GitLab will continue to be upgraded from 11.4.5-ce.0 on November 25th 2019 at 4.00pm (AEDT) to 5.00pm (AEDT) due to Critical Security Patch Availability. During the update, GitLab and Mattermost services will not be available.

inst_sel.rs 189 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;
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 28
use utils::math;

29
use std::collections::HashMap;
qinsoon's avatar
qinsoon committed
30
use std::any::Any;
31

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
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))
            })
        ]))
    });
84

85 86
    pub static ref FPTOUI_C_DOUBLE : P<Value> = P(Value{
        hdr: MuEntityHeader::named(new_internal_id(), Mu("FPTOUI_C_DOUBLE")),
87 88 89
        ty : UINT64_TYPE.clone(),
        v  : Value_::Constant(Constant::Int(4890909195324358656u64))
    });
90 91 92 93 94 95

    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))
    });
96 97
}

98 99
const INLINE_FASTPATH : bool = false;

100
pub struct InstructionSelection {
101
    name: &'static str,
102
    backend: Box<CodeGenerator>,
103 104

    current_fv_id: MuID,
qinsoon's avatar
qinsoon committed
105
    current_callsite_id: usize,
qinsoon's avatar
qinsoon committed
106 107
    current_frame: Option<Frame>,
    current_block: Option<MuName>,
qinsoon's avatar
qinsoon committed
108 109 110 111
    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
112 113 114 115
    current_exn_blocks: HashMap<MuID, ValueLocation>,

    current_constants: HashMap<MuID, P<Value>>,
    current_constants_locs: HashMap<MuID, P<Value>>
116 117
}

118
impl <'a> InstructionSelection {
qinsoon's avatar
qinsoon committed
119
    #[cfg(feature = "aot")]
120
    pub fn new() -> InstructionSelection {
121 122
        InstructionSelection{
            name: "Instruction Selection (x64)",
123
            backend: Box::new(ASMCodeGen::new()),
124

125
            current_fv_id: 0,
qinsoon's avatar
qinsoon committed
126
            current_callsite_id: 0,
qinsoon's avatar
qinsoon committed
127 128 129
            current_frame: None,
            current_block: None,
            current_func_start: None,
qinsoon's avatar
qinsoon committed
130 131
            // key: block id, val: callsite that names the block as exception block
            current_exn_callsites: HashMap::new(), 
132 133 134 135
            current_exn_blocks: HashMap::new(),

            current_constants: HashMap::new(),
            current_constants_locs: HashMap::new()
136 137
        }
    }
qinsoon's avatar
qinsoon committed
138 139 140 141 142

    #[cfg(feature = "jit")]
    pub fn new() -> InstructionSelection {
        unimplemented!()
    }
143 144
    
    // in this pass, we assume that
qinsoon's avatar
qinsoon committed
145
    // * we do not need to backup/restore caller-saved registers
146 147
    // if any of these assumption breaks, we will need to re-emit the code
    #[allow(unused_variables)]
148
    fn instruction_select(&mut self, node: &'a TreeNode, f_content: &FunctionContent, f_context: &mut FunctionContext, vm: &VM) {
149
        trace!("instsel on node#{} {}", node.id(), node);
qinsoon's avatar
qinsoon committed
150 151
        
        match node.v {
152 153
            TreeNode_::Instruction(ref inst) => {
                match inst.v {
qinsoon's avatar
qinsoon committed
154
                    Instruction_::Branch2{cond, ref true_dest, ref false_dest, true_prob} => {
qinsoon's avatar
qinsoon committed
155
                        trace!("instsel on BRANCH2");
156 157
                        // '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)
158 159 160 161 162
                        let (fallthrough_dest, branch_dest, branch_if_true) = {
                            if true_prob > 0.5f32 {
                                (true_dest, false_dest, false)
                            } else {
                                (false_dest, true_dest, true)
163
                            }
164
                        };
165
                        
qinsoon's avatar
qinsoon committed
166
                        let ops = inst.ops.read().unwrap();
167
                        
168 169
                        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
170
                        
171
                        let branch_target = f_content.get_block(branch_dest.target).name().unwrap();
172 173 174
    
                        let ref cond = ops[cond];
                        
qinsoon's avatar
qinsoon committed
175
                        if self.match_cmp_res(cond) {
176
                            trace!("emit cmp_res-branch2");
177
                            match self.emit_cmp_res(cond, f_content, f_context, vm) {
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
                                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
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293

                                // 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
294 295 296 297
                                _ => unimplemented!()
                            }
                        } else if self.match_ireg(cond) {
                            trace!("emit ireg-branch2");
298
                            
299
                            let cond_reg = self.emit_ireg(cond, f_content, f_context, vm);
300
                            
qinsoon's avatar
qinsoon committed
301
                            // emit: cmp cond_reg 1
302
                            self.backend.emit_cmp_imm_r(1, &cond_reg);
qinsoon's avatar
qinsoon committed
303
                            // emit: je #branch_dest
304 305 306 307 308
                            if branch_if_true {
                                self.backend.emit_je(branch_target);
                            } else {
                                self.backend.emit_jne(branch_target);
                            }
qinsoon's avatar
qinsoon committed
309 310
                        } else {
                            unimplemented!();
311
                        }
312
                    },
qinsoon's avatar
qinsoon committed
313 314

                    Instruction_::Select{cond, true_val, false_val} => {
qinsoon's avatar
qinsoon committed
315 316
                        use ast::op::CmpOp::*;

qinsoon's avatar
qinsoon committed
317
                        trace!("instsel on SELECT");
qinsoon's avatar
qinsoon committed
318 319 320 321 322 323 324 325 326
                        let ops = inst.ops.read().unwrap();

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

                        if self.match_ireg(true_val) {
                            // moving integers/pointers
                            let tmp_res   = self.get_result_value(node);
327

qinsoon's avatar
qinsoon committed
328 329 330
                            // generate compare
                            let cmpop = if self.match_cmp_res(cond) {
                                self.emit_cmp_res(cond, f_content, f_context, vm)
qinsoon's avatar
qinsoon committed
331 332 333
                            } else if self.match_ireg(cond) {
                                let tmp_cond = self.emit_ireg(cond, f_content, f_context, vm);
                                // emit: cmp cond_reg 1
334
                                self.backend.emit_cmp_imm_r(1, &tmp_cond);
qinsoon's avatar
qinsoon committed
335

qinsoon's avatar
qinsoon committed
336
                                EQ
qinsoon's avatar
qinsoon committed
337
                            } else {
qinsoon's avatar
qinsoon committed
338 339 340
                                panic!("expected ireg, found {}", cond)
                            };

341
                            // use cmov for 16/32/64bit integer
qinsoon's avatar
qinsoon committed
342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
                            // use jcc  for 8 bit
                            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
                                _ => {
376 377
                                    let blk_true  = format!("{}_select_true", node.id());
                                    let blk_false = format!("{}_select_false", node.id());
qinsoon's avatar
qinsoon committed
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402
                                    let blk_end   = format!("{}_select_end", 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!()
                                    }

403 404 405 406 407 408 409 410
                                    // finishing current block
                                    let cur_block = self.current_block.as_ref().unwrap().clone();
                                    self.backend.end_block(cur_block.clone());

                                    // blk_false:
                                    self.current_block = Some(blk_false.clone());
                                    self.backend.start_block(blk_false.clone());

qinsoon's avatar
qinsoon committed
411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
                                    // 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
                                    let cur_block = self.current_block.as_ref().unwrap().clone();
                                    self.backend.end_block(cur_block.clone());

                                    // blk_true:
                                    self.current_block = Some(blk_true.clone());
                                    self.backend.start_block(blk_true.clone());
                                    // 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:
                                    self.backend.start_block(blk_end.clone());
                                    self.current_block = Some(blk_end.clone());
                                }
qinsoon's avatar
qinsoon committed
433 434 435 436 437 438 439
                            }
                        } else {
                            // moving vectors, floatingpoints
                            unimplemented!()
                        }
                    },

440
                    Instruction_::CmpOp(op, op1, op2) => {
qinsoon's avatar
qinsoon committed
441 442
                        use ast::op::CmpOp::*;

qinsoon's avatar
qinsoon committed
443
                        trace!("instsel on CMPOP");
444 445 446 447
                        let ops = inst.ops.read().unwrap();
                        let ref op1 = ops[op1];
                        let ref op2 = ops[op2];

qinsoon's avatar
qinsoon committed
448
                        let tmp_res = self.get_result_value(node);
449 450 451
                        
                        debug_assert!(tmp_res.ty.get_int_length().is_some());
                        debug_assert!(tmp_res.ty.get_int_length().unwrap() == 1);
452

453
                        // set byte to result
qinsoon's avatar
qinsoon committed
454
                        match self.emit_cmp_res(node, f_content, f_context, vm) {
455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471
                            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
472 473

                            _ => unimplemented!()
474 475 476
                        }
                    }

qinsoon's avatar
qinsoon committed
477
                    Instruction_::Branch1(ref dest) => {
qinsoon's avatar
qinsoon committed
478
                        trace!("instsel on BRANCH1");
qinsoon's avatar
qinsoon committed
479
                        let ops = inst.ops.read().unwrap();
480
                                            
481
                        self.process_dest(&ops, dest, f_content, f_context, vm);
482
                        
483
                        let target = f_content.get_block(dest.target).name().unwrap();
qinsoon's avatar
qinsoon committed
484
                        
qinsoon's avatar
qinsoon committed
485
                        trace!("emit branch1");
486
                        // jmp
qinsoon's avatar
qinsoon committed
487
                        self.backend.emit_jmp(target);
488
                    },
qinsoon's avatar
qinsoon committed
489 490

                    Instruction_::Switch{cond, ref default, ref branches} => {
qinsoon's avatar
qinsoon committed
491
                        trace!("instsel on SWITCH");
qinsoon's avatar
qinsoon committed
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511
                        let ops = inst.ops.read().unwrap();

                        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
512
                                    self.backend.emit_cmp_imm_r(imm, &tmp_cond);
qinsoon's avatar
qinsoon committed
513 514 515 516 517 518
                                    // 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
519
                                    self.backend.emit_cmp_r_r(&tmp_case_op, &tmp_cond);
qinsoon's avatar
qinsoon committed
520 521 522 523 524
                                    // je dest
                                    self.backend.emit_je(target);
                                } else {
                                    panic!("expecting ireg cond to be either iimm or ireg: {}", cond);
                                }
525 526 527

                                self.finish_block();
                                self.start_block(format!("{}_switch_not_met_case_{}", node.id(), case_op_index));
qinsoon's avatar
qinsoon committed
528 529 530 531 532 533 534 535 536 537 538
                            }

                            // 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);
                        }
                    }
539
                    
qinsoon's avatar
qinsoon committed
540
                    Instruction_::ExprCall{ref data, is_abort} => {
qinsoon's avatar
qinsoon committed
541 542
                        trace!("instsel on EXPRCALL");

qinsoon's avatar
qinsoon committed
543 544
                        if is_abort {
                            unimplemented!()
545
                        }
546
                        
qinsoon's avatar
qinsoon committed
547 548 549 550 551
                        self.emit_mu_call(
                            inst, // inst: &Instruction,
                            data, // calldata: &CallData,
                            None, // resumption: Option<&ResumptionData>,
                            node, // cur_node: &TreeNode, 
552
                            f_content, f_context, vm);
553 554
                    },
                    
qinsoon's avatar
qinsoon committed
555
                    Instruction_::Call{ref data, ref resume} => {
qinsoon's avatar
qinsoon committed
556 557
                        trace!("instsel on CALL");

qinsoon's avatar
qinsoon committed
558 559 560 561 562 563
                        self.emit_mu_call(
                            inst, 
                            data, 
                            Some(resume), 
                            node, 
                            f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
564 565 566
                    },

                    Instruction_::ExprCCall{ref data, is_abort} => {
qinsoon's avatar
qinsoon committed
567 568
                        trace!("instsel on EXPRCCALL");

qinsoon's avatar
qinsoon committed
569 570 571 572 573 574 575 576
                        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
577 578
                        trace!("instsel on CCALL");

qinsoon's avatar
qinsoon committed
579
                        self.emit_c_call_ir(inst, data, Some(resume), node, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
580 581
                    }
                    
582
                    Instruction_::Return(_) => {
qinsoon's avatar
qinsoon committed
583 584
                        trace!("instsel on RETURN");

585
                        self.emit_common_epilogue(inst, f_content, f_context, vm);
586
                        
qinsoon's avatar
qinsoon committed
587
                        self.backend.emit_ret();
588 589
                    },
                    
qinsoon's avatar
qinsoon committed
590
                    Instruction_::BinOp(op, op1, op2) => {
qinsoon's avatar
qinsoon committed
591 592
                        trace!("instsel on BINOP");

qinsoon's avatar
qinsoon committed
593 594
                        self.emit_binop(node, inst, op, op1, op2, f_content, f_context, vm);
                    },
qinsoon's avatar
qinsoon committed
595

qinsoon's avatar
qinsoon committed
596 597
                    Instruction_::BinOpWithStatus(op, status, op1, op2) => {
                        trace!("instsel on BINOP_STATUS");
qinsoon's avatar
qinsoon committed
598

qinsoon's avatar
qinsoon committed
599
                        self.emit_binop(node, inst, op, op1, op2, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
600

qinsoon's avatar
qinsoon committed
601 602
                        let values = inst.value.as_ref().unwrap();
                        let mut status_value_index = 1;
qinsoon's avatar
qinsoon committed
603

604
                        // status flags only works with int operations
605
                        if RegGroup::get_from_value(&values[0]) == RegGroup::GPR {
606 607 608 609 610 611 612
                            // 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
613

614 615 616 617
                            // zero flag
                            if status.flag_z {
                                let tmp_status = values[status_value_index].clone();
                                status_value_index += 1;
qinsoon's avatar
qinsoon committed
618

619 620
                                self.backend.emit_setz_r8(&tmp_status);
                            }
qinsoon's avatar
qinsoon committed
621

622 623 624 625
                            // unsigned overflow
                            if status.flag_c {
                                let tmp_status = values[status_value_index].clone();
                                status_value_index += 1;
qinsoon's avatar
qinsoon committed
626

627 628 629 630 631
                                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
632 633
                                }
                            }
634

635 636 637
                            // signed overflow
                            if status.flag_v {
                                let tmp_status = values[status_value_index].clone();
638

639 640 641 642 643
                                match op {
                                    BinOp::Add | BinOp::Sub | BinOp::Mul => {
                                        self.backend.emit_seto_r8(&tmp_status);
                                    }
                                    _ => panic!("Only Add/Sub/Mul has #V flag")
644 645
                                }
                            }
646 647
                        }
                    }
qinsoon's avatar
qinsoon committed
648 649

                    Instruction_::ConvOp{operation, ref from_ty, ref to_ty, operand} => {
qinsoon's avatar
qinsoon committed
650 651
                        trace!("instsel on CONVOP");

qinsoon's avatar
qinsoon committed
652 653 654 655 656 657 658 659
                        let ops = inst.ops.read().unwrap();

                        let ref op = ops[operand];

                        match operation {
                            op::ConvOp::TRUNC => {
                                if self.match_ireg(op) {
                                    let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
660

qinsoon's avatar
qinsoon committed
661
                                    let tmp_res = self.get_result_value(node);
662 663 664 665
                                    let dst_length = match tmp_res.ty.get_int_length() {
                                        Some(len) => len,
                                        None => panic!("dst of TRUNC is not int: {}", tmp_res)
                                    };
qinsoon's avatar
qinsoon committed
666 667

                                    // mov op -> result
668 669 670 671 672 673 674 675
                                    match dst_length {
                                        1  => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT8_TYPE.clone())}),
                                        8  => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT8_TYPE.clone())}),
                                        16 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT16_TYPE.clone())}),
                                        32 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT32_TYPE.clone())}),
                                        64 => self.backend.emit_mov_r_r(&tmp_res, unsafe {&tmp_op.as_type(UINT64_TYPE.clone())}),
                                        _  => panic!("unsupported int length: {}", dst_length)
                                    }
qinsoon's avatar
qinsoon committed
676 677 678 679 680
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
                                }
                            }
                            op::ConvOp::ZEXT => {
681 682 683
                                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
684

685 686 687
                                    // 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
688

689
                                    if from_ty_size != to_ty_size {
qinsoon's avatar
qinsoon committed
690 691 692 693 694 695 696 697 698 699 700 701 702
                                        if from_ty_size == 4 && to_ty_size == 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);
                                        } else {
                                            self.backend.emit_movz_r_r(&tmp_res, &tmp_op);
                                        }
703
                                    } else {
704
                                        self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
705
                                    }
706 707
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
qinsoon's avatar
qinsoon committed
708 709 710
                                }
                            },
                            op::ConvOp::SEXT => {
711 712 713
                                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
714

715 716 717
                                    // 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;
718

719
                                    if from_ty_size != to_ty_size {
720 721
                                        self.backend.emit_movs_r_r(&tmp_res, &tmp_op);
                                    } else {
722
                                        self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
723
                                    }
724 725
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op)
qinsoon's avatar
qinsoon committed
726 727
                                }
                            }
728 729 730 731 732 733
                            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);
734
                                    self.backend.emit_mov_r_r(&tmp_res, &tmp_op);
735 736 737 738
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op)
                                }
                            }
739 740 741
                            op::ConvOp::SITOFP => {
                                let tmp_res = self.get_result_value(node);

742 743 744 745 746 747 748
                                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)
749 750 751 752 753
                                }
                            }
                            op::ConvOp::FPTOSI => {
                                let tmp_res = self.get_result_value(node);

754 755 756 757
                                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() {
758
                                    self.backend.emit_cvtsd2si_r_f64(&tmp_res, &tmp_op);
759 760
                                } else if from_ty.is_float() {
                                    self.backend.emit_cvtss2si_r_f32(&tmp_res, &tmp_op);
761
                                } else {
762
                                    panic!("expected fp type as from type in FPTOSI, found {}", to_ty)
763 764
                                }
                            }
765 766 767
                            op::ConvOp::UITOFP => {
                                let tmp_res = self.get_result_value(node);

768 769
                                assert!(self.match_ireg(op), "unexpected op (expected ireg): {}", op);
                                let tmp_op = self.emit_ireg(op, f_content, f_context, vm);
770

771
                                let op_ty_size = vm.get_backend_type_info(tmp_op.ty.id()).size;
772

773 774 775 776 777
                                if to_ty.is_double() {
                                    match op_ty_size {
                                        8 => {
                                            // movd/movq op -> res
                                            self.backend.emit_mov_fpr_r64(&tmp_res, &tmp_op);
778

779 780 781 782
                                            // 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);
783

784 785 786
                                            // 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);
787

788 789 790 791 792
                                            // 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);
793

794 795 796
                                            // movl op -> tmp(32)
                                            let tmp32 = unsafe { tmp.as_type(UINT32_TYPE.clone()) };
                                            self.backend.emit_mov_r_r(&tmp32, &tmp_op);
797

798 799 800 801 802 803 804 805
                                            // 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)
806
                                    }
807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
                                } 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)
871
                                    }
872 873
                                } else {
                                    panic!("expect double or float")
874 875 876 877 878
                                }
                            }
                            op::ConvOp::FPTOUI => {
                                let tmp_res = self.get_result_value(node);

879
                                assert!(self.match_fpreg(op), "unexpected op (expected fpreg): {}", op);
880
                                let tmp_op = self.emit_fpreg(op, f_content, f_context, vm);
881
                                let res_ty_size = vm.get_backend_type_info(tmp_res.ty.id()).size;
882

883 884 885 886 887
                                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);
888

889 890 891
                                            // 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);
892

893 894
                                            // movapd %tmp_op -> %tmp2
                                            self.backend.emit_movapd_f64_f64(&tmp2, &tmp_op);
895

896 897
                                            // subsd %tmp1, %tmp2 -> %tmp2
                                            self.backend.emit_subsd_f64_f64(&tmp2, &tmp1);
898

899 900
                                            // cvttsd2si %tmp2 -> %tmp_res
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res, &tmp2);
901

902 903 904
                                            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);
905

906 907
                                            // xor %tmp_res, %tmp_const -> %tmp_const
                                            self.backend.emit_xor_r_r(&tmp_const, &tmp_res);
qinsoon's avatar
qinsoon committed
908

909 910
                                            // cvttsd2si %tmp_op -> %tmp_res
                                            self.backend.emit_cvttsd2si_r_f64(&tmp_res, &tmp_op);
911

912 913
                                            // ucomisd %tmp_op %tmp1
                                            self.backend.emit_ucomisd_f64_f64(&tmp1, &tmp_op);
914

915 916 917 918 919
                                            // 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()) };
920

921 922 923 924 925 926 927 928 929 930 931 932 933
                                            // 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)
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
                                } 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);
960

961 962
                                            // cvttss2si %tmp_op -> %tmp_res
                                            self.backend.emit_cvttss2si_r_f32(&tmp_res, &tmp_op);
963

964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985
                                            // 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)
986
                                    }
987 988
                                } else {
                                    panic!("expect double or float")
989 990
                                }
                            }
qinsoon's avatar
qinsoon committed
991 992 993
                            _ => unimplemented!()
                        }
                    }
994
                    
995 996
                    // load on x64 generates mov inst (no matter what order is specified)
                    // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
997
                    Instruction_::Load{is_ptr, order, mem_loc} => {
qinsoon's avatar
qinsoon committed
998 999
                        trace!("instsel on LOAD");

qinsoon's avatar
qinsoon committed
1000
                        let ops = inst.ops.read().unwrap();
1001
                        let ref loc_op = ops[mem_loc];
1002 1003 1004 1005 1006 1007
                        
                        // check order
                        match order {
                            MemoryOrder::Relaxed 
                            | MemoryOrder::Consume 
                            | MemoryOrder::Acquire
1008 1009
                            | MemoryOrder::SeqCst
                            | MemoryOrder::NotAtomic => {},
1010 1011
                            _ => panic!("didnt expect order {:?} with store inst", order)
                        }                        
1012

1013
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1014
                        let res_temp = self.get_result_value(node);
1015

1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026
                        if self.match_ireg(node) {
                            self.backend.emit_mov_r_mem(&res_temp, &resolved_loc);
                        } 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!()
                        }
1027 1028 1029
                    }
                    
                    Instruction_::Store{is_ptr, order, mem_loc, value} => {
qinsoon's avatar
qinsoon committed
1030 1031
                        trace!("instsel on STORE");

qinsoon's avatar
qinsoon committed
1032
                        let ops = inst.ops.read().unwrap();
1033 1034 1035 1036 1037
                        let ref loc_op = ops[mem_loc];
                        let ref val_op = ops[value];
                        
                        let generate_plain_mov : bool = {
                            match order {
1038 1039 1040
                                MemoryOrder::Relaxed
                                | MemoryOrder::Release
                                | MemoryOrder::NotAtomic => true,
1041 1042 1043 1044 1045
                                MemoryOrder::SeqCst => false,
                                _ => panic!("didnt expect order {:?} with store inst", order)
                            }
                        };
                        
1046 1047 1048
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, f_content, f_context, vm);

                        if self.match_iimm(val_op) {
1049
                            let (val, len) = self.node_iimm_to_i32_with_len(val_op);
1050
                            if generate_plain_mov {
1051
                                self.backend.emit_mov_mem_imm(&resolved_loc, val, len);
1052 1053 1054
                            } else {
                                unimplemented!()
                            }
1055 1056
                        } else if self.match_ireg(val_op) {
                            let val = self.emit_ireg(val_op, f_content, f_context, vm);
1057
                            if generate_plain_mov {
1058
                                self.backend.emit_mov_mem_r(&resolved_loc, &val);
1059 1060 1061
                            } else {
                                unimplemented!()
                            }
1062
                        } else if self.match_fpreg(val_op) {
qinsoon's avatar
qinsoon committed
1063 1064
                            let val = self.emit_fpreg(val_op, f_content, f_context, vm);

1065 1066 1067 1068 1069 1070 1071
                            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!()
1072 1073
                        }
                    }
1074

1075 1076 1077 1078
                    // memory insts: calculate the address, then lea
                    Instruction_::GetIRef(_)
                    | Instruction_::GetFieldIRef{..}
                    | Instruction_::GetVarPartIRef{..}
1079 1080 1081
                    | Instruction_::ShiftIRef{..}
                    | Instruction_::GetElementIRef{..} => {
                        trace!("instsel on GET/FIELD/VARPART/SHIFT/ELEM IREF");
qinsoon's avatar
qinsoon committed
1082

1083 1084
                        let mem_addr = self.emit_get_mem_from_inst(node, f_content, f_context, vm);
                        let tmp_res  = self.get_result_value(node);
1085

1086
                        self.backend.emit_lea_r64(&tmp_res, &mem_addr);
1087
                    }
1088
                    
1089
                    Instruction_::ThreadExit => {
qinsoon's avatar
qinsoon committed
1090
                        trace!("instsel on THREADEXIT");
1091
                        // emit a call to swap_back_to_native_stack(sp_loc: Address)
1092 1093
                        
                        // get thread local and add offset to get sp_loc
qinsoon's avatar
qinsoon committed
1094
                        let tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);
1095
                        self.backend.emit_add_r_imm(&tl, *thread::NATIVE_SP_LOC_OFFSET as i32);
1096
                        
qinsoon's avatar
qinsoon committed
1097
                        self.emit_runtime_entry(&entrypoints::SWAP_BACK_TO_NATIVE_STACK, vec![tl.clone()], None, Some(node), f_content, f_context, vm);
1098
                    }
1099 1100

                    Instruction_::CommonInst_GetThreadLocal => {
qinsoon's avatar
qinsoon committed
1101
                        trace!("instsel on GETTHREADLOCAL");
1102 1103 1104 1105 1106 1107 1108 1109 1110
                        // 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
1111 1112
                        trace!("instsel on SETTHREADLOCAL");

1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
                        let ops = inst.ops.read().unwrap();
                        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);
                    }
1126

1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
                    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)
                            let ops = inst.ops.read().unwrap();
                            let ref op = ops[op];

                            let tmp_res = self.get_result_value(node);

                            self.emit_move_node_to_value(&tmp_res, op, f_content, f_context, vm);
                        } else {
                            unimplemented!()
                        }
                    }
                    Instruction_::CommonInst_Unpin(_) => {
                        trace!("instsel on UNPIN");

                        if !mm::GC_MOVES_OBJECT {
                            // do nothing
                        } else {
                            unimplemented!()
                        }
                    }


1153
                    Instruction_::Move(op) => {
qinsoon's avatar
qinsoon committed
1154 1155
                        trace!("instsel on MOVE (internal IR)");

1156 1157 1158 1159 1160 1161 1162
                        let ops = inst.ops.read().unwrap();
                        let ref op = ops[op];

                        let tmp_res = self.get_result_value(node);

                        self.emit_move_node_to_value(&tmp_res, op, f_content, f_context, vm);
                    }
1163 1164
                    
                    Instruction_::New(ref ty) => {
qinsoon's avatar
qinsoon committed
1165 1166
                        trace!("instsel on NEW");

qinsoon's avatar
qinsoon committed
1167 1168 1169 1170 1171 1172 1173
                        if cfg!(debug_assertions) {
                            match ty.v {
                                MuType_::Hybrid(_) => panic!("cannot use NEW for hybrid, use NEWHYBRID instead"),
                                _ => {}
                            }
                        }

1174
                        let ty_info = vm.get_backend_type_info(ty.id());
qinsoon's avatar
qinsoon committed
1175
                        let size = ty_info.size;
1176
                        let ty_align= ty_info.alignment;
qinsoon's avatar
qinsoon committed
1177 1178

                        let const_size = self.make_value_int_const(size as u64, vm);
qinsoon's avatar
qinsoon committed
1179 1180

                        let tmp_allocator = self.emit_get_allocator(node, f_content, f_context, vm);
1181
                        let tmp_res = self.emit_alloc_sequence(tmp_allocator.clone(), const_size, ty_align, node, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1182 1183 1184 1185 1186 1187 1188 1189 1190

                        // ASM: call muentry_init_object(%allocator, %tmp_res, %encode)
                        let encode = self.make_value_int_const(mm::get_gc_type_encode(ty_info.gc_type.id), vm);
                        self.emit_runtime_entry(
                            &entrypoints::INIT_OBJ,
                            vec![tmp_allocator.clone(), tmp_res.clone(), encode],
                            None,
                            Some(node), f_content, f_context, vm
                        );
qinsoon's avatar
qinsoon committed
1191 1192 1193
                    }

                    Instruction_::NewHybrid(ref ty, var_len) => {
qinsoon's avatar
qinsoon committed
1194 1195
                        trace!("instsel on NEWHYBRID");

qinsoon's avatar
qinsoon committed
1196 1197 1198 1199 1200
                        if cfg!(debug_assertions) {
                            match ty.v {
                                MuType_::Hybrid(_) => {},
                                _ => panic!("NEWHYBRID is only for allocating hybrid types, use NEW for others")
                            }
1201
                        }
qinsoon's avatar
qinsoon committed
1202 1203 1204 1205

                        let ty_info = vm.get_backend_type_info(ty.id());
                        let ty_align = ty_info.alignment;
                        let fix_part_size = ty_info.size;
1206 1207 1208
                        let var_ty_size = match ty_info.elem_padded_size {
                            Some(sz) => sz,
                            None => panic!("expect HYBRID type here with elem_padded_size, found {}", ty_info)
qinsoon's avatar
qinsoon committed
1209 1210 1211
                        };

                        // actual size = fix_part_size + var_ty_size * len
qinsoon's avatar
qinsoon committed
1212
                        let (actual_size, length) = {
qinsoon's avatar
qinsoon committed
1213 1214 1215 1216 1217 1218 1219
                            let ops = inst.ops.read().unwrap();
                            let ref var_len = ops[var_len];

                            if self.match_iimm(var_len) {
                                let var_len = self.node_iimm_to_i32(var_len);
                                let actual_size = fix_part_size + var_ty_size * (var_len as usize);

qinsoon's avatar
qinsoon committed
1220 1221 1222 1223
                                (
                                    self.make_value_int_const(actual_size as u64, vm),
                                    self.make_value_int_const(var_len as u64, vm)
                                )
qinsoon's avatar
qinsoon committed
1224 1225 1226 1227
                            } else {
                                let tmp_actual_size = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
                                let tmp_var_len = self.emit_ireg(var_len, f_content, f_context, vm);

1228
                                match math::is_power_of_two(var_ty_size) {
qinsoon's avatar
qinsoon committed
1229
                                    Some(shift) => {
1230 1231 1232
                                        // use tmp_actual_size as result - we do not want to change tmp_var_len
                                        self.backend.emit_mov_r_r(&tmp_actual_size, &tmp_var_len);

qinsoon's avatar
qinsoon committed
1233 1234
                                        if shift != 0 {
                                            // a shift-left will get the total size of var part
1235
                                            self.backend.emit_shl_r_imm8(&tmp_actual_size, shift as i8);
qinsoon's avatar
qinsoon committed
1236
                                        }
qinsoon's avatar
qinsoon committed
1237 1238

                                        // add with fix-part size
1239
                                        self.backend.emit_add_r_imm(&tmp_actual_size, fix_part_size as i32);
qinsoon's avatar
qinsoon committed
1240 1241 1242 1243 1244
                                    }
                                    None => {
                                        // we need to do a multiply

                                        // mov var_ty_size -> rax
1245
                                        self.backend.emit_mov_r_imm(&x86_64::RAX, var_ty_size as i32);
qinsoon's avatar
qinsoon committed
1246 1247

                                        // mul tmp_var_len, rax -> rdx:rax
1248
                                        self.backend.emit_mul_r(&tmp_var_len);
qinsoon's avatar
qinsoon committed
1249 1250

                                        // add with fix-part size
1251
                                        self.backend.emit_add_r_imm(&x86_64::RAX, fix_part_size as i32);
qinsoon's avatar
qinsoon committed
1252 1253

                                        // mov result to tmp_actual_size
1254
                                        self.backend.emit_mov_r_r(&tmp_actual_size, &x86_64::RAX);
qinsoon's avatar
qinsoon committed
1255 1256 1257
                                    }
                                }

qinsoon's avatar
qinsoon committed
1258
                                (tmp_actual_size, tmp_var_len)
qinsoon's avatar
qinsoon committed
1259 1260 1261
                            }
                        };

qinsoon's avatar
qinsoon committed
1262
                        let tmp_allocator = self.emit_get_allocator(node, f_content, f_context, vm);
1263
                        let tmp_res = self.emit_alloc_sequence(tmp_allocator.clone(), actual_size, ty_align, node, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1264 1265 1266 1267 1268 1269 1270 1271 1272

                        // ASM: call muentry_init_object(%allocator, %tmp_res, %encode)
                        let encode = self.make_value_int_const(mm::get_gc_type_encode(ty_info.gc_type.id), vm);
                        self.emit_runtime_entry(
                            &entrypoints::INIT_HYBRID,
                            vec![tmp_allocator.clone(), tmp_res.clone(), encode, length],
                            None,
                            Some(node), f_content, f_context, vm
                        );
1273
                    }
qinsoon's avatar
qinsoon committed
1274 1275
                    
                    Instruction_::Throw(op_index) => {
qinsoon's avatar
qinsoon committed
1276 1277
                        trace!("instsel on THROW");

qinsoon's avatar
qinsoon committed
1278 1279 1280 1281 1282 1283 1284
                        let ops = inst.ops.read().unwrap();
                        let ref exception_obj = ops[op_index];
                        
                        self.emit_runtime_entry(
                            &entrypoints::THROW_EXCEPTION, 
                            vec![exception_obj.clone_value()], 
                            None,
qinsoon's avatar
qinsoon committed
1285
                            Some(node), f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
1286
                    }
1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300

                    Instruction_::PrintHex(index) => {
                        trace!("instsel on PRINTHEX");

                        let ops = inst.ops.read().unwrap();
                        let ref op = ops[index];

                        self.emit_runtime_entry(
                            &entrypoints::PRINT_HEX,
                            vec![op.clone_value()],
                            None,
                            Some(node), f_content, f_context, vm
                        );
                    }
1301 1302 1303 1304 1305 1306
    
                    _ => unimplemented!()
                } // main switch
            },
            
            TreeNode_::Value(ref p) => {
1307
        
1308 1309 1310 1311
            }
        }
    }
    
1312 1313 1314 1315
    fn make_temporary(&mut self, f_context: &mut FunctionContext, ty: P<MuType>, vm: &VM) -> P<Value> {
        f_context.make_temporary(vm.next_id(), ty).clone_value()
    }
    
1316
    fn make_memory_op_base_offset (&mut self, base: &P<Value>, offset: i32, ty: P<MuType>, vm: &VM) -> P<Value> {
1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327
        P(Value{
            hdr: MuEntityHeader::unnamed(vm.next_id()),
            ty: ty.clone(),
            v: Value_::Memory(MemoryLocation::Address{
                base: base.clone(),
                offset: Some(self.make_value_int_const(offset as u64, vm)),
                index: None,
                scale: None
            })
        })
    }
qinsoon's avatar
qinsoon committed
1328

1329
    fn make_memory_op_base_index(&mut self, base: &P<Value>, index: &P<Value>, scale: u8, ty: P<MuType>, vm: &VM) -> P<Value> {
qinsoon's avatar
qinsoon committed
1330 1331 1332 1333 1334
        P(Value{
            hdr: MuEntityHeader::unnamed(vm.next_id()),
            ty: ty.clone(),
            v: Value_::Memory(MemoryLocation::Address{
                base: base.clone(),
1335 1336 1337
                offset: None,
                index: Some(index.clone()),
                scale: Some(scale)
qinsoon's avatar
qinsoon committed
1338 1339 1340
            })
        })
    }
1341 1342 1343 1344
    
    fn make_value_int_const (&mut self, val: u64, vm: &VM) -> P<Value> {
        P(Value{
            hdr: MuEntityHeader::unnamed(vm.next_id()),
qinsoon's avatar
qinsoon committed
1345
            ty: UINT64_TYPE.clone(),
1346 1347
            v: Value_::Constant(Constant::Int(val))
        })
qinsoon's avatar
qinsoon committed
1348 1349
    }

qinsoon's avatar
qinsoon committed
1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
    fn emit_binop (&mut self, node: &TreeNode, inst: &Instruction, op: BinOp, op1: OpIndex, op2: OpIndex, f_content: &FunctionContent, f_context: &mut FunctionContext, vm: &VM) {
        let ops = inst.ops.read().unwrap();

        let res_tmp = self.get_result_value(node);

        match op {
            op::BinOp::Add => {
                if self.match_ireg(&ops[op1]) && self.match_iimm(&ops[op2]) {
                    trace!("emit add-ireg-imm");

                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
                    let reg_op2 = self.node_iimm_to_i32(&ops[op2]);

                    // mov op1, res
                    self.backend.emit_mov_r_r(&res_tmp, &reg_op1);
                    // add op2, res
                    self.backend.emit_add_r_imm(&res_tmp, reg_op2);
                } else if self.match_ireg(&ops[op1]) && self.match_mem(&ops[op2]) {
                    trace!("emit add-ireg-mem");

                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
                    let reg_op2 = self.emit_mem(&ops[op2], vm);

                    // mov op1, res
                    self.backend.emit_mov_r_r(&res_tmp, &reg_op1);
                    // add op2 res
                    self.backend.emit_add_r_mem(&res_tmp, &reg_op2);
                } else if self.match_ireg(&ops[op1]) && self.match_ireg(&ops[op2]) {
                    trace!("emit add-ireg-ireg");

                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
                    let reg_op2 = self.emit_ireg(&ops[op2], f_content, f_context, vm);

                    // mov op1, res
                    self.backend.emit_mov_r_r(&res_tmp, &reg_op1);
                    // add op2 res
                    self.backend.emit_add_r_r(&res_tmp, &reg_op2);
qinsoon's avatar
qinsoon committed
1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
                } else if self.match_ireg_ex(&ops[op1]) && self.match_ireg_ex(&ops[op2]){
                    trace!("emit add-iregex-iregex");

                    let (op1_l, op1_h) = self.emit_ireg_ex(&ops[op1], f_content, f_context, vm);
                    let (op2_l, op2_h) = self.emit_ireg_ex(&ops[op2], f_content, f_context, vm);

                    // make result split
                    // mov op1 to res
                    let (res_l, res_h) = self.split_int128(&res_tmp, f_context, vm);
                    self.backend.emit_mov_r_r(&res_l, &op1_l);
                    self.backend.emit_mov_r_r(&res_h, &op1_h);

                    // add res_l op2_l -> res_l
                    self.backend.emit_add_r_r(&res_l, &op2_l);

                    // adc res_h op2_h -> res_h
                    self.backend.emit_adc_r_r(&res_h, &op2_h);
qinsoon's avatar
qinsoon committed
1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414
                }
            },
            op::BinOp::Sub => {
                if self.match_ireg(&ops[op1]) && self.match_iimm(&ops[op2]) {
                    trace!("emit sub-ireg-imm");

                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
                    let imm_op2 = self.node_iimm_to_i32(&ops[op2]);

                    // mov op1, res
                    self.backend.emit_mov_r_r(&res_tmp, &reg_op1);
1415
                    // sub op2, res
qinsoon's avatar
qinsoon committed
1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597