To protect your data, the CISO officer has suggested users to enable GitLab 2FA as soon as possible.

inst_sel.rs 88.5 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::OpCode;
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
12
13
use runtime::ValueLocation;
use runtime::thread;
use runtime::entrypoints;
use runtime::entrypoints::RuntimeEntrypoint;
14
15

use compiler::CompilerPass;
16
use compiler::backend;
qinsoon's avatar
qinsoon committed
17
use compiler::backend::PROLOGUE_BLOCK_NAME;
qinsoon's avatar
qinsoon committed
18
19
20
use compiler::backend::x86_64;
use compiler::backend::x86_64::CodeGenerator;
use compiler::backend::x86_64::ASMCodeGen;
qinsoon's avatar
qinsoon committed
21
22
use compiler::machine_code::CompiledFunction;
use compiler::frame::Frame;
23

24
use std::collections::HashMap;
qinsoon's avatar
qinsoon committed
25
use std::any::Any;
26

27
pub struct InstructionSelection {
28
    name: &'static str,
29
30
    backend: Box<CodeGenerator>,
    
qinsoon's avatar
qinsoon committed
31
    current_callsite_id: usize,
qinsoon's avatar
qinsoon committed
32
33
    current_frame: Option<Frame>,
    current_block: Option<MuName>,
qinsoon's avatar
qinsoon committed
34
35
36
37
38
    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
    current_exn_blocks: HashMap<MuID, ValueLocation>     
39
40
}

41
impl <'a> InstructionSelection {
qinsoon's avatar
qinsoon committed
42
    #[cfg(feature = "aot")]
43
    pub fn new() -> InstructionSelection {
44
45
        InstructionSelection{
            name: "Instruction Selection (x64)",
46
            backend: Box::new(ASMCodeGen::new()),
qinsoon's avatar
qinsoon committed
47
            
qinsoon's avatar
qinsoon committed
48
            current_callsite_id: 0,
qinsoon's avatar
qinsoon committed
49
50
51
            current_frame: None,
            current_block: None,
            current_func_start: None,
qinsoon's avatar
qinsoon committed
52
53
54
            // key: block id, val: callsite that names the block as exception block
            current_exn_callsites: HashMap::new(), 
            current_exn_blocks: HashMap::new()
55
56
        }
    }
qinsoon's avatar
qinsoon committed
57
58
59
60
61

    #[cfg(feature = "jit")]
    pub fn new() -> InstructionSelection {
        unimplemented!()
    }
62
63
64
65
66
67
68
    
    // in this pass, we assume that
    // 1. all temporaries will use 64bit registers
    // 2. we do not need to backup/restore caller-saved registers
    // 3. we need to backup/restore all the callee-saved registers
    // if any of these assumption breaks, we will need to re-emit the code
    #[allow(unused_variables)]
69
    fn instruction_select(&mut self, node: &'a TreeNode, f_content: &FunctionContent, f_context: &mut FunctionContext, vm: &VM) {
qinsoon's avatar
qinsoon committed
70
71
72
        trace!("instsel on node {}", node);
        
        match node.v {
73
74
            TreeNode_::Instruction(ref inst) => {
                match inst.v {
qinsoon's avatar
qinsoon committed
75
                    Instruction_::Branch2{cond, ref true_dest, ref false_dest, true_prob} => {
76
77
                        // '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)
78
79
80
81
82
                        let (fallthrough_dest, branch_dest, branch_if_true) = {
                            if true_prob > 0.5f32 {
                                (true_dest, false_dest, false)
                            } else {
                                (false_dest, true_dest, true)
83
                            }
84
                        };
85
                        
qinsoon's avatar
qinsoon committed
86
                        let ops = inst.ops.read().unwrap();
87
                        
88
89
                        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
90
                        
91
                        let branch_target = f_content.get_block(branch_dest.target).name().unwrap();
92
93
94
    
                        let ref cond = ops[cond];
                        
qinsoon's avatar
qinsoon committed
95
96
                        if self.match_cmp_res(cond) {
                            trace!("emit cmp_eq-branch2");
97
                            match self.emit_cmp_res(cond, f_content, f_context, vm) {
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
                                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
168
169
170
171
                                _ => unimplemented!()
                            }
                        } else if self.match_ireg(cond) {
                            trace!("emit ireg-branch2");
172
                            
173
                            let cond_reg = self.emit_ireg(cond, f_content, f_context, vm);
174
                            
qinsoon's avatar
qinsoon committed
175
                            // emit: cmp cond_reg 1
176
                            self.backend.emit_cmp_imm32_r64(1, &cond_reg);
qinsoon's avatar
qinsoon committed
177
                            // emit: je #branch_dest
178
                            self.backend.emit_je(branch_target);
qinsoon's avatar
qinsoon committed
179
180
                        } else {
                            unimplemented!();
181
                        }
182
183
                    },
                    
qinsoon's avatar
qinsoon committed
184
                    Instruction_::Branch1(ref dest) => {
qinsoon's avatar
qinsoon committed
185
                        let ops = inst.ops.read().unwrap();
186
                                            
187
                        self.process_dest(&ops, dest, f_content, f_context, vm);
188
                        
189
                        let target = f_content.get_block(dest.target).name().unwrap();
qinsoon's avatar
qinsoon committed
190
                        
qinsoon's avatar
qinsoon committed
191
                        trace!("emit branch1");
192
                        // jmp
qinsoon's avatar
qinsoon committed
193
                        self.backend.emit_jmp(target);
194
195
                    },
                    
qinsoon's avatar
qinsoon committed
196
                    Instruction_::ExprCall{ref data, is_abort} => {
qinsoon's avatar
qinsoon committed
197
198
                        if is_abort {
                            unimplemented!()
199
                        }
200
                        
qinsoon's avatar
qinsoon committed
201
202
203
204
205
                        self.emit_mu_call(
                            inst, // inst: &Instruction,
                            data, // calldata: &CallData,
                            None, // resumption: Option<&ResumptionData>,
                            node, // cur_node: &TreeNode, 
206
                            f_content, f_context, vm);
207
208
                    },
                    
qinsoon's avatar
qinsoon committed
209
210
211
212
213
214
215
216
217
                    Instruction_::Call{ref data, ref resume} => {
                        self.emit_mu_call(
                            inst, 
                            data, 
                            Some(resume), 
                            node, 
                            f_content, f_context, vm);
                    }
                    
218
                    Instruction_::Return(_) => {
219
                        self.emit_common_epilogue(inst, f_content, f_context, vm);
220
                        
qinsoon's avatar
qinsoon committed
221
                        self.backend.emit_ret();
222
223
                    },
                    
qinsoon's avatar
qinsoon committed
224
                    Instruction_::BinOp(op, op1, op2) => {
qinsoon's avatar
qinsoon committed
225
                        let ops = inst.ops.read().unwrap();
226
227

                        let res_tmp = self.get_result_value(node);
qinsoon's avatar
qinsoon committed
228
                        
229
230
                        match op {
                            op::BinOp::Add => {
qinsoon's avatar
qinsoon committed
231
                                if self.match_ireg(&ops[op1]) && self.match_iimm(&ops[op2]) {
qinsoon's avatar
qinsoon committed
232
233
                                    trace!("emit add-ireg-imm");
                                    
234
                                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
235
                                    let reg_op2 = self.node_iimm_to_i32(&ops[op2]);
qinsoon's avatar
qinsoon committed
236
237
238
239
240
241
242
243
                                    
                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &reg_op1);
                                    // add op2, res
                                    self.backend.emit_add_r64_imm32(&res_tmp, reg_op2);
                                } else if self.match_ireg(&ops[op1]) && self.match_mem(&ops[op2]) {
                                    trace!("emit add-ireg-mem");
                                    
244
                                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
qinsoon's avatar
shl    
qinsoon committed
245
                                    let reg_op2 = self.emit_mem(&ops[op2], vm);
qinsoon's avatar
qinsoon committed
246
247
248
249
250
                                    
                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &reg_op1);
                                    // add op2 res
                                    self.backend.emit_add_r64_mem64(&res_tmp, &reg_op2);
qinsoon's avatar
qinsoon committed
251
252
253
254
255
256
257
258
259
260
                                } 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_r64_r64(&res_tmp, &reg_op1);
                                    // add op2 res
                                    self.backend.emit_add_r64_r64(&res_tmp, &reg_op2);
qinsoon's avatar
qinsoon committed
261
262
263
                                } else {
                                    unimplemented!()
                                }
264
265
                            },
                            op::BinOp::Sub => {
266
                                if self.match_ireg(&ops[op1]) && self.match_iimm(&ops[op2]) {
qinsoon's avatar
qinsoon committed
267
268
                                    trace!("emit sub-ireg-imm");

269
                                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
270
                                    let imm_op2 = self.node_iimm_to_i32(&ops[op2]);
qinsoon's avatar
qinsoon committed
271
272
273
274
                                    
                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &reg_op1);
                                    // add op2, res
275
                                    self.backend.emit_sub_r64_imm32(&res_tmp, imm_op2);
qinsoon's avatar
qinsoon committed
276
277
278
                                } else if self.match_ireg(&ops[op1]) && self.match_mem(&ops[op2]) {
                                    trace!("emit sub-ireg-mem");
                                    
279
                                    let reg_op1 = self.emit_ireg(&ops[op1], f_content, f_context, vm);
qinsoon's avatar
shl    
qinsoon committed
280
                                    let mem_op2 = self.emit_mem(&ops[op2], vm);
qinsoon's avatar
qinsoon committed
281
282
283
284
                                    
                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &reg_op1);
                                    // sub op2 res
285
                                    self.backend.emit_sub_r64_mem64(&res_tmp, &mem_op2);
286
287
                                } else if self.match_ireg(&ops[op1]) && self.match_ireg(&ops[op2]) {
                                    trace!("emit sub-ireg-ireg");
288

289
290
                                    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);
291

292
293
294
295
                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &reg_op1);
                                    // add op2 res
                                    self.backend.emit_sub_r64_r64(&res_tmp, &reg_op2);
qinsoon's avatar
qinsoon committed
296
297
298
                                } else {
                                    unimplemented!()
                                }
299
                            },
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
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
                            op::BinOp::And => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

                                if self.match_ireg(op1) && self.match_iimm(op2) {
                                    trace!("emit and-ireg-iimm");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let imm_op2 = self.node_iimm_to_i32(op2);

                                    // mov op1 -> res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // and op2, res -> res
                                    self.backend.emit_and_r64_imm32(&res_tmp, imm_op2);
                                } else if self.match_ireg(op1) && self.match_mem(op2) {
                                    trace!("emit and-ireg-mem");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let mem_op2 = self.emit_mem(op2, vm);

                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // and op2, res -> res
                                    self.backend.emit_and_r64_mem64(&res_tmp, &mem_op2);
                                } else if self.match_ireg(op1) && self.match_ireg(op2) {
                                    trace!("emit and-ireg-ireg");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let tmp_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // and op2, res -> res
                                    self.backend.emit_and_r64_r64(&res_tmp, &tmp_op2);
                                } else {
                                    unimplemented!()
                                }
                            },
                            op::BinOp::Xor => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

                                if self.match_ireg(op1) && self.match_iimm(op2) {
                                    trace!("emit xor-ireg-iimm");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let imm_op2 = self.node_iimm_to_i32(op2);

                                    // mov op1 -> res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // xor op2, res -> res
                                    self.backend.emit_xor_r64_imm32(&res_tmp, imm_op2);
                                } else if self.match_ireg(op1) && self.match_mem(op2) {
                                    trace!("emit xor-ireg-mem");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let mem_op2 = self.emit_mem(op2, vm);

                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // xor op2, res -> res
                                    self.backend.emit_xor_r64_mem64(&res_tmp, &mem_op2);
                                } else if self.match_ireg(op1) && self.match_ireg(op2) {
                                    trace!("emit xor-ireg-ireg");

                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);
                                    let tmp_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                    // mov op1, res
                                    self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    // xor op2, res -> res
                                    self.backend.emit_xor_r64_r64(&res_tmp, &tmp_op2);
                                } else {
                                    unimplemented!()
                                }
                            }
376
                            op::BinOp::Mul => {
377
378
379
                                // mov op1 -> rax
                                let rax = x86_64::RAX.clone();
                                let op1 = &ops[op1];
380
                                if self.match_iimm(op1) {
381
                                    let imm_op1 = self.node_iimm_to_i32(op1);
382
383
384
                                    
                                    self.backend.emit_mov_r64_imm32(&rax, imm_op1);
                                } else if self.match_mem(op1) {
qinsoon's avatar
shl    
qinsoon committed
385
                                    let mem_op1 = self.emit_mem(op1, vm);
386
387
                                    
                                    self.backend.emit_mov_r64_mem64(&rax, &mem_op1);
388
389
390
391
                                } else if self.match_ireg(op1) {
                                    let reg_op1 = self.emit_ireg(op1, f_content, f_context, vm);

                                    self.backend.emit_mov_r64_r64(&rax, &reg_op1);
392
393
394
395
396
397
                                } else {
                                    unimplemented!();
                                }
                                
                                // mul op2 -> rax
                                let op2 = &ops[op2];
398
                                if self.match_iimm(op2) {
399
                                    let imm_op2 = self.node_iimm_to_i32(op2);
400
401
402
403
404
405
406
                                    
                                    // put imm in a temporary
                                    // here we use result reg as temporary
                                    self.backend.emit_mov_r64_imm32(&res_tmp, imm_op2);
                                    
                                    self.backend.emit_mul_r64(&res_tmp);
                                } else if self.match_mem(op2) {
qinsoon's avatar
shl    
qinsoon committed
407
                                    let mem_op2 = self.emit_mem(op2, vm);
408
409
                                    
                                    self.backend.emit_mul_mem64(&mem_op2);
410
411
412
413
                                } else if self.match_ireg(op2) {
                                    let reg_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                    self.backend.emit_mul_r64(&reg_op2);
414
415
416
417
418
419
                                } else {
                                    unimplemented!();
                                }
                                
                                // mov rax -> result
                                self.backend.emit_mov_r64_r64(&res_tmp, &rax);
420
                            },
qinsoon's avatar
qinsoon committed
421
422
423
424
                            op::BinOp::Udiv => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

qinsoon's avatar
sdiv    
qinsoon committed
425
                                self.emit_udiv(op1, op2, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
426

qinsoon's avatar
sdiv    
qinsoon committed
427
428
429
430
431
432
                                // mov rax -> result
                                self.backend.emit_mov_r64_r64(&res_tmp, &x86_64::RAX);
                            },
                            op::BinOp::Sdiv => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];
qinsoon's avatar
qinsoon committed
433

qinsoon's avatar
sdiv    
qinsoon committed
434
                                self.emit_idiv(op1, op2, f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
435
436

                                // mov rax -> result
qinsoon's avatar
sdiv    
qinsoon committed
437
438
                                self.backend.emit_mov_r64_r64(&res_tmp, &x86_64::RAX);
                            },
qinsoon's avatar
qinsoon committed
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
                            op::BinOp::Urem => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

                                self.emit_udiv(op1, op2, f_content, f_context, vm);

                                // mov rdx -> result
                                self.backend.emit_mov_r64_r64(&res_tmp, &x86_64::RDX);
                            },
                            op::BinOp::Srem => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

                                self.emit_idiv(op1, op2, f_content, f_context, vm);

                                // mov rdx -> result
                                self.backend.emit_mov_r64_r64(&res_tmp, &x86_64::RDX);
                            },
qinsoon's avatar
qinsoon committed
457

qinsoon's avatar
shl    
qinsoon committed
458
459
460
461
                            op::BinOp::Shl => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

462
463
464
                                if self.match_mem(op1) {
                                    unimplemented!()
                                } else if self.match_ireg(op1) {
qinsoon's avatar
shl    
qinsoon committed
465
466
                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);

467
468
469
470
471
472
473
474
475
                                    if self.match_iimm(op2) {
                                        let imm_op2 = self.node_iimm_to_i32(op2) as i8;

                                        // shl op1, op2 -> op1
                                        self.backend.emit_shl_r64_imm8(&tmp_op1, imm_op2);

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    } else if self.match_ireg(op2) {
qinsoon's avatar
shl    
qinsoon committed
476
477
478
479
480
481
                                        let tmp_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                        // mov op2 -> rcx
                                        self.backend.emit_mov_r64_r64(&x86_64::RCX, &tmp_op2);

                                        // shl op1, cl -> op1
qinsoon's avatar
fix    
qinsoon committed
482
                                        self.backend.emit_shl_r64_cl(&tmp_op1);
qinsoon's avatar
shl    
qinsoon committed
483
484
485
486
487
488

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    } else {
                                        panic!("unexpected op2 (not ireg not iimm): {}", op2);
                                    }
489
490
                                } else {
                                    panic!("unexpected op1 (not ireg not mem): {}", op1);
qinsoon's avatar
shl    
qinsoon committed
491
                                }
qinsoon's avatar
qinsoon committed
492
493
494
495
496
                            },
                            op::BinOp::Lshr => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

497
498
499
                                if self.match_mem(op1) {
                                    unimplemented!()
                                } else if self.match_ireg(op1) {
qinsoon's avatar
qinsoon committed
500
501
                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);

502
503
504
505
506
507
508
509
510
                                    if self.match_iimm(op2) {
                                        let imm_op2 = self.node_iimm_to_i32(op2) as i8;

                                        // shr op1, op2 -> op1
                                        self.backend.emit_shr_r64_imm8(&tmp_op1, imm_op2);

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    } else if self.match_ireg(op2) {
qinsoon's avatar
qinsoon committed
511
512
513
514
515
516
517
518
519
520
521
522
523
                                        let tmp_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                        // mov op2 -> rcx
                                        self.backend.emit_mov_r64_r64(&x86_64::RCX, &tmp_op2);

                                        // shr op1, cl -> op1
                                        self.backend.emit_shr_r64_cl(&tmp_op1);

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    } else {
                                        panic!("unexpected op2 (not ireg not iimm): {}", op2);
                                    }
524
525
                                } else {
                                    panic!("unexpected op1 (not ireg not mem): {}", op1);
qinsoon's avatar
qinsoon committed
526
527
528
529
530
531
                                }
                            },
                            op::BinOp::Ashr => {
                                let op1 = &ops[op1];
                                let op2 = &ops[op2];

532
533
534
                                if self.match_mem(op1) {
                                    unimplemented!()
                                } else if self.match_ireg(op1) {
qinsoon's avatar
qinsoon committed
535
536
                                    let tmp_op1 = self.emit_ireg(op1, f_content, f_context, vm);

537
538
539
540
541
542
543
544
545
                                    if self.match_iimm(op2) {
                                        let imm_op2 = self.node_iimm_to_i32(op2) as i8;

                                        // sar op1, op2 -> op1
                                        self.backend.emit_sar_r64_imm8(&tmp_op1, imm_op2);

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
                                    } else if self.match_ireg(op2) {
qinsoon's avatar
qinsoon committed
546
547
548
549
550
551
552
553
554
555
                                        let tmp_op2 = self.emit_ireg(op2, f_content, f_context, vm);

                                        // mov op2 -> rcx
                                        self.backend.emit_mov_r64_r64(&x86_64::RCX, &tmp_op2);

                                        // sar op1, cl -> op1
                                        self.backend.emit_sar_r64_cl(&tmp_op1);

                                        // mov op1 -> result
                                        self.backend.emit_mov_r64_r64(&res_tmp, &tmp_op1);
556
                                    } else  {
qinsoon's avatar
qinsoon committed
557
558
                                        panic!("unexpected op2 (not ireg not iimm): {}", op2);
                                    }
559
560
                                } else {
                                    panic!("unexpected op1 (not ireg not mem): {}", op1);
qinsoon's avatar
qinsoon committed
561
562
563
                                }
                            },

qinsoon's avatar
shl    
qinsoon committed
564

qinsoon's avatar
qinsoon committed
565
566
                            // floating point
                            op::BinOp::FAdd => {
567
                                if self.match_fpreg(&ops[op1]) && self.match_mem(&ops[op2]) {
qinsoon's avatar
qinsoon committed
568
569
570
                                    trace!("emit add-fpreg-mem");

                                    let reg_op1 = self.emit_fpreg(&ops[op1], f_content, f_context, vm);
qinsoon's avatar
shl    
qinsoon committed
571
                                    let mem_op2 = self.emit_mem(&ops[op2], vm);
qinsoon's avatar
qinsoon committed
572
573
574
575
576

                                    // mov op1, res
                                    self.backend.emit_movsd_f64_f64(&res_tmp, &reg_op1);
                                    // sub op2 res
                                    self.backend.emit_addsd_f64_mem64(&res_tmp, &mem_op2);
577
578
579
580
581
582
583
584
585
586
                                } else if self.match_fpreg(&ops[op1]) && self.match_fpreg(&ops[op2]) {
                                    trace!("emit add-fpreg-fpreg");

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

                                    // movsd op1, res
                                    self.backend.emit_movsd_f64_f64(&res_tmp, &reg_op1);
                                    // add op2 res
                                    self.backend.emit_addsd_f64_f64(&res_tmp, &reg_op2);
qinsoon's avatar
qinsoon committed
587
588
589
590
                                } else {
                                    unimplemented!()
                                }
                            }
591
592
                            
                            _ => unimplemented!()
593
                        }
594
595
596
597
598

                        // truncate result
                        if res_tmp.is_int_reg() {
                            self.emit_truncate_result(&UINT64_TYPE, &res_tmp.ty, &res_tmp);
                        }
599
                    }
qinsoon's avatar
qinsoon committed
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624

                    Instruction_::ConvOp{operation, ref from_ty, ref to_ty, operand} => {
                        let ops = inst.ops.read().unwrap();

                        let ref op = ops[operand];

                        let extract_int_len = |x: &P<MuType>| {
                            match x.v {
                                MuType_::Int(len) => len,
                                _ => panic!("only expect int types, found: {}", x)
                            }
                        };

                        match operation {
                            op::ConvOp::TRUNC => {
                                // currently only use 64bits register
                                // so only keep what is needed in the register (set others to 0)

                                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);

                                    // mov op -> result
                                    self.backend.emit_mov_r64_r64(&tmp_res, &tmp_op);

625
626
                                    // truncate result
                                    self.emit_truncate_result(from_ty, to_ty, &tmp_res);
qinsoon's avatar
qinsoon committed
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
                                }
                            }
                            op::ConvOp::ZEXT => {
                                // currently only use 64bits register
                                // so set irrelevant bits to 0
                                let from_ty_len = extract_int_len(from_ty);
                                let to_ty_len   = extract_int_len(to_ty);

                                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);

                                    let mask = match from_ty_len {
                                        8  => 0xFFi32,
                                        16 => 0xFFFFi32,
644
                                        32 => 0xFFFFFFFFi32,
qinsoon's avatar
qinsoon committed
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
                                        _ => unimplemented!()
                                    };

                                    // mov op -> result
                                    self.backend.emit_mov_r64_r64(&tmp_res, &tmp_op);

                                    // and mask result -> result
                                    self.backend.emit_and_r64_imm32(&tmp_res, mask);
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op);
                                }
                            },
                            op::ConvOp::SEXT => {
                                // currently only use 64bits register
                                // we left shift the value, then arithmetic right shift back
                                let from_ty_len = extract_int_len(from_ty);
                                let to_ty_len   = extract_int_len(to_ty);

                                let shift : i8 = (to_ty_len - from_ty_len) as i8;

                                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);

                                    // mov op -> result
                                    self.backend.emit_mov_r64_r64(&tmp_res, &tmp_op);

                                    // shl result, shift -> result
                                    self.backend.emit_shl_r64_imm8(&tmp_res, shift);

                                    // sar result, shift -> result
                                    self.backend.emit_sar_r64_imm8(&tmp_res, shift);
                                } else {
                                    panic!("unexpected op (expect ireg): {}", op)
                                }
                            }

                            _ => unimplemented!()
                        }
                    }
685
                    
686
687
                    // load on x64 generates mov inst (no matter what order is specified)
                    // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
688
                    Instruction_::Load{is_ptr, order, mem_loc} => {
qinsoon's avatar
qinsoon committed
689
                        let ops = inst.ops.read().unwrap();
690
                        let ref loc_op = ops[mem_loc];
691
692
693
694
695
696
697
698
699
                        
                        // check order
                        match order {
                            MemoryOrder::Relaxed 
                            | MemoryOrder::Consume 
                            | MemoryOrder::Acquire
                            | MemoryOrder::SeqCst => {},
                            _ => panic!("didnt expect order {:?} with store inst", order)
                        }                        
700

701
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, vm);
qinsoon's avatar
qinsoon committed
702
                        let res_temp = self.get_result_value(node);
703
704
705
                        
                        if self.match_ireg(node) {
                            // emit mov(GPR)
706
707
708
709
710
711
712
713
                            self.backend.emit_mov_r64_mem64(&res_temp, &resolved_loc);
                        } else {
                            // emit mov(FPR)
                            unimplemented!()
                        }
                    }
                    
                    Instruction_::Store{is_ptr, order, mem_loc, value} => {
qinsoon's avatar
qinsoon committed
714
                        let ops = inst.ops.read().unwrap();
715
716
717
718
719
720
721
722
723
724
725
                        let ref loc_op = ops[mem_loc];
                        let ref val_op = ops[value];
                        
                        let generate_plain_mov : bool = {
                            match order {
                                MemoryOrder::Relaxed | MemoryOrder::Release => true,
                                MemoryOrder::SeqCst => false,
                                _ => panic!("didnt expect order {:?} with store inst", order)
                            }
                        };
                        
726
                        let resolved_loc = self.emit_node_addr_to_value(loc_op, vm);
727
728
                        
                        if self.match_ireg(val_op) {
729
                            let val = self.emit_ireg(val_op, f_content, f_context, vm);
730
731
732
733
734
735
                            if generate_plain_mov {
                                self.backend.emit_mov_mem64_r64(&resolved_loc, &val);
                            } else {
                                unimplemented!()
                            }
                        } else if self.match_iimm(val_op) {
736
                            let val = self.node_iimm_to_i32(val_op);
737
738
739
740
741
                            if generate_plain_mov {
                                self.backend.emit_mov_mem64_imm32(&resolved_loc, val);
                            } else {
                                unimplemented!()
                            }
742
743
744
745
746
                        } else {
                            // emit mov(FPR)
                            unimplemented!()
                        }
                    }
747
                    
748
749
750
751
                    Instruction_::GetIRef(op_index) => {
                        let ops = inst.ops.read().unwrap();
                        
                        let ref op = ops[op_index];
qinsoon's avatar
qinsoon committed
752
                        let res_tmp = self.get_result_value(node);
753
                        
754
755
                        let hdr_size = mm::objectmodel::OBJECT_HEADER_SIZE;
                        if hdr_size == 0 {
qinsoon's avatar
qinsoon committed
756
                            self.emit_move_node_to_value(&res_tmp, &op, f_content, f_context, vm);
757
758
759
                        } else {
                            self.emit_lea_base_offset(&res_tmp, &op.clone_value(), hdr_size as i32, vm);
                        }
760
761
                    }
                    
762
                    Instruction_::ThreadExit => {
763
                        // emit a call to swap_back_to_native_stack(sp_loc: Address)
764
765
                        
                        // get thread local and add offset to get sp_loc
qinsoon's avatar
qinsoon committed
766
                        let tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);
767
                        self.backend.emit_add_r64_imm32(&tl, *thread::NATIVE_SP_LOC_OFFSET as i32);
768
                        
qinsoon's avatar
qinsoon committed
769
                        self.emit_runtime_entry(&entrypoints::SWAP_BACK_TO_NATIVE_STACK, vec![tl.clone()], None, Some(node), f_content, f_context, vm);
770
                    }
qinsoon's avatar
qinsoon committed
771
772
773
                    
                    Instruction_::New(ref ty) => {
                        let ty_info = vm.get_backend_type_info(ty.id());
774
775
                        let ty_size = ty_info.size;
                        let ty_align= ty_info.alignment;
qinsoon's avatar
qinsoon committed
776
                        
777
                        if ty_size > mm::LARGE_OBJECT_THRESHOLD {
qinsoon's avatar
qinsoon committed
778
779
780
                            // emit large object allocation
                            unimplemented!()
                        } else {
781
782
783
                            // emit immix allocation fast path
                            
                            // ASM: %tl = get_thread_local()
qinsoon's avatar
qinsoon committed
784
                            let tmp_tl = self.emit_get_threadlocal(Some(node), f_content, f_context, vm);
785
786
787
                            
                            // ASM: mov [%tl + allocator_offset + cursor_offset] -> %cursor
                            let cursor_offset = *thread::ALLOCATOR_OFFSET + *mm::ALLOCATOR_CURSOR_OFFSET;
qinsoon's avatar
qinsoon committed
788
                            let tmp_cursor = self.make_temporary(f_context, ADDRESS_TYPE.clone(), vm);
789
                            self.emit_load_base_offset(&tmp_cursor, &tmp_tl, cursor_offset as i32, vm);
790
791
792
                            
                            // alignup cursor (cursor + align - 1 & !(align - 1))
                            // ASM: lea align-1(%cursor) -> %start
793
                            let align = ty_info.alignment as i32;
qinsoon's avatar
qinsoon committed
794
                            let tmp_start = self.make_temporary(f_context, ADDRESS_TYPE.clone(), vm);
795
                            self.emit_lea_base_offset(&tmp_start, &tmp_cursor, align - 1, vm);
796
                            // ASM: and %start, !(align-1) -> %start
797
                            self.backend.emit_and_r64_imm32(&tmp_start, !(align - 1) as i32);
798
799
800
                            
                            // bump cursor
                            // ASM: lea size(%start) -> %end
qinsoon's avatar
qinsoon committed
801
                            let tmp_end = self.make_temporary(f_context, ADDRESS_TYPE.clone(), vm);
802
                            self.emit_lea_base_offset(&tmp_end, &tmp_start, ty_size as i32, vm);
803
804
805
806
                            
                            // check with limit
                            // ASM: cmp %end, [%tl + allocator_offset + limit_offset]
                            let limit_offset = *thread::ALLOCATOR_OFFSET + *mm::ALLOCATOR_LIMIT_OFFSET;
qinsoon's avatar
qinsoon committed
807
                            let mem_limit = self.make_memory_op_base_offset(&tmp_tl, limit_offset as i32, ADDRESS_TYPE.clone(), vm);
808
                            self.backend.emit_cmp_mem64_r64(&mem_limit, &tmp_end);
809
                            
810
811
                            // branch to slow path if end > limit (end - limit > 0)
                            // ASM: jg alloc_slow
812
                            let slowpath = format!("{}_allocslow", node.id());
813
                            self.backend.emit_jg(slowpath.clone());
814
815
                            
                            // update cursor
816
817
                            // ASM: mov %end -> [%tl + allocator_offset + cursor_offset]
                            self.emit_store_base_offset(&tmp_tl, cursor_offset as i32, &tmp_end, vm);
818
819
820
                            
                            // put start as result
                            // ASM: mov %start -> %result
qinsoon's avatar
qinsoon committed
821
                            let tmp_res = self.get_result_value(node);
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
                            self.backend.emit_mov_r64_r64(&tmp_res, &tmp_start);
                            
                            // ASM jmp alloc_end
                            let allocend = format!("{}_allocend", node.id());
                            self.backend.emit_jmp(allocend.clone());
                            
                            // finishing current block
                            self.backend.end_block(self.current_block.as_ref().unwrap().clone());
                            
                            // alloc_slow: 
                            // call alloc_slow(size, align) -> %ret
                            // new block (no livein)
                            self.current_block = Some(slowpath.clone());
                            self.backend.start_block(slowpath.clone());
                            self.backend.set_block_livein(slowpath.clone(), &vec![]); 
837
838
839

                            // arg1: allocator address                            
                            let allocator_offset = *thread::ALLOCATOR_OFFSET;
qinsoon's avatar
qinsoon committed
840
                            let tmp_allocator = self.make_temporary(f_context, ADDRESS_TYPE.clone(), vm);
841
842
                            self.emit_lea_base_offset(&tmp_allocator, &tmp_tl, allocator_offset as i32, vm);
                            // arg2: size                            
843
                            let const_size = self.make_value_int_const(ty_size as u64, vm);
844
                            // arg3: align
845
                            let const_align= self.make_value_int_const(ty_align as u64, vm);
846
                            
847
848
                            let rets = self.emit_runtime_entry(
                                &entrypoints::ALLOC_SLOW,
849
                                vec![tmp_allocator, const_size, const_align],
850
851
852
                                Some(vec![
                                    tmp_res.clone()
                                ]),
qinsoon's avatar
qinsoon committed
853
                                Some(node), f_content, f_context, vm
854
855
                            );
                            
856
                            // end block (no liveout other than result)
857
858
859
860
861
862
                            self.backend.end_block(slowpath.clone());
                            self.backend.set_block_liveout(slowpath.clone(), &vec![tmp_res.clone()]);
                            
                            // block: alloc_end
                            self.backend.start_block(allocend.clone());
                            self.current_block = Some(allocend.clone());
qinsoon's avatar
qinsoon committed
863
864
                        }
                    }
qinsoon's avatar
qinsoon committed
865
866
867
868
869
870
871
872
873
                    
                    Instruction_::Throw(op_index) => {
                        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
874
                            Some(node), f_content, f_context, vm);
qinsoon's avatar
qinsoon committed
875
                    }
876
877
878
879
880
881
    
                    _ => unimplemented!()
                } // main switch
            },
            
            TreeNode_::Value(ref p) => {
882
        
883
884
885
886
            }
        }
    }
    
887
888
889
890
    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()
    }
    
891
    fn make_memory_op_base_offset (&mut self, base: &P<Value>, offset: i32, ty: P<MuType>, vm: &VM) -> P<Value> {
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
        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
            })
        })
    }
    
    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
907
            ty: UINT64_TYPE.clone(),
908
909
910
            v: Value_::Constant(Constant::Int(val))
        })
    } 
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932

    fn emit_truncate_result (&mut self, from_ty: &P<MuType>, to_ty: &P<MuType>, op: &P<Value>) {
        // currently only use 64bits register
        // so only keep what is needed in the register (set others to 0)
        let from_ty_len = match from_ty.v {
            MuType_::Int(len) => len,
            _ => panic!("only expect int types, found: {}", from_ty)
        };
        let to_ty_len   = match to_ty.v {
            MuType_::Int(len) => len,
            _ => panic!("only expect int types, found: {}", to_ty)
        };

        if from_ty_len == to_ty_len {
            return;
        } else {
            debug_assert!(from_ty_len > to_ty_len);

            // ignoring from_ty for now (we use 64bits register for everything)
            let mask = match to_ty_len {
                8 => 0xFFi32,
                16 => 0xFFFFi32,
933
                32 => 0xFFFFFFFFi32,
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
962
963
964
965
966
967
                _ => unimplemented!()
            };

            // and mask, result -> result
            self.backend.emit_and_r64_imm32(&op, mask);
        }
    }

    fn emit_sign_extend_operand (&mut self, from_ty: &P<MuType>, to_ty: &P<MuType>, op: &P<Value>) {
        // currently only use 64bits register
        // we left shift the value, then arithmetic right shift back
        let from_ty_len = match from_ty.v {
            MuType_::Int(len) => len,
            _ => panic!("only expect int types, found: {}", from_ty)
        };
        let to_ty_len   = match to_ty.v {
            MuType_::Int(len) => len,
            _ => panic!("only expect int types, found: {}", to_ty)
        };

        if from_ty_len == to_ty_len {
            return;
        } else {
            debug_assert!(to_ty_len > from_ty_len);

            let shift : i8 = (to_ty_len - from_ty_len) as i8;

            // shl result, shift -> result
            self.backend.emit_shl_r64_imm8(&op, shift);
            // sar result, shift -> result
            self.backend.emit_sar_r64_imm8(&op, shift);
        }
    }

968
    fn emit_load_base_offset (&mut self, dest: &P<Value>, base: &P<Value>, offset: i32, vm: &VM) {
969
        let mem = self.make_memory_op_base_offset(base, offset, dest.ty.clone(), vm);
qinsoon's avatar
qinsoon committed
970
971
972
973
974
975
976
977

        if dest.is_int_reg() {
            self.backend.emit_mov_r64_mem64(dest, &mem);
        } else if dest.is_fp_reg() {
            self.backend.emit_movsd_f64_mem64(dest, &mem);
        } else {
            unimplemented!();
        }
978
979
    }
    
980
    fn emit_store_base_offset (&mut self, base: &P<Value>, offset: i32, src: &P<Value>, vm: &VM) {
981
982
983
984
985
        let mem = self.make_memory_op_base_offset(base, offset, src.ty.clone(), vm);
        
        self.backend.emit_mov_mem64_r64(&mem, src);
    }
    
986
    fn emit_lea_base_offset (&mut self, dest: &P<Value>, base: &P<Value>, offset: i32, vm: &VM) {
qinsoon's avatar
qinsoon committed
987
        let mem = self.make_memory_op_base_offset(base, offset, ADDRESS_TYPE.clone(), vm);
988
989
990
        
        self.backend.emit_lea_r64(dest, &mem);
    }
qinsoon's avatar
sdiv    
qinsoon committed
991
992
993
994
995
996
997
998
999

    fn emit_udiv (
        &mut self,
        op1: &P<TreeNode>, op2: &P<TreeNode>,
        f_content: &FunctionContent,
        f_context: &mut FunctionContext,
        vm: &VM)
    {
        let rax = x86_64::RAX.clone();
1000
1001
1002
1003

        debug_assert!(self.match_ireg(op1));
        let reg_op1 = self.emit_ireg(op1, f_content, f_context, vm);
        self.emit_move_value_to_value(&rax, &reg_op1);
qinsoon's avatar
sdiv    
qinsoon committed
1004
1005
1006
1007
1008
1009

        // xorq rdx, rdx -> rdx
        let rdx = x86_64::RDX.clone();
        self.backend.emit_xor_r64_r64(&rdx, &rdx);

        // div op2
1010
        if self.match_mem(op2) {
qinsoon's avatar
shl    
qinsoon committed
1011
            let mem_op2 = self.emit_mem(op2, vm);
qinsoon's avatar
sdiv    
qinsoon committed
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021

            self.backend.emit_div_mem64(&mem_op2);
        } else if self.match_iimm(op2) {
            let imm = self.node_iimm_to_i32(op2);
            // moving to a temp
            let temp = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
            self.backend.emit_mov_r64_imm32(&temp, imm);

            // div tmp
            self.backend.emit_div_r64(&temp);
1022
1023
1024
1025
        } else if self.match_ireg(op2) {
            let reg_op2 = self.emit_ireg(op2, f_content, f_context, vm);

            self.backend.emit_div_r64(&reg_op2);
qinsoon's avatar
sdiv    
qinsoon committed
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
        } else {
            unimplemented!();
        }
    }

    fn emit_idiv (
        &mut self,
        op1: &P<TreeNode>, op2: &P<TreeNode>,
        f_content: &FunctionContent,
        f_context: &mut FunctionContext,
        vm: &VM)
    {
        let rax = x86_64::RAX.clone();
1039

1040
1041
1042
        debug_assert!(self.match_ireg(op1));
        let reg_op1 = self.emit_ireg(op1, f_content, f_context, vm);
        self.emit_move_value_to_value(&rax, &reg_op1);
1043
1044

        // sign extend rax
1045
        self.emit_sign_extend_operand(&reg_op1.ty, &UINT64_TYPE, &rax);
qinsoon's avatar
sdiv    
qinsoon committed
1046
1047
1048
1049
1050

        // cqo
        self.backend.emit_cqo();

        // idiv op2
1051
        if self.match_mem(op2) {
qinsoon's avatar
shl    
qinsoon committed
1052
            let mem_op2 = self.emit_mem(op2, vm);
qinsoon's avatar
sdiv    
qinsoon committed
1053
            self.backend.emit_idiv_mem64(&mem_op2);
1054
1055
1056

            // need to sign extend op2
            unimplemented!()
qinsoon's avatar
sdiv    
qinsoon committed
1057
1058
1059
1060
1061
1062
1063
1064
        } else if self.match_iimm(op2) {
            let imm = self.node_iimm_to_i32(op2);
            // moving to a temp
            let temp = self.make_temporary(f_context, UINT64_TYPE.clone(), vm);
            self.backend.emit_mov_r64_imm32(&temp, imm);

            // idiv temp
            self.backend.emit_idiv_r64(&temp);
1065
1066
1067
1068
1069
1070
        } else if self.match_ireg(op2) {
            let reg_op2 = self.emit_ireg(op2, f_content, f_context, vm);

            self.emit_sign_extend_operand(&reg_op2.ty, &UINT64_TYPE, &reg_op2);

            self.backend.emit_idiv_r64(&reg_op2);
qinsoon's avatar
sdiv    
qinsoon committed
1071
1072
1073
1074
        } else {
            unimplemented!();
        }
    }
1075
    
qinsoon's avatar
qinsoon committed
1076
1077
    fn emit_get_threadlocal (
        &mut self, 
qinsoon's avatar
qinsoon committed
1078
        cur_node: Option<&TreeNode>,
qinsoon's avatar
qinsoon committed
1079
1080
1081
1082
        f_content: &FunctionContent, 
        f_context: &mut FunctionContext, 
        vm: &VM) -> P<Value> {
        let mut rets = self.emit_runtime_entry(&entrypoints::GET_THREAD_LOCAL, vec![], None, cur_node, f_content, f_context, vm);
1083
1084
1085
1086
        
        rets.pop().unwrap()
    }
    
1087
1088
1089
1090
    // ret: Option<Vec<P<Value>>
    // if ret is Some, return values will put stored in given temporaries
    // otherwise create temporaries
    // always returns result temporaries (given or created)
qinsoon's avatar
qinsoon committed
1091
1092
1093
1094
1095
    fn emit_runtime_entry (
        &mut self, 
        entry: &RuntimeEntrypoint, 
        args: Vec<P<Value>>, 
        rets: Option<Vec<P<Value>>>,
qinsoon's avatar
qinsoon committed
1096
        cur_node: Option<&TreeNode>, 
qinsoon's avatar
qinsoon committed
1097
1098
1099
        f_content: &FunctionContent, 
        f_context: &mut FunctionContext, 
        vm: &VM) -> Vec<P<Value>> {
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
        let sig = entry.sig.clone();
        
        let entry_name = {
            if vm.is_running() {
                unimplemented!()
            } else {
                let ref entry_loc = entry.aot;
                
                match entry_loc {
                    &ValueLocation::Relocatable(_, ref name) => name.clone(),
                    _ => panic!("expecting a relocatable value")
                }
            }
        };
        
qinsoon's avatar
qinsoon committed
1115
        self.emit_c_call(entry_name, sig, args, rets, cur_node, f_content, f_context, vm)
1116
1117
    }
    
1118
1119
1120
1121
    // returns the stack arg offset - we will need this to collapse stack after the call
    fn emit_precall_convention(
        &mut self,
        args: &Vec<P<Value>>, 
1122
        vm: &VM) -> usize {
1123
1124
1125
1126
1127
1128
        // if we need to save caller saved regs
        // put it here (since this is fastpath compile, we wont have them)
        
        // put args into registers if we can
        // in the meantime record args that do not fit in registers
        let mut stack_args : Vec<P<Value>> = vec![];        
1129
1130
1131
1132
1133
1134
1135
1136
        let mut gpr_arg_count = 0;
        for arg in args.iter() {
            if arg.is_int_reg() {
                if gpr_arg_count < x86_64::ARGUMENT_GPRs.len() {
                    self.backend.emit_mov_r64_r64(&x86_64::ARGUMENT_GPRs[gpr_arg_count], &arg);
                    gpr_arg_count += 1;
                } else {
                    // use stack to pass argument
1137
                    stack_args.push(arg.clone());
1138
1139
1140
                }
            } else if arg.is_int_const() {
                if x86_64::is_valid_x86_imm(arg) {                
1141
                    let int_const = arg.extract_int_const() as i32;
1142
1143
1144
1145
1146
1147
                    
                    if gpr_arg_count < x86_64::ARGUMENT_GPRs.len() {
                        self.backend.emit_mov_r64_imm32(&x86_64::ARGUMENT_GPRs[gpr_arg_count], int_const);
                        gpr_arg_count += 1;
                    } else {
                        // use stack to pass argument
1148
                        stack_args.push(arg.clone());
1149
1150
1151
1152
1153
                    }
                } else {
                    // put the constant to memory
                    unimplemented!()
                }
1154
1155
1156
1157
1158
1159
            } else if arg.is_mem() {
                if gpr_arg_count < x86_64::ARGUMENT_GPRs.len() {
                    self.backend.emit_mov_r64_mem64(&x86_64::ARGUMENT_GPRs[gpr_arg_count], &arg);
                    gpr_arg_count += 1;
                } else {
                    // use stack to pass argument
1160
                    stack_args.push(arg.clone());
1161
                }
1162
1163
1164
1165
1166
            } else {
                // floating point
                unimplemented!()
            }
        }
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197

        if !stack_args.is_empty() {
            // deal with stack arg, put them on stack
            // in reverse order, i.e. push the rightmost arg first to stack
            stack_args.reverse();

            // "The end of the input argument area shall be aligned on a 16
            // (32, if __m256 is passed on stack) byte boundary." - x86 ABI
            // if we need to special align the args, we do it now
            // (then the args will be put to stack following their regular alignment)
            let stack_arg_tys = stack_args.iter().map(|x| x.ty.clone()).collect();
            let (stack_arg_size, _, stack_arg_offsets) = backend::sequetial_layout(&stack_arg_tys, vm);
            let mut stack_arg_size_with_padding = stack_arg_size;
            if stack_arg_size % 16 == 0 {
                // do not need to adjust rsp
            } else if stack_arg_size % 8 == 0 {
                // adjust rsp by -8 (push a random padding value)
                self.backend.emit_push_imm32(0x7777);
                stack_arg_size_with_padding += 8;
            } else {
                panic!("expecting stack arguments to be at least 8-byte aligned, but it has size of {}", stack_arg_size);
            }

            // now, we just put all the args on the stack
            {
                let mut index = 0;
                for arg in stack_args {
                    self.emit_store_base_offset(&x86_64::RSP, - (stack_arg_offsets[index] as i32), &arg, vm);
                    index += 1;
                }

1198
                self.backend.emit_add_r64_imm32(&x86_64::RSP, (- (stack_arg_size as i32)) as i32);
1199
1200
1201
            }

            stack_arg_size_with_padding
1202
        } else {
1203
            0
1204
        }
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226