5 files changed, 26 insertions, 13 deletions
diff --git a/compiler/rustc_codegen_ssa/src/meth.rs b/compiler/rustc_codegen_ssa/src/meth.rs
index 63245a94c8e..b392b2c4ab8 100644
--- a/compiler/rustc_codegen_ssa/src/meth.rs
+++ b/compiler/rustc_codegen_ssa/src/meth.rs
@@ -20,10 +20,11 @@ impl<'a, 'tcx> VirtualIndex {
         // Load the data pointer from the object.
         debug!("get_fn({:?}, {:?})", llvtable, self);
 
-        let llvtable = bx.pointercast(llvtable, bx.type_ptr_to(bx.fn_ptr_backend_type(fn_abi)));
+        let llty = bx.fn_ptr_backend_type(fn_abi);
+        let llvtable = bx.pointercast(llvtable, bx.type_ptr_to(llty));
         let ptr_align = bx.tcx().data_layout.pointer_align.abi;
         let gep = bx.inbounds_gep(llvtable, &[bx.const_usize(self.0)]);
-        let ptr = bx.load(gep, ptr_align);
+        let ptr = bx.load(llty, gep, ptr_align);
         bx.nonnull_metadata(ptr);
         // Vtable loads are invariant.
         bx.set_invariant_load(ptr);
@@ -38,10 +39,11 @@ impl<'a, 'tcx> VirtualIndex {
         // Load the data pointer from the object.
         debug!("get_int({:?}, {:?})", llvtable, self);
 
-        let llvtable = bx.pointercast(llvtable, bx.type_ptr_to(bx.type_isize()));
+        let llty = bx.type_isize();
+        let llvtable = bx.pointercast(llvtable, bx.type_ptr_to(llty));
         let usize_align = bx.tcx().data_layout.pointer_align.abi;
         let gep = bx.inbounds_gep(llvtable, &[bx.const_usize(self.0)]);
-        let ptr = bx.load(gep, usize_align);
+        let ptr = bx.load(llty, gep, usize_align);
         // Vtable loads are invariant.
         bx.set_invariant_load(ptr);
         ptr
diff --git a/compiler/rustc_codegen_ssa/src/mir/block.rs b/compiler/rustc_codegen_ssa/src/mir/block.rs
index 2cb28d7361c..b584801a62d 100644
--- a/compiler/rustc_codegen_ssa/src/mir/block.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/block.rs
@@ -260,7 +260,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             PassMode::Direct(_) | PassMode::Pair(..) => {
                 let op = self.codegen_consume(&mut bx, mir::Place::return_place().as_ref());
                 if let Ref(llval, _, align) = op.val {
-                    bx.load(llval, align)
+                    bx.load(bx.backend_type(op.layout), llval, align)
                 } else {
                     op.immediate_or_packed_pair(&mut bx)
                 }
@@ -287,8 +287,9 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                         llval
                     }
                 };
-                let addr = bx.pointercast(llslot, bx.type_ptr_to(bx.cast_backend_type(&cast_ty)));
-                bx.load(addr, self.fn_abi.ret.layout.align.abi)
+                let ty = bx.cast_backend_type(&cast_ty);
+                let addr = bx.pointercast(llslot, bx.type_ptr_to(ty));
+                bx.load(ty, addr, self.fn_abi.ret.layout.align.abi)
             }
         };
         bx.ret(llval);
@@ -1086,15 +1087,16 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         if by_ref && !arg.is_indirect() {
             // Have to load the argument, maybe while casting it.
             if let PassMode::Cast(ty) = arg.mode {
-                let addr = bx.pointercast(llval, bx.type_ptr_to(bx.cast_backend_type(&ty)));
-                llval = bx.load(addr, align.min(arg.layout.align.abi));
+                let llty = bx.cast_backend_type(&ty);
+                let addr = bx.pointercast(llval, bx.type_ptr_to(llty));
+                llval = bx.load(llty, addr, align.min(arg.layout.align.abi));
             } else {
                 // We can't use `PlaceRef::load` here because the argument
                 // may have a type we don't treat as immediate, but the ABI
                 // used for this call is passing it by-value. In that case,
                 // the load would just produce `OperandValue::Ref` instead
                 // of the `OperandValue::Immediate` we need for the call.
-                llval = bx.load(llval, align);
+                llval = bx.load(bx.backend_type(arg.layout), llval, align);
                 if let abi::Abi::Scalar(ref scalar) = arg.layout.abi {
                     if scalar.is_bool() {
                         bx.range_metadata(llval, 0..2);
diff --git a/compiler/rustc_codegen_ssa/src/mir/operand.rs b/compiler/rustc_codegen_ssa/src/mir/operand.rs
index 25e84c38ed3..3481b36bcc0 100644
--- a/compiler/rustc_codegen_ssa/src/mir/operand.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/operand.rs
@@ -289,6 +289,15 @@ impl<'a, 'tcx, V: CodegenObject> OperandValue<V> {
         }
         match self {
             OperandValue::Ref(r, None, source_align) => {
+                if flags.contains(MemFlags::NONTEMPORAL) {
+                    // HACK(nox): This is inefficient but there is no nontemporal memcpy.
+                    // FIXME: Don't access pointer element type.
+                    let ty = bx.element_type(bx.val_ty(r));
+                    let val = bx.load(ty, r, source_align);
+                    let ptr = bx.pointercast(dest.llval, bx.type_ptr_to(ty));
+                    bx.store_with_flags(val, ptr, dest.align, flags);
+                    return;
+                }
                 base::memcpy_ty(bx, dest.llval, dest.align, r, source_align, dest.layout, flags)
             }
             OperandValue::Ref(_, Some(_), _) => {
diff --git a/compiler/rustc_codegen_ssa/src/mir/place.rs b/compiler/rustc_codegen_ssa/src/mir/place.rs
index a9e7ebf6d43..9906cf59e9a 100644
--- a/compiler/rustc_codegen_ssa/src/mir/place.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/place.rs
@@ -407,7 +407,7 @@ impl<'a, 'tcx, V: CodegenObject> PlaceRef<'tcx, V> {
         let layout = bx.layout_of(target_ty.ty);
 
         PlaceRef {
-            llval: bx.load(self.llval, self.align),
+            llval: bx.load(bx.backend_type(layout), self.llval, self.align),
             llextra: None,
             layout,
             align: layout.align.abi,
diff --git a/compiler/rustc_codegen_ssa/src/traits/builder.rs b/compiler/rustc_codegen_ssa/src/traits/builder.rs
index c5d7c4f86c5..f0c232a97bc 100644
--- a/compiler/rustc_codegen_ssa/src/traits/builder.rs
+++ b/compiler/rustc_codegen_ssa/src/traits/builder.rs
@@ -137,8 +137,8 @@ pub trait BuilderMethods<'a, 'tcx>:
     fn dynamic_alloca(&mut self, ty: Self::Type, align: Align) -> Self::Value;
     fn array_alloca(&mut self, ty: Self::Type, len: Self::Value, align: Align) -> Self::Value;
 
-    fn load(&mut self, ptr: Self::Value, align: Align) -> Self::Value;
-    fn volatile_load(&mut self, ptr: Self::Value) -> Self::Value;
+    fn load(&mut self, ty: Self::Type, ptr: Self::Value, align: Align) -> Self::Value;
+    fn volatile_load(&mut self, ty: Self::Type, ptr: Self::Value) -> Self::Value;
     fn atomic_load(
         &mut self,
         ty: Self::Type,