343e94a257
Currently the wakeup function for the PerfectSwitch contains three loops - loop on number of virtual networks loop on number of incoming links loop till all messages for this (link, network) have been routed With an 8 processor mesh network and Hammer protocol, about 11-12% of the was observed to have been spent in this function, which is the highest amongst all the functions. It was found that the innermost loop is executed about 45 times per invocation of the wakeup function, when each invocation of the wakeup function processes just about one message. The patch tries to do away with the redundant executions of the innermost loop. Counters have been added for each virtual network that record the number of messages that need to be routed for that virtual network. The inner loops are only executed when the number of messages for that particular virtual network > 0. This does away with almost 80% of the executions of the innermost loop. The function now consumes about 5-6% of the total execution time.
350 lines
12 KiB
C++
350 lines
12 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.
|
|
*/
|
|
|
|
#include <algorithm>
|
|
|
|
#include "mem/protocol/Protocol.hh"
|
|
#include "mem/ruby/buffers/MessageBuffer.hh"
|
|
#include "mem/ruby/network/simple/PerfectSwitch.hh"
|
|
#include "mem/ruby/network/simple/SimpleNetwork.hh"
|
|
#include "mem/ruby/profiler/Profiler.hh"
|
|
#include "mem/ruby/slicc_interface/NetworkMessage.hh"
|
|
#include "mem/ruby/system/System.hh"
|
|
|
|
using namespace std;
|
|
|
|
const int PRIORITY_SWITCH_LIMIT = 128;
|
|
|
|
// Operator for helper class
|
|
bool
|
|
operator<(const LinkOrder& l1, const LinkOrder& l2)
|
|
{
|
|
return (l1.m_value < l2.m_value);
|
|
}
|
|
|
|
PerfectSwitch::PerfectSwitch(SwitchID sid, SimpleNetwork* network_ptr)
|
|
{
|
|
m_virtual_networks = network_ptr->getNumberOfVirtualNetworks();
|
|
m_switch_id = sid;
|
|
m_round_robin_start = 0;
|
|
m_network_ptr = network_ptr;
|
|
m_wakeups_wo_switch = 0;
|
|
|
|
for(int i = 0;i < m_virtual_networks;++i)
|
|
{
|
|
m_pending_message_count.push_back(0);
|
|
}
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::addInPort(const vector<MessageBuffer*>& in)
|
|
{
|
|
assert(in.size() == m_virtual_networks);
|
|
NodeID port = m_in.size();
|
|
m_in.push_back(in);
|
|
|
|
for (int j = 0; j < m_virtual_networks; j++) {
|
|
m_in[port][j]->setConsumer(this);
|
|
string desc = csprintf("[Queue from port %s %s %s to PerfectSwitch]",
|
|
NodeIDToString(m_switch_id), NodeIDToString(port),
|
|
NodeIDToString(j));
|
|
m_in[port][j]->setDescription(desc);
|
|
m_in[port][j]->setIncomingLink(port);
|
|
m_in[port][j]->setVnet(j);
|
|
}
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::addOutPort(const vector<MessageBuffer*>& out,
|
|
const NetDest& routing_table_entry)
|
|
{
|
|
assert(out.size() == m_virtual_networks);
|
|
|
|
// Setup link order
|
|
LinkOrder l;
|
|
l.m_value = 0;
|
|
l.m_link = m_out.size();
|
|
m_link_order.push_back(l);
|
|
|
|
// Add to routing table
|
|
m_out.push_back(out);
|
|
m_routing_table.push_back(routing_table_entry);
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::clearRoutingTables()
|
|
{
|
|
m_routing_table.clear();
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::clearBuffers()
|
|
{
|
|
for (int i = 0; i < m_in.size(); i++){
|
|
for(int vnet = 0; vnet < m_virtual_networks; vnet++) {
|
|
m_in[i][vnet]->clear();
|
|
}
|
|
}
|
|
|
|
for (int i = 0; i < m_out.size(); i++){
|
|
for(int vnet = 0; vnet < m_virtual_networks; vnet++) {
|
|
m_out[i][vnet]->clear();
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::reconfigureOutPort(const NetDest& routing_table_entry)
|
|
{
|
|
m_routing_table.push_back(routing_table_entry);
|
|
}
|
|
|
|
PerfectSwitch::~PerfectSwitch()
|
|
{
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::wakeup()
|
|
{
|
|
DPRINTF(RubyNetwork, "m_switch_id: %d\n",m_switch_id);
|
|
|
|
MsgPtr msg_ptr;
|
|
|
|
// Give the highest numbered link priority most of the time
|
|
m_wakeups_wo_switch++;
|
|
int highest_prio_vnet = m_virtual_networks-1;
|
|
int lowest_prio_vnet = 0;
|
|
int decrementer = 1;
|
|
NetworkMessage* net_msg_ptr = NULL;
|
|
|
|
// invert priorities to avoid starvation seen in the component network
|
|
if (m_wakeups_wo_switch > PRIORITY_SWITCH_LIMIT) {
|
|
m_wakeups_wo_switch = 0;
|
|
highest_prio_vnet = 0;
|
|
lowest_prio_vnet = m_virtual_networks-1;
|
|
decrementer = -1;
|
|
}
|
|
|
|
// For all components incoming queues
|
|
for (int vnet = highest_prio_vnet;
|
|
(vnet * decrementer) >= (decrementer * lowest_prio_vnet);
|
|
vnet -= decrementer) {
|
|
|
|
// This is for round-robin scheduling
|
|
int incoming = m_round_robin_start;
|
|
m_round_robin_start++;
|
|
if (m_round_robin_start >= m_in.size()) {
|
|
m_round_robin_start = 0;
|
|
}
|
|
|
|
if(m_pending_message_count[vnet] > 0) {
|
|
// for all input ports, use round robin scheduling
|
|
for (int counter = 0; counter < m_in.size(); counter++) {
|
|
// Round robin scheduling
|
|
incoming++;
|
|
if (incoming >= m_in.size()) {
|
|
incoming = 0;
|
|
}
|
|
|
|
// temporary vectors to store the routing results
|
|
vector<LinkID> output_links;
|
|
vector<NetDest> output_link_destinations;
|
|
|
|
// Is there a message waiting?
|
|
while (m_in[incoming][vnet]->isReady()) {
|
|
DPRINTF(RubyNetwork, "incoming: %d\n", incoming);
|
|
|
|
// Peek at message
|
|
msg_ptr = m_in[incoming][vnet]->peekMsgPtr();
|
|
net_msg_ptr = safe_cast<NetworkMessage*>(msg_ptr.get());
|
|
DPRINTF(RubyNetwork, "Message: %s\n", (*net_msg_ptr));
|
|
|
|
output_links.clear();
|
|
output_link_destinations.clear();
|
|
NetDest msg_dsts =
|
|
net_msg_ptr->getInternalDestination();
|
|
|
|
// Unfortunately, the token-protocol sends some
|
|
// zero-destination messages, so this assert isn't valid
|
|
// assert(msg_dsts.count() > 0);
|
|
|
|
assert(m_link_order.size() == m_routing_table.size());
|
|
assert(m_link_order.size() == m_out.size());
|
|
|
|
if (m_network_ptr->getAdaptiveRouting()) {
|
|
if (m_network_ptr->isVNetOrdered(vnet)) {
|
|
// Don't adaptively route
|
|
for (int out = 0; out < m_out.size(); out++) {
|
|
m_link_order[out].m_link = out;
|
|
m_link_order[out].m_value = 0;
|
|
}
|
|
} else {
|
|
// Find how clogged each link is
|
|
for (int out = 0; out < m_out.size(); out++) {
|
|
int out_queue_length = 0;
|
|
for (int v = 0; v < m_virtual_networks; v++) {
|
|
out_queue_length += m_out[out][v]->getSize();
|
|
}
|
|
int value =
|
|
(out_queue_length << 8) | (random() & 0xff);
|
|
m_link_order[out].m_link = out;
|
|
m_link_order[out].m_value = value;
|
|
}
|
|
|
|
// Look at the most empty link first
|
|
sort(m_link_order.begin(), m_link_order.end());
|
|
}
|
|
}
|
|
|
|
for (int i = 0; i < m_routing_table.size(); i++) {
|
|
// pick the next link to look at
|
|
int link = m_link_order[i].m_link;
|
|
NetDest dst = m_routing_table[link];
|
|
DPRINTF(RubyNetwork, "dst: %s\n", dst);
|
|
|
|
if (!msg_dsts.intersectionIsNotEmpty(dst))
|
|
continue;
|
|
|
|
// Remember what link we're using
|
|
output_links.push_back(link);
|
|
|
|
// Need to remember which destinations need this
|
|
// message in another vector. This Set is the
|
|
// intersection of the routing_table entry and the
|
|
// current destination set. The intersection must
|
|
// not be empty, since we are inside "if"
|
|
output_link_destinations.push_back(msg_dsts.AND(dst));
|
|
|
|
// Next, we update the msg_destination not to
|
|
// include those nodes that were already handled
|
|
// by this link
|
|
msg_dsts.removeNetDest(dst);
|
|
}
|
|
|
|
assert(msg_dsts.count() == 0);
|
|
//assert(output_links.size() > 0);
|
|
|
|
// Check for resources - for all outgoing queues
|
|
bool enough = true;
|
|
for (int i = 0; i < output_links.size(); i++) {
|
|
int outgoing = output_links[i];
|
|
if (!m_out[outgoing][vnet]->areNSlotsAvailable(1))
|
|
enough = false;
|
|
DPRINTF(RubyNetwork, "Checking if node is blocked\n"
|
|
"outgoing: %d, vnet: %d, enough: %d\n",
|
|
outgoing, vnet, enough);
|
|
}
|
|
|
|
// There were not enough resources
|
|
if (!enough) {
|
|
g_eventQueue_ptr->scheduleEvent(this, 1);
|
|
DPRINTF(RubyNetwork, "Can't deliver message since a node "
|
|
"is blocked\n"
|
|
"Message: %s\n", (*net_msg_ptr));
|
|
break; // go to next incoming port
|
|
}
|
|
|
|
MsgPtr unmodified_msg_ptr;
|
|
|
|
if (output_links.size() > 1) {
|
|
// If we are sending this message down more than
|
|
// one link (size>1), we need to make a copy of
|
|
// the message so each branch can have a different
|
|
// internal destination we need to create an
|
|
// unmodified MsgPtr because the MessageBuffer
|
|
// enqueue func will modify the message
|
|
|
|
// This magic line creates a private copy of the
|
|
// message
|
|
unmodified_msg_ptr = msg_ptr->clone();
|
|
}
|
|
|
|
// Enqueue it - for all outgoing queues
|
|
for (int i=0; i<output_links.size(); i++) {
|
|
int outgoing = output_links[i];
|
|
|
|
if (i > 0) {
|
|
// create a private copy of the unmodified
|
|
// message
|
|
msg_ptr = unmodified_msg_ptr->clone();
|
|
}
|
|
|
|
// Change the internal destination set of the
|
|
// message so it knows which destinations this
|
|
// link is responsible for.
|
|
net_msg_ptr = safe_cast<NetworkMessage*>(msg_ptr.get());
|
|
net_msg_ptr->getInternalDestination() =
|
|
output_link_destinations[i];
|
|
|
|
// Enqeue msg
|
|
DPRINTF(RubyNetwork, "Switch: %d enqueuing net msg from "
|
|
"inport[%d][%d] to outport [%d][%d] time: %lld.\n",
|
|
m_switch_id, incoming, vnet, outgoing, vnet,
|
|
g_eventQueue_ptr->getTime());
|
|
|
|
m_out[outgoing][vnet]->enqueue(msg_ptr);
|
|
}
|
|
|
|
// Dequeue msg
|
|
m_in[incoming][vnet]->pop();
|
|
m_pending_message_count[vnet]--;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::storeEventInfo(int info)
|
|
{
|
|
m_pending_message_count[info]++;
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::printStats(std::ostream& out) const
|
|
{
|
|
out << "PerfectSwitch printStats" << endl;
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::clearStats()
|
|
{
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::printConfig(std::ostream& out) const
|
|
{
|
|
}
|
|
|
|
void
|
|
PerfectSwitch::print(std::ostream& out) const
|
|
{
|
|
out << "[PerfectSwitch " << m_switch_id << "]";
|
|
}
|
|
|