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
use super::operand::OperandValue;
use super::{FunctionCx, LocalRef};

use crate::common::IntPredicate;
use crate::glue;
use crate::traits::*;

use rustc_middle::mir;
use rustc_middle::mir::tcx::PlaceTy;
use rustc_middle::ty::layout::{HasTyCtxt, LayoutOf, TyAndLayout};
use rustc_middle::ty::{self, Ty};
use rustc_target::abi::{Abi, Align, FieldsShape, Int, Pointer, TagEncoding};
use rustc_target::abi::{VariantIdx, Variants};

#[derive(Copy, Clone, Debug)]
pub struct PlaceRef<'tcx, V> {
    /// A pointer to the contents of the place.
    pub llval: V,

    /// This place's extra data if it is unsized, or `None` if null.
    pub llextra: Option<V>,

    /// The monomorphized type of this place, including variant information.
    pub layout: TyAndLayout<'tcx>,

    /// The alignment we know for this place.
    pub align: Align,
}

impl<'a, 'tcx, V: CodegenObject> PlaceRef<'tcx, V> {
    pub fn new_sized(llval: V, layout: TyAndLayout<'tcx>) -> PlaceRef<'tcx, V> {
        assert!(layout.is_sized());
        PlaceRef { llval, llextra: None, layout, align: layout.align.abi }
    }

    pub fn new_sized_aligned(
        llval: V,
        layout: TyAndLayout<'tcx>,
        align: Align,
    ) -> PlaceRef<'tcx, V> {
        assert!(layout.is_sized());
        PlaceRef { llval, llextra: None, layout, align }
    }

    // FIXME(eddyb) pass something else for the name so no work is done
    // unless LLVM IR names are turned on (e.g. for `--emit=llvm-ir`).
    pub fn alloca<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        bx: &mut Bx,
        layout: TyAndLayout<'tcx>,
    ) -> Self {
        Self::alloca_aligned(bx, layout, layout.align.abi)
    }

    pub fn alloca_aligned<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        bx: &mut Bx,
        layout: TyAndLayout<'tcx>,
        align: Align,
    ) -> Self {
        assert!(layout.is_sized(), "tried to statically allocate unsized place");
        let tmp = bx.alloca(bx.cx().backend_type(layout), align);
        Self::new_sized_aligned(tmp, layout, align)
    }

    /// Returns a place for an indirect reference to an unsized place.
    // FIXME(eddyb) pass something else for the name so no work is done
    // unless LLVM IR names are turned on (e.g. for `--emit=llvm-ir`).
    pub fn alloca_unsized_indirect<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        bx: &mut Bx,
        layout: TyAndLayout<'tcx>,
    ) -> Self {
        assert!(layout.is_unsized(), "tried to allocate indirect place for sized values");
        let ptr_ty = Ty::new_mut_ptr(bx.cx().tcx(), layout.ty);
        let ptr_layout = bx.cx().layout_of(ptr_ty);
        Self::alloca(bx, ptr_layout)
    }

    pub fn len<Cx: ConstMethods<'tcx, Value = V>>(&self, cx: &Cx) -> V {
        if let FieldsShape::Array { count, .. } = self.layout.fields {
            if self.layout.is_unsized() {
                assert_eq!(count, 0);
                self.llextra.unwrap()
            } else {
                cx.const_usize(count)
            }
        } else {
            bug!("unexpected layout `{:#?}` in PlaceRef::len", self.layout)
        }
    }
}

impl<'a, 'tcx, V: CodegenObject> PlaceRef<'tcx, V> {
    /// Access a field, at a point when the value's case is known.
    pub fn project_field<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        self,
        bx: &mut Bx,
        ix: usize,
    ) -> Self {
        let field = self.layout.field(bx.cx(), ix);
        let offset = self.layout.fields.offset(ix);
        let effective_field_align = self.align.restrict_for_offset(offset);

        let mut simple = || {
            let llval = match self.layout.abi {
                _ if offset.bytes() == 0 => {
                    // Unions and newtypes only use an offset of 0.
                    // Also handles the first field of Scalar, ScalarPair, and Vector layouts.
                    self.llval
                }
                Abi::ScalarPair(a, b)
                    if offset == a.size(bx.cx()).align_to(b.align(bx.cx()).abi) =>
                {
                    // Offset matches second field.
                    let ty = bx.backend_type(self.layout);
                    bx.struct_gep(ty, self.llval, 1)
                }
                Abi::Scalar(_) | Abi::ScalarPair(..) | Abi::Vector { .. } if field.is_zst() => {
                    // ZST fields (even some that require alignment) are not included in Scalar,
                    // ScalarPair, and Vector layouts, so manually offset the pointer.
                    bx.gep(bx.cx().type_i8(), self.llval, &[bx.const_usize(offset.bytes())])
                }
                Abi::Scalar(_) | Abi::ScalarPair(..) => {
                    // All fields of Scalar and ScalarPair layouts must have been handled by this point.
                    // Vector layouts have additional fields for each element of the vector, so don't panic in that case.
                    bug!(
                        "offset of non-ZST field `{:?}` does not match layout `{:#?}`",
                        field,
                        self.layout
                    );
                }
                _ => {
                    let ty = bx.backend_type(self.layout);
                    bx.struct_gep(ty, self.llval, bx.cx().backend_field_index(self.layout, ix))
                }
            };
            PlaceRef {
                llval,
                llextra: if bx.cx().type_has_metadata(field.ty) { self.llextra } else { None },
                layout: field,
                align: effective_field_align,
            }
        };

        // Simple cases, which don't need DST adjustment:
        //   * no metadata available - just log the case
        //   * known alignment - sized types, `[T]`, `str` or a foreign type
        //   * packed struct - there is no alignment padding
        match field.ty.kind() {
            _ if self.llextra.is_none() => {
                debug!(
                    "unsized field `{}`, of `{:?}` has no metadata for adjustment",
                    ix, self.llval
                );
                return simple();
            }
            _ if field.is_sized() => return simple(),
            ty::Slice(..) | ty::Str | ty::Foreign(..) => return simple(),
            ty::Adt(def, _) => {
                if def.repr().packed() {
                    // FIXME(eddyb) generalize the adjustment when we
                    // start supporting packing to larger alignments.
                    assert_eq!(self.layout.align.abi.bytes(), 1);
                    return simple();
                }
            }
            _ => {}
        }

        // We need to get the pointer manually now.
        // We do this by casting to a `*i8`, then offsetting it by the appropriate amount.
        // We do this instead of, say, simply adjusting the pointer from the result of a GEP
        // because the field may have an arbitrary alignment in the LLVM representation
        // anyway.
        //
        // To demonstrate:
        //
        //     struct Foo<T: ?Sized> {
        //         x: u16,
        //         y: T
        //     }
        //
        // The type `Foo<Foo<Trait>>` is represented in LLVM as `{ u16, { u16, u8 }}`, meaning that
        // the `y` field has 16-bit alignment.

        let meta = self.llextra;

        let unaligned_offset = bx.cx().const_usize(offset.bytes());

        // Get the alignment of the field
        let (_, unsized_align) = glue::size_and_align_of_dst(bx, field.ty, meta);

        // Bump the unaligned offset up to the appropriate alignment
        let offset = round_up_const_value_to_alignment(bx, unaligned_offset, unsized_align);

        debug!("struct_field_ptr: DST field offset: {:?}", offset);

        // Adjust pointer.
        let ptr = bx.gep(bx.cx().type_i8(), self.llval, &[offset]);

        PlaceRef { llval: ptr, llextra: self.llextra, layout: field, align: effective_field_align }
    }

    /// Obtain the actual discriminant of a value.
    #[instrument(level = "trace", skip(bx))]
    pub fn codegen_get_discr<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        self,
        bx: &mut Bx,
        cast_to: Ty<'tcx>,
    ) -> V {
        let dl = &bx.tcx().data_layout;
        let cast_to_layout = bx.cx().layout_of(cast_to);
        let cast_to = bx.cx().immediate_backend_type(cast_to_layout);
        if self.layout.abi.is_uninhabited() {
            return bx.cx().const_poison(cast_to);
        }
        let (tag_scalar, tag_encoding, tag_field) = match self.layout.variants {
            Variants::Single { index } => {
                let discr_val = self
                    .layout
                    .ty
                    .discriminant_for_variant(bx.cx().tcx(), index)
                    .map_or(index.as_u32() as u128, |discr| discr.val);
                return bx.cx().const_uint_big(cast_to, discr_val);
            }
            Variants::Multiple { tag, ref tag_encoding, tag_field, .. } => {
                (tag, tag_encoding, tag_field)
            }
        };

        // Read the tag/niche-encoded discriminant from memory.
        let tag = self.project_field(bx, tag_field);
        let tag_op = bx.load_operand(tag);
        let tag_imm = tag_op.immediate();

        // Decode the discriminant (specifically if it's niche-encoded).
        match *tag_encoding {
            TagEncoding::Direct => {
                let signed = match tag_scalar.primitive() {
                    // We use `i1` for bytes that are always `0` or `1`,
                    // e.g., `#[repr(i8)] enum E { A, B }`, but we can't
                    // let LLVM interpret the `i1` as signed, because
                    // then `i1 1` (i.e., `E::B`) is effectively `i8 -1`.
                    Int(_, signed) => !tag_scalar.is_bool() && signed,
                    _ => false,
                };
                bx.intcast(tag_imm, cast_to, signed)
            }
            TagEncoding::Niche { untagged_variant, ref niche_variants, niche_start } => {
                // Cast to an integer so we don't have to treat a pointer as a
                // special case.
                let (tag, tag_llty) = match tag_scalar.primitive() {
                    // FIXME(erikdesjardins): handle non-default addrspace ptr sizes
                    Pointer(_) => {
                        let t = bx.type_from_integer(dl.ptr_sized_integer());
                        let tag = bx.ptrtoint(tag_imm, t);
                        (tag, t)
                    }
                    _ => (tag_imm, bx.cx().immediate_backend_type(tag_op.layout)),
                };

                let relative_max = niche_variants.end().as_u32() - niche_variants.start().as_u32();

                // We have a subrange `niche_start..=niche_end` inside `range`.
                // If the value of the tag is inside this subrange, it's a
                // "niche value", an increment of the discriminant. Otherwise it
                // indicates the untagged variant.
                // A general algorithm to extract the discriminant from the tag
                // is:
                // relative_tag = tag - niche_start
                // is_niche = relative_tag <= (ule) relative_max
                // discr = if is_niche {
                //     cast(relative_tag) + niche_variants.start()
                // } else {
                //     untagged_variant
                // }
                // However, we will likely be able to emit simpler code.
                let (is_niche, tagged_discr, delta) = if relative_max == 0 {
                    // Best case scenario: only one tagged variant. This will
                    // likely become just a comparison and a jump.
                    // The algorithm is:
                    // is_niche = tag == niche_start
                    // discr = if is_niche {
                    //     niche_start
                    // } else {
                    //     untagged_variant
                    // }
                    let niche_start = bx.cx().const_uint_big(tag_llty, niche_start);
                    let is_niche = bx.icmp(IntPredicate::IntEQ, tag, niche_start);
                    let tagged_discr =
                        bx.cx().const_uint(cast_to, niche_variants.start().as_u32() as u64);
                    (is_niche, tagged_discr, 0)
                } else {
                    // The special cases don't apply, so we'll have to go with
                    // the general algorithm.
                    let relative_discr = bx.sub(tag, bx.cx().const_uint_big(tag_llty, niche_start));
                    let cast_tag = bx.intcast(relative_discr, cast_to, false);
                    let is_niche = bx.icmp(
                        IntPredicate::IntULE,
                        relative_discr,
                        bx.cx().const_uint(tag_llty, relative_max as u64),
                    );
                    (is_niche, cast_tag, niche_variants.start().as_u32() as u128)
                };

                let tagged_discr = if delta == 0 {
                    tagged_discr
                } else {
                    bx.add(tagged_discr, bx.cx().const_uint_big(cast_to, delta))
                };

                let discr = bx.select(
                    is_niche,
                    tagged_discr,
                    bx.cx().const_uint(cast_to, untagged_variant.as_u32() as u64),
                );

                // In principle we could insert assumes on the possible range of `discr`, but
                // currently in LLVM this seems to be a pessimization.

                discr
            }
        }
    }

    /// Sets the discriminant for a new value of the given case of the given
    /// representation.
    pub fn codegen_set_discr<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        &self,
        bx: &mut Bx,
        variant_index: VariantIdx,
    ) {
        if self.layout.for_variant(bx.cx(), variant_index).abi.is_uninhabited() {
            // We play it safe by using a well-defined `abort`, but we could go for immediate UB
            // if that turns out to be helpful.
            bx.abort();
            return;
        }
        match self.layout.variants {
            Variants::Single { index } => {
                assert_eq!(index, variant_index);
            }
            Variants::Multiple { tag_encoding: TagEncoding::Direct, tag_field, .. } => {
                let ptr = self.project_field(bx, tag_field);
                let to =
                    self.layout.ty.discriminant_for_variant(bx.tcx(), variant_index).unwrap().val;
                bx.store(
                    bx.cx().const_uint_big(bx.cx().backend_type(ptr.layout), to),
                    ptr.llval,
                    ptr.align,
                );
            }
            Variants::Multiple {
                tag_encoding:
                    TagEncoding::Niche { untagged_variant, ref niche_variants, niche_start },
                tag_field,
                ..
            } => {
                if variant_index != untagged_variant {
                    let niche = self.project_field(bx, tag_field);
                    let niche_llty = bx.cx().immediate_backend_type(niche.layout);
                    let niche_value = variant_index.as_u32() - niche_variants.start().as_u32();
                    let niche_value = (niche_value as u128).wrapping_add(niche_start);
                    // FIXME(eddyb): check the actual primitive type here.
                    let niche_llval = if niche_value == 0 {
                        // HACK(eddyb): using `c_null` as it works on all types.
                        bx.cx().const_null(niche_llty)
                    } else {
                        bx.cx().const_uint_big(niche_llty, niche_value)
                    };
                    OperandValue::Immediate(niche_llval).store(bx, niche);
                }
            }
        }
    }

    pub fn project_index<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        &self,
        bx: &mut Bx,
        llindex: V,
    ) -> Self {
        // Statically compute the offset if we can, otherwise just use the element size,
        // as this will yield the lowest alignment.
        let layout = self.layout.field(bx, 0);
        let offset = if let Some(llindex) = bx.const_to_opt_uint(llindex) {
            layout.size.checked_mul(llindex, bx).unwrap_or(layout.size)
        } else {
            layout.size
        };

        PlaceRef {
            llval: bx.inbounds_gep(
                bx.cx().backend_type(self.layout),
                self.llval,
                &[bx.cx().const_usize(0), llindex],
            ),
            llextra: None,
            layout,
            align: self.align.restrict_for_offset(offset),
        }
    }

    pub fn project_downcast<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        &self,
        bx: &mut Bx,
        variant_index: VariantIdx,
    ) -> Self {
        let mut downcast = *self;
        downcast.layout = self.layout.for_variant(bx.cx(), variant_index);
        downcast
    }

    pub fn project_type<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
        &self,
        bx: &mut Bx,
        ty: Ty<'tcx>,
    ) -> Self {
        let mut downcast = *self;
        downcast.layout = bx.cx().layout_of(ty);
        downcast
    }

    pub fn storage_live<Bx: BuilderMethods<'a, 'tcx, Value = V>>(&self, bx: &mut Bx) {
        bx.lifetime_start(self.llval, self.layout.size);
    }

    pub fn storage_dead<Bx: BuilderMethods<'a, 'tcx, Value = V>>(&self, bx: &mut Bx) {
        bx.lifetime_end(self.llval, self.layout.size);
    }
}

impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
    #[instrument(level = "trace", skip(self, bx))]
    pub fn codegen_place(
        &mut self,
        bx: &mut Bx,
        place_ref: mir::PlaceRef<'tcx>,
    ) -> PlaceRef<'tcx, Bx::Value> {
        let cx = self.cx;
        let tcx = self.cx.tcx();

        let mut base = 0;
        let mut cg_base = match self.locals[place_ref.local] {
            LocalRef::Place(place) => place,
            LocalRef::UnsizedPlace(place) => bx.load_operand(place).deref(cx),
            LocalRef::Operand(..) => {
                if place_ref.is_indirect_first_projection() {
                    base = 1;
                    let cg_base = self.codegen_consume(
                        bx,
                        mir::PlaceRef { projection: &place_ref.projection[..0], ..place_ref },
                    );
                    cg_base.deref(bx.cx())
                } else {
                    bug!("using operand local {:?} as place", place_ref);
                }
            }
            LocalRef::PendingOperand => {
                bug!("using still-pending operand local {:?} as place", place_ref);
            }
        };
        for elem in place_ref.projection[base..].iter() {
            cg_base = match *elem {
                mir::ProjectionElem::Deref => bx.load_operand(cg_base).deref(bx.cx()),
                mir::ProjectionElem::Field(ref field, _) => {
                    cg_base.project_field(bx, field.index())
                }
                mir::ProjectionElem::OpaqueCast(ty) => {
                    bug!("encountered OpaqueCast({ty}) in codegen")
                }
                mir::ProjectionElem::Subtype(ty) => cg_base.project_type(bx, self.monomorphize(ty)),
                mir::ProjectionElem::Index(index) => {
                    let index = &mir::Operand::Copy(mir::Place::from(index));
                    let index = self.codegen_operand(bx, index);
                    let llindex = index.immediate();
                    cg_base.project_index(bx, llindex)
                }
                mir::ProjectionElem::ConstantIndex { offset, from_end: false, min_length: _ } => {
                    let lloffset = bx.cx().const_usize(offset as u64);
                    cg_base.project_index(bx, lloffset)
                }
                mir::ProjectionElem::ConstantIndex { offset, from_end: true, min_length: _ } => {
                    let lloffset = bx.cx().const_usize(offset as u64);
                    let lllen = cg_base.len(bx.cx());
                    let llindex = bx.sub(lllen, lloffset);
                    cg_base.project_index(bx, llindex)
                }
                mir::ProjectionElem::Subslice { from, to, from_end } => {
                    let mut subslice = cg_base.project_index(bx, bx.cx().const_usize(from as u64));
                    let projected_ty =
                        PlaceTy::from_ty(cg_base.layout.ty).projection_ty(tcx, *elem).ty;
                    subslice.layout = bx.cx().layout_of(self.monomorphize(projected_ty));

                    if subslice.layout.is_unsized() {
                        assert!(from_end, "slice subslices should be `from_end`");
                        subslice.llextra = Some(bx.sub(
                            cg_base.llextra.unwrap(),
                            bx.cx().const_usize((from as u64) + (to as u64)),
                        ));
                    }

                    subslice
                }
                mir::ProjectionElem::Downcast(_, v) => cg_base.project_downcast(bx, v),
            };
        }
        debug!("codegen_place(place={:?}) => {:?}", place_ref, cg_base);
        cg_base
    }

    pub fn monomorphized_place_ty(&self, place_ref: mir::PlaceRef<'tcx>) -> Ty<'tcx> {
        let tcx = self.cx.tcx();
        let place_ty = place_ref.ty(self.mir, tcx);
        self.monomorphize(place_ty.ty)
    }
}

fn round_up_const_value_to_alignment<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
    bx: &mut Bx,
    value: Bx::Value,
    align: Bx::Value,
) -> Bx::Value {
    // In pseudo code:
    //
    //     if value & (align - 1) == 0 {
    //         value
    //     } else {
    //         (value & !(align - 1)) + align
    //     }
    //
    // Usually this is written without branches as
    //
    //     (value + align - 1) & !(align - 1)
    //
    // But this formula cannot take advantage of constant `value`. E.g. if `value` is known
    // at compile time to be `1`, this expression should be optimized to `align`. However,
    // optimization only holds if `align` is a power of two. Since the optimizer doesn't know
    // that `align` is a power of two, it cannot perform this optimization.
    //
    // Instead we use
    //
    //     value + (-value & (align - 1))
    //
    // Since `align` is used only once, the expression can be optimized. For `value = 0`
    // its optimized to `0` even in debug mode.
    //
    // NB: The previous version of this code used
    //
    //     (value + align - 1) & -align
    //
    // Even though `-align == !(align - 1)`, LLVM failed to optimize this even for
    // `value = 0`. Bug report: https://bugs.llvm.org/show_bug.cgi?id=48559
    let one = bx.const_usize(1);
    let align_minus_1 = bx.sub(align, one);
    let neg_value = bx.neg(value);
    let offset = bx.and(neg_value, align_minus_1);
    bx.add(value, offset)
}