atomicrmw on pointers: move integer-pointer cast hacks into backend

RalfJung · RalfJung · commit a5a87d322b9d · 2025-07-21T08:52:39.000+02:00
diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs
@@ -982,7 +982,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
             let old =
                 fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::Add, ptr, amount);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_xsub => {
@@ -1004,7 +1004,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
             let old =
                 fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::Sub, ptr, amount);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_and => {
@@ -1025,7 +1025,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             let old = fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::And, ptr, src);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_or => {
@@ -1046,7 +1046,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             let old = fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::Or, ptr, src);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_xor => {
@@ -1067,7 +1067,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             let old = fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::Xor, ptr, src);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_nand => {
@@ -1088,7 +1088,7 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             let old = fx.bcx.ins().atomic_rmw(ty, MemFlags::trusted(), AtomicRmwOp::Nand, ptr, src);
 
-            let old = CValue::by_val(old, layout);
+            let old = CValue::by_val(old, ret.layout());
             ret.write_cvalue(fx, old);
         }
         sym::atomic_max => {
diff --git a/compiler/rustc_codegen_gcc/src/builder.rs b/compiler/rustc_codegen_gcc/src/builder.rs
@@ -1656,6 +1656,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         dst: RValue<'gcc>,
         src: RValue<'gcc>,
         order: AtomicOrdering,
+        ret_ptr: bool,
     ) -> RValue<'gcc> {
         let size = get_maybe_pointer_size(src);
         let name = match op {
@@ -1683,14 +1684,18 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         let atomic_function = self.context.get_builtin_function(name);
         let order = self.context.new_rvalue_from_int(self.i32_type, order.to_gcc());
 
+        // FIXME: If `ret_ptr` is true and `src` is an integer, we should really tell GCC
+        // that this is a pointer operation that needs to preserve provenance -- but like LLVM,
+        // GCC does not currently seems to support that.
         let void_ptr_type = self.context.new_type::<*mut ()>();
         let volatile_void_ptr_type = void_ptr_type.make_volatile();
         let dst = self.context.new_cast(self.location, dst, volatile_void_ptr_type);
         // FIXME(antoyo): not sure why, but we have the wrong type here.
         let new_src_type = atomic_function.get_param(1).to_rvalue().get_type();
         let src = self.context.new_bitcast(self.location, src, new_src_type);
         let res = self.context.new_call(self.location, atomic_function, &[dst, src, order]);
-        self.context.new_cast(self.location, res, src.get_type())
+        let res_type = if ret_ptr { void_ptr_type } else { src.get_type() };
+        self.context.new_cast(self.location, res, res_type)
     }
 
     fn atomic_fence(&mut self, order: AtomicOrdering, scope: SynchronizationScope) {
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -1257,15 +1257,13 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         &mut self,
         op: rustc_codegen_ssa::common::AtomicRmwBinOp,
         dst: &'ll Value,
-        mut src: &'ll Value,
+        src: &'ll Value,
         order: rustc_middle::ty::AtomicOrdering,
+        ret_ptr: bool,
     ) -> &'ll Value {
-        // The only RMW operation that LLVM supports on pointers is compare-exchange.
-        let requires_cast_to_int = self.val_ty(src) == self.type_ptr()
-            && op != rustc_codegen_ssa::common::AtomicRmwBinOp::AtomicXchg;
-        if requires_cast_to_int {
-            src = self.ptrtoint(src, self.type_isize());
-        }
+        // FIXME: If `ret_ptr` is true and `src` is not a pointer, we *should* tell LLVM that the
+        // LHS is a pointer and the operation should be provenance-preserving, but LLVM does not
+        // currently support that (https://github.com/llvm/llvm-project/issues/120837).
         let mut res = unsafe {
             llvm::LLVMBuildAtomicRMW(
                 self.llbuilder,
@@ -1276,7 +1274,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
                 llvm::False, // SingleThreaded
             )
         };
-        if requires_cast_to_int {
+        if ret_ptr && self.val_ty(res) != self.type_ptr() {
             res = self.inttoptr(res, self.type_ptr());
         }
         res
diff --git a/compiler/rustc_codegen_ssa/messages.ftl b/compiler/rustc_codegen_ssa/messages.ftl
@@ -97,6 +97,8 @@ codegen_ssa_invalid_monomorphization_basic_float_type = invalid monomorphization
 
 codegen_ssa_invalid_monomorphization_basic_integer_type = invalid monomorphization of `{$name}` intrinsic: expected basic integer type, found `{$ty}`
 
+codegen_ssa_invalid_monomorphization_basic_integer_or_ptr_type = invalid monomorphization of `{$name}` intrinsic: expected basic integer or pointer type, found `{$ty}`
+
 codegen_ssa_invalid_monomorphization_cannot_return = invalid monomorphization of `{$name}` intrinsic: cannot return `{$ret_ty}`, expected `u{$expected_int_bits}` or `[u8; {$expected_bytes}]`
 
 codegen_ssa_invalid_monomorphization_cast_wide_pointer = invalid monomorphization of `{$name}` intrinsic: cannot cast wide pointer `{$ty}`
diff --git a/compiler/rustc_codegen_ssa/src/errors.rs b/compiler/rustc_codegen_ssa/src/errors.rs
@@ -764,6 +764,14 @@ pub enum InvalidMonomorphization<'tcx> {
         ty: Ty<'tcx>,
     },
 
+    #[diag(codegen_ssa_invalid_monomorphization_basic_integer_or_ptr_type, code = E0511)]
+    BasicIntegerOrPtrType {
+        #[primary_span]
+        span: Span,
+        name: Symbol,
+        ty: Ty<'tcx>,
+    },
+
     #[diag(codegen_ssa_invalid_monomorphization_basic_float_type, code = E0511)]
     BasicFloatType {
         #[primary_span]
diff --git a/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs b/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs
@@ -92,6 +92,13 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         let invalid_monomorphization_int_type = |ty| {
             bx.tcx().dcx().emit_err(InvalidMonomorphization::BasicIntegerType { span, name, ty });
         };
+        let invalid_monomorphization_int_or_ptr_type = |ty| {
+            bx.tcx().dcx().emit_err(InvalidMonomorphization::BasicIntegerOrPtrType {
+                span,
+                name,
+                ty,
+            });
+        };
 
         let parse_atomic_ordering = |ord: ty::Value<'tcx>| {
             let discr = ord.valtree.unwrap_branch()[0].unwrap_leaf();
@@ -351,7 +358,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             sym::atomic_load => {
                 let ty = fn_args.type_at(0);
                 if !(int_type_width_signed(ty, bx.tcx()).is_some() || ty.is_raw_ptr()) {
-                    invalid_monomorphization_int_type(ty);
+                    invalid_monomorphization_int_or_ptr_type(ty);
                     return Ok(());
                 }
                 let ordering = fn_args.const_at(1).to_value();
@@ -367,7 +374,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             sym::atomic_store => {
                 let ty = fn_args.type_at(0);
                 if !(int_type_width_signed(ty, bx.tcx()).is_some() || ty.is_raw_ptr()) {
-                    invalid_monomorphization_int_type(ty);
+                    invalid_monomorphization_int_or_ptr_type(ty);
                     return Ok(());
                 }
                 let ordering = fn_args.const_at(1).to_value();
@@ -377,10 +384,11 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 bx.atomic_store(val, ptr, parse_atomic_ordering(ordering), size);
                 return Ok(());
             }
+            // These are all AtomicRMW ops
             sym::atomic_cxchg | sym::atomic_cxchgweak => {
                 let ty = fn_args.type_at(0);
                 if !(int_type_width_signed(ty, bx.tcx()).is_some() || ty.is_raw_ptr()) {
-                    invalid_monomorphization_int_type(ty);
+                    invalid_monomorphization_int_or_ptr_type(ty);
                     return Ok(());
                 }
                 let succ_ordering = fn_args.const_at(1).to_value();
@@ -407,7 +415,6 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
 
                 return Ok(());
             }
-            // These are all AtomicRMW ops
             sym::atomic_max | sym::atomic_min => {
                 let atom_op = if name == sym::atomic_max {
                     AtomicRmwBinOp::AtomicMax
@@ -420,7 +427,13 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                     let ordering = fn_args.const_at(1).to_value();
                     let ptr = args[0].immediate();
                     let val = args[1].immediate();
-                    bx.atomic_rmw(atom_op, ptr, val, parse_atomic_ordering(ordering))
+                    bx.atomic_rmw(
+                        atom_op,
+                        ptr,
+                        val,
+                        parse_atomic_ordering(ordering),
+                        /* ret_ptr */ false,
+                    )
                 } else {
                     invalid_monomorphization_int_type(ty);
                     return Ok(());
@@ -438,21 +451,44 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                     let ordering = fn_args.const_at(1).to_value();
                     let ptr = args[0].immediate();
                     let val = args[1].immediate();
-                    bx.atomic_rmw(atom_op, ptr, val, parse_atomic_ordering(ordering))
+                    bx.atomic_rmw(
+                        atom_op,
+                        ptr,
+                        val,
+                        parse_atomic_ordering(ordering),
+                        /* ret_ptr */ false,
+                    )
                 } else {
                     invalid_monomorphization_int_type(ty);
                     return Ok(());
                 }
             }
-            sym::atomic_xchg
-            | sym::atomic_xadd
+            sym::atomic_xchg => {
+                let ty = fn_args.type_at(0);
+                let ordering = fn_args.const_at(fn_args.len() - 1).to_value();
+                let ptr = args[0].immediate();
+                let val = args[1].immediate();
+                if int_type_width_signed(ty, bx.tcx()).is_some() || ty.is_raw_ptr() {
+                    let atomic_op = AtomicRmwBinOp::AtomicXchg;
+                    bx.atomic_rmw(
+                        atomic_op,
+                        ptr,
+                        val,
+                        parse_atomic_ordering(ordering),
+                        /* ret_ptr */ ty.is_raw_ptr(),
+                    )
+                } else {
+                    invalid_monomorphization_int_or_ptr_type(ty);
+                    return Ok(());
+                }
+            }
+            sym::atomic_xadd
             | sym::atomic_xsub
             | sym::atomic_and
             | sym::atomic_nand
             | sym::atomic_or
             | sym::atomic_xor => {
                 let atom_op = match name {
-                    sym::atomic_xchg => AtomicRmwBinOp::AtomicXchg,
                     sym::atomic_xadd => AtomicRmwBinOp::AtomicAdd,
                     sym::atomic_xsub => AtomicRmwBinOp::AtomicSub,
                     sym::atomic_and => AtomicRmwBinOp::AtomicAnd,
@@ -462,14 +498,28 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                     _ => unreachable!(),
                 };
 
-                let ty = fn_args.type_at(0);
-                if int_type_width_signed(ty, bx.tcx()).is_some() || ty.is_raw_ptr() {
-                    let ordering = fn_args.const_at(1).to_value();
-                    let ptr = args[0].immediate();
-                    let val = args[1].immediate();
-                    bx.atomic_rmw(atom_op, ptr, val, parse_atomic_ordering(ordering))
+                // The type of the in-memory data.
+                let ty_mem = fn_args.type_at(0);
+                // The type of the 2nd operand, given by-value.
+                let ty_op = fn_args.type_at(1);
+
+                let ordering = fn_args.const_at(2).to_value();
+                let ptr = args[0].immediate(); // of type "pointer to `ty_mem`"
+                let val = args[1].immediate(); // of type `ty_op`
+                // We require either both arguments to have the same integer type, or the first to
+                // be a pointer and the second to be `usize`.
+                if (int_type_width_signed(ty_mem, bx.tcx()).is_some() && ty_op == ty_mem)
+                    || (ty_mem.is_raw_ptr() && ty_op == bx.tcx().types.usize)
+                {
+                    bx.atomic_rmw(
+                        atom_op,
+                        ptr,
+                        val,
+                        parse_atomic_ordering(ordering),
+                        /* ret_ptr */ ty_mem.is_raw_ptr(),
+                    )
                 } else {
-                    invalid_monomorphization_int_type(ty);
+                    invalid_monomorphization_int_or_ptr_type(ty_mem);
                     return Ok(());
                 }
             }
diff --git a/compiler/rustc_codegen_ssa/src/traits/builder.rs b/compiler/rustc_codegen_ssa/src/traits/builder.rs
@@ -548,12 +548,15 @@ pub trait BuilderMethods<'a, 'tcx>:
         failure_order: AtomicOrdering,
         weak: bool,
     ) -> (Self::Value, Self::Value);
+    /// `ret_ptr` indicates whether the return type (which is also the type `dst` points to)
+    /// is a pointer or the same type as `src`.
     fn atomic_rmw(
         &mut self,
         op: AtomicRmwBinOp,
         dst: Self::Value,
         src: Self::Value,
         order: AtomicOrdering,
+        ret_ptr: bool,
     ) -> Self::Value;
     fn atomic_fence(&mut self, order: AtomicOrdering, scope: SynchronizationScope);
     fn set_invariant_load(&mut self, load: Self::Value);
diff --git a/compiler/rustc_hir_analysis/src/check/intrinsic.rs b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
@@ -652,16 +652,16 @@ pub(crate) fn check_intrinsic_type(
         sym::atomic_store => (1, 1, vec![Ty::new_mut_ptr(tcx, param(0)), param(0)], tcx.types.unit),
 
         sym::atomic_xchg
-        | sym::atomic_xadd
-        | sym::atomic_xsub
-        | sym::atomic_and
-        | sym::atomic_nand
-        | sym::atomic_or
-        | sym::atomic_xor
         | sym::atomic_max
         | sym::atomic_min
         | sym::atomic_umax
         | sym::atomic_umin => (1, 1, vec![Ty::new_mut_ptr(tcx, param(0)), param(0)], param(0)),
+        sym::atomic_xadd
+        | sym::atomic_xsub
+        | sym::atomic_and
+        | sym::atomic_nand
+        | sym::atomic_or
+        | sym::atomic_xor => (2, 1, vec![Ty::new_mut_ptr(tcx, param(0)), param(1)], param(0)),
         sym::atomic_fence | sym::atomic_singlethreadfence => (0, 1, Vec::new(), tcx.types.unit),
 
         other => {
diff --git a/library/core/src/intrinsics/mod.rs b/library/core/src/intrinsics/mod.rs
diff --git a/library/core/src/sync/atomic.rs b/library/core/src/sync/atomic.rs
diff --git a/src/tools/miri/src/intrinsics/atomic.rs b/src/tools/miri/src/intrinsics/atomic.rs
diff --git a/src/tools/miri/src/operator.rs b/src/tools/miri/src/operator.rs
