| Markus Weinhardt , "Pipeline Synthesis and Optimization for Reconfigurable Custom Computing Machines", Universitaet Karlsruhe, Fakultaet fuer Informatik D-76128 Karlsruhe, Germany January 3, 1997. |
....3. Pipelined Version of Average Filter. The problem of determining where to insert the registers can be mapped into an Integer Linear Programming problem. The magnitudes considered are area and clock rate. One magnitude is used as objective function, and the other as a constraint as described in [Wein97] and [Shen96] The datapath in Figure 3 can be further improved. For this particular example the compiler can recognize commutative and associative operator and abstract n ary operators like the one in Figure 4. Figure 4. Three Input Adder from Queue. The original C code contains two binary ....
Markus Weinhardt , "Pipeline Synthesis and Optimization for Reconfigurable Custom Computing Machines", Universitaet Karlsruhe, Fakultaet fuer Informatik D-76128 Karlsruhe, Germany January 3, 1997.
....from an input register to an output. And register insertion is forbidden in feedback cycles because this would change the circuit s behaviour. We developed a method based on integer linear programming which inserts the minimal number of FPGA flipflops necessary to achieve a given cycle time [4]. The cycle time is determined by the number of ports (see section 4) and by the delay in the feedback cycles. Figure 4 shows the example DG with two pipeline stages. Additionally, a chain of flipflops is inserted to control the feedback register and the delayed output values. DIV 4 SCALAR IN X ....
M. Weinhardt. Pipeline Synthesis and Optimization for Reconfigurable Custom Computing Machines. Interner Bericht 1/97, Universitat Karlsruhe, Fakultat fur Informatik, 1997.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC