diff --git a/compiler/rustc_codegen_gcc/src/builder.rs b/compiler/rustc_codegen_gcc/src/builder.rs
index 7671d2e026b03..bea3911686f5b 100644
--- a/compiler/rustc_codegen_gcc/src/builder.rs
+++ b/compiler/rustc_codegen_gcc/src/builder.rs
@@ -18,7 +18,7 @@ use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue};
 use rustc_codegen_ssa::mir::place::PlaceRef;
 use rustc_codegen_ssa::traits::{
     BackendTypes, BaseTypeCodegenMethods, BuilderMethods, ConstCodegenMethods,
-    LayoutTypeCodegenMethods, OverflowOp, StaticBuilderMethods,
+    LayoutTypeCodegenMethods, OverflowOp, ReturnSlot, StaticBuilderMethods,
 };
 use rustc_data_structures::fx::FxHashSet;
 use rustc_middle::bug;
@@ -603,6 +603,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         func: RValue<'gcc>,
+        return_slot: ReturnSlot<RValue<'gcc>>,
         args: &[RValue<'gcc>],
         then: Block<'gcc>,
         catch: Block<'gcc>,
@@ -613,7 +614,8 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
 
         let current_block = self.block;
         self.block = try_block;
-        let call = self.call(typ, fn_attrs, fn_abi, func, args, None, instance); // FIXME(antoyo): use funclet here?
+        // FIXME(antoyo): use funclet here?
+        let call = self.call(typ, fn_attrs, fn_abi, func, return_slot, args, None, instance);
         self.block = current_block;
 
         let return_value =
@@ -641,13 +643,14 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         func: RValue<'gcc>,
+        return_slot: ReturnSlot<RValue<'gcc>>,
         args: &[RValue<'gcc>],
         then: Block<'gcc>,
         catch: Block<'gcc>,
         _funclet: Option<&Funclet>,
         instance: Option<Instance<'tcx>>,
     ) -> RValue<'gcc> {
-        let call_site = self.call(typ, fn_attrs, fn_abi, func, args, None, instance);
+        let call_site = self.call(typ, fn_attrs, fn_abi, func, return_slot, args, None, instance);
         let condition = self.context.new_rvalue_from_int(self.bool_type, 1);
         self.llbb().end_with_conditional(self.location, condition, then, catch);
         if let Some(_fn_abi) = fn_abi {
@@ -1772,19 +1775,30 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         _fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         func: RValue<'gcc>,
+        return_slot: ReturnSlot<RValue<'gcc>>,
         args: &[RValue<'gcc>],
         funclet: Option<&Funclet>,
         _instance: Option<Instance<'tcx>>,
     ) -> RValue<'gcc> {
+        // FIXME: change this in the `rustc_codegen_gcc` repo after the sync, to use the `libgccjit` indirect return suppport.
+        let args = match return_slot {
+            ReturnSlot::Direct => args.to_vec(),
+            ReturnSlot::Indirect(sret_ptr) => {
+                let mut args = args.to_vec();
+                // Prepend the indirect return pointer
+                args.insert(0, sret_ptr);
+                args
+            }
+        };
         // FIXME(antoyo): remove when having a proper API.
         let gcc_func = unsafe { std::mem::transmute::<RValue<'gcc>, Function<'gcc>>(func) };
         let call = if self.functions.borrow().values().any(|value| *value == gcc_func) {
             // FIXME(antoyo): remove when the API supports a different type for functions.
             let func: Function<'gcc> = self.cx.rvalue_as_function(func);
-            self.function_call(func, args, funclet)
+            self.function_call(func, &args, funclet)
         } else {
             // If it's a not function that was defined, it's a function pointer.
-            self.function_ptr_call(typ, fn_abi, func, args, funclet)
+            self.function_ptr_call(typ, fn_abi, func, &args, funclet)
         };
         if let Some(_fn_abi) = fn_abi {
             // FIXME(bjorn3): Apply function attributes
@@ -1798,6 +1812,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         _fn_attrs: Option<&CodegenFnAttrs>,
         _fn_abi: &FnAbi<'tcx, Ty<'tcx>>,
         _llfn: Self::Value,
+        _return_slot: ReturnSlot<Self::Value>,
         _args: &[Self::Value],
         _funclet: Option<&Self::Funclet>,
         _instance: Option<Instance<'tcx>>,
diff --git a/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs b/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
index 986c04e6b4813..5244b771724a5 100644
--- a/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
+++ b/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
@@ -16,7 +16,7 @@ use rustc_codegen_ssa::mir::place::{PlaceRef, PlaceValue};
 use rustc_codegen_ssa::traits::MiscCodegenMethods;
 use rustc_codegen_ssa::traits::{
     ArgAbiBuilderMethods, BaseTypeCodegenMethods, BuilderMethods, ConstCodegenMethods,
-    IntrinsicCallBuilderMethods, LayoutTypeCodegenMethods,
+    IntrinsicCallBuilderMethods, LayoutTypeCodegenMethods, ReturnSlot,
 };
 use rustc_codegen_ssa::{MemFlags, RetagInfo};
 use rustc_data_structures::fx::FxHashSet;
@@ -656,7 +656,8 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tc
         }
 
         // FIXME directly use the llvm intrinsic adjustment functions here
-        let llret = self.call(fn_ty, None, None, fn_ptr, &call_args, None, None);
+        let llret =
+            self.call(fn_ty, None, None, fn_ptr, ReturnSlot::Direct, &call_args, None, None);
         if is_cleanup {
             self.apply_attrs_to_cleanup_callsite(llret);
         }
@@ -667,7 +668,7 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tc
     fn abort(&mut self) {
         let func = self.context.get_builtin_function("abort");
         let func: RValue<'gcc> = unsafe { std::mem::transmute(func) };
-        self.call(self.type_void(), None, None, func, &[], None, None);
+        self.call(self.type_void(), None, None, func, ReturnSlot::Direct, &[], None, None);
     }
 
     fn assume(&mut self, value: Self::Value) {
@@ -1348,7 +1349,7 @@ fn try_intrinsic<'a, 'b, 'gcc, 'tcx>(
         let param_type = bx.u8_type.make_pointer();
         let fn_type =
             bx.context.new_function_pointer_type(None, bx.type_void(), &[param_type], false);
-        bx.call(fn_type, None, None, try_func, &[data], None, None);
+        bx.call(fn_type, None, None, try_func, ReturnSlot::Direct, &[data], None, None);
         // Return 0 unconditionally from the intrinsic call;
         // we can never unwind.
         OperandValue::Immediate(bx.const_bool(false)).store(bx, dest);
@@ -1421,21 +1422,41 @@ fn codegen_gnu_try<'gcc, 'tcx>(
         let zero = bx.cx.context.new_rvalue_zero(bx.int_type);
         let ptr = bx.cx.context.new_call(None, eh_pointer_builtin, &[zero]);
         let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None, None);
+        bx.call(catch_ty, None, None, catch_func, ReturnSlot::Direct, &[data, ptr], None, None);
         bx.ret(bx.const_bool(true));
 
         // NOTE: the blocks must be filled before adding the try/catch, otherwise gcc will not
         // generate a try/catch.
         // FIXME(antoyo): add a check in the libgccjit API to prevent this.
         bx.switch_to_block(current_block);
-        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None, None);
+        bx.invoke(
+            try_func_ty,
+            None,
+            None,
+            try_func,
+            ReturnSlot::Direct,
+            &[data],
+            then,
+            catch,
+            None,
+            None,
+        );
     });
 
     let func = unsafe { std::mem::transmute::<Function<'gcc>, RValue<'gcc>>(func) };
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, func, &[try_func, data, catch_func], None, None);
+    let ret = bx.call(
+        llty,
+        None,
+        None,
+        func,
+        ReturnSlot::Direct,
+        &[try_func, data, catch_func],
+        None,
+        None,
+    );
     OperandValue::Immediate(ret).store(bx, dest);
 }
 
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
index 1efab9b7c496d..78f550ad1cedb 100644
--- a/compiler/rustc_codegen_llvm/src/asm.rs
+++ b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -562,9 +562,20 @@ pub(crate) fn inline_asm_call<'ll>(
         assert!(catch_funclet.is_none());
         bx.callbr(fty, None, None, v, inputs, dest.unwrap(), labels, None, None)
     } else if let Some((catch, funclet)) = catch_funclet {
-        bx.invoke(fty, None, None, v, inputs, dest.unwrap(), catch, funclet, None)
+        bx.invoke(
+            fty,
+            None,
+            None,
+            v,
+            ReturnSlot::Direct,
+            inputs,
+            dest.unwrap(),
+            catch,
+            funclet,
+            None,
+        )
     } else {
-        bx.call(fty, None, None, v, inputs, None, None)
+        bx.call(fty, None, None, v, ReturnSlot::Direct, inputs, None, None)
     };
 
     // Store mark in a metadata node so we can map LLVM errors
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index 55bb6f1abeb96..ce5ecc2032ff9 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -451,15 +451,25 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         llfn: &'ll Value,
+        return_slot: ReturnSlot<&'ll Value>,
         args: &[&'ll Value],
         then: &'ll BasicBlock,
         catch: &'ll BasicBlock,
         funclet: Option<&Funclet<'ll>>,
         instance: Option<Instance<'tcx>>,
     ) -> &'ll Value {
+        // If this function returns indirectly (`PassMode::Indirect`),
+        // the `return_slot` should be the first argument.
+        let args = match return_slot {
+            ReturnSlot::Direct => args.to_vec(),
+            ReturnSlot::Indirect(sret_ptr) => {
+                let mut args = args.to_vec();
+                args.insert(0, sret_ptr);
+                args
+            }
+        };
         debug!("invoke {:?} with args ({:?})", llfn, args);
-
-        let args = self.check_call("invoke", llty, llfn, args);
+        let args = self.check_call("invoke", llty, llfn, &args);
         let funclet_bundle = funclet.map(|funclet| funclet.bundle());
         let mut bundles: SmallVec<[_; 2]> = SmallVec::new();
         if let Some(funclet_bundle) = funclet_bundle {
@@ -1452,13 +1462,23 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         caller_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         llfn: &'ll Value,
+        return_slot: ReturnSlot<&'ll Value>,
         args: &[&'ll Value],
         funclet: Option<&Funclet<'ll>>,
         callee_instance: Option<Instance<'tcx>>,
     ) -> &'ll Value {
+        // If this function returns indirectly (`PassMode::Indirect`),
+        // the `return_slot` should be the first argument.
+        let args = match return_slot {
+            ReturnSlot::Direct => args.to_vec(),
+            ReturnSlot::Indirect(sret_ptr) => {
+                let mut args = args.to_vec();
+                args.insert(0, sret_ptr);
+                args
+            }
+        };
         debug!("call {:?} with args ({:?})", llfn, args);
-
-        let args = self.check_call("call", llty, llfn, args);
+        let args = self.check_call("call", llty, llfn, &args);
         let funclet_bundle = funclet.map(|funclet| funclet.bundle());
         let mut bundles: SmallVec<[_; 2]> = SmallVec::new();
         if let Some(funclet_bundle) = funclet_bundle {
@@ -1519,12 +1539,21 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         caller_attrs: Option<&CodegenFnAttrs>,
         fn_abi: &FnAbi<'tcx, Ty<'tcx>>,
         llfn: Self::Value,
+        return_slot: ReturnSlot<Self::Value>,
         args: &[Self::Value],
         funclet: Option<&Self::Funclet>,
         callee_instance: Option<Instance<'tcx>>,
     ) {
-        let call =
-            self.call(llty, caller_attrs, Some(fn_abi), llfn, args, funclet, callee_instance);
+        let call = self.call(
+            llty,
+            caller_attrs,
+            Some(fn_abi),
+            llfn,
+            return_slot,
+            args,
+            funclet,
+            callee_instance,
+        );
         llvm::LLVMSetTailCallKind(call, llvm::TailCallKind::MustTail);
 
         match &fn_abi.ret.mode {
@@ -1819,7 +1848,8 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         args: &[&'ll Value],
     ) -> &'ll Value {
         let (ty, f) = self.cx.get_intrinsic(base_name.into(), type_params);
-        self.call(ty, None, None, f, args, None, None)
+        // No LLVM intrinsic returns its data indirectly (via `sret`).
+        self.call(ty, None, None, f, ReturnSlot::Direct, args, None, None)
     }
 
     fn call_lifetime_intrinsic(&mut self, intrinsic: &'static str, ptr: &'ll Value, size: Size) {
diff --git a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
index b0ac5c096efdf..8c0cad2ed7f72 100644
--- a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
+++ b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
@@ -6,7 +6,7 @@ use rustc_codegen_ssa::common::TypeKind;
 use rustc_codegen_ssa::mir::IntrinsicResult;
 use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue};
 use rustc_codegen_ssa::mir::place::PlaceValue;
-use rustc_codegen_ssa::traits::{BaseTypeCodegenMethods, BuilderMethods};
+use rustc_codegen_ssa::traits::{BaseTypeCodegenMethods, BuilderMethods, ReturnSlot};
 use rustc_data_structures::thin_vec::ThinVec;
 use rustc_hir::attrs::RustcAutodiff;
 use rustc_middle::ty::{PseudoCanonicalInput, Ty, TyCtxt, TypingEnv};
@@ -377,7 +377,7 @@ pub(crate) fn generate_enzyme_call<'ll, 'tcx>(
         crate::typetree::add_tt(&bx, fn_to_diff, fnc_tree);
     }
 
-    let call = bx.call(enzyme_ty, None, None, ad_fn, &args, None, None);
+    let call = bx.call(enzyme_ty, None, None, ad_fn, ReturnSlot::Direct, &args, None, None);
 
     let fn_ret_ty = bx.cx.val_ty(call);
     if fn_ret_ty == bx.cx.type_void() || fn_ret_ty == bx.cx.type_struct(&[], false) {
diff --git a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
index 0b009321802cf..10081cec2ad56 100644
--- a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
+++ b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
@@ -6,7 +6,7 @@ use rustc_abi::Align;
 use rustc_codegen_ssa::MemFlags;
 use rustc_codegen_ssa::common::TypeKind;
 use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue};
-use rustc_codegen_ssa::traits::{BaseTypeCodegenMethods, BuilderMethods};
+use rustc_codegen_ssa::traits::{BaseTypeCodegenMethods, BuilderMethods, ReturnSlot};
 use rustc_middle::bug;
 use rustc_middle::ty::offload_meta::{MappingFlags, OffloadMetadata, OffloadSize};
 
@@ -109,9 +109,18 @@ pub(crate) fn register_offload<'ll>(cx: &CodegenCx<'ll, '_>) {
     // }
     let bb = Builder::append_block(cx, desc_reg_fn, "entry");
     let mut a = Builder::build(cx, bb);
-    a.call(reg_lib_decl, None, None, register_lib, &[omp_descriptor], None, None);
-    a.call(init_ty, None, None, init_rtls, &[], None, None);
-    a.call(atexit, None, None, atexit_fn, &[desc_unreg_fn], None, None);
+    a.call(
+        reg_lib_decl,
+        None,
+        None,
+        register_lib,
+        ReturnSlot::Direct,
+        &[omp_descriptor],
+        None,
+        None,
+    );
+    a.call(init_ty, None, None, init_rtls, ReturnSlot::Direct, &[], None, None);
+    a.call(atexit, None, None, atexit_fn, ReturnSlot::Direct, &[desc_unreg_fn], None, None);
     a.ret_void();
 
     // define internal void @.omp_offloading.descriptor_unreg() section ".text.startup" {
@@ -121,7 +130,16 @@ pub(crate) fn register_offload<'ll>(cx: &CodegenCx<'ll, '_>) {
     // }
     let bb = Builder::append_block(cx, desc_unreg_fn, "entry");
     let mut a = Builder::build(cx, bb);
-    a.call(reg_lib_decl, None, None, unregister_lib, &[omp_descriptor], None, None);
+    a.call(
+        reg_lib_decl,
+        None,
+        None,
+        unregister_lib,
+        ReturnSlot::Direct,
+        &[omp_descriptor],
+        None,
+        None,
+    );
     a.ret_void();
 
     // @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 101, ptr @.omp_offloading.descriptor_reg, ptr null }]
@@ -740,7 +758,7 @@ pub(crate) fn gen_call_handling<'ll, 'tcx>(
         let num_args = cx.get_const_i32(num_args);
         let args =
             vec![s_ident_t, i64_max, num_args, geps[0], geps[1], geps[2], o_type, nullptr, nullptr];
-        builder.call(fn_ty, None, None, fn_to_call, &args, None, None);
+        builder.call(fn_ty, None, None, fn_to_call, ReturnSlot::Direct, &args, None, None);
     }
 
     // Step 2)
@@ -784,7 +802,7 @@ pub(crate) fn gen_call_handling<'ll, 'tcx>(
         region_id,
         a5,
     ];
-    builder.call(tgt_target_kernel_ty, None, None, tgt_decl, &args, None, None);
+    builder.call(tgt_target_kernel_ty, None, None, tgt_decl, ReturnSlot::Direct, &args, None, None);
     // %41 = call i32 @__tgt_target_kernel(ptr @1, i64 -1, i32 2097152, i32 256, ptr @.kernel_1.region_id, ptr %kernel_args)
 
     // Step 4)
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 4f80e5c6e81e1..a5419e3a024f1 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -1340,7 +1340,7 @@ fn catch_unwind_intrinsic<'ll, 'tcx>(
 ) -> &'ll Value {
     if !bx.sess().panic_strategy().unwinds() {
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.call(try_func_ty, None, None, try_func, &[data], None, None);
+        bx.call(try_func_ty, None, None, try_func, ReturnSlot::Direct, &[data], None, None);
         // Return 0 unconditionally from the intrinsic call;
         // we can never unwind.
         bx.const_bool(false)
@@ -1438,7 +1438,18 @@ fn codegen_msvc_try<'ll, 'tcx>(
         let ptr_align = bx.tcx().data_layout.pointer_align().abi;
         let slot = bx.alloca(ptr_size, ptr_align);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None, None);
+        bx.invoke(
+            try_func_ty,
+            None,
+            None,
+            try_func,
+            ReturnSlot::Direct,
+            &[data],
+            normal,
+            catchswitch,
+            None,
+            None,
+        );
 
         bx.switch_to_block(normal);
         bx.ret(bx.const_bool(false));
@@ -1486,7 +1497,16 @@ fn codegen_msvc_try<'ll, 'tcx>(
         let funclet = bx.catch_pad(cs, &[tydesc, flags, slot]);
         let ptr = bx.load(bx.type_ptr(), slot, ptr_align);
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet), None);
+        bx.call(
+            catch_ty,
+            None,
+            None,
+            catch_func,
+            ReturnSlot::Direct,
+            &[data, ptr],
+            Some(&funclet),
+            None,
+        );
         bx.catch_ret(&funclet, caught);
 
         // The flag value of 64 indicates a "catch-all".
@@ -1494,7 +1514,16 @@ fn codegen_msvc_try<'ll, 'tcx>(
         let flags = bx.const_i32(64);
         let null = bx.const_null(bx.type_ptr());
         let funclet = bx.catch_pad(cs, &[null, flags, null]);
-        bx.call(catch_ty, None, None, catch_func, &[data, null], Some(&funclet), None);
+        bx.call(
+            catch_ty,
+            None,
+            None,
+            catch_func,
+            ReturnSlot::Direct,
+            &[data, null],
+            Some(&funclet),
+            None,
+        );
         bx.catch_ret(&funclet, caught);
 
         bx.switch_to_block(caught);
@@ -1503,7 +1532,16 @@ fn codegen_msvc_try<'ll, 'tcx>(
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
+    let ret = bx.call(
+        llty,
+        None,
+        None,
+        llfn,
+        ReturnSlot::Direct,
+        &[try_func, data, catch_func],
+        None,
+        None,
+    );
     ret
 }
 
@@ -1550,7 +1588,18 @@ fn codegen_wasm_try<'ll, 'tcx>(
         //   }
         //
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None, None);
+        bx.invoke(
+            try_func_ty,
+            None,
+            None,
+            try_func,
+            ReturnSlot::Direct,
+            &[data],
+            normal,
+            catchswitch,
+            None,
+            None,
+        );
 
         bx.switch_to_block(normal);
         bx.ret(bx.const_bool(false));
@@ -1566,7 +1615,16 @@ fn codegen_wasm_try<'ll, 'tcx>(
         let _sel = bx.call_intrinsic("llvm.wasm.get.ehselector", &[], &[funclet.cleanuppad()]);
 
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet), None);
+        bx.call(
+            catch_ty,
+            None,
+            None,
+            catch_func,
+            ReturnSlot::Direct,
+            &[data, ptr],
+            Some(&funclet),
+            None,
+        );
         bx.catch_ret(&funclet, caught);
 
         bx.switch_to_block(caught);
@@ -1575,7 +1633,16 @@ fn codegen_wasm_try<'ll, 'tcx>(
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
+    let ret = bx.call(
+        llty,
+        None,
+        None,
+        llfn,
+        ReturnSlot::Direct,
+        &[try_func, data, catch_func],
+        None,
+        None,
+    );
     ret
 }
 
@@ -1616,7 +1683,18 @@ fn codegen_gnu_try<'ll, 'tcx>(
         let data = llvm::get_param(bx.llfn(), 1);
         let catch_func = llvm::get_param(bx.llfn(), 2);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None, None);
+        bx.invoke(
+            try_func_ty,
+            None,
+            None,
+            try_func,
+            ReturnSlot::Direct,
+            &[data],
+            then,
+            catch,
+            None,
+            None,
+        );
 
         bx.switch_to_block(then);
         bx.ret(bx.const_bool(false));
@@ -1634,13 +1712,22 @@ fn codegen_gnu_try<'ll, 'tcx>(
         bx.add_clause(vals, tydesc);
         let ptr = bx.extract_value(vals, 0);
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None, None);
+        bx.call(catch_ty, None, None, catch_func, ReturnSlot::Direct, &[data, ptr], None, None);
         bx.ret(bx.const_bool(true));
     });
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
+    let ret = bx.call(
+        llty,
+        None,
+        None,
+        llfn,
+        ReturnSlot::Direct,
+        &[try_func, data, catch_func],
+        None,
+        None,
+    );
     ret
 }
 
diff --git a/compiler/rustc_codegen_llvm/src/mono_item.rs b/compiler/rustc_codegen_llvm/src/mono_item.rs
index b746bab643a34..0374af7dcf81c 100644
--- a/compiler/rustc_codegen_llvm/src/mono_item.rs
+++ b/compiler/rustc_codegen_llvm/src/mono_item.rs
@@ -198,12 +198,21 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
                 args.push(llvm::get_param(alias_lldecl, index));
             }
 
+            // For an indirect return, the alias's own first parameter is the
+            // caller-provided return slot: forward it to the aliasee as such.
+            let (return_slot, args) = if fn_abi.ret.is_indirect() {
+                let (sret_ptr, rest) = args.split_first().unwrap();
+                (ReturnSlot::Indirect(*sret_ptr), rest)
+            } else {
+                (ReturnSlot::Direct, &args[..])
+            };
             let call = start_bx.call(
                 fn_ty,
                 Some(attrs),
                 Some(fn_abi),
                 aliasee,
-                &args,
+                return_slot,
+                args,
                 None,
                 Some(aliasee_instance),
             );
diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs
index 339c0a968e9d3..4e1faa58dfda9 100644
--- a/compiler/rustc_codegen_ssa/src/base.rs
+++ b/compiler/rustc_codegen_ssa/src/base.rs
@@ -592,7 +592,8 @@ pub fn maybe_create_entry_wrapper<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
             )
         };
 
-        let result = bx.call(start_ty, None, None, start_fn, &args, None, instance);
+        let result =
+            bx.call(start_ty, None, None, start_fn, ReturnSlot::Direct, &args, None, instance);
         if cx.sess().target.os == Os::Uefi {
             bx.ret(result);
         } else {
diff --git a/compiler/rustc_codegen_ssa/src/mir/block.rs b/compiler/rustc_codegen_ssa/src/mir/block.rs
index 20d6c01f12f73..4228d39039e87 100644
--- a/compiler/rustc_codegen_ssa/src/mir/block.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/block.rs
@@ -165,12 +165,16 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
 
     /// Call `fn_ptr` of `fn_abi` with the arguments `llargs`, the optional
     /// return destination `destination` and the unwind action `unwind`.
+    /// The `return_slot` is [`ReturnSlot::Indirect`] for functions returning
+    /// via `PassMode::Indirect`, and points to a buffer where the return value
+    /// shall be stored.
     fn do_call<Bx: BuilderMethods<'a, 'tcx>>(
         &self,
         fx: &mut FunctionCx<'a, 'tcx, Bx>,
         bx: &mut Bx,
         fn_abi: &'tcx FnAbi<'tcx, Ty<'tcx>>,
         fn_ptr: Bx::Value,
+        return_slot: ReturnSlot<Bx::Value>,
         llargs: &[Bx::Value],
         destination: Option<(ReturnDest<'tcx, Bx::Value>, mir::BasicBlock)>,
         mut unwind: mir::UnwindAction,
@@ -244,8 +248,23 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
             }
         };
 
+        debug_assert_eq!(
+            return_slot.is_indirect(),
+            fn_abi.ret.is_indirect(),
+            "a return slot must be provided if and only if the return is `PassMode::Indirect`",
+        );
+
         if kind == CallKind::Tail {
-            bx.tail_call(fn_ty, caller_attrs, fn_abi, fn_ptr, llargs, self.funclet(fx), instance);
+            bx.tail_call(
+                fn_ty,
+                caller_attrs,
+                fn_abi,
+                fn_ptr,
+                return_slot,
+                llargs,
+                self.funclet(fx),
+                instance,
+            );
             return MergingSucc::False;
         }
 
@@ -260,6 +279,7 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
                 caller_attrs,
                 Some(fn_abi),
                 fn_ptr,
+                return_slot,
                 llargs,
                 ret_llbb,
                 unwind_block,
@@ -291,6 +311,7 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
                 caller_attrs,
                 Some(fn_abi),
                 fn_ptr,
+                return_slot,
                 llargs,
                 self.funclet(fx),
                 instance,
@@ -721,6 +742,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             bx,
             fn_abi,
             drop_fn,
+            ReturnSlot::Direct,
             args,
             Some((ReturnDest::Nothing, target)),
             unwind,
@@ -825,6 +847,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             bx,
             fn_abi,
             llfn,
+            ReturnSlot::Direct,
             &args,
             None,
             unwind,
@@ -856,6 +879,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             bx,
             fn_abi,
             llfn,
+            ReturnSlot::Direct,
             &[],
             None,
             mir::UnwindAction::Unreachable,
@@ -925,6 +949,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             bx,
             fn_abi,
             llfn,
+            ReturnSlot::Direct,
             &[msg.0, msg.1],
             target.as_ref().map(|bb| (ReturnDest::Nothing, *bb)),
             unwind,
@@ -1205,21 +1230,24 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         // We still need to call `make_return_dest` even if there's no `target`, since
         // `fn_abi.ret` could be `PassMode::Indirect`, even if it is uninhabited,
         // and `make_return_dest` adds the return-place indirect pointer to `llargs`.
-        let destination = match kind {
+        let (destination, return_slot) = match kind {
             CallKind::Normal => {
-                let return_dest = self.make_return_dest(bx, destination, &fn_abi.ret, &mut llargs);
-                target.map(|target| (return_dest, target))
+                let (return_dest, return_slot) =
+                    self.make_return_dest(bx, destination, &fn_abi.ret);
+                (target.map(|target| (return_dest, target)), return_slot)
             }
             CallKind::Tail => {
-                if fn_abi.ret.is_indirect() {
-                    match self.make_return_dest(bx, destination, &fn_abi.ret, &mut llargs) {
-                        ReturnDest::Nothing => {}
+                let return_slot = if fn_abi.ret.is_indirect() {
+                    match self.make_return_dest(bx, destination, &fn_abi.ret) {
+                        (ReturnDest::Nothing, return_slot) => return_slot,
                         _ => bug!(
                             "tail calls to functions with indirect returns cannot store into a destination"
                         ),
                     }
-                }
-                None
+                } else {
+                    ReturnSlot::Direct
+                };
+                (None, return_slot)
             }
         };
 
@@ -1444,6 +1472,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             bx,
             fn_abi,
             fn_ptr,
+            return_slot,
             &llargs,
             destination,
             unwind,
@@ -2369,7 +2398,16 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         } else {
             let fn_ty = bx.fn_decl_backend_type(fn_abi);
 
-            let llret = bx.call(fn_ty, None, Some(fn_abi), fn_ptr, &[], funclet.as_ref(), None);
+            let llret = bx.call(
+                fn_ty,
+                None,
+                Some(fn_abi),
+                fn_ptr,
+                ReturnSlot::Direct,
+                &[],
+                funclet.as_ref(),
+                None,
+            );
             bx.apply_attrs_to_cleanup_callsite(llret);
         }
 
@@ -2405,11 +2443,10 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         bx: &mut Bx,
         dest: mir::Place<'tcx>,
         fn_ret: &ArgAbi<'tcx, Ty<'tcx>>,
-        llargs: &mut Vec<Bx::Value>,
-    ) -> ReturnDest<'tcx, Bx::Value> {
+    ) -> (ReturnDest<'tcx, Bx::Value>, ReturnSlot<Bx::Value>) {
         // If the return is ignored, we can just return a do-nothing `ReturnDest`.
         if fn_ret.is_ignore() {
-            return ReturnDest::Nothing;
+            return (ReturnDest::Nothing, ReturnSlot::Direct);
         }
         let dest = if let Some(index) = dest.as_local() {
             match self.locals[index] {
@@ -2423,10 +2460,12 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                         // but the calling convention has an indirect return.
                         let tmp = PlaceRef::alloca(bx, fn_ret.layout);
                         tmp.storage_live(bx);
-                        llargs.push(tmp.val.llval);
-                        ReturnDest::IndirectOperand(tmp, index)
+                        (
+                            ReturnDest::IndirectOperand(tmp, index),
+                            ReturnSlot::Indirect(tmp.val.llval),
+                        )
                     } else {
-                        ReturnDest::DirectOperand(index)
+                        (ReturnDest::DirectOperand(index), ReturnSlot::Direct)
                     };
                 }
                 LocalRef::Operand(_) => {
@@ -2446,10 +2485,9 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 // to create a temporary.
                 span_bug!(self.mir.span, "can't directly store to unaligned value");
             }
-            llargs.push(dest.val.llval);
-            ReturnDest::Nothing
+            (ReturnDest::Nothing, ReturnSlot::Indirect(dest.val.llval))
         } else {
-            ReturnDest::Store(dest)
+            (ReturnDest::Store(dest), ReturnSlot::Direct)
         }
     }
 
diff --git a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
index 4270073b29894..559eae3edbf03 100644
--- a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
@@ -760,6 +760,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                         fn_attrs.as_deref(),
                         Some(fn_abi),
                         fn_ptr,
+                        ReturnSlot::Direct,
                         &[],
                         None,
                         Some(instance),
diff --git a/compiler/rustc_codegen_ssa/src/size_of_val.rs b/compiler/rustc_codegen_ssa/src/size_of_val.rs
index 52ffc321cbb6f..e25421b7778a0 100644
--- a/compiler/rustc_codegen_ssa/src/size_of_val.rs
+++ b/compiler/rustc_codegen_ssa/src/size_of_val.rs
@@ -77,6 +77,7 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
                 /* fn_attrs */ None,
                 Some(fn_abi),
                 llfn,
+                ReturnSlot::Direct, // we know the ABI here
                 &[msg.0, msg.1],
                 None,
                 None,
diff --git a/compiler/rustc_codegen_ssa/src/traits/builder.rs b/compiler/rustc_codegen_ssa/src/traits/builder.rs
index caf63abeef3ad..0eb0ed79fccf1 100644
--- a/compiler/rustc_codegen_ssa/src/traits/builder.rs
+++ b/compiler/rustc_codegen_ssa/src/traits/builder.rs
@@ -34,6 +34,20 @@ pub enum OverflowOp {
     Mul,
 }
 
+/// The location of the return value for the call.
+#[derive(Copy, Clone, Debug)]
+pub enum ReturnSlot<V> {
+    Direct,
+    /// The return value will be passed via sret (e.g. `PassMode::Indirect`).
+    Indirect(V),
+}
+
+impl<V> ReturnSlot<V> {
+    pub fn is_indirect(&self) -> bool {
+        matches!(self, ReturnSlot::Indirect(_))
+    }
+}
+
 pub trait BuilderMethods<'a, 'tcx>:
     Sized
     + LayoutOf<'tcx, LayoutOfResult = TyAndLayout<'tcx>>
@@ -135,6 +149,7 @@ pub trait BuilderMethods<'a, 'tcx>:
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         llfn: Self::Value,
+        return_slot: ReturnSlot<Self::Value>,
         args: &[Self::Value],
         then: Self::BasicBlock,
         catch: Self::BasicBlock,
@@ -640,16 +655,20 @@ pub trait BuilderMethods<'a, 'tcx>:
     /// The typical case that they are None is during the codegen of intrinsics and lang-items,
     /// as those are "fake functions" with only a trivial ABI if any, et cetera.
     ///
+    /// `return_slot` must be `ReturnSlot::Indirect` if an argument uses `PassMode::Indirect`.
+    ///
     /// ## Return
     ///
-    /// Must return the value the function will return so it can be written to the destination,
-    /// assuming the function does not explicitly pass the destination as a pointer in `args`.
+    /// Must return the value the function will return so it can be written to the destination.
+    /// For calls with an indirect return, the returned value is meaningless and must not be
+    /// used: the return value lives in the return slot.
     fn call(
         &mut self,
         llty: Self::FunctionSignature,
         caller_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         fn_val: Self::Value,
+        return_slot: ReturnSlot<Self::Value>,
         args: &[Self::Value],
         funclet: Option<&Self::Funclet>,
         callee_instance: Option<Instance<'tcx>>,
@@ -661,6 +680,7 @@ pub trait BuilderMethods<'a, 'tcx>:
         caller_attrs: Option<&CodegenFnAttrs>,
         fn_abi: &FnAbi<'tcx, Ty<'tcx>>,
         llfn: Self::Value,
+        return_slot: ReturnSlot<Self::Value>,
         args: &[Self::Value],
         funclet: Option<&Self::Funclet>,
         callee_instance: Option<Instance<'tcx>>,
diff --git a/compiler/rustc_codegen_ssa/src/traits/mod.rs b/compiler/rustc_codegen_ssa/src/traits/mod.rs
index f46d07ea5008e..1f013cb122070 100644
--- a/compiler/rustc_codegen_ssa/src/traits/mod.rs
+++ b/compiler/rustc_codegen_ssa/src/traits/mod.rs
@@ -36,7 +36,7 @@ pub use self::asm::{
     AsmBuilderMethods, AsmCodegenMethods, GlobalAsmOperandRef, InlineAsmOperandRef,
 };
 pub use self::backend::{BackendTypes, CodegenBackend, ExtraBackendMethods};
-pub use self::builder::{BuilderMethods, OverflowOp};
+pub use self::builder::{BuilderMethods, OverflowOp, ReturnSlot};
 pub use self::consts::ConstCodegenMethods;
 pub use self::coverageinfo::CoverageInfoBuilderMethods;
 pub use self::debuginfo::{DebugInfoBuilderMethods, DebugInfoCodegenMethods};