ARM VERSION 1.2 - page 244

Vector Floating-point Programming

6.2.2 Vectors

A vector can use up to eight single-precision registers, or four double-precision

registers, from the same bank. The number of registers used by a vector is controlled by

the

LEN

bits in the

FPSCR

(see FPSCR, the floating-point status and control register on

page6-10).

A vector can start from any register. The first register used by a vector is specified in the

Vector wrap-around

If the vector extends beyond the end of a bank, it wraps around to the beginning of the

same bank, for example:

•a vector of length 6 starting at s5 is {s5, s6, s7, s0, s1, s2}

•a vector of length 3 starting at s15 is {s15, s8, s9}

•a vector of length 4 starting at s22 is {s22, s23, s16, s17}

•a vector of length 2 starting at d7 is {d7, d4}

•a vector of length 3 starting at d10 is {d10, d11, d8}.

A vector cannot contain registers from more than one bank.

Vector stride

Vectors can occupy consecutive registers, as in the examples above, or they can occupy

alternate registers. This is con trolled by the

STRIDE

bits in the

FPSCR

(see FPSCR, the

floating-point status and control register on page6-10). For example:

•a vector of length 3, stride 2, starting at s1, is {s1, s3, s5}

•a vector of length 4, stride 2, starting at s6, is {s6, s0, s2, s4}

•a vector of length 2, stride 2, starting at d1, is {d1, d3}.

Restriction on vector length

A vector cannot use the same register twice. Allowing for vector wrap-around, this

means that you cannot have:

•a single-precision vector with length > 4 and stride = 2

•a double-precision vector with length > 4 and stride = 1

•a double-precision vector with length > 2 and stride = 2.