---

yaml
---
r: 229185
b: refs/heads/try
c: 21be094
h: refs/heads/master
i:
  229183: 80eba7c
v: v3

Author: pnkfelix

[refs]

@@ -1,7 +1,7 @@
 ---
 refs/heads/master: aca2057ed5fb7af3f8905b2bc01f72fa001c35c8
 refs/heads/snap-stage3: 1af31d4974e33027a68126fa5a5a3c2c6491824f
-refs/heads/try: 26f4ebe7a0b166cbfb04f6e8df16cb799f81058b
+refs/heads/try: 21be09448b21c25570f28ddd597f3d30ccc6fabf
 refs/tags/release-0.1: 1f5c5126e96c79d22cb7862f75304136e204f105
 refs/tags/release-0.2: c870d2dffb391e14efb05aa27898f1f6333a9596
 refs/tags/release-0.3: b5f0d0f648d9a6153664837026ba1be43d3e2503

branches/try/src/librustc_trans/trans/glue.rs

@@ -475,21 +475,13 @@ pub fn size_and_align_of_dst<'blk, 'tcx>(bcx: Block<'blk, 'tcx>, t: Ty<'tcx>, in
             //
             //   `size + ((size & (align-1)) ? align : 0)`
             //
-            // Currently I am emulating the above via:
+            // emulated via the semi-standard fast bit trick:
             //
-            //   `size + ((size & (align-1)) * align-(size & (align-1)))`
-            //
-            // because I am not sure which is cheaper between a branch
-            // or a multiply.
-
-            let mask = Sub(bcx, align, C_uint(bcx.ccx(), 1_u64), dbloc);
-            let lowbits = And(bcx, size, mask, DebugLoc::None);
-            let nonzero = ICmp(bcx, llvm::IntNE, lowbits, C_uint(bcx.ccx(), 0_u64), dbloc);
-            let add_size = Mul(bcx,
-                               ZExt(bcx, nonzero, Type::i64(bcx.ccx())),
-                               Sub(bcx, align, lowbits, dbloc),
-                               dbloc);
-            let size = Add(bcx, size, add_size, dbloc);
+            //   `(size + (align-1)) & !align`
+
+            let addend = Sub(bcx, align, C_uint(bcx.ccx(), 1_u64), dbloc);
+            let size = And(
+                bcx, Add(bcx, size, addend, dbloc), Neg(bcx, align, dbloc), dbloc);
 
             (size, align)
         }