winch-codegen 0.22.2

Winch code generation library
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
84
85
86
87
88
89
90
91
92
93
94
95
96
97
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
168
169
170
171
172
173
174
175
176
177
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
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
294
295
296
297
298
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
376
377
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
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
use wasmtime_environ::{VMOffsets, WasmHeapType, WasmValType};

use super::ControlStackFrame;
use crate::{
    abi::{scratch, vmctx, ABIOperand, ABIResults, RetArea},
    frame::Frame,
    isa::reg::RegClass,
    masm::{MacroAssembler, OperandSize, RegImm, SPOffset, ShiftKind, StackSlot},
    reg::Reg,
    regalloc::RegAlloc,
    stack::{Stack, TypedReg, Val},
};

/// The code generation context.
/// The code generation context is made up of three
/// essential data structures:
///
/// * The register allocator, in charge of keeping the inventory of register
///   availability.
/// * The value stack, which keeps track of the state of the values
///   after each operation.
/// * The current function's frame.
///
/// These data structures normally require cooperating with each other
/// to perform most of the operations needed during the code
/// generation process. The code generation context should
/// be generally used as the single entry point to access
/// the compound functionality provided by its elements.
pub(crate) struct CodeGenContext<'a> {
    /// The register allocator.
    pub regalloc: RegAlloc,
    /// The value stack.
    pub stack: Stack,
    /// The current function's frame.
    pub frame: Frame,
    /// Reachability state.
    pub reachable: bool,
    /// A reference to the VMOffsets.
    pub vmoffsets: &'a VMOffsets<u8>,
}

impl<'a> CodeGenContext<'a> {
    /// Create a new code generation context.
    pub fn new(
        regalloc: RegAlloc,
        stack: Stack,
        frame: Frame,
        vmoffsets: &'a VMOffsets<u8>,
    ) -> Self {
        Self {
            regalloc,
            stack,
            frame,
            reachable: true,
            vmoffsets,
        }
    }

    /// Request a specific register to the register allocator,
    /// spilling if not available.
    pub fn reg<M: MacroAssembler>(&mut self, named: Reg, masm: &mut M) -> Reg {
        self.regalloc.reg(named, |regalloc| {
            Self::spill_impl(&mut self.stack, regalloc, &self.frame, masm)
        })
    }

    /// Allocate a register for the given WebAssembly type.
    pub fn reg_for_type<M: MacroAssembler>(&mut self, ty: WasmValType, masm: &mut M) -> Reg {
        use WasmValType::*;
        match ty {
            I32 | I64 => self.reg_for_class(RegClass::Int, masm),
            F32 | F64 => self.reg_for_class(RegClass::Float, masm),
            // All of our supported architectures use the float registers for vector operations.
            V128 => self.reg_for_class(RegClass::Float, masm),
            Ref(rt) => match rt.heap_type {
                WasmHeapType::Func => self.reg_for_class(RegClass::Int, masm),
                ht => unimplemented!("Support for WasmHeapType: {ht}"),
            },
        }
    }

    /// Request the register allocator to provide the next available
    /// register of the specified class.
    pub fn reg_for_class<M: MacroAssembler>(&mut self, class: RegClass, masm: &mut M) -> Reg {
        self.regalloc.reg_for_class(class, &mut |regalloc| {
            Self::spill_impl(&mut self.stack, regalloc, &self.frame, masm)
        })
    }

    /// Convenience wrapper around `CodeGenContext::reg_for_class`, to
    /// request the next available general purpose register.
    pub fn any_gpr<M: MacroAssembler>(&mut self, masm: &mut M) -> Reg {
        self.reg_for_class(RegClass::Int, masm)
    }

    /// Convenience wrapper around `CodeGenContext::reg_for_class`, to
    /// request the next available floating point register.
    pub fn any_fpr<M: MacroAssembler>(&mut self, masm: &mut M) -> Reg {
        self.reg_for_class(RegClass::Float, masm)
    }

    /// Executes the provided function, guaranteeing that the specified set of
    /// registers, if any, remain unallocatable throughout the function's
    /// execution.
    pub fn without<'r, T, M, F>(
        &mut self,
        regs: impl IntoIterator<Item = &'r Reg> + Copy,
        masm: &mut M,
        mut f: F,
    ) -> T
    where
        M: MacroAssembler,
        F: FnMut(&mut Self, &mut M) -> T,
    {
        for r in regs {
            self.reg(*r, masm);
        }

        let result = f(self, masm);

        for r in regs {
            self.free_reg(*r);
        }

        result
    }

    /// Free the given register.
    pub fn free_reg(&mut self, reg: impl Into<Reg>) {
        let reg: Reg = reg.into();
        self.regalloc.free(reg);
    }

    /// Loads the stack top value into the next available register, if
    /// it isn't already one; spilling if there are no registers
    /// available.  Optionally the caller may specify a specific
    /// destination register.
    /// When a named register is requested and it's not at the top of the
    /// stack a move from register to register might happen, in which case
    /// the source register will be freed.
    pub fn pop_to_reg<M: MacroAssembler>(&mut self, masm: &mut M, named: Option<Reg>) -> TypedReg {
        let typed_reg = if let Some(dst) = named {
            self.stack.pop_named_reg(dst)
        } else {
            self.stack.pop_reg()
        };

        if let Some(dst) = typed_reg {
            return dst;
        }

        let val = self.stack.pop().expect("a value at stack top");
        let reg = if let Some(r) = named {
            self.reg(r, masm)
        } else {
            self.reg_for_type(val.ty(), masm)
        };

        if val.is_mem() {
            let mem = val.unwrap_mem();
            debug_assert_eq!(mem.slot.offset.as_u32(), masm.sp_offset().as_u32());
            masm.pop(reg, val.ty().into());
        } else {
            self.move_val_to_reg(&val, reg, masm);
            // Free the source value if it is a register.
            if val.is_reg() {
                self.free_reg(val.unwrap_reg());
            }
        }

        TypedReg::new(val.ty(), reg)
    }

    /// Pops the value stack top and stores it at the specified address.
    pub fn pop_to_addr<M: MacroAssembler>(&mut self, masm: &mut M, addr: M::Address) {
        let val = self.stack.pop().expect("a value at stack top");
        let ty = val.ty();
        let size: OperandSize = ty.into();
        match val {
            Val::Reg(tr) => {
                masm.store(tr.reg.into(), addr, size);
                self.free_reg(tr.reg);
            }
            Val::I32(v) => masm.store(RegImm::i32(v), addr, size),
            Val::I64(v) => masm.store(RegImm::i64(v), addr, size),
            Val::F32(v) => masm.store(RegImm::f32(v.bits()), addr, size),
            Val::F64(v) => masm.store(RegImm::f64(v.bits()), addr, size),
            Val::V128(v) => masm.store(RegImm::v128(v), addr, size),
            Val::Local(local) => {
                let slot = self.frame.get_wasm_local(local.index);
                let scratch = scratch!(M);
                let local_addr = masm.local_address(&slot);
                masm.load(local_addr, scratch, size);
                masm.store(scratch.into(), addr, size);
            }
            Val::Memory(_) => {
                let scratch = scratch!(M, &ty);
                masm.pop(scratch, size);
                masm.store(scratch.into(), addr, size);
            }
        }
    }

    /// Move a stack value to the given register.
    pub fn move_val_to_reg<M: MacroAssembler>(&self, src: &Val, dst: Reg, masm: &mut M) {
        let size: OperandSize = src.ty().into();
        match src {
            Val::Reg(tr) => masm.mov(RegImm::reg(tr.reg), dst, size),
            Val::I32(imm) => masm.mov(RegImm::i32(*imm), dst, size),
            Val::I64(imm) => masm.mov(RegImm::i64(*imm), dst, size),
            Val::F32(imm) => masm.mov(RegImm::f32(imm.bits()), dst, size),
            Val::F64(imm) => masm.mov(RegImm::f64(imm.bits()), dst, size),
            Val::V128(imm) => masm.mov(RegImm::v128(*imm), dst, size),
            Val::Local(local) => {
                let slot = self.frame.get_wasm_local(local.index);
                let addr = masm.local_address(&slot);
                masm.load(addr, dst, size);
            }
            Val::Memory(mem) => {
                let addr = masm.address_from_sp(mem.slot.offset);
                masm.load(addr, dst, size);
            }
        }
    }

    /// Prepares arguments for emitting a unary operation.
    ///
    /// The `emit` function returns the `TypedReg` to put on the value stack.
    pub fn unop<F, M>(&mut self, masm: &mut M, size: OperandSize, emit: &mut F)
    where
        F: FnMut(&mut M, Reg, OperandSize) -> TypedReg,
        M: MacroAssembler,
    {
        let typed_reg = self.pop_to_reg(masm, None);
        let dst = emit(masm, typed_reg.reg, size);
        self.stack.push(dst.into());
    }

    /// Prepares arguments for emitting a binary operation.
    ///
    /// The `emit` function returns the `TypedReg` to put on the value stack.
    pub fn binop<F, M>(&mut self, masm: &mut M, size: OperandSize, mut emit: F)
    where
        F: FnMut(&mut M, Reg, Reg, OperandSize) -> TypedReg,
        M: MacroAssembler,
    {
        let src = self.pop_to_reg(masm, None);
        let dst = self.pop_to_reg(masm, None);
        let dst = emit(masm, dst.reg, src.reg.into(), size);
        self.free_reg(src);
        self.stack.push(dst.into());
    }

    /// Prepares arguments for emitting an f32 or f64 comparison operation.
    pub fn float_cmp_op<F, M>(&mut self, masm: &mut M, size: OperandSize, mut emit: F)
    where
        F: FnMut(&mut M, Reg, Reg, Reg, OperandSize),
        M: MacroAssembler,
    {
        let src2 = self.pop_to_reg(masm, None);
        let src1 = self.pop_to_reg(masm, None);
        let dst = self.any_gpr(masm);
        emit(masm, dst, src1.reg, src2.reg, size);
        self.free_reg(src1);
        self.free_reg(src2);

        let dst = match size {
            // Float comparison operators are defined as
            // [f64 f64] -> i32
            // https://webassembly.github.io/spec/core/appendix/index-instructions.html
            OperandSize::S32 | OperandSize::S64 => TypedReg::i32(dst),
            OperandSize::S8 | OperandSize::S16 | OperandSize::S128 => unreachable!(),
        };
        self.stack.push(dst.into());
    }

    /// Prepares arguments for emitting an i32 binary operation.
    ///
    /// The `emit` function returns the `TypedReg` to put on the value stack.
    pub fn i32_binop<F, M>(&mut self, masm: &mut M, mut emit: F)
    where
        F: FnMut(&mut M, Reg, RegImm, OperandSize) -> TypedReg,
        M: MacroAssembler,
    {
        let top = self.stack.peek().expect("value at stack top");

        if top.is_i32_const() {
            let val = self
                .stack
                .pop_i32_const()
                .expect("i32 const value at stack top");
            let typed_reg = self.pop_to_reg(masm, None);
            let dst = emit(masm, typed_reg.reg, RegImm::i32(val), OperandSize::S32);
            self.stack.push(dst.into());
        } else {
            self.binop(masm, OperandSize::S32, |masm, dst, src, size| {
                emit(masm, dst, src.into(), size)
            });
        }
    }

    /// Prepares arguments for emitting an i64 binary operation.
    ///
    /// The `emit` function returns the `TypedReg` to put on the value stack.
    pub fn i64_binop<F, M>(&mut self, masm: &mut M, mut emit: F)
    where
        F: FnMut(&mut M, Reg, RegImm, OperandSize) -> TypedReg,
        M: MacroAssembler,
    {
        let top = self.stack.peek().expect("value at stack top");
        if top.is_i64_const() {
            let val = self
                .stack
                .pop_i64_const()
                .expect("i64 const value at stack top");
            let typed_reg = self.pop_to_reg(masm, None);
            let dst = emit(masm, typed_reg.reg, RegImm::i64(val), OperandSize::S64);
            self.stack.push(dst.into());
        } else {
            self.binop(masm, OperandSize::S64, |masm, dst, src, size| {
                emit(masm, dst, src.into(), size)
            });
        };
    }

    /// Prepares arguments for emitting an i32 shift operation.
    pub fn i32_shift<M>(&mut self, masm: &mut M, kind: ShiftKind)
    where
        M: MacroAssembler,
    {
        let top = self.stack.peek().expect("value at stack top");

        if top.is_i32_const() {
            let val = self
                .stack
                .pop_i32_const()
                .expect("i32 const value at stack top");
            let typed_reg = self.pop_to_reg(masm, None);
            masm.shift_ir(
                typed_reg.reg,
                val as u64,
                typed_reg.reg,
                kind,
                OperandSize::S32,
            );
            self.stack.push(typed_reg.into());
        } else {
            masm.shift(self, kind, OperandSize::S32);
        }
    }

    /// Prepares arguments for emitting an i64 binary operation.
    pub fn i64_shift<M>(&mut self, masm: &mut M, kind: ShiftKind)
    where
        M: MacroAssembler,
    {
        let top = self.stack.peek().expect("value at stack top");
        if top.is_i64_const() {
            let val = self
                .stack
                .pop_i64_const()
                .expect("i64 const value at stack top");
            let typed_reg = self.pop_to_reg(masm, None);
            masm.shift_ir(
                typed_reg.reg,
                val as u64,
                typed_reg.reg,
                kind,
                OperandSize::S64,
            );
            self.stack.push(typed_reg.into());
        } else {
            masm.shift(self, kind, OperandSize::S64);
        };
    }

    /// Prepares arguments for emitting a convert operation.
    pub fn convert_op<F, M>(&mut self, masm: &mut M, dst_ty: WasmValType, mut emit: F)
    where
        F: FnMut(&mut M, Reg, Reg, OperandSize),
        M: MacroAssembler,
    {
        let src = self.pop_to_reg(masm, None);
        let dst = self.reg_for_type(dst_ty, masm);
        let dst_size = match dst_ty {
            WasmValType::I32 => OperandSize::S32,
            WasmValType::I64 => OperandSize::S64,
            WasmValType::F32 => OperandSize::S32,
            WasmValType::F64 => OperandSize::S64,
            WasmValType::V128 => unreachable!(),
            WasmValType::Ref(_) => unreachable!(),
        };

        emit(masm, dst, src.into(), dst_size);

        self.free_reg(src);
        self.stack.push(TypedReg::new(dst_ty, dst).into());
    }

    /// Prepares arguments for emitting a convert operation with a temporary
    /// register.
    pub fn convert_op_with_tmp_reg<F, M>(
        &mut self,
        masm: &mut M,
        dst_ty: WasmValType,
        tmp_reg_class: RegClass,
        mut emit: F,
    ) where
        F: FnMut(&mut M, Reg, Reg, Reg, OperandSize),
        M: MacroAssembler,
    {
        let tmp_gpr = self.reg_for_class(tmp_reg_class, masm);
        self.convert_op(masm, dst_ty, |masm, dst, src, dst_size| {
            emit(masm, dst, src, tmp_gpr, dst_size);
        });
        self.free_reg(tmp_gpr);
    }

    /// Drops the last `n` elements of the stack, calling the provided
    /// function for each `n` stack value.
    /// The values are dropped in top-to-bottom order.
    pub fn drop_last<F>(&mut self, last: usize, mut f: F)
    where
        F: FnMut(&mut RegAlloc, &Val),
    {
        if last > 0 {
            let len = self.stack.len();
            assert!(last <= len);
            let truncate = self.stack.len() - last;
            let stack_mut = self.stack.inner_mut();

            // Invoke the callback in top-to-bottom order.
            for v in stack_mut[truncate..].into_iter().rev() {
                f(&mut self.regalloc, v)
            }
            stack_mut.truncate(truncate);
        }
    }

    /// Convenience wrapper around [`Self::spill_callback`].
    ///
    /// This function exists for cases in which triggering an unconditional
    /// spill is needed, like before entering control flow.
    pub fn spill<M: MacroAssembler>(&mut self, masm: &mut M) {
        Self::spill_impl(&mut self.stack, &mut self.regalloc, &mut self.frame, masm);
    }

    /// Prepares the compiler to emit an uncoditional jump to the given
    /// destination branch.  This process involves:
    /// * Balancing the machine
    ///   stack pointer and value stack by popping it to match the destination
    ///   branch.
    /// * Updating the reachability state.
    /// * Marking the destination frame as a destination target.
    pub fn unconditional_jump<M, F>(&mut self, dest: &mut ControlStackFrame, masm: &mut M, mut f: F)
    where
        M: MacroAssembler,
        F: FnMut(&mut M, &mut Self, &mut ControlStackFrame),
    {
        let state = dest.stack_state();
        let target_offset = state.target_offset;
        let base_offset = state.base_offset;
        // Invariant: The SP, must be greater or equal to the target
        // SP, given that we haven't popped any results by this point
        // yet. But it may happen in the callback.
        assert!(masm.sp_offset().as_u32() >= base_offset.as_u32());
        f(masm, self, dest);

        // The following snippet, pops the stack pointer to ensure that it
        // is correctly placed according to the expectations of the destination
        // branch.
        //
        // This is done in the context of unconditional jumps, as the machine
        // stack might be left unbalanced at the jump site, due to register
        // spills. Note that in some cases the stack pointer offset might be
        // already less than or equal to the original stack pointer offset
        // registered when entering the destination control stack frame, which
        // effectively means that when reaching the jump site no extra space was
        // allocated similar to what would happen in a fall through in which we
        // assume that the program has allocated and deallocated the right
        // amount of stack space.
        //
        // More generally speaking the current stack pointer will be less than
        // the original stack pointer offset in cases in which the top value in
        // the value stack is a memory entry which needs to be popped into the
        // return location according to the ABI (a register for single value
        // returns and a memory slot for 1+ returns). This could happen in the
        // callback invocation above if the callback invokes
        // `ControlStackFrame::pop_abi_results` (e.g. `br` instruction).
        //
        // After an unconditional jump, the compiler will enter in an
        // unreachable state; instead of immediately truncating the value stack
        // to the expected length of the destination branch, we let the
        // reachability analysis code decide what should happen with the length
        // of the value stack once reachability is actually restored. At that
        // point, the right stack pointer offset will also be restored, which
        // should match the contents of the value stack.
        masm.ensure_sp_for_jump(target_offset);
        dest.set_as_target();
        masm.jmp(*dest.label());
        self.reachable = false;
    }

    /// Push the ABI representation of the results stack.
    pub fn push_abi_results<M, F>(
        &mut self,
        results: &ABIResults,
        masm: &mut M,
        mut calculate_ret_area: F,
    ) where
        M: MacroAssembler,
        F: FnMut(&ABIResults, &mut CodeGenContext, &mut M) -> Option<RetArea>,
    {
        let area = results
            .on_stack()
            .then(|| calculate_ret_area(&results, self, masm).unwrap());

        for operand in results.operands().iter() {
            match operand {
                ABIOperand::Reg { reg, ty, .. } => {
                    assert!(self.regalloc.reg_available(*reg));
                    let typed_reg = TypedReg::new(*ty, self.reg(*reg, masm));
                    self.stack.push(typed_reg.into());
                }
                ABIOperand::Stack { ty, offset, size } => match area.unwrap() {
                    RetArea::SP(sp_offset) => {
                        let slot =
                            StackSlot::new(SPOffset::from_u32(sp_offset.as_u32() - offset), *size);
                        self.stack.push(Val::mem(*ty, slot));
                    }
                    // This function is only expected to be called when dealing
                    // with control flow and when calling functions; as a
                    // callee, only [Self::pop_abi_results] is needed when
                    // finalizing the function compilation.
                    _ => unreachable!(),
                },
            }
        }
    }

    /// Truncates the value stack to the specified target.
    /// This function is intended to only be used when restoring the code
    /// generation's reachability state, when handling an unreachable end or
    /// else.
    pub fn truncate_stack_to(&mut self, target: usize) {
        if self.stack.len() > target {
            self.drop_last(self.stack.len() - target, |regalloc, val| match val {
                Val::Reg(tr) => regalloc.free(tr.reg),
                _ => {}
            });
        }
    }

    /// Load the [VMContext] pointer into the designated pinned register.
    pub fn load_vmctx<M>(&mut self, masm: &mut M)
    where
        M: MacroAssembler,
    {
        let addr = masm.local_address(&self.frame.vmctx_slot);
        masm.load_ptr(addr, vmctx!(M));
    }

    /// Spill locals and registers to memory.
    // TODO: optimize the spill range;
    // At any point in the program, the stack might already contain memory
    // entries; we could effectively ignore that range; only focusing on the
    // range that contains spillable values.
    fn spill_impl<M: MacroAssembler>(
        stack: &mut Stack,
        regalloc: &mut RegAlloc,
        frame: &Frame,
        masm: &mut M,
    ) {
        stack.inner_mut().iter_mut().for_each(|v| match v {
            Val::Reg(r) => {
                let slot = masm.push(r.reg, r.ty.into());
                regalloc.free(r.reg);
                *v = Val::mem(r.ty, slot);
            }
            Val::Local(local) => {
                let slot = frame.get_wasm_local(local.index);
                let addr = masm.local_address(&slot);
                let scratch = scratch!(M, &slot.ty);
                masm.load(addr, scratch, slot.ty.into());
                let stack_slot = masm.push(scratch, slot.ty.into());
                *v = Val::mem(slot.ty, stack_slot);
            }
            _ => {}
        });
    }
}