make the memcpy/memmove intrinsics higher-level

This allows them to make use of the type's alignment, instead of being
pessimistic and assuming it is only 1.

2 files changed, 58 insertions, 23 deletions

diff --git a/src/libstd/ptr.rs b/src/libstd/ptr.rs
index d1c0ffe7953..97a7d0e11a6 100644
--- a/src/libstd/ptr.rs
+++ b/src/libstd/ptr.rs
@@ -27,12 +27,6 @@ pub mod libc_ {
     #[abi = "cdecl"]
     pub extern {
         #[rust_stack]
-        unsafe fn memmove(dest: *mut c_void,
-                          src: *const c_void,
-                          n: libc::size_t)
-                       -> *c_void;
-
-        #[rust_stack]
         unsafe fn memset(dest: *mut c_void,
                          c: libc::c_int,
                          len: libc::size_t)
@@ -97,15 +91,28 @@ pub fn is_not_null<T>(ptr: *const T) -> bool { !is_null(ptr) }
  * and destination may overlap.
  */
 #[inline(always)]
-#[cfg(target_word_size = "32")]
+#[cfg(target_word_size = "32", stage0)]
 pub unsafe fn copy_memory<T>(dst: *mut T, src: *const T, count: uint) {
     use unstable::intrinsics::memmove32;
     let n = count * sys::size_of::<T>();
     memmove32(dst as *mut u8, src as *u8, n as u32);
 }
 
+/**
+ * Copies data from one location to another
+ *
+ * Copies `count` elements (not bytes) from `src` to `dst`. The source
+ * and destination may overlap.
+ */
+#[inline(always)]
+#[cfg(target_word_size = "32", not(stage0))]
+pub unsafe fn copy_memory<T>(dst: *mut T, src: *const T, count: uint) {
+    use unstable::intrinsics::memmove32;
+    memmove32(dst, src, count as u32);
+}
+
 #[inline(always)]
-#[cfg(target_word_size = "64")]
+#[cfg(target_word_size = "64", stage0)]
 pub unsafe fn copy_memory<T>(dst: *mut T, src: *const T, count: uint) {
     use unstable::intrinsics::memmove64;
     let n = count * sys::size_of::<T>();
@@ -113,25 +120,40 @@ pub unsafe fn copy_memory<T>(dst: *mut T, src: *const T, count: uint) {
 }
 
 #[inline(always)]
-#[cfg(target_word_size = "32")]
+#[cfg(target_word_size = "64", not(stage0))]
+pub unsafe fn copy_memory<T>(dst: *mut T, src: *const T, count: uint) {
+    use unstable::intrinsics::memmove64;
+    memmove64(dst, src as *T, count as u64);
+}
+
+#[inline(always)]
+#[cfg(target_word_size = "32", stage0)]
+pub unsafe fn copy_nonoverlapping_memory<T>(dst: *mut T, src: *const T, count: uint) {
+    use unstable::intrinsics::memmove32;
+    let n = count * sys::size_of::<T>();
+    memmove32(dst as *mut u8, src as *u8, n as u32);
+}
+
+#[inline(always)]
+#[cfg(target_word_size = "32", not(stage0))]
 pub unsafe fn copy_nonoverlapping_memory<T>(dst: *mut T, src: *const T, count: uint) {
-    #[cfg(stage0)]
-    use memcpy32 = unstable::intrinsics::memmove32;
-    #[cfg(not(stage0))]
     use unstable::intrinsics::memcpy32;
+    memcpy32(dst, src as *T, count as u32);
+}
+
+#[inline(always)]
+#[cfg(target_word_size = "64", stage0)]
+pub unsafe fn copy_nonoverlapping_memory<T>(dst: *mut T, src: *const T, count: uint) {
+    use unstable::intrinsics::memmove64;
     let n = count * sys::size_of::<T>();
-    memcpy32(dst as *mut u8, src as *u8, n as u32);
+    memmove64(dst as *mut u8, src as *u8, n as u64);
 }
 
 #[inline(always)]
-#[cfg(target_word_size = "64")]
+#[cfg(target_word_size = "64", not(stage0))]
 pub unsafe fn copy_nonoverlapping_memory<T>(dst: *mut T, src: *const T, count: uint) {
-    #[cfg(stage0)]
-    use memcpy64 = unstable::intrinsics::memmove64;
-    #[cfg(not(stage0))]
     use unstable::intrinsics::memcpy64;
-    let n = count * sys::size_of::<T>();
-    memcpy64(dst as *mut u8, src as *u8, n as u64);
+    memcpy64(dst, src as *T, count as u64);
 }
 
 #[inline(always)]
diff --git a/src/libstd/unstable/intrinsics.rs b/src/libstd/unstable/intrinsics.rs
index 521708621fc..c0975d6a526 100644
--- a/src/libstd/unstable/intrinsics.rs
+++ b/src/libstd/unstable/intrinsics.rs
@@ -127,18 +127,31 @@ pub extern "rust-intrinsic" {
     /// Get the address of the `__morestack` stack growth function.
     pub fn morestack_addr() -> *();
 
-    /// Equivalent to the `llvm.memcpy.p0i8.0i8.i32` intrinsic.
+    /// Equivalent to the `llvm.memcpy.p0i8.0i8.i32` intrinsic, with a size of
+    /// `count` * `size_of::<T>()` and an alignment of `min_align_of::<T>()`
     #[cfg(not(stage0))]
-    pub fn memcpy32(dst: *mut u8, src: *u8, size: u32);
-    /// Equivalent to the `llvm.memcpy.p0i8.0i8.i64` intrinsic.
+    pub fn memcpy32<T>(dst: *mut T, src: *T, count: u32);
+    /// Equivalent to the `llvm.memcpy.p0i8.0i8.i64` intrinsic, with a size of
+    /// `count` * `size_of::<T>()` and an alignment of `min_align_of::<T>()`
     #[cfg(not(stage0))]
-    pub fn memcpy64(dst: *mut u8, src: *u8, size: u64);
+    pub fn memcpy64<T>(dst: *mut T, src: *T, count: u64);
 
     /// Equivalent to the `llvm.memmove.p0i8.0i8.i32` intrinsic.
+    #[cfg(stage0)]
     pub fn memmove32(dst: *mut u8, src: *u8, size: u32);
     /// Equivalent to the `llvm.memmove.p0i8.0i8.i64` intrinsic.
+    #[cfg(stage0)]
     pub fn memmove64(dst: *mut u8, src: *u8, size: u64);
 
+    /// Equivalent to the `llvm.memmove.p0i8.0i8.i32` intrinsic, with a size of
+    /// `count` * `size_of::<T>()` and an alignment of `min_align_of::<T>()`
+    #[cfg(not(stage0))]
+    pub fn memmove32<T>(dst: *mut T, src: *T, count: u32);
+    /// Equivalent to the `llvm.memmove.p0i8.0i8.i64` intrinsic, with a size of
+    /// `count` * `size_of::<T>()` and an alignment of `min_align_of::<T>()`
+    #[cfg(not(stage0))]
+    pub fn memmove64<T>(dst: *mut T, src: *T, count: u64);
+
     pub fn sqrtf32(x: f32) -> f32;
     pub fn sqrtf64(x: f64) -> f64;