gccrs: Add move_val_init intrinsic

This implements it as a builtin memcpy using the generic param T for the
size hint.

Fixes #1902

gcc/rust/ChangeLog:

	* backend/rust-compile-intrinsic.cc (move_val_init_handler): new intrinsice
	(uninit_handler): use a builtin memcpy

gcc/testsuite/ChangeLog:

	* rust/compile/issue-1981.rs: New test.

Signed-off-by: Philip Herron <herron.philip@googlemail.com>

2 files changed, 187 insertions, 1 deletions

diff --git a/gcc/rust/backend/rust-compile-intrinsic.cc b/gcc/rust/backend/rust-compile-intrinsic.cc
index 85f6e1c..b9951a0 100644
--- a/gcc/rust/backend/rust-compile-intrinsic.cc
+++ b/gcc/rust/backend/rust-compile-intrinsic.cc
@@ -83,6 +83,8 @@ static tree
 op_with_overflow_inner (Context *ctx, TyTy::FnType *fntype, tree_code op);
 static tree
 uninit_handler (Context *ctx, TyTy::FnType *fntype);
+static tree
+move_val_init_handler (Context *ctx, TyTy::FnType *fntype);
 
 enum class Prefetch
 {
@@ -205,6 +207,7 @@ static const std::map<std::string,
     {"unchecked_shl", unchecked_op_handler (LSHIFT_EXPR)},
     {"unchecked_shr", unchecked_op_handler (RSHIFT_EXPR)},
     {"uninit", uninit_handler},
+    {"move_val_init", move_val_init_handler},
 };
 
 Intrinsics::Intrinsics (Context *ctx) : ctx (ctx) {}
@@ -1001,7 +1004,7 @@ uninit_handler (Context *ctx, TyTy::FnType *fntype)
 
   auto fndecl = compile_intrinsic_function (ctx, fntype);
 
-  // get the template parameter type tree fn size_of<T>();
+  // get the template parameter type tree fn uninit<T>();
   rust_assert (fntype->get_num_substitutions () == 1);
   auto &param_mapping = fntype->get_substs ().at (0);
   const TyTy::ParamType *param_tyty = param_mapping.get_param_ty ();
@@ -1042,5 +1045,56 @@ uninit_handler (Context *ctx, TyTy::FnType *fntype)
   return fndecl;
 }
 
+static tree
+move_val_init_handler (Context *ctx, TyTy::FnType *fntype)
+{
+  rust_assert (fntype->get_params ().size () == 2);
+
+  tree lookup = NULL_TREE;
+  if (check_for_cached_intrinsic (ctx, fntype, &lookup))
+    return lookup;
+
+  auto fndecl = compile_intrinsic_function (ctx, fntype);
+
+  // get the template parameter type tree fn size_of<T>();
+  rust_assert (fntype->get_num_substitutions () == 1);
+  auto &param_mapping = fntype->get_substs ().at (0);
+  const TyTy::ParamType *param_tyty = param_mapping.get_param_ty ();
+  TyTy::BaseType *resolved_tyty = param_tyty->resolve ();
+  tree template_parameter_type
+    = TyTyResolveCompile::compile (ctx, resolved_tyty);
+
+  std::vector<Bvariable *> param_vars;
+  compile_fn_params (ctx, fntype, fndecl, &param_vars);
+
+  if (!ctx->get_backend ()->function_set_parameters (fndecl, param_vars))
+    return error_mark_node;
+
+  enter_intrinsic_block (ctx, fndecl);
+
+  // BUILTIN size_of FN BODY BEGIN
+
+  tree dst = ctx->get_backend ()->var_expression (param_vars[0], Location ());
+  tree src = ctx->get_backend ()->var_expression (param_vars[1], Location ());
+  tree size = TYPE_SIZE_UNIT (template_parameter_type);
+
+  tree memcpy_builtin = error_mark_node;
+  BuiltinsContext::get ().lookup_simple_builtin ("memcpy", &memcpy_builtin);
+  rust_assert (memcpy_builtin != error_mark_node);
+
+  src = build_fold_addr_expr_loc (BUILTINS_LOCATION, src);
+  tree memset_call = build_call_expr_loc (BUILTINS_LOCATION, memcpy_builtin, 3,
+					  dst, src, size);
+  TREE_READONLY (memset_call) = 0;
+  TREE_SIDE_EFFECTS (memset_call) = 1;
+
+  ctx->add_statement (memset_call);
+  // BUILTIN size_of FN BODY END
+
+  finalize_intrinsic_block (ctx, fndecl);
+
+  return fndecl;
+}
+
 } // namespace Compile
 } // namespace Rust
diff --git a/gcc/testsuite/rust/compile/issue-1981.rs b/gcc/testsuite/rust/compile/issue-1981.rs
new file mode 100644
index 0000000..e3f1723
--- /dev/null
+++ b/gcc/testsuite/rust/compile/issue-1981.rs
@@ -0,0 +1,132 @@
+mod intrinsics {
+    extern "rust-intrinsic" {
+        pub fn offset<T>(ptr: *const T, count: isize) -> *const T;
+        pub fn copy_nonoverlapping<T>(src: *const T, dst: *mut T, count: usize);
+        pub fn move_val_init<T>(dst: *mut T, src: T);
+        pub fn uninit<T>() -> T;
+    }
+}
+
+mod ptr {
+    #[lang = "const_ptr"]
+    impl<T> *const T {
+        pub unsafe fn offset(self, count: isize) -> *const T {
+            intrinsics::offset(self, count)
+        }
+    }
+
+    #[lang = "mut_ptr"]
+    impl<T> *mut T {
+        pub unsafe fn offset(self, count: isize) -> *mut T {
+            intrinsics::offset(self, count) as *mut T
+        }
+    }
+
+    pub unsafe fn swap_nonoverlapping<T>(x: *mut T, y: *mut T, count: usize) {
+        let x = x as *mut T;
+        let y = y as *mut T;
+        let len = mem::size_of::<T>() * count;
+        swap_nonoverlapping_bytes(x, y, len)
+    }
+
+    pub unsafe fn swap_nonoverlapping_one<T>(x: *mut T, y: *mut T) {
+        // For types smaller than the block optimization below,
+        // just swap directly to avoid pessimizing codegen.
+        if mem::size_of::<T>() < 32 {
+            let z = read(x);
+            intrinsics::copy_nonoverlapping(y, x, 1);
+            write(y, z);
+        } else {
+            swap_nonoverlapping(x, y, 1);
+        }
+    }
+
+    pub unsafe fn write<T>(dst: *mut T, src: T) {
+        intrinsics::move_val_init(&mut *dst, src)
+    }
+
+    pub unsafe fn read<T>(src: *const T) -> T {
+        let mut tmp: T = mem::uninitialized();
+        intrinsics::copy_nonoverlapping(src, &mut tmp, 1);
+        tmp
+    }
+
+    unsafe fn swap_nonoverlapping_bytes(x: *mut u8, y: *mut u8, len: usize) {
+        struct Block(u64, u64, u64, u64);
+        struct UnalignedBlock(u64, u64, u64, u64);
+
+        let block_size = mem::size_of::<Block>();
+
+        // Loop through x & y, copying them `Block` at a time
+        // The optimizer should unroll the loop fully for most types
+        // N.B. We can't use a for loop as the `range` impl calls `mem::swap` recursively
+        let mut i = 0;
+        while i + block_size <= len {
+            // Create some uninitialized memory as scratch space
+            // Declaring `t` here avoids aligning the stack when this loop is unused
+            let mut t: Block = mem::uninitialized();
+            let t = &mut t as *mut _ as *mut u8;
+            let x = x.offset(i as isize);
+            let y = y.offset(i as isize);
+
+            // Swap a block of bytes of x & y, using t as a temporary buffer
+            // This should be optimized into efficient SIMD operations where available
+            intrinsics::copy_nonoverlapping(x, t, block_size);
+            intrinsics::copy_nonoverlapping(y, x, block_size);
+            intrinsics::copy_nonoverlapping(t, y, block_size);
+            i += block_size;
+        }
+
+        if i < len {
+            // Swap any remaining bytes
+            let mut t: UnalignedBlock = mem::uninitialized();
+            let rem = len - i;
+
+            let t = &mut t as *mut _ as *mut u8;
+            let x = x.offset(i as isize);
+            let y = y.offset(i as isize);
+
+            intrinsics::copy_nonoverlapping(x, t, rem);
+            intrinsics::copy_nonoverlapping(y, x, rem);
+            intrinsics::copy_nonoverlapping(t, y, rem);
+        }
+    }
+}
+
+mod mem {
+    extern "rust-intrinsic" {
+        pub fn transmute<T, U>(_: T) -> U;
+        pub fn size_of<T>() -> usize;
+    }
+
+    pub fn swap<T>(x: &mut T, y: &mut T) {
+        unsafe {
+            ptr::swap_nonoverlapping_one(x, y);
+        }
+    }
+
+    pub fn replace<T>(dest: &mut T, mut src: T) -> T {
+        swap(dest, &mut src);
+        src
+    }
+
+    pub unsafe fn uninitialized<T>() -> T {
+        intrinsics::uninit()
+    }
+}
+
+trait Step {
+    fn replace_zero(&mut self) -> Self;
+}
+
+impl Step for i32 {
+    fn replace_zero(&mut self) -> Self {
+        mem::replace(self, 0)
+    }
+}
+
+fn main() -> i32 {
+    let a = 123;
+    a.replace_zero();
+    a
+}