410
                                    self.finish_block();
411 412

                                    // blk_false:
qinsoon's avatar
qinsoon committed
413
                                    self.start_block(blk_false.clone());
qinsoon's avatar
qinsoon committed
414 415 416 417 418 419
                                    // mov false result here
                                    self.emit_move_node_to_value(&tmp_res, &false_val, f_content, f_context, vm);
                                    // jmp to end
                                    self.backend.emit_jmp(blk_end.clone());

                                    // finishing current block
qinsoon's avatar
qinsoon committed
420
                                    self.finish_block();
qinsoon's avatar
qinsoon committed
421 422

                                    // blk_true:
qinsoon's avatar
qinsoon committed
423
                                    self.start_block(blk_true.clone());
qinsoon's avatar
qinsoon committed
424 425 426 427 428 429
                                    // mov true value -> result
                                    self.emit_move_node_to_value(&tmp_res, &true_val, f_content, f_context, vm);

                                    self.backend.end_block(blk_true.clone());

                                    // blk_end:
qinsoon's avatar
qinsoon committed
430
                                    self.finish_block();
qinsoon's avatar
qinsoon committed
431
                                }
qinsoon's avatar
qinsoon committed
432
                            }
433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
                        } else if self.match_fpreg(true_val) {
                            let tmp_res = self.get_result_value(node);

                            let blk_true  = format!("{}_{}_select_true",  self.current_fv_id, node.id());
                            let blk_false = format!("{}_{}_select_false", self.current_fv_id, node.id());
                            let blk_end   = format!("{}_{}_select_end",   self.current_fv_id, node.id());

                            // jump to blk_true if true
                            match cmpop {
                                EQ  => self.backend.emit_je (blk_true.clone()),
                                NE  => self.backend.emit_jne(blk_true.clone()),
                                SGE => self.backend.emit_jge(blk_true.clone()),
                                SGT => self.backend.emit_jg (blk_true.clone()),
                                SLE => self.backend.emit_jle(blk_true.clone()),
                                SLT => self.backend.emit_jl (blk_true.clone()),
                                UGE => self.backend.emit_jae(blk_true.clone()),
                                UGT => self.backend.emit_ja (blk_true.clone()),
                                ULE => self.backend.emit_jbe(blk_true.clone()),
                                ULT => self.backend.emit_jb (blk_true.clone()),

                                FOEQ | FUEQ => self.backend.emit_je (blk_true.clone()),
                                FONE | FUNE => self.backend.emit_jne(blk_true.clone()),
                                FOGT | FUGT => self.backend.emit_ja (blk_true.clone()),
                                FOGE | FUGE => self.backend.emit_jae(blk_true.clone()),
                                FOLT | FULT => self.backend.emit_jb (blk_true.clone()),
                                FOLE | FULE => self.backend.emit_jbe(blk_true.clone()),

                                _ => unimplemented!()
                            }

                            // finishing current block
qinsoon's avatar
qinsoon committed
464
                            self.finish_block();
465 466

                            // blk_false:
qinsoon's avatar
qinsoon committed
467
                            self.start_block(blk_false.clone());
468 469 470 471 472 473
                            // mov false result here
                            self.emit_move_node_to_value(&tmp_res, &false_val, f_content, f_context, vm);
                            // jmp to end
                            self.backend.emit_jmp(blk_end.clone());

                            // finishing current block
qinsoon's avatar
qinsoon committed
474
                            self.finish_block();
475 476

                            // blk_true:
qinsoon's avatar
qinsoon committed
477
                            self.start_block(blk_true.clone());
478 479
                            // mov true value -> result
                            self.emit_move_node_to_value(&tmp_res, &true_val, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
480
                            self.finish_block();
481 482

                            // blk_end:
qinsoon's avatar
qinsoon committed
483
                            self.start_block(blk_end.clone());
qinsoon's avatar
qinsoon committed
484 485 486 487 488
                        } else {
                            unimplemented!()
                        }
                    },

489
                    Instruction_::CmpOp(op, op1, op2) => {
qinsoon's avatar
qinsoon committed
490 491
                        use ast::op::CmpOp::*;

qinsoon's avatar
qinsoon committed
492
                        trace!("instsel on CMPOP");
493
                        let ref ops = inst.ops;
494 495 496
                        let ref op1 = ops[op1];
                        let ref op2 = ops[op2];

qinsoon's avatar
qinsoon committed
497
                        let tmp_res = self.get_result_value(node);
498 499 500
                        
                        debug_assert!(tmp_res.ty.get_int_length().is_some());
                        debug_assert!(tmp_res.ty.get_int_length().unwrap() == 1);
qinsoon's avatar
qinsoon committed
501

502
                        // set byte to result
qinsoon's avatar
qinsoon committed
503
                        match self.emit_cmp_res(node, f_content, f_context, vm) {
504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
                            EQ  => self.backend.emit_sete_r (&tmp_res),
                            NE  => self.backend.emit_setne_r(&tmp_res),
                            SGE => self.backend.emit_setge_r(&tmp_res),
                            SGT => self.backend.emit_setg_r (&tmp_res),
                            SLE => self.backend.emit_setle_r(&tmp_res),
                            SLT => self.backend.emit_setl_r (&tmp_res),
                            UGE => self.backend.emit_setae_r(&tmp_res),
                            UGT => self.backend.emit_seta_r (&tmp_res),
                            ULE => self.backend.emit_setbe_r(&tmp_res),
                            ULT => self.backend.emit_setb_r (&tmp_res),

                            FOEQ | FUEQ => self.backend.emit_sete_r (&tmp_res),
                            FONE | FUNE => self.backend.emit_setne_r(&tmp_res),
                            FOGT | FUGT => self.backend.emit_seta_r (&tmp_res),
                            FOGE | FUGE => self.backend.emit_setae_r(&tmp_res),
                            FOLT | FULT => self.backend.emit_setb_r (&tmp_res),
                            FOLE | FULE => self.backend.emit_setbe_r(&tmp_res),
qinsoon's avatar
qinsoon committed
521 522

                            _ => unimplemented!()
523 524 525
                        }
                    }

qinsoon's avatar
qinsoon committed
526
                    Instruction_::Branch1(ref dest) => {
qinsoon's avatar
qinsoon committed
527
                        trace!("instsel on BRANCH1");
528
                        let ref ops = inst.ops;
529
                                            
530
                        self.process_dest(&ops, dest, f_content, f_context, vm);
531
                        
532
                        let target = f_content.get_block(dest.target).name().unwrap();
qinsoon's avatar
qinsoon committed
533
                        
qinsoon's avatar
qinsoon committed
534
                        trace!("emit branch1");
535
                        // jmp
qinsoon's avatar
qinsoon committed
536
                        self.backend.emit_jmp(target);
537
                    },
qinsoon's avatar
qinsoon committed
538 539

                    Instruction_::Switch{cond, ref default, ref branches} => {
qinsoon's avatar
qinsoon committed
540
                        trace!("instsel on SWITCH");
541
                        let ref ops = inst.ops;
qinsoon's avatar
qinsoon committed
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560

                        let ref cond = ops[cond];

                        if self.match_ireg(cond) {
                            let tmp_cond = self.emit_ireg(cond, f_content, f_context, vm);

                            // emit each branch
                            for &(case_op_index, ref case_dest) in branches {
                                let ref case_op = ops[case_op_index];

                                // process dest
                                self.process_dest(&ops, case_dest, f_content, f_context, vm);

                                let target = f_content.get_block(case_dest.target).name().unwrap();

                                if self.match_iimm(case_op) {
                                    let imm = self.node_iimm_to_i32(case_op);

                                    // cmp case cond
qinsoon's avatar
qinsoon committed
561
                                    self.backend.emit_cmp_imm_r(imm, &tmp_cond);
qinsoon's avatar
qinsoon committed
562 563 564 565 566 567
                                    // je dest
                                    self.backend.emit_je(target);
                                } else if self.match_ireg(case_op) {
                                    let tmp_case_op = self.emit_ireg(case_op, f_content, f_context, vm);

                                    // cmp case cond
qinsoon's avatar
qinsoon committed
568
                                    self.backend.emit_cmp_r_r(&tmp_case_op, &tmp_cond);
qinsoon's avatar
qinsoon committed
569 570 571 572 573
                                    // je dest
                                    self.backend.emit_je(target);
                                } else {
                                    panic!("expecting ireg cond to be either iimm or ireg: {}", cond);
                                }
574 575 576

                                self.finish_block();
                                self.start_block(format!("{}_switch_not_met_case_{}", node.id(), case_op_index));
qinsoon's avatar
qinsoon committed
577 578 579 580 581 582 583 584 585 586 587
                            }

                            // emit default
                            self.process_dest(&ops, default, f_content, f_context, vm);
                            
                            let default_target = f_content.get_block(default.target).name().unwrap();
                            self.backend.emit_jmp(default_target);
                        } else {
                            panic!("expecting cond in switch to be ireg: {}", cond);
                        }
                    }
588
                    
qinsoon's avatar
qinsoon committed
589
                    Instruction_::ExprCall{ref data, is_abort} => {
qinsoon's avatar
qinsoon committed
590 591
                        trace!("instsel on EXPRCALL");

qinsoon's avatar
qinsoon committed
592 593
                        if is_abort {
                            unimplemented!()
594
                        }
595
                        
qinsoon's avatar
qinsoon committed
596 597 598 599 600
                        self.emit_mu_call(
                            inst, // inst: &Instruction,
                            data, // calldata: &CallData,
                            None, // resumption: Option<&ResumptionData>,
                            node, // cur_node: &TreeNode, 
601
                            f_content, f_context, vm);
602 603
                    },
                    
qinsoon's avatar
qinsoon committed
604
                    Instruction_::Call{ref data, ref resume} => {
qinsoon's avatar
qinsoon committed
605 606
                        trace!("instsel on CALL");

qinsoon's avatar
qinsoon committed
607 608 609 610 611 612
                        self.emit_mu_call(
                            inst, 
                            data, 
                            Some(resume), 
                            node, 
                            f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
613 614 615
                    },

                    Instruction_::ExprCCall{ref data, is_abort} => {
qinsoon's avatar
qinsoon committed
616 617
                        trace!("instsel on EXPRCCALL");

qinsoon's avatar
qinsoon committed
618 619 620 621 622 623 624 625
                        if is_abort {
                            unimplemented!()
                        }

                        self.emit_c_call_ir(inst, data, None, node, f_content, f_context, vm);
                    }

                    Instruction_::CCall{ref data, ref resume} => {
qinsoon's avatar
qinsoon committed
626 627
                        trace!("instsel on CCALL");

qinsoon's avatar
qinsoon committed
628
                        self.emit_c_call_ir(inst, data, Some(resume), node, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
629 630
                    }
                    
631
                    Instruction_::Return(_) => {
qinsoon's avatar
qinsoon committed
632 633
                        trace!("instsel on RETURN");

634
                        self.emit_common_epilogue(inst, f_content, f_context, vm);
635
                        
qinsoon's avatar
qinsoon committed
636
                        self.backend.emit_ret();
637 638
                    },
                    
qinsoon's avatar
qinsoon committed
639
                    Instruction_::BinOp(op, op1, op2) => {
qinsoon's avatar
qinsoon committed
640 641
                        trace!("instsel on BINOP");

qinsoon's avatar
qinsoon committed
642 643
                        self.emit_binop(node, inst, op, op1, op2, f_content, f_context, vm);
                    },
qinsoon's avatar
qinsoon committed
644

qinsoon's avatar
qinsoon committed
645 646
                    Instruction_::BinOpWithStatus(op, status, op1, op2) => {
                        trace!("instsel on BINOP_STATUS");
qinsoon's avatar
qinsoon committed
647

qinsoon's avatar
qinsoon committed
648
                        self.emit_binop(node, inst, op, op1, op2, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
649

qinsoon's avatar
qinsoon committed
650 651
                        let values = inst.value.as_ref().unwrap();
                        let mut status_value_index = 1;
qinsoon's avatar
qinsoon committed
652

653
                        // status flags only works with int operations
654
                        if RegGroup::get_from_value(&values[0]) == RegGroup::GPR {
655 656 657 658 659 660 661
                            // negative flag
                            if status.flag_n {
                                let tmp_status = values[status_value_index].clone();
                                status_value_index += 1;

                                self.backend.emit_sets_r8(&tmp_status);
                            }
qinsoon's avatar
qinsoon committed
662

663 664 665 666
                            // zero flag
                            if status.flag_z {
                                let tmp_status = values[status_value_index].clone();
                                status_value_index += 1;
qinsoon's avatar
qinsoon committed
667

668 669
                                self.backend.emit_setz_r8(&tmp_status);
                            }
qinsoon's avatar
qinsoon committed
670

671 672 673 674
                            // unsigned overflow
                            if status.flag_c {
                                let tmp_status = values[status_value_index].clone();
                                status_value_index += 1;
qinsoon's avatar
qinsoon committed
675

676 677 678 679 680
                                match op {
                                    BinOp::Add | BinOp::Sub | BinOp::Mul => {
                                        self.backend.emit_setb_r8(&tmp_status);
                                    }
                                    _ => panic!("Only Add/Sub/Mul has #C flag")
qinsoon's avatar
qinsoon committed
681 682
                                }
                            }
qinsoon's avatar
qinsoon committed
683

684 685 686
                            // signed overflow
                            if status.flag_v {
                                let tmp_status = values[status_value_index].clone();
qinsoon's avatar
qinsoon committed
687

688 689 690 691 692
                                match op {
                                    BinOp::Add | BinOp::Sub | BinOp::Mul => {
                                        self.backend.emit_seto_r8(&tmp_status);
                                    }
                                    _ => panic!("Only Add/Sub/Mul has #V flag")
qinsoon's avatar
qinsoon committed
693 694
                                }
                            }
695 696
                        }
                    }
qinsoon's avatar
qinsoon committed
697 698

                    Instruction_::ConvOp{operation, ref from_ty, ref to_ty, operand} => {
qinsoon's avatar
qinsoon committed
699 700
                        trace!("instsel on CONVOP");

701
                        let ref ops = inst.ops;
qinsoon's avatar
qinsoon committed
702 703 704 705 706

                        let ref op = ops[operand];

                        match operation {
                            op::ConvOp::TRUNC => {
qinsoon's avatar
qinsoon committed
707 708 709
                                let tmp_res = self.get_result_value(node);
                                let to_ty_size = vm.get_backend_type_info(tmp_res.ty.id()).size;

qinsoon's avatar
qinsoon committed
710 711
                                if self.match_ireg(op) {
                                    let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
712

qinsoon's avatar
qinsoon committed
713
                                    // mov op -> result
qinsoon's avatar
qinsoon committed
714 715 716 717 718 719 720 721 722 723 724 725 726 727
                                    match to_ty_size {
                                        1 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT8_TYPE.clone())}),
                                        2 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT16_TYPE.clone())}),
                                        4 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT32_TYPE.clone())}),
                                        8 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT64_TYPE.clone())}),
                                        _  => panic!("unsupported int size: {}", to_ty_size)
                                    }
                                } else if self.match_ireg_ex(op) {
                                    let (op_l, op_h) = self.emit_ireg_ex(op, f_content, f_context, vm);

                                    match to_ty_size {
                                        1 | 2 => self.backend.emit_movz_r_r(unsafe {&tmp_res.as_type(UINT32_TYPE.clone())}, &op_l),
                                        4 | 8 => self.backend.emit_mov_r_r(&tmp_res, &op_l),
                                        _ => panic!("unsupported int size: {}", to_ty_size)
728
                                    }
qinsoon's avatar
qinsoon committed
729 730 731 732 733
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
                                }
                            }
                            op::ConvOp::ZEXT => {
734 735 736
                                if self.match_ireg(op) {
                                    let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
                                    let tmp_res = self.get_result_value(node);
qinsoon's avatar
qinsoon committed
737

738 739 740
                                    // movz op -> result
                                    let from_ty_size = vm.get_backend_type_info(from_ty.id()).size;
                                    let to_ty_size   = vm.get_backend_type_info(to_ty.id()).size;
qinsoon's avatar
qinsoon committed
741

742
                                    if from_ty_size != to_ty_size {
qinsoon's avatar
qinsoon committed
743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
                                        match (from_ty_size, to_ty_size) {
                                            // int32 to int64
                                            (4, 8) => {
                                                // zero extend from 32 bits to 64 bits is a mov instruction
                                                // x86 does not have movzlq (32 to 64)

                                                // tmp_op is int32, but tmp_res is int64
                                                // we want to force a 32-to-32 mov, so high bits of the destination will be zeroed

                                                let tmp_res32 = unsafe {tmp_res.as_type(UINT32_TYPE.clone())};

                                                self.backend.emit_mov_r_r(&tmp_res32, &tmp_op);
                                            }
                                            // any int to int128
                                            (_, 16) => {
                                                let (res_l, res_h) = self.split_int128(&tmp_res, f_context, vm);

                                                self.backend.emit_mov_r_r(&res_l, unsafe {&tmp_op.as_type(UINT64_TYPE.clone())});
                                                self.backend.emit_mov_r_imm(&res_h, 0);
                                            }
                                            // else
                                            _ => {
                                                self.backend.emit_movz_r_r(&tmp_res, &tmp_op);
                                            }
qinsoon's avatar
qinsoon committed
767
                                        }
qinsoon's avatar
qinsoon committed
768
                                    } else {
769
                                        self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
qinsoon's avatar
qinsoon committed
770
                                    }
771 772
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
qinsoon's avatar
qinsoon committed
773 774 775
                                }
                            },
                            op::ConvOp::SEXT => {
776 777 778
                                if self.match_ireg(op) {
                                    let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
                                    let tmp_res = self.get_result_value(node);
qinsoon's avatar
qinsoon committed
779

780 781 782
                                    // movs op -> result
                                    let from_ty_size = vm.get_backend_type_info(from_ty.id()).size;
                                    let to_ty_size   = vm.get_backend_type_info(to_ty.id()).size;
qinsoon's avatar
qinsoon committed
783

784
                                    if from_ty_size != to_ty_size {
qinsoon's avatar
qinsoon committed
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811
                                        match (from_ty_size, to_ty_size) {
                                            // int64 to int128
                                            (8, 16) => {
                                                let (res_l, res_h) = self.split_int128(&tmp_res, f_context, vm);

                                                // mov tmp_op -> res_h
                                                // sar res_h 63
                                                self.backend.emit_mov_r_r(&res_h, &tmp_op);
                                                self.backend.emit_sar_r_imm8(&res_h, 63i8);

                                                // mov tmp_op -> res_l
                                                self.backend.emit_mov_r_r(&res_l, &tmp_op);
                                            }
                                            // int32 to int128
                                            (_, 16) => {
                                                let (res_l, res_h) = self.split_int128(&tmp_res, f_context, vm);

                                                // movs tmp_op -> res_l
                                                self.backend.emit_movs_r_r(&res_l, &tmp_op);

                                                // mov res_l -> res_h
                                                // sar res_h 63
                                                self.backend.emit_mov_r_r(&res_h, unsafe {&res_l.as_type(UINT32_TYPE.clone())});
                                                self.backend.emit_sar_r_imm8(&res_h, 63i8);
                                            }
                                            _ => self.backend.emit_movs_r_r(&tmp_res, &tmp_op)
                                        }
qinsoon's avatar
qinsoon committed
812
                                    } else {
813
                                        self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
qinsoon's avatar
qinsoon committed
814
                                    }
815 816
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op)
qinsoon's avatar
qinsoon committed
817 818
                                }
                            }
819 820 821 822 823 824
                            op::ConvOp::REFCAST | op::ConvOp::PTRCAST => {
                                // just a mov (and hopefully reg alloc will coalesce it)
                                let tmp_res = self.get_result_value(node);

                                if self.match_ireg(op) {
                                    let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
825
                                    self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
826 827 828 829
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op)
                                }
                            }
qinsoon's avatar
qinsoon committed
830 831 832
                            op::ConvOp::SITOFP => {
                                let tmp_res = self.get_result_value(node);

833 834 835 836 837 838 839
                                assert!(self.match_ireg(op), "unexpected op (expected ireg): {}", op);
                                let tmp_op = self.emit_ireg(op, f_content, f_context, vm);

                                match to_ty.v {
                                    MuType_::Double => self.backend.emit_cvtsi2sd_f64_r(&tmp_res, &tmp_op),
                                    MuType_::Float  => self.backend.emit_cvtsi2ss_f32_r(&tmp_res, &tmp_op),
                                    _ => panic!("expecing fp type as to type in SITOFP, found {}", to_ty)
qinsoon's avatar
qinsoon committed
840 841 842 843 844
                                }
                            }
                            op::ConvOp::FPTOSI => {
                                let tmp_res = self.get_result_value(node);

845 846 847 848
                                assert!(self.match_fpreg(op), "unexpected op (expected fpreg): {}", op);
                                let tmp_op = self.emit_fpreg(op, f_content, f_context, vm);

                                if from_ty.is_double() {
qinsoon's avatar
qinsoon committed
849
                                    self.backend.emit_cvtsd2si_r_f64(&tmp_res, &tmp_op);
850 851
                                } else if from_ty.is_float() {
                                    self.backend.emit_cvtss2si_r_f32(&tmp_res, &tmp_op);
qinsoon's avatar
qinsoon committed
852
                                } else {
853
                                    panic!("expected fp type as from type in FPTOSI, found {}", to_ty)
qinsoon's avatar
qinsoon committed
854 855
                                }
                            }
856 857 858
                            op::ConvOp::UITOFP => {
                                let tmp_res = self.get_result_value(node);

859 860
                                assert!(self.match_ireg(op), "unexpected op (expected ireg): {}", op);
                                let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
861

862
                                let op_ty_size = vm.get_backend_type_info(tmp_op.ty.id()).size;
863

qinsoon's avatar
qinsoon committed
864 865 866 867 868
                                if to_ty.is_double() {
                                    match op_ty_size {
                                        8 => {
                                            // movd/movq op -> res
                                            self.backend.emit_mov_fpr_r64(&tmp_res, &tmp_op);
869

qinsoon's avatar
qinsoon committed
870 871 872 873
                                            // punpckldq UITOFP_C0, tmp_res -> tmp_res
                                            // (interleaving low bytes: xmm = xmm[0] mem[0] xmm[1] mem[1]
                                            let mem_c0 = self.get_mem_for_const(UITOFP_C0.clone(), vm);
                                            self.backend.emit_punpckldq_f64_mem128(&tmp_res, &mem_c0);
874

qinsoon's avatar
qinsoon committed
875 876 877
                                            // subpd UITOFP_C1, tmp_res -> tmp_res
                                            let mem_c1 = self.get_mem_for_const(UITOFP_C1.clone(), vm);
                                            self.backend.emit_subpd_f64_mem128(&tmp_res, &mem_c1);
878

qinsoon's avatar
qinsoon committed
879 880 881 882 883
                                            // haddpd tmp_res, tmp_res -> tmp_res
                                            self.backend.emit_haddpd_f64_f64(&tmp_res, &tmp_res);
                                        }
                                        4 => {
                                            let tmp = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
884

qinsoon's avatar
qinsoon committed
885 886 887
                                            // movl op -> tmp(32)
                                            let tmp32 = unsafe { tmp.as_type(UINT32_TYPE.clone()) };
                                            self.backend.emit_mov_r_r(&tmp32, &tmp_op);
888

qinsoon's avatar
qinsoon committed
889 890 891 892 893 894 895 896
                                            // cvtsi2sd %tmp(64) -> %tmp_res
                                            self.backend.emit_cvtsi2sd_f64_r(&tmp_res, &tmp);
                                        }
                                        2 | 1 => {
                                            let tmp_op32 = unsafe { tmp_op.as_type(UINT32_TYPE.clone()) };
                                            self.backend.emit_cvtsi2sd_f64_r(&tmp_res, &tmp_op32);
                                        }
                                        _ => panic!("not implemented int length {}", op_ty_size)
897
                                    }
qinsoon's avatar
qinsoon committed
898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
                                } else if to_ty.is_float() {
                                    match op_ty_size {
                                        8 => {
                                            // movl %tmp_op -> %tmp1
                                            let tmp1 = self.make_temporary(f_context, UINT32_TYPE.clone(), vm);
                                            self.backend.emit_mov_r_r(&tmp1, unsafe {&tmp_op.as_type(UINT32_TYPE.clone())});

                                            // andl %tmp1 $1 -> %tmp1
                                            self.backend.emit_and_r_imm(&tmp1, 1);

                                            // testq %tmp_op %tmp_op
                                            self.backend.emit_test_r_r(&tmp_op, &tmp_op);

                                            let blk_if_signed     = format!("{}_{}_uitofp_float_if_signed", self.current_fv_id, node.id());
                                            let blk_if_not_signed = format!("{}_{}_uitofp_float_if_not_signed", self.current_fv_id, node.id());
                                            let blk_done          = format!("{}_{}_uitofp_float_done", self.current_fv_id, node.id());

                                            // js %if_signed
                                            self.backend.emit_js(blk_if_signed.clone());
                                            self.finish_block();

                                            // blk_if_not_signed:
                                            self.start_block(blk_if_not_signed);

                                            // cvtsi2ss %tmp_op -> %tmp_res
                                            self.backend.emit_cvtsi2ss_f32_r(&tmp_res, &tmp_op);

                                            // jmp blk_done
                                            self.backend.emit_jmp(blk_done.clone());
                                            self.finish_block();

                                            // blk_if_signed:
                                            self.start_block(blk_if_signed);

                                            // shr %tmp_op $1 -> %tmp_op
                                            self.backend.emit_shr_r_imm8(&tmp_op, 1);

                                            // or %tmp_op %tmp1 -> %tmp1
                                            self.backend.emit_or_r_r(unsafe {&tmp1.as_type(UINT64_TYPE.clone())}, &tmp_op);

                                            // cvtsi2ss %tmp1 -> %tmp_res
                                            self.backend.emit_cvtsi2ss_f32_r(&tmp_res, &tmp1);

                                            // addss %tmp_res %tmp_res -> %tmp_res
                                            self.backend.emit_addss_f32_f32(&tmp_res, &tmp_res);
                                            self.finish_block();

                                            self.start_block(blk_done);
                                        }
                                        4 => {
                                            // movl %tmp_op -> %tmp1
                                            let tmp1 = self.make_temporary(f_context, UINT32_TYPE.clone(), vm);
                                            self.backend.emit_mov_r_r(&tmp1, &tmp_op);

                                            // cvtsi2ssq %tmp1(64) -> %tmp_res
                                            self.backend.emit_cvtsi2ss_f32_r(&tmp_res, unsafe {&tmp1.as_type(UINT64_TYPE.clone())});
                                        }
                                        2 | 1 => {
                                            let tmp_op32 = unsafe {tmp_op.as_type(UINT32_TYPE.clone())};

                                            // cvtsi2ss %tmp_op32 -> %tmp_res
                                            self.backend.emit_cvtsi2ss_f32_r(&tmp_res, &tmp_op32);
                                        }
                                        _ => panic!("not implemented int length {}", op_ty_size)
962
                                    }
qinsoon's avatar
qinsoon committed
963 964
                                } else {
                                    panic!("expect double or float")
965 966 967 968 969
                                }
                            }
                            op::ConvOp::FPTOUI => {
                                let tmp_res = self.get_result_value(node);

970
                                assert!(self.match_fpreg(op), "unexpected op (expected fpreg): {}", op);
971
                                let tmp_op = self.emit_fpreg(op, f_content, f_context, vm);
972
                                let res_ty_size = vm.get_backend_type_info(tmp_res.ty.id()).size;
973

qinsoon's avatar
qinsoon committed
974 975 976 977 978
                                if from_ty.is_double() {
                                    match res_ty_size {
                                        8 => {
                                            let tmp1 = self.make_temporary(f_context, DOUBLE_TYPE.clone(), vm);
                                            let tmp2 = self.make_temporary(f_context, DOUBLE_TYPE.clone(), vm);
979

qinsoon's avatar
qinsoon committed
980 981 982
                                            // movsd FPTOUI_C_DOUBLE -> %tmp1
                                            let mem_c = self.get_mem_for_const(FPTOUI_C_DOUBLE.clone(), vm);
                                            self.backend.emit_movsd_f64_mem64(&tmp1, &mem_c);
983

qinsoon's avatar
qinsoon committed
984 985
                                            // movapd %tmp_op -> %tmp2
                                            self.backend.emit_movapd_f64_f64(&tmp2, &tmp_op);
986

qinsoon's avatar
qinsoon committed
987 988
                                            // subsd %tmp1, %tmp2 -> %tmp2
                                            self.backend.emit_subsd_f64_f64(&tmp2, &tmp1);
989

qinsoon's avatar
qinsoon committed
990 991
                                            // cvttsd2si %tmp2 -> %tmp_res
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res, &tmp2);
992

qinsoon's avatar
qinsoon committed
993 994 995
                                            let tmp_const = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
                                            // mov 0x8000000000000000 -> %tmp_const
                                            self.backend.emit_mov_r64_imm64(&tmp_const, -9223372036854775808i64);
996

qinsoon's avatar
qinsoon committed
997 998
                                            // xor %tmp_res, %tmp_const -> %tmp_const
                                            self.backend.emit_xor_r_r(&tmp_const, &tmp_res);
qinsoon's avatar
qinsoon committed
999

qinsoon's avatar
qinsoon committed
1000 1001
                                            // cvttsd2si %tmp_op -> %tmp_res
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res, &tmp_op);
1002

qinsoon's avatar
qinsoon committed
1003 1004
                                            // ucomisd %tmp_op %tmp1
                                            self.backend.emit_ucomisd_f64_f64(&tmp1, &tmp_op);
1005

qinsoon's avatar
qinsoon committed
1006 1007 1008 1009 1010
                                            // cmovaeq %tmp_const -> %tmp_res
                                            self.backend.emit_cmovae_r_r(&tmp_res, &tmp_const);
                                        }
                                        4 => {
                                            let tmp_res64 = unsafe { tmp_res.as_type(UINT64_TYPE.clone()) };
1011

qinsoon's avatar
qinsoon committed
1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
                                            // cvttsd2si %tmp_op -> %tmp_res(64)
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res64, &tmp_op);
                                        }
                                        2 | 1 => {
                                            let tmp_res32 = unsafe { tmp_res.as_type(UINT32_TYPE.clone()) };

                                            // cvttsd2si %tmp_op -> %tmp_res(32)
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res32, &tmp_op);

                                            // movz %tmp_res -> %tmp_res(32)
                                            self.backend.emit_movz_r_r(&tmp_res32, &tmp_res);
                                        }
                                        _ => panic!("not implemented int length {}", res_ty_size)
1025
                                    }
qinsoon's avatar
qinsoon committed
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
                                } else if from_ty.is_float() {
                                    match res_ty_size {
                                        8 => {
                                            let tmp1 = self.make_temporary(f_context, FLOAT_TYPE.clone(), vm);
                                            let tmp2 = self.make_temporary(f_context, FLOAT_TYPE.clone(), vm);

                                            // movss FPTOUI_C_FLOAT -> %tmp1
                                            let mem_c = self.get_mem_for_const(FPTOUI_C_FLOAT.clone(), vm);
                                            self.backend.emit_movss_f32_mem32(&tmp1, &mem_c);

                                            // movaps %tmp_op -> %tmp2
                                            self.backend.emit_movaps_f32_f32(&tmp2, &tmp_op);

                                            // subss %tmp1, %tmp2 -> %tmp2
                                            self.backend.emit_subss_f32_f32(&tmp2, &tmp1);

                                            // cvttss2si %tmp2 -> %tmp_res
                                            self.backend.emit_cvttss2si_r_f32(&tmp_res, &tmp2);

                                            let tmp_const = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
                                            // mov 0x8000000000000000 -> %tmp_const
                                            self.backend.emit_mov_r64_imm64(&tmp_const, -9223372036854775808i64);

                                            // xor %tmp_res, %tmp_const -> %tmp_const
                                            self.backend.emit_xor_r_r(&tmp_const, &tmp_res);
1051

qinsoon's avatar
qinsoon committed
1052 1053
                                            // cvttss2si %tmp_op -> %tmp_res
                                            self.backend.emit_cvttss2si_r_f32(&tmp_res, &tmp_op);
1054

qinsoon's avatar
qinsoon committed
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076
                                            // ucomiss %tmp_op %tmp1
                                            self.backend.emit_ucomiss_f32_f32(&tmp1, &tmp_op);

                                            // cmovaeq %tmp_const -> %tmp_res
                                            self.backend.emit_cmovae_r_r(&tmp_res, &tmp_const);
                                        }
                                        4 => {
                                            let tmp_res64 = unsafe { tmp_res.as_type(UINT64_TYPE.clone())};

                                            // cvttss2si %tmp_op -> %tmp_res(64)
                                            self.backend.emit_cvttss2si_r_f32(&tmp_res64, &tmp_op);
                                        }
                                        2 | 1 => {
                                            let tmp_res32 = unsafe {tmp_res.as_type(UINT32_TYPE.clone())};

                                            // cvttss2si %tmp_op -> %tmp_res(32)
                                            self.backend.emit_cvttss2si_r_f32(&tmp_res32, &tmp_op);

                                            // movz %tmp_res(32) -> %tmp_res
                                            self.backend.emit_movz_r_r(&tmp_res32, &tmp_res);
                                        }
                                        _ => panic!("not implemented int length {}", res_ty_size)
1077
                                    }
qinsoon's avatar
qinsoon committed
1078 1079
                                } else {
                                    panic!("expect double or float")
1080 1081
                                }
                            }
qinsoon's avatar
qinsoon committed
1082 1083 1084
                            _ => unimplemented!()
                        }
                    }
1085
                    
1086 1087
                    // load on x64 generates mov inst (no matter what order is specified)
                    // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
1088
                    Instruction_::Load{is_ptr, order, mem_loc} => {
qinsoon's avatar
qinsoon committed
1089 1090
                        trace!("instsel on LOAD");

1091
                        let ref ops = inst.ops;
1092
                        let ref loc_op = ops[mem_loc];
1093 1094 1095 1096 1097 1098
                        
                        // check order
                        match order {
                            MemoryOrder::Relaxed 
                            | MemoryOrder::Consume 
                            | MemoryOrder::Acquire
1099 1100
                            | MemoryOrder::SeqCst
                            | MemoryOrder::NotAtomic => {},
1101 1102
                            _ => panic!("didnt expect order {:?} with store inst", order)
                        }                        
1103

1104
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1105
                        let res_temp = self.get_result_value(node);
1106

qinsoon's avatar
qinsoon committed
1107 1108
                        if self.match_ireg(node) {
                            self.backend.emit_mov_r_mem(&res_temp, &resolved_loc);
qinsoon's avatar
qinsoon committed
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118
                        } else if self.match_ireg_ex(node) {
                            let (res_l, res_h) = self.split_int128(&res_temp, f_context, vm);

                            // load lower half
                            self.backend.emit_mov_r_mem(&res_l, &resolved_loc);

                            // shift ptr, and load higher half
                            let loc = self.addr_const_offset_adjust(resolved_loc.extract_memory_location().unwrap(), POINTER_SIZE as u64, vm);
                            let val_loc = self.make_value_memory_location(loc, vm);
                            self.backend.emit_mov_r_mem(&res_h, &val_loc);
qinsoon's avatar
qinsoon committed
1119 1120 1121 1122 1123 1124 1125 1126 1127
                        } else if self.match_fpreg(node) {
                            match res_temp.ty.v {
                                MuType_::Double => self.backend.emit_movsd_f64_mem64(&res_temp, &resolved_loc),
                                MuType_::Float  => self.backend.emit_movss_f32_mem32(&res_temp, &resolved_loc),
                                _ => panic!("expect double or float")
                            }
                        } else {
                            unimplemented!()
                        }
1128 1129 1130
                    }
                    
                    Instruction_::Store{is_ptr, order, mem_loc, value} => {
qinsoon's avatar
qinsoon committed
1131 1132
                        trace!("instsel on STORE");

1133
                        let ref ops = inst.ops;
1134 1135 1136 1137 1138
                        let ref loc_op = ops[mem_loc];
                        let ref val_op = ops[value];
                        
                        let generate_plain_mov : bool = {
                            match order {
1139 1140 1141
                                MemoryOrder::Relaxed
                                | MemoryOrder::Release
                                | MemoryOrder::NotAtomic => true,
1142 1143 1144 1145 1146
                                MemoryOrder::SeqCst => false,
                                _ => panic!("didnt expect order {:?} with store inst", order)
                            }
                        };
                        
1147 1148 1149
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, f_content, f_context, vm);

                        if self.match_iimm(val_op) {
1150
                            let (val, len) = self.node_iimm_to_i32_with_len(val_op);
1151
                            if generate_plain_mov {
1152
                                self.backend.emit_mov_mem_imm(&resolved_loc, val, len);
1153 1154 1155
                            } else {
                                unimplemented!()
                            }
1156 1157
                        } else if self.match_ireg(val_op) {
                            let val = self.emit_ireg(val_op, f_content, f_context, vm);
1158
                            if generate_plain_mov {
qinsoon's avatar
qinsoon committed
1159
                                self.backend.emit_mov_mem_r(&resolved_loc, &val);
1160 1161 1162
                            } else {
                                unimplemented!()
                            }
qinsoon's avatar
qinsoon committed
1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
                        } else if self.match_ireg_ex(val_op) {
                            let (val_l, val_h) = self.emit_ireg_ex(val_op, f_content, f_context, vm);
                            if generate_plain_mov {
                                // store lower half
                                self.backend.emit_mov_mem_r(&resolved_loc, &val_l);

                                // shift pointer, and store higher hal
                                let loc = self.addr_const_offset_adjust(resolved_loc.extract_memory_location().unwrap(),
                                                                        POINTER_SIZE as u64, vm);
                                let loc_val = self.make_value_memory_location(loc, vm);

                                self.backend.emit_mov_mem_r(&loc_val, &val_h);
                            } else {
                                unimplemented!()
                            }
1178
                        } else if self.match_fpreg(val_op) {
qinsoon's avatar
qinsoon committed
1179 1180
                            let val = self.emit_fpreg(val_op, f_content, f_context, vm);

1181 1182 1183 1184 1185 1186 1187
                            match val.ty.v {
                                MuType_::Double => self.backend.emit_movsd_mem64_f64(&resolved_loc, &val),
                                MuType_::Float  => self.backend.emit_movss_mem32_f32(&resolved_loc, &val),
                                _ => panic!("unexpected fp type: {}", val.ty)
                            }
                        } else {
                            unimplemented!()
1188 1189
                        }
                    }
qinsoon's avatar
qinsoon committed
1190

1191 1192 1193 1194
                    // memory insts: calculate the address, then lea
                    Instruction_::GetIRef(_)
                    | Instruction_::GetFieldIRef{..}
                    | Instruction_::GetVarPartIRef{..}
1195 1196 1197
                    | Instruction_::ShiftIRef{..}
                    | Instruction_::GetElementIRef{..} => {
                        trace!("instsel on GET/FIELD/VARPART/SHIFT/ELEM IREF");
qinsoon's avatar
qinsoon committed
1198

1199 1200
                        let mem_addr = self.emit_get_mem_from_inst(node, f_content, f_context, vm);
                        let tmp_res  = self.get_result_value(node);
qinsoon's avatar
qinsoon committed
1201

1202
                        self.backend.emit_lea_r64(&tmp_res, &mem_addr);
qinsoon's avatar
qinsoon committed
1203
                    }
1204
                    
1205
                    Instruction_::ThreadExit => {
qinsoon's avatar
qinsoon committed
1206
                        trace!("instsel on THREADEXIT");
1207
                        // emit a call to swap_back_to_native_stack(sp_loc: Address)
1208 1209
                        
                        // get thread local and add offset to get sp_loc
qinsoon's avatar
qinsoon committed
1210
                        let tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1211
                        self.backend.emit_add_r_imm(&tl, *thread::NATIVE_SP_LOC_OFFSET as i32);
1212
                        
qinsoon's avatar
qinsoon committed
1213
                        self.emit_runtime_entry(&entrypoints::SWAP_BACK_TO_NATIVE_STACK, vec![tl.clone()], None, Some(node), f_content, f_context, vm);
1214
                    }
1215 1216

                    Instruction_::CommonInst_GetThreadLocal => {
qinsoon's avatar
qinsoon committed
1217
                        trace!("instsel on GETTHREADLOCAL");
1218 1219 1220 1221 1222 1223 1224 1225 1226
                        // get thread local
                        let tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);

                        let tmp_res = self.get_result_value(node);

                        // load [tl + USER_TLS_OFFSET] -> tmp_res
                        self.emit_load_base_offset(&tmp_res, &tl, *thread::USER_TLS_OFFSET as i32, vm);
                    }
                    Instruction_::CommonInst_SetThreadLocal(op) => {
qinsoon's avatar
qinsoon committed
1227 1228
                        trace!("instsel on SETTHREADLOCAL");

1229
                        let ref ops = inst.ops;
1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241
                        let ref op = ops[op];

                        debug_assert!(self.match_ireg(op));

                        let tmp_op = self.emit_ireg(op, f_content, f_context, vm);

                        // get thread local
                        let tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);

                        // store tmp_op -> [tl + USER_TLS_OFFSTE]
                        self.emit_store_base_offset(&tl, *thread::USER_TLS_OFFSET as i32, &tmp_op, vm);
                    }
1242

qinsoon's avatar
qinsoon committed
1243 1244 1245 1246 1247
                    Instruction_::CommonInst_Pin(op) => {
                        trace!("instsel on PIN");

                        if !mm::GC_MOVES_OBJECT {
                            // non-moving GC: pin is a nop (move from op to result)
1248
                            let ref ops