6.54 Using Vector Instructions through Built-in Functions

On some targets, the instruction set contains SIMD vector instructions which operate on multiple values contained in one large register at the same time. For example, on the x86 the MMX, 3DNow! and SSE extensions can be used this way.

The first step in using these extensions is to provide the necessary data types. This should be done using an appropriate typedef:

typedef int v4si __attribute__ ((vector_size (16)));

The int type specifies the base type (which can be a typedef), while the attribute specifies the vector size for the variable, measured in bytes. For example, the declaration above causes the compiler to set the mode for the v4si type to be 16 bytes wide and divided into int sized units. For a 32-bit int this means a vector of 4 units of 4 bytes, and the corresponding mode of foo is V4SI.

The vector_size attribute is only applicable to integral and floating scalars, although arrays, pointers, and function return values are allowed in conjunction with this construct. Only sizes that are positive power-of-two multiples of the base type size are currently allowed.

All the basic integer types can be used as base types, both as signed and as unsigned: char, short, int, long, long long. In addition, float and double can be used to build floating-point vector types.

Specifying a combination that is not valid for the current architecture causes GCC to synthesize the instructions using a narrower mode. For example, if you specify a variable of type V4SI and your architecture does not allow for this specific SIMD type, GCC produces code that uses 4 SIs.

The types defined in this manner can be used with a subset of normal C operations. Currently, GCC allows using the following operators on these types: +, -, *, /, unary minus, ^, |, &, ~, %.

The operations behave like C++ valarrays. Addition is defined as the addition of the corresponding elements of the operands. For example, in the code below, each of the 4 elements in a is added to the corresponding 4 elements in b and the resulting vector is stored in c.

typedef int v4si __attribute__ ((vector_size (16)));

v4si a, b, c;

c = a + b;

Subtraction, multiplication, division, and the logical operations operate in a similar manner. Likewise, the result of using the unary minus or complement operators on a vector type is a vector whose elements are the negative or complemented values of the corresponding elements in the operand.

It is possible to use shifting operators <<, >> on integer-type vectors. The operation is defined as following: {a0, a1, …, an} >> {b0, b1, …, bn} == {a0 >> b0, a1 >> b1, …, an >> bn}. Unlike OpenCL, values of b are not implicitly taken modulo bit width of the base type B, and the behavior is undefined if any bi is greater than or equal to B.

In contrast to scalar operations in C and C++, operands of integer vector operations do not undergo integer promotions.

Operands of binary vector operations must have the same number of elements.

For convenience, it is allowed to use a binary vector operation where one operand is a scalar. In that case the compiler transforms the scalar operand into a vector where each element is the scalar from the operation. The transformation happens only if the scalar could be safely converted to the vector-element type. Consider the following code.

typedef int v4si __attribute__ ((vector_size (16)));

v4si a, b, c;
long l;

a = b + 1;    /* a = b + {1,1,1,1}; */
a = 2 * b;    /* a = {2,2,2,2} * b; */

a = l + a;    /* Error, cannot convert long to int. */

Vectors can be subscripted as if the vector were an array with the same number of elements and base type. Out of bound accesses invoke undefined behavior at run time. Warnings for out of bound accesses for vector subscription can be enabled with -Warray-bounds.

Vector comparison is supported with standard comparison operators: ==, !=, <, <=, >, >=. Comparison operands can be vector expressions of integer-type or real-type. Comparison between integer-type vectors and real-type vectors are not supported. The result of the comparison is a vector of the same width and number of elements as the comparison operands with a signed integral element type.

Vectors are compared element-wise producing 0 when comparison is false and -1 (constant of the appropriate type where all bits are set) otherwise. Consider the following example.

typedef int v4si __attribute__ ((vector_size (16)));

v4si a = {1,2,3,4};
v4si b = {3,2,1,4};
v4si c;

c = a >  b;     /* The result would be {0, 0,-1, 0}  */
c = a == b;     /* The result would be {0,-1, 0,-1}  */

In C++, the ternary operator ?: is available. a?b:c, where b and c are vectors of the same type and a is an integer vector with the same number of elements of the same size as b and c, computes all three arguments and creates a vector {a[0]?b[0]:c[0], a[1]?b[1]:c[1], …}. Note that unlike in OpenCL, a is thus interpreted as a != 0 and not a < 0. As in the case of binary operations, this syntax is also accepted when one of b or c is a scalar that is then transformed into a vector. If both b and c are scalars and the type of true?b:c has the same size as the element type of a, then b and c are converted to a vector type whose elements have this type and with the same number of elements as a.

In C++, the logic operators !, &&, || are available for vectors. !v is equivalent to v == 0, a && b is equivalent to a!=0 & b!=0 and a || b is equivalent to a!=0 | b!=0. For mixed operations between a scalar s and a vector v, s && v is equivalent to s?v!=0:0 (the evaluation is short-circuit) and v && s is equivalent to v!=0 & (s?-1:0).

Vector shuffling is available using functions __builtin_shuffle (vec, mask) and __builtin_shuffle (vec0, vec1, mask). Both functions construct a permutation of elements from one or two vectors and return a vector of the same type as the input vector(s). The mask is an integral vector with the same width (W) and element count (N) as the output vector.

The elements of the input vectors are numbered in memory ordering of vec0 beginning at 0 and vec1 beginning at N. The elements of mask are considered modulo N in the single-operand case and modulo 2*N in the two-operand case.

Consider the following example,

typedef int v4si __attribute__ ((vector_size (16)));

v4si a = {1,2,3,4};
v4si b = {5,6,7,8};
v4si mask1 = {0,1,1,3};
v4si mask2 = {0,4,2,5};
v4si res;

res = __builtin_shuffle (a, mask1);       /* res is {1,2,2,4}  */
res = __builtin_shuffle (a, b, mask2);    /* res is {1,5,3,6}  */

Note that __builtin_shuffle is intentionally semantically compatible with the OpenCL shuffle and shuffle2 functions.

You can declare variables and use them in function calls and returns, as well as in assignments and some casts. You can specify a vector type as a return type for a function. Vector types can also be used as function arguments. It is possible to cast from one vector type to another, provided they are of the same size (in fact, you can also cast vectors to and from other datatypes of the same size).

You cannot operate between vectors of different lengths or different signedness without a cast.

Vector shuffling is available using the __builtin_shufflevector (vec1, vec2, index...) function. vec1 and vec2 must be expressions with vector type with a compatible element type. The result of __builtin_shufflevector is a vector with the same element type as vec1 and vec2 but that has an element count equal to the number of indices specified.

The index arguments are a list of integers that specify the elements indices of the first two vectors that should be extracted and returned in a new vector. These element indices are numbered sequentially starting with the first vector, continuing into the second vector. An index of -1 can be used to indicate that the corresponding element in the returned vector is a don’t care and can be freely chosen to optimized the generated code sequence performing the shuffle operation.

Consider the following example,

typedef int v4si __attribute__ ((vector_size (16)));
typedef int v8si __attribute__ ((vector_size (32)));

v8si a = {1,-2,3,-4,5,-6,7,-8};
v4si b = __builtin_shufflevector (a, a, 0, 2, 4, 6); /* b is {1,3,5,7} */
v4si c = {-2,-4,-6,-8};
v8si d = __builtin_shufflevector (c, b, 4, 0, 5, 1, 6, 2, 7, 3); /* d is a */

Vector conversion is available using the __builtin_convertvector (vec, vectype) function. vec must be an expression with integral or floating vector type and vectype an integral or floating vector type with the same number of elements. The result has vectype type and value of a C cast of every element of vec to the element type of vectype.

Consider the following example,

typedef int v4si __attribute__ ((vector_size (16)));
typedef float v4sf __attribute__ ((vector_size (16)));
typedef double v4df __attribute__ ((vector_size (32)));
typedef unsigned long long v4di __attribute__ ((vector_size (32)));

v4si a = {1,-2,3,-4};
v4sf b = {1.5f,-2.5f,3.f,7.f};
v4di c = {1ULL,5ULL,0ULL,10ULL};
v4sf d = __builtin_convertvector (a, v4sf); /* d is {1.f,-2.f,3.f,-4.f} */
/* Equivalent of:
   v4sf d = { (float)a[0], (float)a[1], (float)a[2], (float)a[3] }; */
v4df e = __builtin_convertvector (a, v4df); /* e is {1.,-2.,3.,-4.} */
v4df f = __builtin_convertvector (b, v4df); /* f is {1.5,-2.5,3.,7.} */
v4si g = __builtin_convertvector (f, v4si); /* g is {1,-2,3,7} */
v4si h = __builtin_convertvector (c, v4si); /* h is {1,5,0,10} */

Sometimes it is desirable to write code using a mix of generic vector operations (for clarity) and machine-specific vector intrinsics (to access vector instructions that are not exposed via generic built-ins). On x86, intrinsic functions for integer vectors typically use the same vector type __m128i irrespective of how they interpret the vector, making it necessary to cast their arguments and return values from/to other vector types. In C, you can make use of a union type:

#include <immintrin.h>

typedef unsigned char u8x16 __attribute__ ((vector_size (16)));
typedef unsigned int  u32x4 __attribute__ ((vector_size (16)));

typedef union {
        __m128i mm;
        u8x16   u8;
        u32x4   u32;
} v128;

for variables that can be used with both built-in operators and x86 intrinsics:

v128 x, y = { 0 };
memcpy (&x, ptr, sizeof x);
y.u8  += 0x80;
x.mm  = _mm_adds_epu8 (x.mm, y.mm);
x.u32 &= 0xffffff;

/* Instead of a variable, a compound literal may be used to pass the
   return value of an intrinsic call to a function expecting the union: */
v128 foo (v128);
x = foo ((v128) {_mm_adds_epu8 (x.mm, y.mm)});