Method optimize_array

array.cc:83–295

    }

void ArrayST::optimize_array()
{
	list<uca_org_t > candidate_solutions(0);
	list<uca_org_t >::iterator candidate_iter, min_dynamic_energy_iter;

	uca_org_t * temp_res = 0;
	local_result.valid=false;

	double 	throughput=l_ip.throughput, latency=l_ip.latency;
	double  area_efficiency_threshold = 20.0;
	bool 	throughput_overflow=true, latency_overflow=true;
	int     optimization_end = 20;
	compute_base_power();

	if ((local_result.cycle_time - throughput) <= 1e-10 )
		throughput_overflow=false;
	if ((local_result.access_time - latency)<= 1e-10)
		latency_overflow=false;

	if ((opt_for_clk && opt_local) && ((l_ip.cache_sz>2048 && l_ip.assoc!=0) ||(l_ip.cache_sz>256 && l_ip.assoc==0)))//over opt small array lead to sub-optimal solutions
	{
		if (throughput_overflow || latency_overflow)
		{
			l_ip.ed=0;

			l_ip.delay_wt                = 100;//Fixed number, make sure timing can be satisfied.
			l_ip.cycle_time_wt           = 1000;

			l_ip.area_wt                 = 10;//Fixed number, This is used to exhaustive search for individual components.
			l_ip.dynamic_power_wt        = 10;//Fixed number, This is used to exhaustive search for individual components.
			l_ip.leakage_power_wt        = 10;

			l_ip.delay_dev               = 1000000;//Fixed number, make sure timing can be satisfied.
			l_ip.cycle_time_dev          = 100;

			l_ip.area_dev                = 1000000;//Fixed number, This is used to exhaustive search for individual components.
			l_ip.dynamic_power_dev       = 1000000;//Fixed number, This is used to exhaustive search for individual components.
			l_ip.leakage_power_dev       = 1000000;

			throughput_overflow=true; //Reset overflow flag before start optimization iterations
			latency_overflow=true;

			temp_res = &local_result; //Clean up the result for optimized for ED^2P
			temp_res->cleanup();
		}


		while ((throughput_overflow || latency_overflow)&&l_ip.cycle_time_dev > optimization_end)// l_ip.delay_dev <40 will have over-opt results
		{
			compute_base_power();

			l_ip.cycle_time_dev-=10;//This is the time_dev to be used for next iteration

			//		from best area to worst area -->worst timing to best timing
			if ((((local_result.cycle_time - throughput) <= 1e-10 ) && (local_result.access_time - latency)<= 1e-10)||
					(local_result.data_array2->area_efficiency < area_efficiency_threshold && l_ip.assoc == 0))
			{  //if no satisfiable solution is found,the most aggressive one is left
				candidate_solutions.push_back(local_result);