Processor Description Languages

8.7 Custom Data Interfaces 205

instructions.The primary advantage of this implementation is the reduced hardware

cost, and the disadvantage is the fact that the same instruction cannot be issued at

every cycle.

The following example illustrates a vector dot product computation imple-

mented using a single multiplier:

function [31:0] mul16 ([15:0] a, [15:0] b, signed) shared {

assign mul16 = TIEmul(a,b,signed);

}

operation VDOTPROD {out AR acc, in AR m0, in AR m1} {

wire [31:0] prod0 = mul16(m0[15: 0], m1[15: 0], 1’b1);

wire [31:0] prod1 = mul16(m0[31:16], m1[31:16], 1’b1);

assign acc = prod0 + prod1;

}

schedule dotprod {VDOTPROD} { def acc Estage + 2; }