auto merge of #6724 : thestinger/rust/swap_fast, r=thestinger

Passing higher alignment values gives the optimization passes more freedom since it can copy in larger chunks. This change results in rustc outputting the same post-optimization IR as clang for swaps and most copies excluding the lack of information about padding.

Code snippet:

```rust
#[inline(never)]
fn swap<T>(x: &mut T, y: &mut T) {
    util::swap(x, y);
}
```

Original IR (for `int`):

```llvm
define internal fastcc void @_ZN9swap_283417_a71830ca3ed2d65d3_00E(i64*, i64*) #1 {
static_allocas:
  %2 = icmp eq i64* %0, %1
  br i1 %2, label %_ZN4util9swap_283717_a71830ca3ed2d65d3_00E.exit, label %3

; <label>:3                                       ; preds = %static_allocas
  %4 = load i64* %0, align 1
  %5 = load i64* %1, align 1
  store i64 %5, i64* %0, align 1
  store i64 %4, i64* %1, align 1
  br label %_ZN4util9swap_283717_a71830ca3ed2d65d3_00E.exit

_ZN4util9swap_283717_a71830ca3ed2d65d3_00E.exit:  ; preds = %3, %static_allocas
  ret void
}
```

After #6710:

```llvm
define internal fastcc void @_ZN9swap_283017_a71830ca3ed2d65d3_00E(i64* nocapture, i64* nocapture) #1 {
static_allocas:
  %2 = load i64* %0, align 1
  %3 = load i64* %1, align 1
  store i64 %3, i64* %0, align 1
  store i64 %2, i64* %1, align 1
  ret void
}
```

After this change:

```llvm
define internal fastcc void @_ZN9swap_283017_a71830ca3ed2d65d3_00E(i64* nocapture, i64* nocapture) #1 {
static_allocas:
  %2 = load i64* %0, align 8
  %3 = load i64* %1, align 8
  store i64 %3, i64* %0, align 8
  store i64 %2, i64* %1, align 8
  ret void
}
```

Another example:

```rust
#[inline(never)]
fn set<T>(x: &mut T, y: T) {
    *x = y;
}
```

Before, with `(int, int)` (align 1):

```llvm
define internal fastcc void @_ZN8set_282517_8fa972e3f9e451983_00E({ i64, i64 }* nocapture, { i64, i64 }* nocapture) #1 {
static_allocas:
  %2 = bitcast { i64, i64 }* %1 to i8*
  %3 = bitcast { i64, i64 }* %0 to i8*
  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 1, i1 false)
  ret void
}
```

After, with `(int, int)` (align 8):

```llvm
define internal fastcc void @_ZN8set_282617_8fa972e3f9e451983_00E({ i64, i64 }* nocapture, { i64, i64 }* nocapture) #1 {
static_allocas:
  %2 = bitcast { i64, i64 }* %1 to i8*
  %3 = bitcast { i64, i64 }* %0 to i8*
  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
  ret void
}
```

1 files changed, 23 insertions, 1 deletions

diff --git a/src/libstd/cast.rs b/src/libstd/cast.rs
index cde22afd34a..30ad41f0ca2 100644
--- a/src/libstd/cast.rs
+++ b/src/libstd/cast.rs
@@ -14,8 +14,9 @@ use sys;
 use unstable::intrinsics;
 
 /// Casts the value at `src` to U. The two types must have the same length.
+#[cfg(stage0)]
 pub unsafe fn transmute_copy<T, U>(src: &T) -> U {
-    let mut dest: U = intrinsics::init();
+    let mut dest: U = intrinsics::uninit();
     {
         let dest_ptr: *mut u8 = transmute(&mut dest);
         let src_ptr: *u8 = transmute(src);
@@ -26,6 +27,26 @@ pub unsafe fn transmute_copy<T, U>(src: &T) -> U {
     dest
 }
 
+#[cfg(target_word_size = "32", not(stage0))]
+#[inline(always)]
+pub unsafe fn transmute_copy<T, U>(src: &T) -> U {
+    let mut dest: U = intrinsics::uninit();
+    let dest_ptr: *mut u8 = transmute(&mut dest);
+    let src_ptr: *u8 = transmute(src);
+    intrinsics::memcpy32(dest_ptr, src_ptr, sys::size_of::<U>() as u32);
+    dest
+}
+
+#[cfg(target_word_size = "64", not(stage0))]
+#[inline(always)]
+pub unsafe fn transmute_copy<T, U>(src: &T) -> U {
+    let mut dest: U = intrinsics::uninit();
+    let dest_ptr: *mut u8 = transmute(&mut dest);
+    let src_ptr: *u8 = transmute(src);
+    intrinsics::memcpy64(dest_ptr, src_ptr, sys::size_of::<U>() as u64);
+    dest
+}
+
 /**
  * Move a thing into the void
  *
@@ -43,6 +64,7 @@ pub unsafe fn forget<T>(thing: T) { intrinsics::forget(thing); }
  * and/or reinterpret_cast when such calls would otherwise scramble a box's
  * reference count
  */
+#[inline(always)]
 pub unsafe fn bump_box_refcount<T>(t: @T) { forget(t); }
 
 /**