Auto merge of rust-lang#126444 - tesuji:gvn-const-arrays, r=<try>

[WIP] gvn: Promote/propagate const local array

Rewriting of rust-lang#125916 which used PromoteTemps pass.

Fix rust-lang#73825

### Current status

- [ ] Waiting for [consensus](https://rust-lang.zulipchat.com/#narrow/stream/136281-t-opsem/topic/Could.20const.20read-only.20arrays.20be.20const.20promoted.3F).

r? ghost

Author: bors

Diff for: compiler/rustc_mir_transform/src/gvn.rs

@@ -365,8 +365,15 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
 
     #[instrument(level = "trace", skip(self), ret)]
     fn eval_to_const(&mut self, value: VnIndex) -> Option<OpTy<'tcx>> {
+        use abi::HasDataLayout;
         use Value::*;
-        let op = match *self.get(value) {
+        // LLVM optimizes the load of `sizeof(size_t) * 2` as a single `mov`,
+        // which is cheap. Bigger values make more `mov` instructions generated.
+        // After GVN, it becomes a single load (`lea`) of an address in `.rodata`.
+        let stack_threshold = self.tcx.data_layout().pointer_size * 2;
+        let vvalue = self.get(value);
+        debug!(?vvalue);
+        let op = match *vvalue {
             Opaque(_) => return None,
             // Do not bother evaluating repeat expressions. This would uselessly consume memory.
             Repeat(..) => return None,
@@ -375,14 +382,26 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
                 self.ecx.eval_mir_constant(value, DUMMY_SP, None).ok()?
             }
             Aggregate(kind, variant, ref fields) => {
+                debug!(?kind, ?variant, ?fields);
                 let fields = fields
                     .iter()
                     .map(|&f| self.evaluated[f].as_ref())
-                    .collect::<Option<Vec<_>>>()?;
+                    .collect::<Option<Vec<&OpTy<'_>>>>()?;
                 let ty = match kind {
                     AggregateTy::Array => {
-                        assert!(fields.len() > 0);
-                        Ty::new_array(self.tcx, fields[0].layout.ty, fields.len() as u64)
+                        let [field, ..] = fields.as_slice() else {
+                            bug!("fields.len() == 0");
+                        };
+                        let field_ty = field.layout.ty;
+                        // Ignore nested array
+                        if field_ty.is_array() {
+                            trace!(
+                                "ignoring nested array of type: [{field_ty}; {len}]",
+                                len = fields.len(),
+                            );
+                            return None;
+                        }
+                        Ty::new_array(self.tcx, field_ty, fields.len() as u64)
                     }
                     AggregateTy::Tuple => {
                         Ty::new_tup_from_iter(self.tcx, fields.iter().map(|f| f.layout.ty))
@@ -406,6 +425,32 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
                     };
                     let ptr_imm = Immediate::new_pointer_with_meta(data, meta, &self.ecx);
                     ImmTy::from_immediate(ptr_imm, ty).into()
+                } else if matches!(kind, AggregateTy::Array) {
+                    if ty.layout.size() <= stack_threshold {
+                        return None;
+                    }
+                    let mut mplace = None;
+                    let alloc_id = self
+                        .ecx
+                        .intern_with_temp_alloc(ty, |ecx, dest| {
+                            // FIXME: Can we speed it up by using `ecx.write_immediate(.ScalarPair(_), dest)`?
+                            for (field_index, op) in fields.iter().copied().enumerate() {
+                                let field_dest = ecx.project_field(dest, field_index)?;
+                                ecx.copy_op(op, &field_dest)?;
+                            }
+
+                            let dest =
+                                dest.assert_mem_place().map_provenance(|prov| prov.as_immutable());
+                            mplace.replace(dest);
+                            Ok(())
+                        })
+                        .ok()?;
+                    let GlobalAlloc::Memory(_alloc) = self.tcx.global_alloc(alloc_id) else {
+                        bug!()
+                    };
+                    let mplace = mplace.unwrap();
+                    debug!(?mplace);
+                    return Some(mplace.into());
                 } else if matches!(ty.abi, Abi::Scalar(..) | Abi::ScalarPair(..)) {
                     let dest = self.ecx.allocate(ty, MemoryKind::Stack).ok()?;
                     let variant_dest = if let Some(variant) = variant {
@@ -428,6 +473,7 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
             }
 
             Projection(base, elem) => {
+                debug!(?base, ?elem);
                 let value = self.evaluated[base].as_ref()?;
                 let elem = match elem {
                     ProjectionElem::Deref => ProjectionElem::Deref,
@@ -449,6 +495,7 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
                 self.ecx.project(value, elem).ok()?
             }
             Address { place, kind, provenance: _ } => {
+                debug!(?place, ?kind);
                 if !place.is_indirect_first_projection() {
                     return None;
                 }
@@ -708,7 +755,7 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
             place.projection = self.tcx.mk_place_elems(&projection);
         }
 
-        trace!(?place);
+        trace!(after_place = ?place);
     }
 
     /// Represent the *value* which would be read from `place`, and point `place` to a preexisting
@@ -1232,6 +1279,7 @@ impl<'body, 'tcx> VnState<'body, 'tcx> {
     }
 }
 
+#[instrument(level = "trace", skip(ecx), ret)]
 fn op_to_prop_const<'tcx>(
     ecx: &mut InterpCx<'tcx, DummyMachine>,
     op: &OpTy<'tcx>,
@@ -1247,7 +1295,8 @@ fn op_to_prop_const<'tcx>(
     }
 
     // Do not synthetize too large constants. Codegen will just memcpy them, which we'd like to avoid.
-    if !matches!(op.layout.abi, Abi::Scalar(..) | Abi::ScalarPair(..)) {
+    if !(op.layout.ty.is_array() || matches!(op.layout.abi, Abi::Scalar(..) | Abi::ScalarPair(..)))
+    {
         return None;
     }
 
@@ -1301,20 +1350,32 @@ fn op_to_prop_const<'tcx>(
 
 impl<'tcx> VnState<'_, 'tcx> {
     /// If `index` is a `Value::Constant`, return the `Constant` to be put in the MIR.
+    #[instrument(level = "trace", skip(self, index), ret)]
     fn try_as_constant(&mut self, index: VnIndex) -> Option<ConstOperand<'tcx>> {
         // This was already constant in MIR, do not change it.
-        if let Value::Constant { value, disambiguator: _ } = *self.get(index)
-            // If the constant is not deterministic, adding an additional mention of it in MIR will
-            // not give the same value as the former mention.
+        let value = self.get(index);
+        debug!(?index, ?value);
+        // If the constant is not deterministic, adding an additional mention of it in MIR will
+        // not give the same value as the former mention.
+        if let Value::Constant { value, disambiguator: _ } = value
             && value.is_deterministic()
         {
-            return Some(ConstOperand { span: DUMMY_SP, user_ty: None, const_: value });
+            return Some(ConstOperand { span: DUMMY_SP, user_ty: None, const_: *value });
         }
 
         let op = self.evaluated[index].as_ref()?;
-        if op.layout.is_unsized() {
-            // Do not attempt to propagate unsized locals.
-            return None;
+
+        if let Either::Left(mplace) = op.as_mplace_or_imm()
+            && let ty::Array(ty, _const) = mplace.layout.ty.kind()
+        {
+            // ignore nested arrays
+            if ty.is_array() {
+                return None;
+            }
+            // ignore promoted arrays
+            else if let Value::Projection(_index, ProjectionElem::Deref) = value {
+                return None;
+            }
         }
 
         let value = op_to_prop_const(&mut self.ecx, op)?;
@@ -1325,6 +1386,10 @@ impl<'tcx> VnState<'_, 'tcx> {
         assert!(!value.may_have_provenance(self.tcx, op.layout.size));
 
         let const_ = Const::Val(value, op.layout.ty);
+        // cache the propagated const
+        if let Some(new_index) = self.insert_constant(const_) {
+            self.values.swap_indices(index.as_usize(), new_index.as_usize());
+        }
         Some(ConstOperand { span: DUMMY_SP, user_ty: None, const_ })
     }
 
@@ -1352,6 +1417,7 @@ impl<'tcx> MutVisitor<'tcx> for VnState<'_, 'tcx> {
         self.simplify_operand(operand, location);
     }
 
+    #[instrument(level = "trace", skip(self, stmt))]
     fn visit_statement(&mut self, stmt: &mut Statement<'tcx>, location: Location) {
         if let StatementKind::Assign(box (ref mut lhs, ref mut rvalue)) = stmt.kind {
             self.simplify_place_projection(lhs, location);
@@ -1365,8 +1431,10 @@ impl<'tcx> MutVisitor<'tcx> for VnState<'_, 'tcx> {
                 .as_local()
                 .and_then(|local| self.locals[local])
                 .or_else(|| self.simplify_rvalue(rvalue, location));
+            debug!(?value);
             let Some(value) = value else { return };
 
+            debug!(before_rvalue = ?rvalue);
             if let Some(const_) = self.try_as_constant(value) {
                 *rvalue = Rvalue::Use(Operand::Constant(Box::new(const_)));
             } else if let Some(local) = self.try_as_local(value, location)
@@ -1375,6 +1443,7 @@ impl<'tcx> MutVisitor<'tcx> for VnState<'_, 'tcx> {
                 *rvalue = Rvalue::Use(Operand::Copy(local.into()));
                 self.reused_locals.insert(local);
             }
+            debug!(after_rvalue = ?rvalue);
 
             return;
         }

Diff for: tests/codegen/issue-73825-gvn-const-local-array.rs

@@ -0,0 +1,25 @@
+// issue: <https://github.com/rust-lang/rust/issues/73825>
+//@ compile-flags: -C opt-level=1
+#![crate_type = "lib"]
+
+// CHECK-LABEL: @foo
+// CHECK-NEXT: {{.*}}:
+// CHECK-NEXT: and
+// CHECK-NEXT: getelementptr inbounds
+// CHECK-NEXT: load i32
+// CHECK-NEXT: ret i32
+#[no_mangle]
+#[rustfmt::skip]
+pub fn foo(x: usize) -> i32 {
+    let base: [i32; 64] = [
+        67, 754, 860, 559, 368, 870, 548, 972,
+        141, 731, 351, 664, 32, 4, 996, 741,
+        203, 292, 237, 480, 151, 940, 777, 540,
+        143, 587, 747, 65, 152, 517, 882, 880,
+        712, 595, 370, 901, 237, 53, 789, 785,
+        912, 650, 896, 367, 316, 392, 62, 473,
+        675, 691, 281, 192, 445, 970, 225, 425,
+        628, 324, 322, 206, 912, 867, 462, 92
+    ];
+    base[x % 64]
+}

Diff for: tests/mir-opt/const_array_locals.main.GVN.diff

@@ -0,0 +1,138 @@
+- // MIR for `main` before GVN
++ // MIR for `main` after GVN
+  
+  fn main() -> () {
+      let mut _0: ();
+      let _1: [i32; 32];
+      let mut _4: [i32; 12];
+      let mut _5: [i32; 12];
+      let mut _7: &[i32; 32];
+      let _8: [i32; 32];
+      let _9: ();
+      let mut _10: [u16; 32];
+      let mut _12: [f32; 8];
+      let _13: [[i32; 3]; 3];
+      let mut _14: [i32; 3];
+      let mut _15: [i32; 3];
+      let mut _16: [i32; 3];
+      scope 1 {
+          debug _arr => _1;
+          let _2: [i32; 32];
+          scope 2 {
+              debug _barr => _2;
+              let _3: [[i32; 12]; 2];
+              scope 3 {
+                  debug _foo => _3;
+                  let _6: [i32; 32];
+                  let mut _17: &[i32; 32];
+                  scope 4 {
+                      debug _darr => _6;
+                      let _11: F32x8;
+                      scope 5 {
+                          debug _f => _11;
+                      }
+                  }
+              }
+          }
+      }
+  
+      bb0: {
+          StorageLive(_1);
+-         _1 = [const 255_i32, const 105_i32, const 15_i32, const 39_i32, const 62_i32, const 251_i32, const 191_i32, const 178_i32, const 9_i32, const 4_i32, const 56_i32, const 221_i32, const 193_i32, const 164_i32, const 194_i32, const 197_i32, const 6_i32, const 243_i32, const 218_i32, const 171_i32, const 87_i32, const 247_i32, const 104_i32, const 159_i32, const 22_i32, const 157_i32, const 105_i32, const 31_i32, const 96_i32, const 173_i32, const 50_i32, const 1_i32];
++         _1 = const [255_i32, 105_i32, 15_i32, 39_i32, 62_i32, 251_i32, 191_i32, 178_i32, 9_i32, 4_i32, 56_i32, 221_i32, 193_i32, 164_i32, 194_i32, 197_i32, 6_i32, 243_i32, 218_i32, 171_i32, 87_i32, 247_i32, 104_i32, 159_i32, 22_i32, 157_i32, 105_i32, 31_i32, 96_i32, 173_i32, 50_i32, 1_i32];
+          StorageLive(_2);
+-         _2 = [const 255_i32, const 105_i32, const 15_i32, const 39_i32, const 62_i32, const 251_i32, const 191_i32, const 178_i32, const 9_i32, const 4_i32, const 56_i32, const 221_i32, const 193_i32, const 164_i32, const 194_i32, const 197_i32, const 6_i32, const 243_i32, const 218_i32, const 171_i32, const 87_i32, const 247_i32, const 104_i32, const 159_i32, const 22_i32, const 157_i32, const 105_i32, const 31_i32, const 96_i32, const 173_i32, const 50_i32, const 1_i32];
++         _2 = const [255_i32, 105_i32, 15_i32, 39_i32, 62_i32, 251_i32, 191_i32, 178_i32, 9_i32, 4_i32, 56_i32, 221_i32, 193_i32, 164_i32, 194_i32, 197_i32, 6_i32, 243_i32, 218_i32, 171_i32, 87_i32, 247_i32, 104_i32, 159_i32, 22_i32, 157_i32, 105_i32, 31_i32, 96_i32, 173_i32, 50_i32, 1_i32];
+          StorageLive(_3);
+          StorageLive(_4);
+-         _4 = [const 255_i32, const 105_i32, const 15_i32, const 39_i32, const 62_i32, const 251_i32, const 191_i32, const 178_i32, const 9_i32, const 4_i32, const 56_i32, const 221_i32];
++         _4 = const [255_i32, 105_i32, 15_i32, 39_i32, 62_i32, 251_i32, 191_i32, 178_i32, 9_i32, 4_i32, 56_i32, 221_i32];
+          StorageLive(_5);
+-         _5 = [const 193_i32, const 164_i32, const 194_i32, const 197_i32, const 6_i32, const 243_i32, const 218_i32, const 171_i32, const 87_i32, const 247_i32, const 104_i32, const 42_i32];
+-         _3 = [move _4, move _5];
++         _5 = const [193_i32, 164_i32, 194_i32, 197_i32, 6_i32, 243_i32, 218_i32, 171_i32, 87_i32, 247_i32, 104_i32, 42_i32];
++         _3 = [const [255_i32, 105_i32, 15_i32, 39_i32, 62_i32, 251_i32, 191_i32, 178_i32, 9_i32, 4_i32, 56_i32, 221_i32], const [193_i32, 164_i32, 194_i32, 197_i32, 6_i32, 243_i32, 218_i32, 171_i32, 87_i32, 247_i32, 104_i32, 42_i32]];
+          StorageDead(_5);
+          StorageDead(_4);
+          StorageLive(_6);
+          StorageLive(_7);
+          _17 = const main::promoted[0];
+          _7 = &(*_17);
+-         _6 = (*_7);
++         _6 = (*_17);
+          StorageDead(_7);
+          StorageLive(_9);
+          StorageLive(_10);
+-         _10 = [const 255_u16, const 105_u16, const 15_u16, const 39_u16, const 62_u16, const 251_u16, const 191_u16, const 178_u16, const 9_u16, const 4_u16, const 56_u16, const 221_u16, const 193_u16, const 164_u16, const 194_u16, const 197_u16, const 6_u16, const 243_u16, const 218_u16, const 171_u16, const 87_u16, const 247_u16, const 104_u16, const 159_u16, const 22_u16, const 157_u16, const 105_u16, const 31_u16, const 96_u16, const 173_u16, const 50_u16, const 1_u16];
+-         _9 = consume(move _10) -> [return: bb1, unwind continue];
++         _10 = const [255_u16, 105_u16, 15_u16, 39_u16, 62_u16, 251_u16, 191_u16, 178_u16, 9_u16, 4_u16, 56_u16, 221_u16, 193_u16, 164_u16, 194_u16, 197_u16, 6_u16, 243_u16, 218_u16, 171_u16, 87_u16, 247_u16, 104_u16, 159_u16, 22_u16, 157_u16, 105_u16, 31_u16, 96_u16, 173_u16, 50_u16, 1_u16];
++         _9 = consume(const [255_u16, 105_u16, 15_u16, 39_u16, 62_u16, 251_u16, 191_u16, 178_u16, 9_u16, 4_u16, 56_u16, 221_u16, 193_u16, 164_u16, 194_u16, 197_u16, 6_u16, 243_u16, 218_u16, 171_u16, 87_u16, 247_u16, 104_u16, 159_u16, 22_u16, 157_u16, 105_u16, 31_u16, 96_u16, 173_u16, 50_u16, 1_u16]) -> [return: bb1, unwind continue];
+      }
+  
+      bb1: {
+          StorageDead(_10);
+          StorageDead(_9);
+          StorageLive(_11);
+          StorageLive(_12);
+-         _12 = [const 1f32, const 2f32, const 3f32, const 1f32, const 1f32, const 1f32, const 1f32, const 42f32];
+-         _11 = F32x8(move _12);
++         _12 = const [1f32, 2f32, 3f32, 1f32, 1f32, 1f32, 1f32, 42f32];
++         _11 = F32x8(const [1f32, 2f32, 3f32, 1f32, 1f32, 1f32, 1f32, 42f32]);
+          StorageDead(_12);
+          StorageLive(_13);
+          StorageLive(_14);
+          _14 = [const 1_i32, const 0_i32, const 0_i32];
+          StorageLive(_15);
+          _15 = [const 0_i32, const 1_i32, const 0_i32];
+          StorageLive(_16);
+          _16 = [const 0_i32, const 0_i32, const 1_i32];
+          _13 = [move _14, move _15, move _16];
+          StorageDead(_16);
+          StorageDead(_15);
+          StorageDead(_14);
+          StorageDead(_13);
+          _0 = const ();
+          StorageDead(_11);
+          StorageDead(_6);
+          StorageDead(_3);
+          StorageDead(_2);
+          StorageDead(_1);
+          return;
+      }
++ }
++ 
++ ALLOC0 (size: 32, align: 4) {
++     0x00 │ 00 00 80 3f 00 00 00 40 00 00 40 40 00 00 80 3f │ ...?...@..@@...?
++     0x10 │ 00 00 80 3f 00 00 80 3f 00 00 80 3f 00 00 28 42 │ ...?...?...?..(B
++ }
++ 
++ ALLOC1 (size: 64, align: 2) {
++     0x00 │ ff 00 69 00 0f 00 27 00 3e 00 fb 00 bf 00 b2 00 │ ..i...'.>.......
++     0x10 │ 09 00 04 00 38 00 dd 00 c1 00 a4 00 c2 00 c5 00 │ ....8...........
++     0x20 │ 06 00 f3 00 da 00 ab 00 57 00 f7 00 68 00 9f 00 │ ........W...h...
++     0x30 │ 16 00 9d 00 69 00 1f 00 60 00 ad 00 32 00 01 00 │ ....i...`...2...
++ }
++ 
++ ALLOC2 (size: 48, align: 4) {
++     0x00 │ c1 00 00 00 a4 00 00 00 c2 00 00 00 c5 00 00 00 │ ................
++     0x10 │ 06 00 00 00 f3 00 00 00 da 00 00 00 ab 00 00 00 │ ................
++     0x20 │ 57 00 00 00 f7 00 00 00 68 00 00 00 2a 00 00 00 │ W.......h...*...
++ }
++ 
++ ALLOC3 (size: 48, align: 4) {
++     0x00 │ ff 00 00 00 69 00 00 00 0f 00 00 00 27 00 00 00 │ ....i.......'...
++     0x10 │ 3e 00 00 00 fb 00 00 00 bf 00 00 00 b2 00 00 00 │ >...............
++     0x20 │ 09 00 00 00 04 00 00 00 38 00 00 00 dd 00 00 00 │ ........8.......
++ }
++ 
++ ALLOC4 (size: 128, align: 4) {
++     0x00 │ ff 00 00 00 69 00 00 00 0f 00 00 00 27 00 00 00 │ ....i.......'...
++     0x10 │ 3e 00 00 00 fb 00 00 00 bf 00 00 00 b2 00 00 00 │ >...............
++     0x20 │ 09 00 00 00 04 00 00 00 38 00 00 00 dd 00 00 00 │ ........8.......
++     0x30 │ c1 00 00 00 a4 00 00 00 c2 00 00 00 c5 00 00 00 │ ................
++     0x40 │ 06 00 00 00 f3 00 00 00 da 00 00 00 ab 00 00 00 │ ................
++     0x50 │ 57 00 00 00 f7 00 00 00 68 00 00 00 9f 00 00 00 │ W.......h.......
++     0x60 │ 16 00 00 00 9d 00 00 00 69 00 00 00 1f 00 00 00 │ ........i.......
++     0x70 │ 60 00 00 00 ad 00 00 00 32 00 00 00 01 00 00 00 │ `.......2.......
+  }
+