blob: 39444c9023fd66b9f703b3a85aa3cefbe8c34b1f [file] [log] [blame]
/*
* Copyright (c) 2020 Advanced Micro Devices, Inc.
* 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.
*/
#include "mem/ruby/network/Topology.hh"
#include <cassert>
#include "base/trace.hh"
#include "debug/RubyNetwork.hh"
#include "mem/ruby/common/NetDest.hh"
#include "mem/ruby/network/BasicLink.hh"
#include "mem/ruby/network/Network.hh"
#include "mem/ruby/slicc_interface/AbstractController.hh"
namespace gem5
{
namespace ruby
{
const int INFINITE_LATENCY = 10000; // Yes, this is a big hack
// 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.
Topology::Topology(uint32_t num_nodes, uint32_t num_routers,
uint32_t num_vnets,
const std::vector<BasicExtLink *> &ext_links,
const std::vector<BasicIntLink *> &int_links)
: m_nodes(MachineType_base_number(MachineType_NUM)),
m_number_of_switches(num_routers), m_vnets(num_vnets),
m_ext_link_vector(ext_links), m_int_link_vector(int_links)
{
// Total nodes/controllers in network
assert(m_nodes > 1);
// analyze both the internal and external links, create data structures.
// The python created external links are bi-directional,
// and the python created internal links are uni-directional.
// The networks and topology utilize uni-directional links.
// Thus each external link is converted to two calls to addLink,
// one for each direction.
//
// External Links
for (std::vector<BasicExtLink*>::const_iterator i = ext_links.begin();
i != ext_links.end(); ++i) {
BasicExtLink *ext_link = (*i);
AbstractController *abs_cntrl = ext_link->params().ext_node;
BasicRouter *router = ext_link->params().int_node;
int machine_base_idx = MachineType_base_number(abs_cntrl->getType());
int ext_idx1 = machine_base_idx + abs_cntrl->getVersion();
int ext_idx2 = ext_idx1 + m_nodes;
int int_idx = router->params().router_id + 2*m_nodes;
// create the internal uni-directional links in both directions
// ext to int
addLink(ext_idx1, int_idx, ext_link);
// int to ext
addLink(int_idx, ext_idx2, ext_link);
}
// Internal Links
for (std::vector<BasicIntLink*>::const_iterator i = int_links.begin();
i != int_links.end(); ++i) {
BasicIntLink *int_link = (*i);
BasicRouter *router_src = int_link->params().src_node;
BasicRouter *router_dst = int_link->params().dst_node;
PortDirection src_outport = int_link->params().src_outport;
PortDirection dst_inport = int_link->params().dst_inport;
// Store the IntLink pointers for later
m_int_link_vector.push_back(int_link);
int src = router_src->params().router_id + 2*m_nodes;
int dst = router_dst->params().router_id + 2*m_nodes;
// create the internal uni-directional link from src to dst
addLink(src, dst, int_link, src_outport, dst_inport);
}
}
void
Topology::createLinks(Network *net)
{
// Find maximum switchID
SwitchID max_switch_id = 0;
for (LinkMap::const_iterator i = m_link_map.begin();
i != m_link_map.end(); ++i) {
std::pair<SwitchID, SwitchID> src_dest = (*i).first;
max_switch_id = std::max(max_switch_id, src_dest.first);
max_switch_id = std::max(max_switch_id, src_dest.second);
}
// Initialize weight, latency, and inter switched vectors
int num_switches = max_switch_id+1;
Matrix topology_weights(m_vnets,
std::vector<std::vector<int>>(num_switches,
std::vector<int>(num_switches, INFINITE_LATENCY)));
Matrix component_latencies(num_switches,
std::vector<std::vector<int>>(num_switches,
std::vector<int>(m_vnets, -1)));
Matrix component_inter_switches(num_switches,
std::vector<std::vector<int>>(num_switches,
std::vector<int>(m_vnets, 0)));
// Set identity weights to zero
for (int i = 0; i < topology_weights[0].size(); i++) {
for (int v = 0; v < m_vnets; v++) {
topology_weights[v][i][i] = 0;
}
}
// Fill in the topology weights and bandwidth multipliers
for (auto link_group : m_link_map) {
std::pair<int, int> src_dest = link_group.first;
std::vector<bool> vnet_done(m_vnets, 0);
int src = src_dest.first;
int dst = src_dest.second;
// Iterate over all links for this source and destination
std::vector<LinkEntry> link_entries = link_group.second;
for (int l = 0; l < link_entries.size(); l++) {
BasicLink* link = link_entries[l].link;
if (link->mVnets.size() == 0) {
for (int v = 0; v < m_vnets; v++) {
// Two links connecting same src and destination
// cannot carry same vnets.
fatal_if(vnet_done[v], "Two links connecting same src"
" and destination cannot support same vnets");
component_latencies[src][dst][v] = link->m_latency;
topology_weights[v][src][dst] = link->m_weight;
vnet_done[v] = true;
}
} else {
for (int v = 0; v < link->mVnets.size(); v++) {
int vnet = link->mVnets[v];
fatal_if(vnet >= m_vnets, "Not enough virtual networks "
"(setting latency and weight for vnet %d)", vnet);
// Two links connecting same src and destination
// cannot carry same vnets.
fatal_if(vnet_done[vnet], "Two links connecting same src"
" and destination cannot support same vnets");
component_latencies[src][dst][vnet] = link->m_latency;
topology_weights[vnet][src][dst] = link->m_weight;
vnet_done[vnet] = true;
}
}
}
}
// Walk topology and hookup the links
Matrix dist = shortest_path(topology_weights, component_latencies,
component_inter_switches);
for (int i = 0; i < topology_weights[0].size(); i++) {
for (int j = 0; j < topology_weights[0][i].size(); j++) {
std::vector<NetDest> routingMap;
routingMap.resize(m_vnets);
// Not all sources and destinations are connected
// by direct links. We only construct the links
// which have been configured in topology.
bool realLink = false;
for (int v = 0; v < m_vnets; v++) {
int weight = topology_weights[v][i][j];
if (weight > 0 && weight != INFINITE_LATENCY) {
realLink = true;
routingMap[v] =
shortest_path_to_node(i, j, topology_weights, dist, v);
}
}
// Make one link for each set of vnets between
// a given source and destination. We do not
// want to create one link for each vnet.
if (realLink) {
makeLink(net, i, j, routingMap);
}
}
}
}
void
Topology::addLink(SwitchID src, SwitchID dest, BasicLink* link,
PortDirection src_outport_dirn,
PortDirection dst_inport_dirn)
{
assert(src <= m_number_of_switches+m_nodes+m_nodes);
assert(dest <= m_number_of_switches+m_nodes+m_nodes);
std::pair<int, int> src_dest_pair;
src_dest_pair.first = src;
src_dest_pair.second = dest;
LinkEntry link_entry;
link_entry.link = link;
link_entry.src_outport_dirn = src_outport_dirn;
link_entry.dst_inport_dirn = dst_inport_dirn;
auto lit = m_link_map.find(src_dest_pair);
if (lit != m_link_map.end()) {
// HeteroGarnet allows multiple links between
// same source-destination pair supporting
// different vnets. If there is a link already
// between a given pair of source and destination
// add this new link to it.
lit->second.push_back(link_entry);
} else {
std::vector<LinkEntry> links;
links.push_back(link_entry);
m_link_map[src_dest_pair] = links;
}
}
void
Topology::makeLink(Network *net, SwitchID src, SwitchID dest,
std::vector<NetDest>& routing_table_entry)
{
// Make sure we're not trying to connect two end-point nodes
// directly together
assert(src >= 2 * m_nodes || dest >= 2 * m_nodes);
std::pair<int, int> src_dest;
LinkEntry link_entry;
if (src < m_nodes) {
src_dest.first = src;
src_dest.second = dest;
std::vector<LinkEntry> links = m_link_map[src_dest];
for (int l = 0; l < links.size(); l++) {
link_entry = links[l];
std::vector<NetDest> linkRoute;
linkRoute.resize(m_vnets);
BasicLink *link = link_entry.link;
if (link->mVnets.size() == 0) {
net->makeExtInLink(src, dest - (2 * m_nodes), link,
routing_table_entry);
} else {
for (int v = 0; v< link->mVnets.size(); v++) {
int vnet = link->mVnets[v];
linkRoute[vnet] = routing_table_entry[vnet];
}
net->makeExtInLink(src, dest - (2 * m_nodes), link,
linkRoute);
}
}
} else if (dest < 2*m_nodes) {
assert(dest >= m_nodes);
NodeID node = dest - m_nodes;
src_dest.first = src;
src_dest.second = dest;
std::vector<LinkEntry> links = m_link_map[src_dest];
for (int l = 0; l < links.size(); l++) {
link_entry = links[l];
std::vector<NetDest> linkRoute;
linkRoute.resize(m_vnets);
BasicLink *link = link_entry.link;
if (link->mVnets.size() == 0) {
net->makeExtOutLink(src - (2 * m_nodes), node, link,
routing_table_entry);
} else {
for (int v = 0; v< link->mVnets.size(); v++) {
int vnet = link->mVnets[v];
linkRoute[vnet] = routing_table_entry[vnet];
}
net->makeExtOutLink(src - (2 * m_nodes), node, link,
linkRoute);
}
}
} else {
assert((src >= 2 * m_nodes) && (dest >= 2 * m_nodes));
src_dest.first = src;
src_dest.second = dest;
std::vector<LinkEntry> links = m_link_map[src_dest];
for (int l = 0; l < links.size(); l++) {
link_entry = links[l];
std::vector<NetDest> linkRoute;
linkRoute.resize(m_vnets);
BasicLink *link = link_entry.link;
if (link->mVnets.size() == 0) {
net->makeInternalLink(src - (2 * m_nodes),
dest - (2 * m_nodes), link, routing_table_entry,
link_entry.src_outport_dirn,
link_entry.dst_inport_dirn);
} else {
for (int v = 0; v< link->mVnets.size(); v++) {
int vnet = link->mVnets[v];
linkRoute[vnet] = routing_table_entry[vnet];
}
net->makeInternalLink(src - (2 * m_nodes),
dest - (2 * m_nodes), link, linkRoute,
link_entry.src_outport_dirn,
link_entry.dst_inport_dirn);
}
}
}
}
// The following all-pairs shortest path algorithm is based on the
// discussion from Cormen et al., Chapter 26.1.
void
Topology::extend_shortest_path(Matrix &current_dist, Matrix &latencies,
Matrix &inter_switches)
{
int nodes = current_dist[0].size();
// We find the shortest path for each vnet for a given pair of
// source and destinations. This is done simply by traversing via
// all other nodes and finding the minimum distance.
for (int v = 0; v < m_vnets; v++) {
// There is a different topology for each vnet. Here we try to
// build a topology by finding the minimum number of intermediate
// switches needed to reach the destination
bool change = true;
while (change) {
change = false;
for (int i = 0; i < nodes; i++) {
for (int j = 0; j < nodes; j++) {
// We follow an iterative process to build the shortest
// path tree:
// 1. Start from the direct connection (if there is one,
// otherwise assume a hypothetical infinite weight link).
// 2. Then we iterate through all other nodes considering
// new potential intermediate switches. If we find any
// lesser weight combination, we set(update) that as the
// new weight between the source and destination.
// 3. Repeat for all pairs of nodes.
// 4. Go to step 1 if there was any new update done in
// Step 2.
int minimum = current_dist[v][i][j];
int previous_minimum = minimum;
int intermediate_switch = -1;
for (int k = 0; k < nodes; k++) {
minimum = std::min(minimum,
current_dist[v][i][k] + current_dist[v][k][j]);
if (previous_minimum != minimum) {
intermediate_switch = k;
inter_switches[i][j][v] =
inter_switches[i][k][v] +
inter_switches[k][j][v] + 1;
}
previous_minimum = minimum;
}
if (current_dist[v][i][j] != minimum) {
change = true;
current_dist[v][i][j] = minimum;
assert(intermediate_switch >= 0);
assert(intermediate_switch < latencies[i].size());
latencies[i][j][v] =
latencies[i][intermediate_switch][v] +
latencies[intermediate_switch][j][v];
}
}
}
}
}
}
Matrix
Topology::shortest_path(const Matrix &weights, Matrix &latencies,
Matrix &inter_switches)
{
Matrix dist = weights;
extend_shortest_path(dist, latencies, inter_switches);
return dist;
}
bool
Topology::link_is_shortest_path_to_node(SwitchID src, SwitchID next,
SwitchID final, const Matrix &weights,
const Matrix &dist, int vnet)
{
return weights[vnet][src][next] + dist[vnet][next][final] ==
dist[vnet][src][final];
}
NetDest
Topology::shortest_path_to_node(SwitchID src, SwitchID next,
const Matrix &weights, const Matrix &dist,
int vnet)
{
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 (NodeID 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, vnet)) {
MachineID mach = {(MachineType)m, i};
result.add(mach);
}
d++;
}
}
DPRINTF(RubyNetwork, "Returning shortest path\n"
"(src-(2*max_machines)): %d, (next-(2*max_machines)): %d, "
"src: %d, next: %d, vnet:%d result: %s\n",
(src-(2*max_machines)), (next-(2*max_machines)),
src, next, vnet, result);
return result;
}
} // namespace ruby
} // namespace gem5