Intel ARCHITECTURE IA-32 - Manual (Page 206)

IA-32 Intel® Architecture Optimization
3-26
In C++ (but not in C) it is also possible to force the alignment of a
class / struct / union type, as in the code that follows:
struct __declspec(align(16)) my_m128
{
float f[4];
};
But, if the data in such a class is going to be used with the Streaming
SIMD Extensions or Streaming SIMD Extensions 2, it is preferable to
use a
union to make this explicit. In C++, an anonymous union can be
used to make this more convenient:
class my_m128 {
union {
__m128 m;
float f[4];
};
};
In this example, because the union is anonymous, the names, m and f ,
can be used as immediate member names of my__m128 . Note that
__declspec(align) has no effect when applied to a class , struct , or
union member in either C or C++.
Alignment by Using __m64 or double Data. In some cases, for better
performance, the compiler will align routines with __m64 or double
data to 16-bytes by default. The command-line switch, -Qsfalign16 ,
can be used to limit the compiler to only align in routines that contain
128-bit data. The default behavior is to use
-Qsfalign8 , which instructs
to align routines with 8- or 16-byte data types to 16-bytes.
For more details, see relevant Intel application notes in the Intel
Architecture Performance Training Center provided with the SDK and
the Intel® C++ Compiler User’s Guide.