iocontrollers.cc (10152:52c552138ba1) | iocontrollers.cc (10234:5cb711fa6176) |
---|---|
1/***************************************************************************** 2 * McPAT 3 * SOFTWARE LICENSE AGREEMENT 4 * Copyright 2012 Hewlett-Packard Development Company, L.P. | 1/***************************************************************************** 2 * McPAT 3 * SOFTWARE LICENSE AGREEMENT 4 * Copyright 2012 Hewlett-Packard Development Company, L.P. |
5 * Copyright (c) 2010-2013 Advanced Micro Devices, Inc. |
|
5 * All Rights Reserved 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are 9 * met: redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer; 11 * redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the --- 7 unchanged lines hidden (view full) --- 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 22 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 23 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 24 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | 6 * All Rights Reserved 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions are 10 * met: redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer; 12 * redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the --- 7 unchanged lines hidden (view full) --- 21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 22 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 23 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 25 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
28 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.” | 29 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
29 * 30 ***************************************************************************/ 31#include <algorithm> 32#include <cassert> 33#include <cmath> 34#include <iostream> 35#include <string> 36 | 30 * 31 ***************************************************************************/ 32#include <algorithm> 33#include <cassert> 34#include <cmath> 35#include <iostream> 36#include <string> 37 |
37#include "XML_Parse.h" | |
38#include "basic_circuit.h" | 38#include "basic_circuit.h" |
39#include "basic_components.h" | 39#include "common.h" |
40#include "const.h" 41#include "io.h" 42#include "iocontrollers.h" 43#include "logic.h" | 40#include "const.h" 41#include "io.h" 42#include "iocontrollers.h" 43#include "logic.h" |
44#include "parameter.h" | |
45 46/* 47SUN Niagara 2 I/O power analysis: 48total signal bits: 711 49Total FBDIMM bits: (14+10)*2*8= 384 50PCIe bits: (8 + 8)*2 = 32 5110Gb NIC: (4*2+4*2)*2 = 32 52Debug I/Os: 168 --- 11 unchanged lines hidden (view full) --- 64 65/* 66 * A bug in Cadence ChipEstimator: After update the clock rate in the clock tab, a user 67 * need to re-select the IP clock (the same clk) and then click Estimate. if not reselect 68 * the new clock rate may not be propogate into the IPs. 69 * 70 */ 71 | 44 45/* 46SUN Niagara 2 I/O power analysis: 47total signal bits: 711 48Total FBDIMM bits: (14+10)*2*8= 384 49PCIe bits: (8 + 8)*2 = 32 5010Gb NIC: (4*2+4*2)*2 = 32 51Debug I/Os: 168 --- 11 unchanged lines hidden (view full) --- 63 64/* 65 * A bug in Cadence ChipEstimator: After update the clock rate in the clock tab, a user 66 * need to re-select the IP clock (the same clk) and then click Estimate. if not reselect 67 * the new clock rate may not be propogate into the IPs. 68 * 69 */ 70 |
72NIUController::NIUController(ParseXML *XML_interface,InputParameter* interface_ip_) 73:XML(XML_interface), 74 interface_ip(*interface_ip_) 75 { 76 local_result = init_interface(&interface_ip); | 71NIUController::NIUController(XMLNode* _xml_data,InputParameter* interface_ip_) 72 : McPATComponent(_xml_data, interface_ip_) { 73 name = "NIU"; 74 set_niu_param(); 75} |
77 | 76 |
78 double frontend_area, phy_area, mac_area, SerDer_area; 79 double frontend_dyn, mac_dyn, SerDer_dyn; 80 double frontend_gates, mac_gates, SerDer_gates; 81 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); 82 double NMOS_sizing, PMOS_sizing; | 77void NIUController::computeArea() { 78 double mac_area; 79 double frontend_area; 80 double SerDer_area; |
83 | 81 |
84 set_niu_param(); | 82 if (niup.type == 0) { //high performance NIU 83 //Area estimation based on average of die photo from Niagara 2 and 84 //Cadence ChipEstimate using 65nm. 85 mac_area = (1.53 + 0.3) / 2 * (interface_ip.F_sz_um / 0.065) * 86 (interface_ip.F_sz_um / 0.065); 87 //Area estimation based on average of die photo from Niagara 2, ISSCC 88 //"An 800mW 10Gb Ethernet Transceiver in 0.13μm CMOS" 89 //and"A 1.2-V-Only 900-mW 10 Gb Ethernet Transceiver and XAUI Interface 90 //With Robust VCO Tuning Technique" Frontend is PCS 91 frontend_area = (9.8 + (6 + 18) * 65 / 130 * 65 / 130) / 3 * 92 (interface_ip.F_sz_um / 0.065) * (interface_ip.F_sz_um / 0.065); 93 //Area estimation based on average of die photo from Niagara 2 and 94 //Cadence ChipEstimate hard IP @65nm. 95 //SerDer is very hard to scale 96 SerDer_area = (1.39 + 0.36) * (interface_ip.F_sz_um / 97 0.065);//* (interface_ip.F_sz_um/0.065); 98 } else { 99 //Low power implementations are mostly from Cadence ChipEstimator; 100 //Ignore the multiple IP effect 101 // ---When there are multiple IP (same kind or not) selected, Cadence 102 //ChipEstimator results are not a simple summation of all IPs. 103 //Ignore this effect 104 mac_area = 0.24 * (interface_ip.F_sz_um / 0.065) * 105 (interface_ip.F_sz_um / 0.065); 106 frontend_area = 0.1 * (interface_ip.F_sz_um / 0.065) * 107 (interface_ip.F_sz_um / 0.065);//Frontend is the PCS layer 108 SerDer_area = 0.35 * (interface_ip.F_sz_um / 0.065) * 109 (interface_ip.F_sz_um/0.065); 110 //Compare 130um implementation in "A 1.2-V-Only 900-mW 10 Gb Ethernet 111 //Transceiver and XAUI Interface With Robust VCO Tuning Technique" 112 //and the ChipEstimator XAUI PHY hard IP, confirm that even PHY can 113 //scale perfectly with the technology 114 } |
85 | 115 |
86 if (niup.type == 0) //high performance NIU 87 { 88 //Area estimation based on average of die photo from Niagara 2 and Cadence ChipEstimate using 65nm. 89 mac_area = (1.53 + 0.3)/2 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 90 //Area estimation based on average of die photo from Niagara 2, ISSCC "An 800mW 10Gb Ethernet Transceiver in 0.13μm CMOS" 91 //and"A 1.2-V-Only 900-mW 10 Gb Ethernet Transceiver and XAUI Interface With Robust VCO Tuning Technique" Frontend is PCS 92 frontend_area = (9.8 + (6 + 18)*65/130*65/130)/3 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 93 //Area estimation based on average of die photo from Niagara 2 and Cadence ChipEstimate hard IP @65nm. 94 //SerDer is very hard to scale 95 SerDer_area = (1.39 + 0.36) * (interface_ip.F_sz_um/0.065);//* (interface_ip.F_sz_um/0.065); 96 phy_area = frontend_area + SerDer_area; 97 //total area 98 area.set_area((mac_area + frontend_area + SerDer_area)*1e6); 99 //Power 100 //Cadence ChipEstimate using 65nm (mac, front_end are all energy. E=P*T = P/F = 1.37/1Ghz = 1.37e-9); 101 mac_dyn = 2.19e-9*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0);//niup.clockRate; //2.19W@1GHz fully active according to Cadence ChipEstimate @65nm 102 //Cadence ChipEstimate using 65nm soft IP; 103 frontend_dyn = 0.27e-9*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0);//niup.clockRate; 104 //according to "A 100mW 9.6Gb/s Transceiver in 90nm CMOS..." ISSCC 2006 105 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 106 SerDer_dyn = 0.01*10*sqrt(interface_ip.F_sz_um/0.09)*g_tp.peri_global.Vdd/1.2*g_tp.peri_global.Vdd/1.2; 107 SerDer_dyn /= niup.clockRate;//covert to energy per clock cycle of whole NIU | 116 //total area 117 output_data.area = (mac_area + frontend_area + SerDer_area) * 1e6; 118 } |
108 | 119 |
109 //Cadence ChipEstimate using 65nm 110 mac_gates = 111700; 111 frontend_gates = 320000; 112 SerDer_gates = 200000; 113 NMOS_sizing = 5*g_tp.min_w_nmos_; 114 PMOS_sizing = 5*g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; | 120void NIUController::computeEnergy() { 121 double mac_dyn; 122 double frontend_dyn; 123 double SerDer_dyn; 124 double frontend_gates; 125 double mac_gates; 126 double SerDer_gates; 127 double NMOS_sizing; 128 double PMOS_sizing; 129 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); |
115 | 130 |
131 if (niup.type == 0) { //high performance NIU 132 //Power 133 //Cadence ChipEstimate using 65nm (mac, front_end are all energy. 134 //E=P*T = P/F = 1.37/1Ghz = 1.37e-9); 135 //2.19W@1GHz fully active according to Cadence ChipEstimate @65nm 136 mac_dyn = 2.19e-9 * g_tp.peri_global.Vdd / 1.1 * g_tp.peri_global.Vdd / 137 1.1 * (interface_ip.F_sz_nm / 65.0);//niup.clockRate; 138 //Cadence ChipEstimate using 65nm soft IP; 139 frontend_dyn = 0.27e-9 * g_tp.peri_global.Vdd / 1.1 * 140 g_tp.peri_global.Vdd / 1.1 * (interface_ip.F_sz_nm / 65.0); 141 //according to "A 100mW 9.6Gb/s Transceiver in 90nm CMOS..." ISSCC 2006 142 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 143 SerDer_dyn = 0.01 * 10 * sqrt(interface_ip.F_sz_um / 0.09) * 144 g_tp.peri_global.Vdd / 1.2 * g_tp.peri_global.Vdd / 1.2; |
|
116 | 145 |
117 } 118 else 119 {//Low power implementations are mostly from Cadence ChipEstimator; Ignore the multiple IP effect 120 // ---When there are multiple IP (same kind or not) selected, Cadence ChipEstimator results are not 121 // a simple summation of all IPs. Ignore this effect 122 mac_area = 0.24 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 123 frontend_area = 0.1 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065);//Frontend is the PCS layer 124 SerDer_area = 0.35 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 125 //Compare 130um implementation in "A 1.2-V-Only 900-mW 10 Gb Ethernet Transceiver and XAUI Interface With Robust VCO Tuning Technique" 126 //and the ChipEstimator XAUI PHY hard IP, confirm that even PHY can scale perfectly with the technology 127 //total area 128 area.set_area((mac_area + frontend_area + SerDer_area)*1e6); 129 //Power 130 //Cadence ChipEstimate using 65nm (mac, front_end are all energy. E=P*T = P/F = 1.37/1Ghz = 1.37e-9); 131 mac_dyn = 1.257e-9*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0);//niup.clockRate; //2.19W@1GHz fully active according to Cadence ChipEstimate @65nm 132 //Cadence ChipEstimate using 65nm soft IP; 133 frontend_dyn = 0.6e-9*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0);//niup.clockRate; 134 //SerDer_dyn is power not energy, scaling from 216mw/10Gb/s @130nm 135 SerDer_dyn = 0.0216*10*(interface_ip.F_sz_um/0.13)*g_tp.peri_global.Vdd/1.2*g_tp.peri_global.Vdd/1.2; 136 SerDer_dyn /= niup.clockRate;//covert to energy per clock cycle of whole NIU | 146 //Cadence ChipEstimate using 65nm 147 mac_gates = 111700; 148 frontend_gates = 320000; 149 SerDer_gates = 200000; 150 NMOS_sizing = 5 * g_tp.min_w_nmos_; 151 PMOS_sizing = 5 * g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; 152 } else { 153 //Power 154 //Cadence ChipEstimate using 65nm (mac, front_end are all energy. 155 ///E=P*T = P/F = 1.37/1Ghz = 1.37e-9); 156 //2.19W@1GHz fully active according to Cadence ChipEstimate @65nm 157 mac_dyn = 1.257e-9 * g_tp.peri_global.Vdd / 1.1 * g_tp.peri_global.Vdd 158 / 1.1 * (interface_ip.F_sz_nm / 65.0);//niup.clockRate; 159 //Cadence ChipEstimate using 65nm soft IP; 160 frontend_dyn = 0.6e-9 * g_tp.peri_global.Vdd / 1.1 * 161 g_tp.peri_global.Vdd / 1.1 * (interface_ip.F_sz_nm / 65.0); 162 //SerDer_dyn is power not energy, scaling from 216mw/10Gb/s @130nm 163 SerDer_dyn = 0.0216 * 10 * (interface_ip.F_sz_um / 0.13) * 164 g_tp.peri_global.Vdd / 1.2 * g_tp.peri_global.Vdd / 1.2; |
137 | 165 |
138 mac_gates = 111700; 139 frontend_gates = 52000; 140 SerDer_gates = 199260; | 166 mac_gates = 111700; 167 frontend_gates = 52000; 168 SerDer_gates = 199260; 169 NMOS_sizing = g_tp.min_w_nmos_; 170 PMOS_sizing = g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; 171 } |
141 | 172 |
142 NMOS_sizing = g_tp.min_w_nmos_; 143 PMOS_sizing = g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; | 173 //covert to energy per clock cycle of whole NIU 174 SerDer_dyn /= niup.clockRate; |
144 | 175 |
145 } | 176 power.readOp.dynamic = mac_dyn + frontend_dyn + SerDer_dyn; 177 power.readOp.leakage = (mac_gates + frontend_gates + frontend_gates) * 178 cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 179 g_tp.peri_global.Vdd;//unit W 180 double long_channel_device_reduction = 181 longer_channel_device_reduction(Uncore_device); 182 power.readOp.longer_channel_leakage = 183 power.readOp.leakage * long_channel_device_reduction; 184 power.readOp.gate_leakage = (mac_gates + frontend_gates + frontend_gates) * 185 cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 186 g_tp.peri_global.Vdd;//unit W |
146 | 187 |
147 power_t.readOp.dynamic = mac_dyn + frontend_dyn + SerDer_dyn; 148 power_t.readOp.leakage = (mac_gates + frontend_gates + frontend_gates)*cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 149 double long_channel_device_reduction = longer_channel_device_reduction(Uncore_device); 150 power_t.readOp.longer_channel_leakage = power_t.readOp.leakage * long_channel_device_reduction; 151 power_t.readOp.gate_leakage = (mac_gates + frontend_gates + frontend_gates)*cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 152 } | 188 // Output power 189 output_data.subthreshold_leakage_power = 190 longer_channel_device ? power.readOp.longer_channel_leakage : 191 power.readOp.leakage; 192 output_data.gate_leakage_power = power.readOp.gate_leakage; 193 output_data.peak_dynamic_power = power.readOp.dynamic * nius.duty_cycle; 194 output_data.runtime_dynamic_energy = power.readOp.dynamic * nius.perc_load; 195} |
153 | 196 |
154void NIUController::computeEnergy(bool is_tdp) 155{ 156 if (is_tdp) 157 { | 197void NIUController::set_niu_param() { 198 int num_children = xml_data->nChildNode("param"); 199 int i; 200 for (i = 0; i < num_children; i++) { 201 XMLNode* paramNode = xml_data->getChildNodePtr("param", &i); 202 XMLCSTR node_name = paramNode->getAttribute("name"); 203 XMLCSTR value = paramNode->getAttribute("value"); |
158 | 204 |
205 if (!node_name) 206 warnMissingParamName(paramNode->getAttribute("id")); |
|
159 | 207 |
160 power = power_t; 161 power.readOp.dynamic *= niup.duty_cycle; | 208 ASSIGN_FP_IF("niu_clockRate", niup.clockRate); 209 ASSIGN_INT_IF("num_units", niup.num_units); 210 ASSIGN_INT_IF("type", niup.type); |
162 | 211 |
212 else { 213 warnUnrecognizedParam(node_name); 214 } |
|
163 } | 215 } |
164 else 165 { 166 rt_power = power_t; 167 rt_power.readOp.dynamic *= niup.perc_load; 168 } 169} | |
170 | 216 |
171void NIUController::displayEnergy(uint32_t indent,int plevel,bool is_tdp) 172{ 173 string indent_str(indent, ' '); 174 string indent_str_next(indent+2, ' '); 175 bool long_channel = XML->sys.longer_channel_device; | 217 // Change from MHz to Hz 218 niup.clockRate *= 1e6; |
176 | 219 |
177 if (is_tdp) 178 { 179 cout << "NIU:" << endl; 180 cout << indent_str<< "Area = " << area.get_area()*1e-6<< " mm^2" << endl; 181 cout << indent_str << "Peak Dynamic = " << power.readOp.dynamic*niup.clockRate << " W" << endl; 182 cout << indent_str<< "Subthreshold Leakage = " 183 << (long_channel? power.readOp.longer_channel_leakage:power.readOp.leakage) <<" W" << endl; 184 //cout << indent_str<< "Subthreshold Leakage = " << power.readOp.longer_channel_leakage <<" W" << endl; 185 cout << indent_str<< "Gate Leakage = " << power.readOp.gate_leakage << " W" << endl; 186 cout << indent_str << "Runtime Dynamic = " << rt_power.readOp.dynamic*niup.clockRate << " W" << endl; 187 cout<<endl; 188 } 189 else 190 { | 220 num_children = xml_data->nChildNode("stat"); 221 for (i = 0; i < num_children; i++) { 222 XMLNode* statNode = xml_data->getChildNodePtr("stat", &i); 223 XMLCSTR node_name = statNode->getAttribute("name"); 224 XMLCSTR value = statNode->getAttribute("value"); |
191 | 225 |
192 } | 226 if (!node_name) 227 warnMissingStatName(statNode->getAttribute("id")); |
193 | 228 |
229 ASSIGN_FP_IF("duty_cycle", nius.duty_cycle); 230 ASSIGN_FP_IF("perc_load", nius.perc_load); 231 232 else { 233 warnUnrecognizedStat(node_name); 234 } 235 } |
|
194} 195 | 236} 237 |
196void NIUController::set_niu_param() 197{ 198 niup.clockRate = XML->sys.niu.clockrate; 199 niup.clockRate *= 1e6; 200 niup.num_units = XML->sys.niu.number_units; 201 niup.duty_cycle = XML->sys.niu.duty_cycle; 202 niup.perc_load = XML->sys.niu.total_load_perc; 203 niup.type = XML->sys.niu.type; 204// niup.executionTime = XML->sys.total_cycles/(XML->sys.target_core_clockrate*1e6); | 238PCIeController::PCIeController(XMLNode* _xml_data, 239 InputParameter* interface_ip_) 240 : McPATComponent(_xml_data, interface_ip_) { 241 name = "PCIe"; 242 set_pcie_param(); |
205} 206 | 243} 244 |
207PCIeController::PCIeController(ParseXML *XML_interface,InputParameter* interface_ip_) 208:XML(XML_interface), 209 interface_ip(*interface_ip_) 210 { 211 local_result = init_interface(&interface_ip); 212 double frontend_area, phy_area, ctrl_area, SerDer_area; 213 double ctrl_dyn, frontend_dyn, SerDer_dyn; 214 double ctrl_gates,frontend_gates, SerDer_gates; 215 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); 216 double NMOS_sizing, PMOS_sizing; | 245void PCIeController::computeArea() { 246 double ctrl_area; 247 double SerDer_area; |
217 | 248 |
218 /* Assuming PCIe is bit-slice based architecture 219 * This is the reason for /8 in both area and power calculation 220 * to get per lane numbers 221 */ | 249 /* Assuming PCIe is bit-slice based architecture 250 * This is the reason for /8 in both area and power calculation 251 * to get per lane numbers 252 */ |
222 | 253 |
223 set_pcie_param(); 224 if (pciep.type == 0) //high performance NIU 225 { 226 //Area estimation based on average of die photo from Niagara 2 and Cadence ChipEstimate @ 65nm. 227 ctrl_area = (5.2 + 0.5)/2 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 228 //Area estimation based on average of die photo from Niagara 2, and Cadence ChipEstimate @ 65nm. 229 frontend_area = (5.2 + 0.1)/2 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 230 //Area estimation based on average of die photo from Niagara 2 and Cadence ChipEstimate hard IP @65nm. 231 //SerDer is very hard to scale 232 SerDer_area = (3.03 + 0.36) * (interface_ip.F_sz_um/0.065);//* (interface_ip.F_sz_um/0.065); 233 phy_area = frontend_area + SerDer_area; 234 //total area 235 //Power 236 //Cadence ChipEstimate using 65nm the controller includes everything: the PHY, the data link and transaction layer 237 ctrl_dyn = 3.75e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 238 // //Cadence ChipEstimate using 65nm soft IP; 239 // frontend_dyn = 0.27e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 240 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 241 SerDer_dyn = 0.01*4*(interface_ip.F_sz_um/0.09)*g_tp.peri_global.Vdd/1.2*g_tp.peri_global.Vdd/1.2;//PCIe 2.0 max per lane speed is 4Gb/s 242 SerDer_dyn /= pciep.clockRate;//covert to energy per clock cycle | 254 if (pciep.type == 0) { //high performance PCIe 255 //Area estimation based on average of die photo from Niagara 2 and 256 //Cadence ChipEstimate @ 65nm. 257 ctrl_area = (5.2 + 0.5) / 2 * (interface_ip.F_sz_um / 0.065) * 258 (interface_ip.F_sz_um / 0.065); 259 //Area estimation based on average of die photo from Niagara 2 and 260 //Cadence ChipEstimate hard IP @65nm. 261 //SerDer is very hard to scale 262 SerDer_area = (3.03 + 0.36) * (interface_ip.F_sz_um / 263 0.065);//* (interface_ip.F_sz_um/0.065); 264 } else { 265 ctrl_area = 0.412 * (interface_ip.F_sz_um / 0.065) * 266 (interface_ip.F_sz_um / 0.065); 267 //Area estimation based on average of die photo from Niagara 2, and 268 //Cadence ChipEstimate @ 65nm. 269 SerDer_area = 0.36 * (interface_ip.F_sz_um / 0.065) * 270 (interface_ip.F_sz_um / 0.065); 271 } |
243 | 272 |
244 //power_t.readOp.dynamic = (ctrl_dyn)*pciep.num_channels; 245 //Cadence ChipEstimate using 65nm 246 ctrl_gates = 900000/8*pciep.num_channels; 247 // frontend_gates = 120000/8; 248 // SerDer_gates = 200000/8; 249 NMOS_sizing = 5*g_tp.min_w_nmos_; 250 PMOS_sizing = 5*g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; 251 } 252 else 253 { 254 ctrl_area = 0.412 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 255 //Area estimation based on average of die photo from Niagara 2, and Cadence ChipEstimate @ 65nm. 256 SerDer_area = 0.36 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 257 //total area 258 //Power 259 //Cadence ChipEstimate using 65nm the controller includes everything: the PHY, the data link and transaction layer 260 ctrl_dyn = 2.21e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 261 // //Cadence ChipEstimate using 65nm soft IP; 262 // frontend_dyn = 0.27e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 263 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 264 SerDer_dyn = 0.01*4*(interface_ip.F_sz_um/0.09)*g_tp.peri_global.Vdd/1.2*g_tp.peri_global.Vdd/1.2;//PCIe 2.0 max per lane speed is 4Gb/s 265 SerDer_dyn /= pciep.clockRate;//covert to energy per clock cycle | 273 // Total area 274 output_data.area = ((ctrl_area + (pciep.withPHY ? SerDer_area : 0)) / 8 * 275 pciep.num_channels) * 1e6; 276} |
266 | 277 |
267 //Cadence ChipEstimate using 65nm 268 ctrl_gates = 200000/8*pciep.num_channels; 269 // frontend_gates = 120000/8; 270 SerDer_gates = 200000/8*pciep.num_channels; 271 NMOS_sizing = g_tp.min_w_nmos_; 272 PMOS_sizing = g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; | 278void PCIeController::computeEnergy() { 279 double ctrl_dyn; 280 double SerDer_dyn; 281 double ctrl_gates; 282 double SerDer_gates = 0; 283 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); 284 double NMOS_sizing; 285 double PMOS_sizing; |
273 | 286 |
274 } 275 area.set_area(((ctrl_area + (pciep.withPHY? SerDer_area:0))/8*pciep.num_channels)*1e6); 276 power_t.readOp.dynamic = (ctrl_dyn + (pciep.withPHY? SerDer_dyn:0))*pciep.num_channels; 277 power_t.readOp.leakage = (ctrl_gates + (pciep.withPHY? SerDer_gates:0))*cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 278 double long_channel_device_reduction = longer_channel_device_reduction(Uncore_device); 279 power_t.readOp.longer_channel_leakage = power_t.readOp.leakage * long_channel_device_reduction; 280 power_t.readOp.gate_leakage = (ctrl_gates + (pciep.withPHY? SerDer_gates:0))*cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 281 } | 287 /* Assuming PCIe is bit-slice based architecture 288 * This is the reason for /8 in both area and power calculation 289 * to get per lane numbers 290 */ |
282 | 291 |
283void PCIeController::computeEnergy(bool is_tdp) 284{ 285 if (is_tdp) 286 { | 292 if (pciep.type == 0) { //high performance PCIe 293 //Power 294 //Cadence ChipEstimate using 65nm the controller includes everything: the PHY, the data link and transaction layer 295 ctrl_dyn = 3.75e-9 / 8 * g_tp.peri_global.Vdd / 1.1 * 296 g_tp.peri_global.Vdd / 1.1 * (interface_ip.F_sz_nm / 65.0); 297 // //Cadence ChipEstimate using 65nm soft IP; 298 // frontend_dyn = 0.27e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 299 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 300 //PCIe 2.0 max per lane speed is 4Gb/s 301 SerDer_dyn = 0.01 * 4 * (interface_ip.F_sz_um /0.09) * 302 g_tp.peri_global.Vdd / 1.2 * g_tp.peri_global.Vdd /1.2; |
287 | 303 |
304 //Cadence ChipEstimate using 65nm 305 ctrl_gates = 900000 / 8 * pciep.num_channels; 306 // frontend_gates = 120000/8; 307 // SerDer_gates = 200000/8; 308 NMOS_sizing = 5 * g_tp.min_w_nmos_; 309 PMOS_sizing = 5 * g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; 310 } else { 311 //Power 312 //Cadence ChipEstimate using 65nm the controller includes everything: the PHY, the data link and transaction layer 313 ctrl_dyn = 2.21e-9 / 8 * g_tp.peri_global.Vdd / 1.1 * 314 g_tp.peri_global.Vdd / 1.1 * (interface_ip.F_sz_nm / 65.0); 315 // //Cadence ChipEstimate using 65nm soft IP; 316 // frontend_dyn = 0.27e-9/8*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 317 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 318 //PCIe 2.0 max per lane speed is 4Gb/s 319 SerDer_dyn = 0.01 * 4 * (interface_ip.F_sz_um / 0.09) * 320 g_tp.peri_global.Vdd / 1.2 * g_tp.peri_global.Vdd /1.2; |
|
288 | 321 |
289 power = power_t; 290 power.readOp.dynamic *= pciep.duty_cycle; | 322 //Cadence ChipEstimate using 65nm 323 ctrl_gates = 200000 / 8 * pciep.num_channels; 324 // frontend_gates = 120000/8; 325 SerDer_gates = 200000 / 8 * pciep.num_channels; 326 NMOS_sizing = g_tp.min_w_nmos_; 327 PMOS_sizing = g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; |
291 292 } | 328 329 } |
293 else 294 { 295 rt_power = power_t; 296 rt_power.readOp.dynamic *= pciep.perc_load; 297 } | 330 331 //covert to energy per clock cycle 332 SerDer_dyn /= pciep.clockRate; 333 334 power.readOp.dynamic = (ctrl_dyn + (pciep.withPHY ? SerDer_dyn : 0)) * 335 pciep.num_channels; 336 power.readOp.leakage = (ctrl_gates + (pciep.withPHY ? SerDer_gates : 0)) * 337 cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 338 g_tp.peri_global.Vdd;//unit W 339 double long_channel_device_reduction = 340 longer_channel_device_reduction(Uncore_device); 341 power.readOp.longer_channel_leakage = 342 power.readOp.leakage * long_channel_device_reduction; 343 power.readOp.gate_leakage = (ctrl_gates + 344 (pciep.withPHY ? SerDer_gates : 0)) * 345 cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 346 g_tp.peri_global.Vdd;//unit W 347 348 // Output power 349 output_data.subthreshold_leakage_power = 350 longer_channel_device ? power.readOp.longer_channel_leakage : 351 power.readOp.leakage; 352 output_data.gate_leakage_power = power.readOp.gate_leakage; 353 output_data.peak_dynamic_power = power.readOp.dynamic * pcies.duty_cycle; 354 output_data.runtime_dynamic_energy = 355 power.readOp.dynamic * pcies.perc_load; |
298} 299 | 356} 357 |
300void PCIeController::displayEnergy(uint32_t indent,int plevel,bool is_tdp) 301{ 302 string indent_str(indent, ' '); 303 string indent_str_next(indent+2, ' '); 304 bool long_channel = XML->sys.longer_channel_device; | 358void PCIeController::set_pcie_param() { 359 int num_children = xml_data->nChildNode("param"); 360 int i; 361 for (i = 0; i < num_children; i++) { 362 XMLNode* paramNode = xml_data->getChildNodePtr("param", &i); 363 XMLCSTR node_name = paramNode->getAttribute("name"); 364 XMLCSTR value = paramNode->getAttribute("value"); |
305 | 365 |
306 if (is_tdp) 307 { 308 cout << "PCIe:" << endl; 309 cout << indent_str<< "Area = " << area.get_area()*1e-6<< " mm^2" << endl; 310 cout << indent_str << "Peak Dynamic = " << power.readOp.dynamic*pciep.clockRate << " W" << endl; 311 cout << indent_str<< "Subthreshold Leakage = " 312 << (long_channel? power.readOp.longer_channel_leakage:power.readOp.leakage) <<" W" << endl; 313 //cout << indent_str<< "Subthreshold Leakage = " << power.readOp.longer_channel_leakage <<" W" << endl; 314 cout << indent_str<< "Gate Leakage = " << power.readOp.gate_leakage << " W" << endl; 315 cout << indent_str << "Runtime Dynamic = " << rt_power.readOp.dynamic*pciep.clockRate << " W" << endl; 316 cout<<endl; 317 } 318 else 319 { | 366 if (!node_name) 367 warnMissingParamName(paramNode->getAttribute("id")); |
320 | 368 |
369 ASSIGN_FP_IF("pcie_clockRate", pciep.clockRate); 370 ASSIGN_INT_IF("num_units", pciep.num_units); 371 ASSIGN_INT_IF("num_channels", pciep.num_channels); 372 ASSIGN_INT_IF("type", pciep.type); 373 ASSIGN_ENUM_IF("withPHY", pciep.withPHY, bool); 374 375 else { 376 warnUnrecognizedParam(node_name); |
|
321 } | 377 } |
378 } |
|
322 | 379 |
323} | 380 // Change from MHz to Hz 381 pciep.clockRate *= 1e6; |
324 | 382 |
325void PCIeController::set_pcie_param() 326{ 327 pciep.clockRate = XML->sys.pcie.clockrate; 328 pciep.clockRate *= 1e6; 329 pciep.num_units = XML->sys.pcie.number_units; 330 pciep.num_channels = XML->sys.pcie.num_channels; 331 pciep.duty_cycle = XML->sys.pcie.duty_cycle; 332 pciep.perc_load = XML->sys.pcie.total_load_perc; 333 pciep.type = XML->sys.pcie.type; 334 pciep.withPHY = XML->sys.pcie.withPHY; 335// pciep.executionTime = XML->sys.total_cycles/(XML->sys.target_core_clockrate*1e6); | 383 num_children = xml_data->nChildNode("stat"); 384 for (i = 0; i < num_children; i++) { 385 XMLNode* statNode = xml_data->getChildNodePtr("stat", &i); 386 XMLCSTR node_name = statNode->getAttribute("name"); 387 XMLCSTR value = statNode->getAttribute("value"); |
336 | 388 |
389 if (!node_name) 390 warnMissingStatName(statNode->getAttribute("id")); 391 392 ASSIGN_FP_IF("duty_cycle", pcies.duty_cycle); 393 ASSIGN_FP_IF("perc_load", pcies.perc_load); 394 395 else { 396 warnUnrecognizedStat(node_name); 397 } 398 } |
|
337} 338 | 399} 400 |
339FlashController::FlashController(ParseXML *XML_interface,InputParameter* interface_ip_) 340:XML(XML_interface), 341 interface_ip(*interface_ip_) 342 { 343 local_result = init_interface(&interface_ip); 344 double frontend_area, phy_area, ctrl_area, SerDer_area; 345 double ctrl_dyn, frontend_dyn, SerDer_dyn; 346 double ctrl_gates,frontend_gates, SerDer_gates; 347 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); 348 double NMOS_sizing, PMOS_sizing; | 401FlashController::FlashController(XMLNode* _xml_data, 402 InputParameter* interface_ip_) 403 : McPATComponent(_xml_data, interface_ip_) { 404 name = "Flash Controller"; 405 set_fc_param(); 406} |
349 | 407 |
350 /* Assuming PCIe is bit-slice based architecture 351 * This is the reason for /8 in both area and power calculation 352 * to get per lane numbers 353 */ | 408void FlashController::computeArea() { 409 double ctrl_area; 410 double SerDer_area; |
354 | 411 |
355 set_fc_param(); 356 if (fcp.type == 0) //high performance NIU 357 { 358 cout<<"Current McPAT does not support high performance flash contorller since even low power designs are enough for maintain throughput"<<endl; 359 exit(0); 360 NMOS_sizing = 5*g_tp.min_w_nmos_; 361 PMOS_sizing = 5*g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; 362 } 363 else 364 { 365 ctrl_area = 0.243 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 366 //Area estimation based on Cadence ChipEstimate @ 65nm: NANDFLASH-CTRL from CAST 367 SerDer_area = 0.36/8 * (interface_ip.F_sz_um/0.065)* (interface_ip.F_sz_um/0.065); 368 //based On PCIe PHY TSMC65GP from Cadence ChipEstimate @ 65nm, it support 8x lanes with each lane 369 //speed up to 250MB/s (PCIe1.1x) This is already saturate the 200MB/s of the flash controller core above. 370 ctrl_gates = 129267; 371 SerDer_gates = 200000/8; 372 NMOS_sizing = g_tp.min_w_nmos_; 373 PMOS_sizing = g_tp.min_w_nmos_*pmos_to_nmos_sizing_r; | 412 /* Assuming Flash is bit-slice based architecture 413 * This is the reason for /8 in both area and power calculation 414 * to get per lane numbers 415 */ |
374 | 416 |
375 //Power 376 //Cadence ChipEstimate using 65nm the controller 125mW for every 200MB/s This is power not energy! 377 ctrl_dyn = 0.125*g_tp.peri_global.Vdd/1.1*g_tp.peri_global.Vdd/1.1*(interface_ip.F_sz_nm/65.0); 378 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 379 SerDer_dyn = 0.01*1.6*(interface_ip.F_sz_um/0.09)*g_tp.peri_global.Vdd/1.2*g_tp.peri_global.Vdd/1.2; 380 //max Per controller speed is 1.6Gb/s (200MB/s) 381 } 382 double number_channel = 1+(fcp.num_channels-1)*0.2; 383 area.set_area((ctrl_area + (fcp.withPHY? SerDer_area:0))*1e6*number_channel); 384 power_t.readOp.dynamic = (ctrl_dyn + (fcp.withPHY? SerDer_dyn:0))*number_channel; 385 power_t.readOp.leakage = ((ctrl_gates + (fcp.withPHY? SerDer_gates:0))*number_channel)*cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 386 double long_channel_device_reduction = longer_channel_device_reduction(Uncore_device); 387 power_t.readOp.longer_channel_leakage = power_t.readOp.leakage * long_channel_device_reduction; 388 power_t.readOp.gate_leakage = ((ctrl_gates + (fcp.withPHY? SerDer_gates:0))*number_channel)*cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand)*g_tp.peri_global.Vdd;//unit W 389 } | 417 if (fcp.type == 0) { //high performance flash controller 418 cout << "Current McPAT does not support high performance flash " 419 << "controller since even low power designs are enough for " 420 << "maintain throughput" <<endl; 421 exit(0); 422 } else { 423 ctrl_area = 0.243 * (interface_ip.F_sz_um / 0.065) * 424 (interface_ip.F_sz_um / 0.065); 425 //Area estimation based on Cadence ChipEstimate @ 65nm: NANDFLASH-CTRL 426 //from CAST 427 SerDer_area = 0.36 / 8 * (interface_ip.F_sz_um / 0.065) * 428 (interface_ip.F_sz_um / 0.065); 429 } |
390 | 430 |
391void FlashController::computeEnergy(bool is_tdp) 392{ 393 if (is_tdp) 394 { | 431 double number_channel = 1 + (fcp.num_channels - 1) * 0.2; 432 output_data.area = (ctrl_area + (fcp.withPHY ? SerDer_area : 0)) * 433 1e6 * number_channel; 434} |
395 | 435 |
436void FlashController::computeEnergy() { 437 double ctrl_dyn; 438 double SerDer_dyn; 439 double ctrl_gates; 440 double SerDer_gates; 441 double pmos_to_nmos_sizing_r = pmos_to_nmos_sz_ratio(); 442 double NMOS_sizing; 443 double PMOS_sizing; |
|
396 | 444 |
397 power = power_t; 398 power.readOp.dynamic *= fcp.duty_cycle; | 445 /* Assuming Flash is bit-slice based architecture 446 * This is the reason for /8 in both area and power calculation 447 * to get per lane numbers 448 */ |
399 | 449 |
450 if (fcp.type == 0) { //high performance flash controller 451 cout << "Current McPAT does not support high performance flash " 452 << "controller since even low power designs are enough for " 453 << "maintain throughput" <<endl; 454 exit(0); 455 NMOS_sizing = 5 * g_tp.min_w_nmos_; 456 PMOS_sizing = 5 * g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; 457 } else { 458 //based On PCIe PHY TSMC65GP from Cadence ChipEstimate @ 65nm, it 459 //support 8x lanes with each lane speed up to 250MB/s (PCIe1.1x). 460 //This is already saturate the 200MB/s of the flash controller core 461 //above. 462 ctrl_gates = 129267; 463 SerDer_gates = 200000 / 8; 464 NMOS_sizing = g_tp.min_w_nmos_; 465 PMOS_sizing = g_tp.min_w_nmos_ * pmos_to_nmos_sizing_r; 466 467 //Power 468 //Cadence ChipEstimate using 65nm the controller 125mW for every 469 //200MB/s This is power not energy! 470 ctrl_dyn = 0.125 * g_tp.peri_global.Vdd / 1.1 * g_tp.peri_global.Vdd / 471 1.1 * (interface_ip.F_sz_nm / 65.0); 472 //SerDer_dyn is power not energy, scaling from 10mw/Gb/s @90nm 473 SerDer_dyn = 0.01 * 1.6 * (interface_ip.F_sz_um / 0.09) * 474 g_tp.peri_global.Vdd / 1.2 * g_tp.peri_global.Vdd / 1.2; 475 //max Per controller speed is 1.6Gb/s (200MB/s) |
|
400 } | 476 } |
401 else 402 { 403 rt_power = power_t; 404 rt_power.readOp.dynamic *= fcp.perc_load; 405 } | 477 478 double number_channel = 1 + (fcp.num_channels - 1) * 0.2; 479 power.readOp.dynamic = (ctrl_dyn + (fcp.withPHY ? SerDer_dyn : 0)) * 480 number_channel; 481 power.readOp.leakage = ((ctrl_gates + (fcp.withPHY ? SerDer_gates : 0)) * 482 number_channel) * 483 cmos_Isub_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 484 g_tp.peri_global.Vdd;//unit W 485 double long_channel_device_reduction = 486 longer_channel_device_reduction(Uncore_device); 487 power.readOp.longer_channel_leakage = 488 power.readOp.leakage * long_channel_device_reduction; 489 power.readOp.gate_leakage = 490 ((ctrl_gates + (fcp.withPHY ? SerDer_gates : 0)) * number_channel) * 491 cmos_Ig_leakage(NMOS_sizing, PMOS_sizing, 2, nand) * 492 g_tp.peri_global.Vdd;//unit W 493 494 // Output power 495 output_data.subthreshold_leakage_power = 496 longer_channel_device ? power.readOp.longer_channel_leakage : 497 power.readOp.leakage; 498 output_data.gate_leakage_power = power.readOp.gate_leakage; 499 output_data.peak_dynamic_power = power.readOp.dynamic * fcs.duty_cycle; 500 output_data.runtime_dynamic_energy = power.readOp.dynamic * fcs.perc_load; |
406} 407 | 501} 502 |
408void FlashController::displayEnergy(uint32_t indent,int plevel,bool is_tdp) | 503void FlashController::set_fc_param() |
409{ | 504{ |
410 string indent_str(indent, ' '); 411 string indent_str_next(indent+2, ' '); 412 bool long_channel = XML->sys.longer_channel_device; | 505 int num_children = xml_data->nChildNode("param"); 506 int i; 507 for (i = 0; i < num_children; i++) { 508 XMLNode* paramNode = xml_data->getChildNodePtr("param", &i); 509 XMLCSTR node_name = paramNode->getAttribute("name"); 510 XMLCSTR value = paramNode->getAttribute("value"); |
413 | 511 |
414 if (is_tdp) 415 { 416 cout << "Flash Controller:" << endl; 417 cout << indent_str<< "Area = " << area.get_area()*1e-6<< " mm^2" << endl; 418 cout << indent_str << "Peak Dynamic = " << power.readOp.dynamic << " W" << endl;//no multiply of clock since this is power already 419 cout << indent_str<< "Subthreshold Leakage = " 420 << (long_channel? power.readOp.longer_channel_leakage:power.readOp.leakage) <<" W" << endl; 421 //cout << indent_str<< "Subthreshold Leakage = " << power.readOp.longer_channel_leakage <<" W" << endl; 422 cout << indent_str<< "Gate Leakage = " << power.readOp.gate_leakage << " W" << endl; 423 cout << indent_str << "Runtime Dynamic = " << rt_power.readOp.dynamic << " W" << endl; 424 cout<<endl; 425 } 426 else 427 { | 512 if (!node_name) 513 warnMissingParamName(paramNode->getAttribute("id")); |
428 | 514 |
515 ASSIGN_INT_IF("num_channels", fcp.num_channels); 516 ASSIGN_INT_IF("type", fcp.type); 517 ASSIGN_ENUM_IF("withPHY", fcp.withPHY, bool); 518 519 else { 520 warnUnrecognizedParam(node_name); |
|
429 } | 521 } |
522 } |
|
430 | 523 |
431} | 524 num_children = xml_data->nChildNode("stat"); 525 for (i = 0; i < num_children; i++) { 526 XMLNode* statNode = xml_data->getChildNodePtr("stat", &i); 527 XMLCSTR node_name = statNode->getAttribute("name"); 528 XMLCSTR value = statNode->getAttribute("value"); |
432 | 529 |
433void FlashController::set_fc_param() 434{ 435// fcp.clockRate = XML->sys.flashc.mc_clock; 436// fcp.clockRate *= 1e6; 437 fcp.peakDataTransferRate = XML->sys.flashc.peak_transfer_rate; 438 fcp.num_channels = ceil(fcp.peakDataTransferRate/200); 439 fcp.num_mcs = XML->sys.flashc.number_mcs; 440 fcp.duty_cycle = XML->sys.flashc.duty_cycle; 441 fcp.perc_load = XML->sys.flashc.total_load_perc; 442 fcp.type = XML->sys.flashc.type; 443 fcp.withPHY = XML->sys.flashc.withPHY; 444// flashcp.executionTime = XML->sys.total_cycles/(XML->sys.target_core_clockrate*1e6); | 530 if (!node_name) 531 warnMissingStatName(statNode->getAttribute("id")); |
445 | 532 |
533 ASSIGN_FP_IF("duty_cycle", fcs.duty_cycle); 534 ASSIGN_FP_IF("perc_load", fcs.perc_load); 535 536 else { 537 warnUnrecognizedStat(node_name); 538 } 539 } |
|
446} | 540} |