diff --git a/scripts/filter_profile.rs b/scripts/filter_profile.rs
index f782671fe36..03912b18ea5 100755
--- a/scripts/filter_profile.rs
+++ b/scripts/filter_profile.rs
@@ -100,9 +100,9 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
             stack = &stack[..index + ENCODE_METADATA.len()];
         }
 
-        const SUBST_AND_NORMALIZE_ERASING_REGIONS: &str = "rustc_middle::ty::normalize_erasing_regions::<impl rustc_middle::ty::context::TyCtxt>::subst_and_normalize_erasing_regions";
-        if let Some(index) = stack.find(SUBST_AND_NORMALIZE_ERASING_REGIONS) {
-            stack = &stack[..index + SUBST_AND_NORMALIZE_ERASING_REGIONS.len()];
+        const INSTANTIATE_AND_NORMALIZE_ERASING_REGIONS: &str = "rustc_middle::ty::normalize_erasing_regions::<impl rustc_middle::ty::context::TyCtxt>::instantiate_and_normalize_erasing_regions";
+        if let Some(index) = stack.find(INSTANTIATE_AND_NORMALIZE_ERASING_REGIONS) {
+            stack = &stack[..index + INSTANTIATE_AND_NORMALIZE_ERASING_REGIONS.len()];
         }
 
         const NORMALIZE_ERASING_LATE_BOUND_REGIONS: &str = "rustc_middle::ty::normalize_erasing_regions::<impl rustc_middle::ty::context::TyCtxt>::normalize_erasing_late_bound_regions";
diff --git a/src/base.rs b/src/base.rs
index 798de7ccc79..43afba6322f 100644
--- a/src/base.rs
+++ b/src/base.rs
@@ -874,7 +874,7 @@ pub(crate) fn codegen_place<'tcx>(
             PlaceElem::Deref => {
                 cplace = cplace.place_deref(fx);
             }
-            PlaceElem::OpaqueCast(ty) => cplace = cplace.place_opaque_cast(fx, ty),
+            PlaceElem::OpaqueCast(ty) => bug!("encountered OpaqueCast({ty}) in codegen"),
             PlaceElem::Field(field, _ty) => {
                 cplace = cplace.place_field(fx, field);
             }
diff --git a/src/common.rs b/src/common.rs
index 2d36a557c8b..8958369267e 100644
--- a/src/common.rs
+++ b/src/common.rs
@@ -358,7 +358,7 @@ pub(crate) fn monomorphize<T>(&self, value: T) -> T
     where
         T: TypeFoldable<TyCtxt<'tcx>> + Copy,
     {
-        self.instance.subst_mir_and_normalize_erasing_regions(
+        self.instance.instantiate_mir_and_normalize_erasing_regions(
             self.tcx,
             ty::ParamEnv::reveal_all(),
             ty::EarlyBinder::bind(value),
diff --git a/src/intrinsics/simd.rs b/src/intrinsics/simd.rs
index 660be9be9bf..256b55a2a01 100644
--- a/src/intrinsics/simd.rs
+++ b/src/intrinsics/simd.rs
@@ -21,7 +21,7 @@ fn report_simd_type_validation_error(
 pub(super) fn codegen_simd_intrinsic_call<'tcx>(
     fx: &mut FunctionCx<'_, '_, 'tcx>,
     intrinsic: Symbol,
-    _args: GenericArgsRef<'tcx>,
+    generic_args: GenericArgsRef<'tcx>,
     args: &[mir::Operand<'tcx>],
     ret: CPlace<'tcx>,
     target: BasicBlock,
@@ -117,6 +117,54 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>(
             });
         }
 
+        // simd_shuffle_generic<T, U, const I: &[u32]>(x: T, y: T) -> U
+        sym::simd_shuffle_generic => {
+            let [x, y] = args else {
+                bug!("wrong number of args for intrinsic {intrinsic}");
+            };
+            let x = codegen_operand(fx, x);
+            let y = codegen_operand(fx, y);
+
+            if !x.layout().ty.is_simd() {
+                report_simd_type_validation_error(fx, intrinsic, span, x.layout().ty);
+                return;
+            }
+
+            let idx = generic_args[2]
+                .expect_const()
+                .eval(fx.tcx, ty::ParamEnv::reveal_all(), Some(span))
+                .unwrap()
+                .unwrap_branch();
+
+            assert_eq!(x.layout(), y.layout());
+            let layout = x.layout();
+
+            let (lane_count, lane_ty) = layout.ty.simd_size_and_type(fx.tcx);
+            let (ret_lane_count, ret_lane_ty) = ret.layout().ty.simd_size_and_type(fx.tcx);
+
+            assert_eq!(lane_ty, ret_lane_ty);
+            assert_eq!(idx.len() as u64, ret_lane_count);
+
+            let total_len = lane_count * 2;
+
+            let indexes =
+                idx.iter().map(|idx| idx.unwrap_leaf().try_to_u16().unwrap()).collect::<Vec<u16>>();
+
+            for &idx in &indexes {
+                assert!(u64::from(idx) < total_len, "idx {} out of range 0..{}", idx, total_len);
+            }
+
+            for (out_idx, in_idx) in indexes.into_iter().enumerate() {
+                let in_lane = if u64::from(in_idx) < lane_count {
+                    x.value_lane(fx, in_idx.into())
+                } else {
+                    y.value_lane(fx, u64::from(in_idx) - lane_count)
+                };
+                let out_lane = ret.place_lane(fx, u64::try_from(out_idx).unwrap());
+                out_lane.write_cvalue(fx, in_lane);
+            }
+        }
+
         // simd_shuffle<T, I, U>(x: T, y: T, idx: I) -> U
         sym::simd_shuffle => {
             let (x, y, idx) = match args {
diff --git a/src/value_and_place.rs b/src/value_and_place.rs
index ef85c6da2d8..bc7a6f1d591 100644
--- a/src/value_and_place.rs
+++ b/src/value_and_place.rs
@@ -674,14 +674,6 @@ fn transmute_scalar<'tcx>(
         }
     }
 
-    pub(crate) fn place_opaque_cast(
-        self,
-        fx: &mut FunctionCx<'_, '_, 'tcx>,
-        ty: Ty<'tcx>,
-    ) -> CPlace<'tcx> {
-        CPlace { inner: self.inner, layout: fx.layout_of(ty) }
-    }
-
     pub(crate) fn place_field(
         self,
         fx: &mut FunctionCx<'_, '_, 'tcx>,