// ScalarizeMaskedLoad() translates masked load intrinsic, like
// <16 x i32 > @llvm.masked.load( <16 x i32>* %addr, i32 align,
// <16 x i1> %mask, <16 x i32> %passthru)
-// to a chain of basic blocks, whith loading element one-by-one if
+// to a chain of basic blocks, with loading element one-by-one if
// the appropriate mask bit is set
//
// %1 = bitcast i8* %addr to i32*
/// \brief Generate a constant vector with \p Val with the same
/// number of elements as the transition.
/// \p UseSplat defines whether or not \p Val should be replicated
- /// accross the whole vector.
+ /// across the whole vector.
/// In other words, if UseSplat == true, we generate <Val, Val, ..., Val>,
/// otherwise we generate a vector with as many undef as possible:
/// <undef, ..., undef, Val, undef, ..., undef> where \p Val is only