gem5/src/mem/ruby/network/simple/Topology.cc
Nathan Binkert 92de70b69a ruby: Import the latest ruby changes from gems.
This was done with an automated process, so there could be things that were
done in this tree in the past that didn't make it.  One known regression
is that atomic memory operations do not seem to work properly anymore.
2009-07-06 15:49:47 -07:00

537 lines
21 KiB
C++

/*
* Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
* Topology.cc
*
* Description: See Topology.hh
*
* $Id$
*
* */
#include "mem/ruby/network/simple/Topology.hh"
#include "mem/ruby/common/NetDest.hh"
#include "mem/ruby/network/Network.hh"
#include "mem/protocol/TopologyType.hh"
//#include "mem/ruby/config/RubyConfig.hh"
#include "mem/gems_common/util.hh"
#include "mem/protocol/MachineType.hh"
#include "mem/protocol/Protocol.hh"
#include "mem/ruby/system/System.hh"
#include <string>
static const int INFINITE_LATENCY = 10000; // Yes, this is a big hack
static const int DEFAULT_BW_MULTIPLIER = 1; // Just to be consistent with above :)
// Note: In this file, we use the first 2*m_nodes SwitchIDs to
// represent the input and output endpoint links. These really are
// not 'switches', as they will not have a Switch object allocated for
// them. The first m_nodes SwitchIDs are the links into the network,
// the second m_nodes set of SwitchIDs represent the the output queues
// of the network.
// Helper functions based on chapter 29 of Cormen et al.
static Matrix extend_shortest_path(const Matrix& current_dist, Matrix& latencies, Matrix& inter_switches);
static Matrix shortest_path(const Matrix& weights, Matrix& latencies, Matrix& inter_switches);
static bool link_is_shortest_path_to_node(SwitchID src, SwitchID next, SwitchID final, const Matrix& weights, const Matrix& dist);
static NetDest shortest_path_to_node(SwitchID src, SwitchID next, const Matrix& weights, const Matrix& dist);
Topology::Topology(const string & name)
: m_name(name)
{
m_network_ptr = NULL;
m_nodes = MachineType_base_number(MachineType_NUM);
m_number_of_switches = 0;
}
void Topology::init(const vector<string> & argv)
{
for (size_t i=0; i<argv.size(); i+=2) {
if (argv[i] == "network")
m_network_ptr = RubySystem::getNetwork();
else if (argv[i] == "connections")
m_connections = argv[i+1];
else if (argv[i] == "print_config") {
m_print_config = string_to_bool(argv[i+1]);
cerr << "print config: " << m_print_config << endl;
}
}
assert(m_network_ptr != NULL);
}
void Topology::makeTopology()
{
/*
if (m_nodes == 1) {
SwitchID id = newSwitchID();
addLink(0, id, m_network_ptr->getOffChipLinkLatency());
addLink(id, 1, m_network_ptr->getOffChipLinkLatency());
return;
}
*/
assert(m_nodes > 1);
Vector< Vector < SwitchID > > nodePairs; // node pairs extracted from the file
Vector<int> latencies; // link latencies for each link extracted
Vector<int> bw_multis; // bw multipliers for each link extracted
Vector<int> weights; // link weights used to enfore e-cube deadlock free routing
Vector< SwitchID > int_network_switches; // internal switches extracted from the file
Vector<bool> endpointConnectionExist; // used to ensure all endpoints are connected to the network
endpointConnectionExist.setSize(m_nodes);
// initialize endpoint check vector
for (int k = 0; k < endpointConnectionExist.size(); k++) {
endpointConnectionExist[k] = false;
}
stringstream networkFile( m_connections );
string line = "";
while (!networkFile.eof()) {
Vector < SwitchID > nodes;
nodes.setSize(2);
int latency = -1; // null latency
int weight = -1; // null weight
int bw_multiplier = DEFAULT_BW_MULTIPLIER; // default multiplier incase the network file doesn't define it
int i = 0; // node pair index
int varsFound = 0; // number of varsFound on the line
int internalNodes = 0; // used to determine if the link is between 2 internal nodes
std::getline(networkFile, line, '\n');
string varStr = string_split(line, ' ');
// parse the current line in the file
while (varStr != "") {
string label = string_split(varStr, ':');
// valid node labels
if (label == "ext_node" || label == "int_node") {
ASSERT(i < 2); // one link between 2 switches per line
varsFound++;
bool isNewIntSwitch = true;
if (label == "ext_node") { // input link to node
MachineType machine = string_to_MachineType(string_split(varStr, ':'));
string nodeStr = string_split(varStr, ':');
nodes[i] = MachineType_base_number(machine)
+ atoi(nodeStr.c_str());
// in nodes should be numbered 0 to m_nodes-1
ASSERT(nodes[i] >= 0 && nodes[i] < m_nodes);
isNewIntSwitch = false;
endpointConnectionExist[nodes[i]] = true;
}
if (label == "int_node") { // interior node
nodes[i] = atoi((string_split(varStr, ':')).c_str())+m_nodes*2;
// in nodes should be numbered >= m_nodes*2
ASSERT(nodes[i] >= m_nodes*2);
for (int k = 0; k < int_network_switches.size(); k++) {
if (int_network_switches[k] == nodes[i]) {
isNewIntSwitch = false;
}
}
if (isNewIntSwitch) { // if internal switch
m_number_of_switches++;
int_network_switches.insertAtBottom(nodes[i]);
}
internalNodes++;
}
i++;
} else if (label == "link_latency") {
latency = atoi((string_split(varStr, ':')).c_str());
varsFound++;
} else if (label == "bw_multiplier") { // not necessary, defaults to DEFAULT_BW_MULTIPLIER
bw_multiplier = atoi((string_split(varStr, ':')).c_str());
} else if (label == "link_weight") { // not necessary, defaults to link_latency
weight = atoi((string_split(varStr, ':')).c_str());
} else {
cerr << "Error: Unexpected Identifier: " << label << endl;
exit(1);
}
varStr = string_split(line, ' ');
}
if (varsFound == 3) { // all three necessary link variables where found so add the link
nodePairs.insertAtBottom(nodes);
latencies.insertAtBottom(latency);
if (weight != -1) {
weights.insertAtBottom(weight);
} else {
weights.insertAtBottom(latency);
}
bw_multis.insertAtBottom(bw_multiplier);
Vector < SwitchID > otherDirectionNodes;
otherDirectionNodes.setSize(2);
otherDirectionNodes[0] = nodes[1];
if (internalNodes == 2) { // this is an internal link
otherDirectionNodes[1] = nodes[0];
} else {
otherDirectionNodes[1] = nodes[0]+m_nodes;
}
nodePairs.insertAtBottom(otherDirectionNodes);
latencies.insertAtBottom(latency);
if (weight != -1) {
weights.insertAtBottom(weight);
} else {
weights.insertAtBottom(latency);
}
bw_multis.insertAtBottom(bw_multiplier);
} else {
if (varsFound != 0) { // if this is not a valid link, then no vars should have been found
cerr << "Error in line: " << line << endl;
exit(1);
}
}
} // end of file
// makes sure all enpoints are connected in the soon to be created network
for (int k = 0; k < endpointConnectionExist.size(); k++) {
if (endpointConnectionExist[k] == false) {
cerr << "Error: Unconnected Endpoint: " << k << endl;
exit(1);
}
}
ASSERT(nodePairs.size() == latencies.size() && latencies.size() == bw_multis.size() && latencies.size() == weights.size())
for (int k = 0; k < nodePairs.size(); k++) {
ASSERT(nodePairs[k].size() == 2);
addLink(nodePairs[k][0], nodePairs[k][1], latencies[k], bw_multis[k], weights[k]);
}
// initialize component latencies record
m_component_latencies.setSize(0);
m_component_inter_switches.setSize(0);
}
void Topology::createLinks(bool isReconfiguration)
{
// Find maximum switchID
SwitchID max_switch_id = 0;
for (int i=0; i<m_links_src_vector.size(); i++) {
max_switch_id = max(max_switch_id, m_links_src_vector[i]);
max_switch_id = max(max_switch_id, m_links_dest_vector[i]);
}
// Initialize weight vector
Matrix topology_weights;
Matrix topology_latency;
Matrix topology_bw_multis;
int num_switches = max_switch_id+1;
topology_weights.setSize(num_switches);
topology_latency.setSize(num_switches);
topology_bw_multis.setSize(num_switches);
m_component_latencies.setSize(num_switches); // FIXME setting the size of a member variable here is a HACK!
m_component_inter_switches.setSize(num_switches); // FIXME setting the size of a member variable here is a HACK!
for(int i=0; i<topology_weights.size(); i++) {
topology_weights[i].setSize(num_switches);
topology_latency[i].setSize(num_switches);
topology_bw_multis[i].setSize(num_switches);
m_component_latencies[i].setSize(num_switches);
m_component_inter_switches[i].setSize(num_switches); // FIXME setting the size of a member variable here is a HACK!
for(int j=0; j<topology_weights[i].size(); j++) {
topology_weights[i][j] = INFINITE_LATENCY;
topology_latency[i][j] = -1; // initialize to an invalid value
topology_bw_multis[i][j] = -1; // initialize to an invalid value
m_component_latencies[i][j] = -1; // initialize to an invalid value
m_component_inter_switches[i][j] = 0; // initially assume direct connections / no intermediate switches between components
}
}
// Set identity weights to zero
for(int i=0; i<topology_weights.size(); i++) {
topology_weights[i][i] = 0;
}
// Fill in the topology weights and bandwidth multipliers
for (int i=0; i<m_links_src_vector.size(); i++) {
topology_weights[m_links_src_vector[i]][m_links_dest_vector[i]] = m_links_weight_vector[i];
topology_latency[m_links_src_vector[i]][m_links_dest_vector[i]] = m_links_latency_vector[i];
m_component_latencies[m_links_src_vector[i]][m_links_dest_vector[i]] = m_links_latency_vector[i]; // initialize to latency vector
topology_bw_multis[m_links_src_vector[i]][m_links_dest_vector[i]] = m_bw_multiplier_vector[i];
}
// Walk topology and hookup the links
Matrix dist = shortest_path(topology_weights, m_component_latencies, m_component_inter_switches);
for(int i=0; i<topology_weights.size(); i++) {
for(int j=0; j<topology_weights[i].size(); j++) {
int weight = topology_weights[i][j];
int bw_multiplier = topology_bw_multis[i][j];
int latency = topology_latency[i][j];
if (weight > 0 && weight != INFINITE_LATENCY) {
NetDest destination_set = shortest_path_to_node(i, j, topology_weights, dist);
assert(latency != -1);
makeLink(i, j, destination_set, latency, weight, bw_multiplier, isReconfiguration);
}
}
}
}
/*
void Topology::makeSwitchesPerChip(Vector< Vector < SwitchID > > &nodePairs, Vector<int> &latencies, Vector<int> &bw_multis, int numberOfChipSwitches)
{
Vector < SwitchID > nodes; // temporary buffer
nodes.setSize(2);
Vector<bool> endpointConnectionExist; // used to ensure all endpoints are connected to the network
endpointConnectionExist.setSize(m_nodes);
// initialize endpoint check vector
for (int k = 0; k < endpointConnectionExist.size(); k++) {
endpointConnectionExist[k] = false;
}
Vector<int> componentCount;
componentCount.setSize(MachineType_NUM);
for (MachineType mType = MachineType_FIRST; mType < MachineType_NUM; ++mType) {
componentCount[mType] = 0;
}
// components to/from network links
// TODO: drh5: bring back chips!!!
for (int chip = 0; chip < RubyConfig::getNumberOfChips(); chip++) {
for (MachineType mType = MachineType_FIRST; mType < MachineType_NUM; ++mType) {
for (int component = 0; component < MachineType_base_count(mType); component++) {
int latency = -1;
int bw_multiplier = -1; // internal link bw multiplier of the global bandwidth
if (mType != MachineType_Directory) {
latency = RubyConfig::getOnChipLinkLatency(); // internal link latency
bw_multiplier = 10; // internal link bw multiplier of the global bandwidth
} else {
latency = RubyConfig::getNetworkLinkLatency(); // local memory latency
bw_multiplier = 1; // local memory link bw multiplier of the global bandwidth
}
nodes[0] = MachineType_base_number(mType)+componentCount[mType];
nodes[1] = chip+m_nodes*2; // this is the chip's internal switch id #
// insert link
nodePairs.insertAtBottom(nodes);
latencies.insertAtBottom(latency);
bw_multis.insertAtBottom(bw_multiplier);
//bw_multis.insertAtBottom(componentCount[mType]+MachineType_base_number((MachineType)mType));
// opposite direction link
Vector < SwitchID > otherDirectionNodes;
otherDirectionNodes.setSize(2);
otherDirectionNodes[0] = nodes[1];
otherDirectionNodes[1] = nodes[0]+m_nodes;
nodePairs.insertAtBottom(otherDirectionNodes);
latencies.insertAtBottom(latency);
bw_multis.insertAtBottom(bw_multiplier);
assert(!endpointConnectionExist[nodes[0]]);
endpointConnectionExist[nodes[0]] = true;
componentCount[mType]++;
}
}
}
// make sure all enpoints are connected in the soon to be created network
for (int k = 0; k < endpointConnectionExist.size(); k++) {
if (endpointConnectionExist[k] == false) {
cerr << "Error: Unconnected Endpoint: " << k << endl;
exit(1);
}
}
// secondary check to ensure we saw the correct machine counts
for (MachineType mType = MachineType_FIRST; mType < MachineType_NUM; ++mType) {
assert(componentCount[mType] == MachineType_base_count((MachineType)mType));
}
}
*/
SwitchID Topology::newSwitchID()
{
m_number_of_switches++;
return m_number_of_switches-1+m_nodes+m_nodes;
}
void Topology::addLink(SwitchID src, SwitchID dest, int link_latency)
{
addLink(src, dest, link_latency, DEFAULT_BW_MULTIPLIER, link_latency);
}
void Topology::addLink(SwitchID src, SwitchID dest, int link_latency, int bw_multiplier)
{
addLink(src, dest, link_latency, bw_multiplier, link_latency);
}
void Topology::addLink(SwitchID src, SwitchID dest, int link_latency, int bw_multiplier, int link_weight)
{
ASSERT(src <= m_number_of_switches+m_nodes+m_nodes);
ASSERT(dest <= m_number_of_switches+m_nodes+m_nodes);
m_links_src_vector.insertAtBottom(src);
m_links_dest_vector.insertAtBottom(dest);
m_links_latency_vector.insertAtBottom(link_latency);
m_links_weight_vector.insertAtBottom(link_weight);
m_bw_multiplier_vector.insertAtBottom(bw_multiplier);
}
void Topology::makeLink(SwitchID src, SwitchID dest, const NetDest& routing_table_entry, int link_latency, int link_weight, int bw_multiplier, bool isReconfiguration)
{
// Make sure we're not trying to connect two end-point nodes directly together
assert((src >= 2*m_nodes) || (dest >= 2*m_nodes));
if (src < m_nodes) {
m_network_ptr->makeInLink(src, dest-(2*m_nodes), routing_table_entry, link_latency, bw_multiplier, isReconfiguration);
} else if (dest < 2*m_nodes) {
assert(dest >= m_nodes);
NodeID node = dest-m_nodes;
m_network_ptr->makeOutLink(src-(2*m_nodes), node, routing_table_entry, link_latency, link_weight, bw_multiplier, isReconfiguration);
} else {
assert((src >= 2*m_nodes) && (dest >= 2*m_nodes));
m_network_ptr->makeInternalLink(src-(2*m_nodes), dest-(2*m_nodes), routing_table_entry, link_latency, link_weight, bw_multiplier, isReconfiguration);
}
}
void Topology::printConfig(ostream& out) const
{
if (m_print_config == false) return;
assert(m_component_latencies.size() > 0);
out << "--- Begin Topology Print ---" << endl;
out << endl;
out << "Topology print ONLY indicates the _NETWORK_ latency between two machines" << endl;
out << "It does NOT include the latency within the machines" << endl;
out << endl;
for (int m=0; m<MachineType_NUM; m++) {
for (int i=0; i<MachineType_base_count((MachineType)m); i++) {
MachineID cur_mach = {(MachineType)m, i};
out << cur_mach << " Network Latencies" << endl;
for (int n=0; n<MachineType_NUM; n++) {
for (int j=0; j<MachineType_base_count((MachineType)n); j++) {
MachineID dest_mach = {(MachineType)n, j};
if (cur_mach != dest_mach) {
int link_latency = m_component_latencies[MachineType_base_number((MachineType)m)+i][MachineType_base_number(MachineType_NUM)+MachineType_base_number((MachineType)n)+j];
int intermediate_switches = m_component_inter_switches[MachineType_base_number((MachineType)m)+i][MachineType_base_number(MachineType_NUM)+MachineType_base_number((MachineType)n)+j];
out << " " << cur_mach << " -> " << dest_mach << " net_lat: "
<< link_latency+intermediate_switches << endl; // NOTE switches are assumed to have single cycle latency
}
}
}
out << endl;
}
}
out << "--- End Topology Print ---" << endl;
}
/**************************************************************************/
// The following all-pairs shortest path algorithm is based on the
// discussion from Cormen et al., Chapter 26.1.
static void extend_shortest_path(Matrix& current_dist, Matrix& latencies, Matrix& inter_switches)
{
bool change = true;
int nodes = current_dist.size();
while (change) {
change = false;
for (int i=0; i<nodes; i++) {
for (int j=0; j<nodes; j++) {
int minimum = current_dist[i][j];
int previous_minimum = minimum;
int intermediate_switch = -1;
for (int k=0; k<nodes; k++) {
minimum = min(minimum, current_dist[i][k] + current_dist[k][j]);
if (previous_minimum != minimum) {
intermediate_switch = k;
inter_switches[i][j] = inter_switches[i][k] + inter_switches[k][j] + 1;
}
previous_minimum = minimum;
}
if (current_dist[i][j] != minimum) {
change = true;
current_dist[i][j] = minimum;
assert(intermediate_switch >= 0);
assert(intermediate_switch < latencies[i].size());
latencies[i][j] = latencies[i][intermediate_switch] + latencies[intermediate_switch][j];
}
}
}
}
}
static Matrix shortest_path(const Matrix& weights, Matrix& latencies, Matrix& inter_switches)
{
Matrix dist = weights;
extend_shortest_path(dist, latencies, inter_switches);
return dist;
}
static bool link_is_shortest_path_to_node(SwitchID src, SwitchID next, SwitchID final,
const Matrix& weights, const Matrix& dist)
{
return (weights[src][next] + dist[next][final] == dist[src][final]);
}
static NetDest shortest_path_to_node(SwitchID src, SwitchID next,
const Matrix& weights, const Matrix& dist)
{
NetDest result;
int d = 0;
int machines;
int max_machines;
machines = MachineType_NUM;
max_machines = MachineType_base_number(MachineType_NUM);
for (int m=0; m<machines; m++) {
for (int i=0; i<MachineType_base_count((MachineType)m); i++) {
// we use "d+max_machines" below since the "destination" switches for the machines are numbered
// [MachineType_base_number(MachineType_NUM)...2*MachineType_base_number(MachineType_NUM)-1]
// for the component network
if (link_is_shortest_path_to_node(src, next,
d+max_machines,
weights, dist)) {
MachineID mach = {(MachineType)m, i};
result.add(mach);
}
d++;
}
}
DEBUG_MSG(NETWORK_COMP, MedPrio, "returning shortest path");
DEBUG_EXPR(NETWORK_COMP, MedPrio, (src-(2*max_machines)));
DEBUG_EXPR(NETWORK_COMP, MedPrio, (next-(2*max_machines)));
DEBUG_EXPR(NETWORK_COMP, MedPrio, src);
DEBUG_EXPR(NETWORK_COMP, MedPrio, next);
DEBUG_EXPR(NETWORK_COMP, MedPrio, result);
DEBUG_NEWLINE(NETWORK_COMP, MedPrio);
return result;
}