llvm-6502/lib/CodeGen/ModuloScheduling/ModuloScheduling.cpp

1402 lines
44 KiB
C++
Raw Normal View History

//===-- ModuloScheduling.cpp - ModuloScheduling ----------------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "ModuloSched"
#include "ModuloScheduling.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Support/CFG.h"
#include "llvm/Target/TargetSchedInfo.h"
#include "Support/Debug.h"
#include "Support/GraphWriter.h"
#include "Support/StringExtras.h"
#include <vector>
#include <utility>
#include <iostream>
#include <fstream>
#include <sstream>
using namespace llvm;
/// Create ModuloSchedulingPass
///
FunctionPass *llvm::createModuloSchedulingPass(TargetMachine & targ) {
DEBUG(std::cerr << "Created ModuloSchedulingPass\n");
return new ModuloSchedulingPass(targ);
}
template<typename GraphType>
static void WriteGraphToFile(std::ostream &O, const std::string &GraphName,
const GraphType &GT) {
std::string Filename = GraphName + ".dot";
O << "Writing '" << Filename << "'...";
std::ofstream F(Filename.c_str());
if (F.good())
WriteGraph(F, GT);
else
O << " error opening file for writing!";
O << "\n";
};
namespace llvm {
template<>
struct DOTGraphTraits<MSchedGraph*> : public DefaultDOTGraphTraits {
static std::string getGraphName(MSchedGraph *F) {
return "Dependence Graph";
}
static std::string getNodeLabel(MSchedGraphNode *Node, MSchedGraph *Graph) {
if (Node->getInst()) {
std::stringstream ss;
ss << *(Node->getInst());
return ss.str(); //((MachineInstr*)Node->getInst());
}
else
return "No Inst";
}
static std::string getEdgeSourceLabel(MSchedGraphNode *Node,
MSchedGraphNode::succ_iterator I) {
//Label each edge with the type of dependence
std::string edgelabel = "";
switch (I.getEdge().getDepOrderType()) {
case MSchedGraphEdge::TrueDep:
edgelabel = "True";
break;
case MSchedGraphEdge::AntiDep:
edgelabel = "Anti";
break;
case MSchedGraphEdge::OutputDep:
edgelabel = "Output";
break;
default:
edgelabel = "Unknown";
break;
}
//FIXME
int iteDiff = I.getEdge().getIteDiff();
std::string intStr = "(IteDiff: ";
intStr += itostr(iteDiff);
intStr += ")";
edgelabel += intStr;
return edgelabel;
}
};
}
/// ModuloScheduling::runOnFunction - main transformation entry point
bool ModuloSchedulingPass::runOnFunction(Function &F) {
bool Changed = false;
DEBUG(std::cerr << "Creating ModuloSchedGraph for each BasicBlock in" + F.getName() + "\n");
//Get MachineFunction
MachineFunction &MF = MachineFunction::get(&F);
//Iterate over BasicBlocks and do ModuloScheduling if they are valid
for (MachineFunction::const_iterator BI = MF.begin(); BI != MF.end(); ++BI) {
if(MachineBBisValid(BI)) {
MSchedGraph *MSG = new MSchedGraph(BI, target);
//Write Graph out to file
DEBUG(WriteGraphToFile(std::cerr, F.getName(), MSG));
//Print out BB for debugging
DEBUG(BI->print(std::cerr));
//Calculate Resource II
int ResMII = calculateResMII(BI);
//Calculate Recurrence II
int RecMII = calculateRecMII(MSG, ResMII);
II = std::max(RecMII, ResMII);
DEBUG(std::cerr << "II starts out as " << II << " ( RecMII=" << RecMII << "and ResMII=" << ResMII << "\n");
//Calculate Node Properties
calculateNodeAttributes(MSG, ResMII);
//Dump node properties if in debug mode
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(), E = nodeToAttributesMap.end(); I !=E; ++I) {
DEBUG(std::cerr << "Node: " << *(I->first) << " ASAP: " << I->second.ASAP << " ALAP: " << I->second.ALAP << " MOB: " << I->second.MOB << " Depth: " << I->second.depth << " Height: " << I->second.height << "\n");
}
//Put nodes in order to schedule them
computePartialOrder();
//Dump out partial order
for(std::vector<std::vector<MSchedGraphNode*> >::iterator I = partialOrder.begin(), E = partialOrder.end(); I !=E; ++I) {
DEBUG(std::cerr << "Start set in PO\n");
for(std::vector<MSchedGraphNode*>::iterator J = I->begin(), JE = I->end(); J != JE; ++J)
DEBUG(std::cerr << "PO:" << **J << "\n");
}
orderNodes();
//Dump out order of nodes
for(std::vector<MSchedGraphNode*>::iterator I = FinalNodeOrder.begin(), E = FinalNodeOrder.end(); I != E; ++I)
DEBUG(std::cerr << "FO:" << **I << "\n");
//Finally schedule nodes
computeSchedule();
DEBUG(schedule.print(std::cerr));
reconstructLoop(BI);
nodeToAttributesMap.clear();
partialOrder.clear();
recurrenceList.clear();
FinalNodeOrder.clear();
schedule.clear();
}
}
return Changed;
}
bool ModuloSchedulingPass::MachineBBisValid(const MachineBasicBlock *BI) {
//Valid basic blocks must be loops and can not have if/else statements or calls.
bool isLoop = false;
//Check first if its a valid loop
for(succ_const_iterator I = succ_begin(BI->getBasicBlock()),
E = succ_end(BI->getBasicBlock()); I != E; ++I) {
if (*I == BI->getBasicBlock()) // has single block loop
isLoop = true;
}
if(!isLoop) {
DEBUG(std::cerr << "Basic Block is not a loop\n");
return false;
}
else
DEBUG(std::cerr << "Basic Block is a loop\n");
//Get Target machine instruction info
/*const TargetInstrInfo& TMI = targ.getInstrInfo();
//Check each instruction and look for calls or if/else statements
unsigned count = 0;
for(MachineBasicBlock::const_iterator I = BI->begin(), E = BI->end(); I != E; ++I) {
//Get opcode to check instruction type
MachineOpCode OC = I->getOpcode();
if(TMI.isControlFlow(OC) && (count+1 < BI->size()))
return false;
count++;
}*/
return true;
}
//ResMII is calculated by determining the usage count for each resource
//and using the maximum.
//FIXME: In future there should be a way to get alternative resources
//for each instruction
int ModuloSchedulingPass::calculateResMII(const MachineBasicBlock *BI) {
const TargetInstrInfo & mii = target.getInstrInfo();
const TargetSchedInfo & msi = target.getSchedInfo();
int ResMII = 0;
//Map to keep track of usage count of each resource
std::map<unsigned, unsigned> resourceUsageCount;
for(MachineBasicBlock::const_iterator I = BI->begin(), E = BI->end(); I != E; ++I) {
//Get resource usage for this instruction
InstrRUsage rUsage = msi.getInstrRUsage(I->getOpcode());
std::vector<std::vector<resourceId_t> > resources = rUsage.resourcesByCycle;
//Loop over resources in each cycle and increments their usage count
for(unsigned i=0; i < resources.size(); ++i)
for(unsigned j=0; j < resources[i].size(); ++j) {
if( resourceUsageCount.find(resources[i][j]) == resourceUsageCount.end()) {
resourceUsageCount[resources[i][j]] = 1;
}
else {
resourceUsageCount[resources[i][j]] = resourceUsageCount[resources[i][j]] + 1;
}
}
}
//Find maximum usage count
//Get max number of instructions that can be issued at once. (FIXME)
int issueSlots = msi.maxNumIssueTotal;
for(std::map<unsigned,unsigned>::iterator RB = resourceUsageCount.begin(), RE = resourceUsageCount.end(); RB != RE; ++RB) {
//Get the total number of the resources in our cpu
int resourceNum = CPUResource::getCPUResource(RB->first)->maxNumUsers;
//Get total usage count for this resources
unsigned usageCount = RB->second;
//Divide the usage count by either the max number we can issue or the number of
//resources (whichever is its upper bound)
double finalUsageCount;
if( resourceNum <= issueSlots)
finalUsageCount = ceil(1.0 * usageCount / resourceNum);
else
finalUsageCount = ceil(1.0 * usageCount / issueSlots);
DEBUG(std::cerr << "Resource ID: " << RB->first << " (usage=" << usageCount << ", resourceNum=X" << ", issueSlots=" << issueSlots << ", finalUsage=" << finalUsageCount << ")\n");
//Only keep track of the max
ResMII = std::max( (int) finalUsageCount, ResMII);
}
DEBUG(std::cerr << "Final Resource MII: " << ResMII << "\n");
return ResMII;
}
int ModuloSchedulingPass::calculateRecMII(MSchedGraph *graph, int MII) {
std::vector<MSchedGraphNode*> vNodes;
//Loop over all nodes in the graph
for(MSchedGraph::iterator I = graph->begin(), E = graph->end(); I != E; ++I) {
findAllReccurrences(I->second, vNodes, MII);
vNodes.clear();
}
int RecMII = 0;
for(std::set<std::pair<int, std::vector<MSchedGraphNode*> > >::iterator I = recurrenceList.begin(), E=recurrenceList.end(); I !=E; ++I) {
std::cerr << "Recurrence: \n";
for(std::vector<MSchedGraphNode*>::const_iterator N = I->second.begin(), NE = I->second.end(); N != NE; ++N) {
std::cerr << **N << "\n";
}
RecMII = std::max(RecMII, I->first);
std::cerr << "End Recurrence with RecMII: " << I->first << "\n";
}
DEBUG(std::cerr << "RecMII: " << RecMII << "\n");
return MII;
}
void ModuloSchedulingPass::calculateNodeAttributes(MSchedGraph *graph, int MII) {
//Loop over the nodes and add them to the map
for(MSchedGraph::iterator I = graph->begin(), E = graph->end(); I != E; ++I) {
//Assert if its already in the map
assert(nodeToAttributesMap.find(I->second) == nodeToAttributesMap.end() && "Node attributes are already in the map");
//Put into the map with default attribute values
nodeToAttributesMap[I->second] = MSNodeAttributes();
}
//Create set to deal with reccurrences
std::set<MSchedGraphNode*> visitedNodes;
//Now Loop over map and calculate the node attributes
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(), E = nodeToAttributesMap.end(); I != E; ++I) {
calculateASAP(I->first, MII, (MSchedGraphNode*) 0);
visitedNodes.clear();
}
int maxASAP = findMaxASAP();
//Calculate ALAP which depends on ASAP being totally calculated
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(), E = nodeToAttributesMap.end(); I != E; ++I) {
calculateALAP(I->first, MII, maxASAP, (MSchedGraphNode*) 0);
visitedNodes.clear();
}
//Calculate MOB which depends on ASAP being totally calculated, also do depth and height
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(), E = nodeToAttributesMap.end(); I != E; ++I) {
(I->second).MOB = std::max(0,(I->second).ALAP - (I->second).ASAP);
DEBUG(std::cerr << "MOB: " << (I->second).MOB << " (" << *(I->first) << ")\n");
calculateDepth(I->first, (MSchedGraphNode*) 0);
calculateHeight(I->first, (MSchedGraphNode*) 0);
}
}
bool ModuloSchedulingPass::ignoreEdge(MSchedGraphNode *srcNode, MSchedGraphNode *destNode) {
if(destNode == 0 || srcNode ==0)
return false;
bool findEdge = edgesToIgnore.count(std::make_pair(srcNode, destNode->getInEdgeNum(srcNode)));
return findEdge;
}
int ModuloSchedulingPass::calculateASAP(MSchedGraphNode *node, int MII, MSchedGraphNode *destNode) {
DEBUG(std::cerr << "Calculating ASAP for " << *node << "\n");
//Get current node attributes
MSNodeAttributes &attributes = nodeToAttributesMap.find(node)->second;
if(attributes.ASAP != -1)
return attributes.ASAP;
int maxPredValue = 0;
//Iterate over all of the predecessors and find max
for(MSchedGraphNode::pred_iterator P = node->pred_begin(), E = node->pred_end(); P != E; ++P) {
//Only process if we are not ignoring the edge
if(!ignoreEdge(*P, node)) {
int predASAP = -1;
predASAP = calculateASAP(*P, MII, node);
assert(predASAP != -1 && "ASAP has not been calculated");
int iteDiff = node->getInEdge(*P).getIteDiff();
int currentPredValue = predASAP + (*P)->getLatency() - (iteDiff * MII);
DEBUG(std::cerr << "pred ASAP: " << predASAP << ", iteDiff: " << iteDiff << ", PredLatency: " << (*P)->getLatency() << ", Current ASAP pred: " << currentPredValue << "\n");
maxPredValue = std::max(maxPredValue, currentPredValue);
}
}
attributes.ASAP = maxPredValue;
DEBUG(std::cerr << "ASAP: " << attributes.ASAP << " (" << *node << ")\n");
return maxPredValue;
}
int ModuloSchedulingPass::calculateALAP(MSchedGraphNode *node, int MII,
int maxASAP, MSchedGraphNode *srcNode) {
DEBUG(std::cerr << "Calculating ALAP for " << *node << "\n");
MSNodeAttributes &attributes = nodeToAttributesMap.find(node)->second;
if(attributes.ALAP != -1)
return attributes.ALAP;
if(node->hasSuccessors()) {
//Trying to deal with the issue where the node has successors, but
//we are ignoring all of the edges to them. So this is my hack for
//now.. there is probably a more elegant way of doing this (FIXME)
bool processedOneEdge = false;
//FIXME, set to something high to start
int minSuccValue = 9999999;
//Iterate over all of the predecessors and fine max
for(MSchedGraphNode::succ_iterator P = node->succ_begin(),
E = node->succ_end(); P != E; ++P) {
//Only process if we are not ignoring the edge
if(!ignoreEdge(node, *P)) {
processedOneEdge = true;
int succALAP = -1;
succALAP = calculateALAP(*P, MII, maxASAP, node);
assert(succALAP != -1 && "Successors ALAP should have been caclulated");
int iteDiff = P.getEdge().getIteDiff();
int currentSuccValue = succALAP - node->getLatency() + iteDiff * MII;
DEBUG(std::cerr << "succ ALAP: " << succALAP << ", iteDiff: " << iteDiff << ", SuccLatency: " << (*P)->getLatency() << ", Current ALAP succ: " << currentSuccValue << "\n");
minSuccValue = std::min(minSuccValue, currentSuccValue);
}
}
if(processedOneEdge)
attributes.ALAP = minSuccValue;
else
attributes.ALAP = maxASAP;
}
else
attributes.ALAP = maxASAP;
DEBUG(std::cerr << "ALAP: " << attributes.ALAP << " (" << *node << ")\n");
if(attributes.ALAP < 0)
attributes.ALAP = 0;
return attributes.ALAP;
}
int ModuloSchedulingPass::findMaxASAP() {
int maxASAP = 0;
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(),
E = nodeToAttributesMap.end(); I != E; ++I)
maxASAP = std::max(maxASAP, I->second.ASAP);
return maxASAP;
}
int ModuloSchedulingPass::calculateHeight(MSchedGraphNode *node,MSchedGraphNode *srcNode) {
MSNodeAttributes &attributes = nodeToAttributesMap.find(node)->second;
if(attributes.height != -1)
return attributes.height;
int maxHeight = 0;
//Iterate over all of the predecessors and find max
for(MSchedGraphNode::succ_iterator P = node->succ_begin(),
E = node->succ_end(); P != E; ++P) {
if(!ignoreEdge(node, *P)) {
int succHeight = calculateHeight(*P, node);
assert(succHeight != -1 && "Successors Height should have been caclulated");
int currentHeight = succHeight + node->getLatency();
maxHeight = std::max(maxHeight, currentHeight);
}
}
attributes.height = maxHeight;
DEBUG(std::cerr << "Height: " << attributes.height << " (" << *node << ")\n");
return maxHeight;
}
int ModuloSchedulingPass::calculateDepth(MSchedGraphNode *node,
MSchedGraphNode *destNode) {
MSNodeAttributes &attributes = nodeToAttributesMap.find(node)->second;
if(attributes.depth != -1)
return attributes.depth;
int maxDepth = 0;
//Iterate over all of the predecessors and fine max
for(MSchedGraphNode::pred_iterator P = node->pred_begin(), E = node->pred_end(); P != E; ++P) {
if(!ignoreEdge(*P, node)) {
int predDepth = -1;
predDepth = calculateDepth(*P, node);
assert(predDepth != -1 && "Predecessors ASAP should have been caclulated");
int currentDepth = predDepth + (*P)->getLatency();
maxDepth = std::max(maxDepth, currentDepth);
}
}
attributes.depth = maxDepth;
DEBUG(std::cerr << "Depth: " << attributes.depth << " (" << *node << "*)\n");
return maxDepth;
}
void ModuloSchedulingPass::addReccurrence(std::vector<MSchedGraphNode*> &recurrence, int II, MSchedGraphNode *srcBENode, MSchedGraphNode *destBENode) {
//Check to make sure that this recurrence is unique
bool same = false;
//Loop over all recurrences already in our list
for(std::set<std::pair<int, std::vector<MSchedGraphNode*> > >::iterator R = recurrenceList.begin(), RE = recurrenceList.end(); R != RE; ++R) {
bool all_same = true;
//First compare size
if(R->second.size() == recurrence.size()) {
for(std::vector<MSchedGraphNode*>::const_iterator node = R->second.begin(), end = R->second.end(); node != end; ++node) {
if(find(recurrence.begin(), recurrence.end(), *node) == recurrence.end()) {
all_same = all_same && false;
break;
}
else
all_same = all_same && true;
}
if(all_same) {
same = true;
break;
}
}
}
if(!same) {
srcBENode = recurrence.back();
destBENode = recurrence.front();
//FIXME
if(destBENode->getInEdge(srcBENode).getIteDiff() == 0) {
//DEBUG(std::cerr << "NOT A BACKEDGE\n");
//find actual backedge HACK HACK
for(unsigned i=0; i< recurrence.size()-1; ++i) {
if(recurrence[i+1]->getInEdge(recurrence[i]).getIteDiff() == 1) {
srcBENode = recurrence[i];
destBENode = recurrence[i+1];
break;
}
}
}
DEBUG(std::cerr << "Back Edge to Remove: " << *srcBENode << " to " << *destBENode << "\n");
edgesToIgnore.insert(std::make_pair(srcBENode, destBENode->getInEdgeNum(srcBENode)));
recurrenceList.insert(std::make_pair(II, recurrence));
}
}
void ModuloSchedulingPass::findAllReccurrences(MSchedGraphNode *node,
std::vector<MSchedGraphNode*> &visitedNodes,
int II) {
if(find(visitedNodes.begin(), visitedNodes.end(), node) != visitedNodes.end()) {
std::vector<MSchedGraphNode*> recurrence;
bool first = true;
int delay = 0;
int distance = 0;
int RecMII = II; //Starting value
MSchedGraphNode *last = node;
MSchedGraphNode *srcBackEdge;
MSchedGraphNode *destBackEdge;
for(std::vector<MSchedGraphNode*>::iterator I = visitedNodes.begin(), E = visitedNodes.end();
I !=E; ++I) {
if(*I == node)
first = false;
if(first)
continue;
delay = delay + (*I)->getLatency();
if(*I != node) {
int diff = (*I)->getInEdge(last).getIteDiff();
distance += diff;
if(diff > 0) {
srcBackEdge = last;
destBackEdge = *I;
}
}
recurrence.push_back(*I);
last = *I;
}
//Get final distance calc
distance += node->getInEdge(last).getIteDiff();
//Adjust II until we get close to the inequality delay - II*distance <= 0
int value = delay-(RecMII * distance);
int lastII = II;
while(value <= 0) {
lastII = RecMII;
RecMII--;
value = delay-(RecMII * distance);
}
DEBUG(std::cerr << "Final II for this recurrence: " << lastII << "\n");
addReccurrence(recurrence, lastII, srcBackEdge, destBackEdge);
assert(distance != 0 && "Recurrence distance should not be zero");
return;
}
for(MSchedGraphNode::succ_iterator I = node->succ_begin(), E = node->succ_end(); I != E; ++I) {
visitedNodes.push_back(node);
findAllReccurrences(*I, visitedNodes, II);
visitedNodes.pop_back();
}
}
void ModuloSchedulingPass::computePartialOrder() {
//Loop over all recurrences and add to our partial order
//be sure to remove nodes that are already in the partial order in
//a different recurrence and don't add empty recurrences.
for(std::set<std::pair<int, std::vector<MSchedGraphNode*> > >::reverse_iterator I = recurrenceList.rbegin(), E=recurrenceList.rend(); I !=E; ++I) {
//Add nodes that connect this recurrence to the previous recurrence
//If this is the first recurrence in the partial order, add all predecessors
for(std::vector<MSchedGraphNode*>::const_iterator N = I->second.begin(), NE = I->second.end(); N != NE; ++N) {
}
std::vector<MSchedGraphNode*> new_recurrence;
//Loop through recurrence and remove any nodes already in the partial order
for(std::vector<MSchedGraphNode*>::const_iterator N = I->second.begin(), NE = I->second.end(); N != NE; ++N) {
bool found = false;
for(std::vector<std::vector<MSchedGraphNode*> >::iterator PO = partialOrder.begin(), PE = partialOrder.end(); PO != PE; ++PO) {
if(find(PO->begin(), PO->end(), *N) != PO->end())
found = true;
}
if(!found) {
new_recurrence.push_back(*N);
if(partialOrder.size() == 0)
//For each predecessors, add it to this recurrence ONLY if it is not already in it
for(MSchedGraphNode::pred_iterator P = (*N)->pred_begin(),
PE = (*N)->pred_end(); P != PE; ++P) {
//Check if we are supposed to ignore this edge or not
if(!ignoreEdge(*P, *N))
//Check if already in this recurrence
if(find(I->second.begin(), I->second.end(), *P) == I->second.end()) {
//Also need to check if in partial order
bool predFound = false;
for(std::vector<std::vector<MSchedGraphNode*> >::iterator PO = partialOrder.begin(), PEND = partialOrder.end(); PO != PEND; ++PO) {
if(find(PO->begin(), PO->end(), *P) != PO->end())
predFound = true;
}
if(!predFound)
if(find(new_recurrence.begin(), new_recurrence.end(), *P) == new_recurrence.end())
new_recurrence.push_back(*P);
}
}
}
}
if(new_recurrence.size() > 0)
partialOrder.push_back(new_recurrence);
}
//Add any nodes that are not already in the partial order
std::vector<MSchedGraphNode*> lastNodes;
for(std::map<MSchedGraphNode*, MSNodeAttributes>::iterator I = nodeToAttributesMap.begin(), E = nodeToAttributesMap.end(); I != E; ++I) {
bool found = false;
//Check if its already in our partial order, if not add it to the final vector
for(std::vector<std::vector<MSchedGraphNode*> >::iterator PO = partialOrder.begin(), PE = partialOrder.end(); PO != PE; ++PO) {
if(find(PO->begin(), PO->end(), I->first) != PO->end())
found = true;
}
if(!found)
lastNodes.push_back(I->first);
}
if(lastNodes.size() > 0)
partialOrder.push_back(lastNodes);
}
void ModuloSchedulingPass::predIntersect(std::vector<MSchedGraphNode*> &CurrentSet, std::vector<MSchedGraphNode*> &IntersectResult) {
//Sort CurrentSet so we can use lowerbound
sort(CurrentSet.begin(), CurrentSet.end());
for(unsigned j=0; j < FinalNodeOrder.size(); ++j) {
for(MSchedGraphNode::pred_iterator P = FinalNodeOrder[j]->pred_begin(),
E = FinalNodeOrder[j]->pred_end(); P != E; ++P) {
//Check if we are supposed to ignore this edge or not
if(ignoreEdge(*P,FinalNodeOrder[j]))
continue;
if(find(CurrentSet.begin(),
CurrentSet.end(), *P) != CurrentSet.end())
if(find(FinalNodeOrder.begin(), FinalNodeOrder.end(), *P) == FinalNodeOrder.end())
IntersectResult.push_back(*P);
}
}
}
void ModuloSchedulingPass::succIntersect(std::vector<MSchedGraphNode*> &CurrentSet, std::vector<MSchedGraphNode*> &IntersectResult) {
//Sort CurrentSet so we can use lowerbound
sort(CurrentSet.begin(), CurrentSet.end());
for(unsigned j=0; j < FinalNodeOrder.size(); ++j) {
for(MSchedGraphNode::succ_iterator P = FinalNodeOrder[j]->succ_begin(),
E = FinalNodeOrder[j]->succ_end(); P != E; ++P) {
//Check if we are supposed to ignore this edge or not
if(ignoreEdge(FinalNodeOrder[j],*P))
continue;
if(find(CurrentSet.begin(),
CurrentSet.end(), *P) != CurrentSet.end())
if(find(FinalNodeOrder.begin(), FinalNodeOrder.end(), *P) == FinalNodeOrder.end())
IntersectResult.push_back(*P);
}
}
}
void dumpIntersection(std::vector<MSchedGraphNode*> &IntersectCurrent) {
std::cerr << "Intersection (";
for(std::vector<MSchedGraphNode*>::iterator I = IntersectCurrent.begin(), E = IntersectCurrent.end(); I != E; ++I)
std::cerr << **I << ", ";
std::cerr << ")\n";
}
void ModuloSchedulingPass::orderNodes() {
int BOTTOM_UP = 0;
int TOP_DOWN = 1;
//Set default order
int order = BOTTOM_UP;
//Loop over all the sets and place them in the final node order
for(std::vector<std::vector<MSchedGraphNode*> >::iterator CurrentSet = partialOrder.begin(), E= partialOrder.end(); CurrentSet != E; ++CurrentSet) {
DEBUG(std::cerr << "Processing set in S\n");
dumpIntersection(*CurrentSet);
//Result of intersection
std::vector<MSchedGraphNode*> IntersectCurrent;
predIntersect(*CurrentSet, IntersectCurrent);
//If the intersection of predecessor and current set is not empty
//sort nodes bottom up
if(IntersectCurrent.size() != 0) {
DEBUG(std::cerr << "Final Node Order Predecessors and Current Set interesection is NOT empty\n");
order = BOTTOM_UP;
}
//If empty, use successors
else {
DEBUG(std::cerr << "Final Node Order Predecessors and Current Set interesection is empty\n");
succIntersect(*CurrentSet, IntersectCurrent);
//sort top-down
if(IntersectCurrent.size() != 0) {
DEBUG(std::cerr << "Final Node Order Successors and Current Set interesection is NOT empty\n");
order = TOP_DOWN;
}
else {
DEBUG(std::cerr << "Final Node Order Successors and Current Set interesection is empty\n");
//Find node with max ASAP in current Set
MSchedGraphNode *node;
int maxASAP = 0;
DEBUG(std::cerr << "Using current set of size " << CurrentSet->size() << "to find max ASAP\n");
for(unsigned j=0; j < CurrentSet->size(); ++j) {
//Get node attributes
MSNodeAttributes nodeAttr= nodeToAttributesMap.find((*CurrentSet)[j])->second;
//assert(nodeAttr != nodeToAttributesMap.end() && "Node not in attributes map!");
DEBUG(std::cerr << "CurrentSet index " << j << "has ASAP: " << nodeAttr.ASAP << "\n");
if(maxASAP < nodeAttr.ASAP) {
maxASAP = nodeAttr.ASAP;
node = (*CurrentSet)[j];
}
}
assert(node != 0 && "In node ordering node should not be null");
IntersectCurrent.push_back(node);
order = BOTTOM_UP;
}
}
//Repeat until all nodes are put into the final order from current set
while(IntersectCurrent.size() > 0) {
if(order == TOP_DOWN) {
DEBUG(std::cerr << "Order is TOP DOWN\n");
while(IntersectCurrent.size() > 0) {
DEBUG(std::cerr << "Intersection is not empty, so find heighest height\n");
int MOB = 0;
int height = 0;
MSchedGraphNode *highestHeightNode = IntersectCurrent[0];
//Find node in intersection with highest heigh and lowest MOB
for(std::vector<MSchedGraphNode*>::iterator I = IntersectCurrent.begin(),
E = IntersectCurrent.end(); I != E; ++I) {
//Get current nodes properties
MSNodeAttributes nodeAttr= nodeToAttributesMap.find(*I)->second;
if(height < nodeAttr.height) {
highestHeightNode = *I;
height = nodeAttr.height;
MOB = nodeAttr.MOB;
}
else if(height == nodeAttr.height) {
if(MOB > nodeAttr.height) {
highestHeightNode = *I;
height = nodeAttr.height;
MOB = nodeAttr.MOB;
}
}
}
//Append our node with greatest height to the NodeOrder
if(find(FinalNodeOrder.begin(), FinalNodeOrder.end(), highestHeightNode) == FinalNodeOrder.end()) {
DEBUG(std::cerr << "Adding node to Final Order: " << *highestHeightNode << "\n");
FinalNodeOrder.push_back(highestHeightNode);
}
//Remove V from IntersectOrder
IntersectCurrent.erase(find(IntersectCurrent.begin(),
IntersectCurrent.end(), highestHeightNode));
//Intersect V's successors with CurrentSet
for(MSchedGraphNode::succ_iterator P = highestHeightNode->succ_begin(),
E = highestHeightNode->succ_end(); P != E; ++P) {
//if(lower_bound(CurrentSet->begin(),
// CurrentSet->end(), *P) != CurrentSet->end()) {
if(find(CurrentSet->begin(), CurrentSet->end(), *P) != CurrentSet->end()) {
if(ignoreEdge(highestHeightNode, *P))
continue;
//If not already in Intersect, add
if(find(IntersectCurrent.begin(), IntersectCurrent.end(), *P) == IntersectCurrent.end())
IntersectCurrent.push_back(*P);
}
}
} //End while loop over Intersect Size
//Change direction
order = BOTTOM_UP;
//Reset Intersect to reflect changes in OrderNodes
IntersectCurrent.clear();
predIntersect(*CurrentSet, IntersectCurrent);
} //End If TOP_DOWN
//Begin if BOTTOM_UP
else {
DEBUG(std::cerr << "Order is BOTTOM UP\n");
while(IntersectCurrent.size() > 0) {
DEBUG(std::cerr << "Intersection of size " << IntersectCurrent.size() << ", finding highest depth\n");
//dump intersection
DEBUG(dumpIntersection(IntersectCurrent));
//Get node with highest depth, if a tie, use one with lowest
//MOB
int MOB = 0;
int depth = 0;
MSchedGraphNode *highestDepthNode = IntersectCurrent[0];
for(std::vector<MSchedGraphNode*>::iterator I = IntersectCurrent.begin(),
E = IntersectCurrent.end(); I != E; ++I) {
//Find node attribute in graph
MSNodeAttributes nodeAttr= nodeToAttributesMap.find(*I)->second;
if(depth < nodeAttr.depth) {
highestDepthNode = *I;
depth = nodeAttr.depth;
MOB = nodeAttr.MOB;
}
else if(depth == nodeAttr.depth) {
if(MOB > nodeAttr.MOB) {
highestDepthNode = *I;
depth = nodeAttr.depth;
MOB = nodeAttr.MOB;
}
}
}
//Append highest depth node to the NodeOrder
if(find(FinalNodeOrder.begin(), FinalNodeOrder.end(), highestDepthNode) == FinalNodeOrder.end()) {
DEBUG(std::cerr << "Adding node to Final Order: " << *highestDepthNode << "\n");
FinalNodeOrder.push_back(highestDepthNode);
}
//Remove heightestDepthNode from IntersectOrder
IntersectCurrent.erase(find(IntersectCurrent.begin(),
IntersectCurrent.end(),highestDepthNode));
//Intersect heightDepthNode's pred with CurrentSet
for(MSchedGraphNode::pred_iterator P = highestDepthNode->pred_begin(),
E = highestDepthNode->pred_end(); P != E; ++P) {
//if(lower_bound(CurrentSet->begin(),
// CurrentSet->end(), *P) != CurrentSet->end()) {
if(find(CurrentSet->begin(), CurrentSet->end(), *P) != CurrentSet->end()) {
if(ignoreEdge(*P, highestDepthNode))
continue;
//If not already in Intersect, add
if(find(IntersectCurrent.begin(),
IntersectCurrent.end(), *P) == IntersectCurrent.end())
IntersectCurrent.push_back(*P);
}
}
} //End while loop over Intersect Size
//Change order
order = TOP_DOWN;
//Reset IntersectCurrent to reflect changes in OrderNodes
IntersectCurrent.clear();
succIntersect(*CurrentSet, IntersectCurrent);
} //End if BOTTOM_DOWN
}
//End Wrapping while loop
}//End for over all sets of nodes
//Return final Order
//return FinalNodeOrder;
}
void ModuloSchedulingPass::computeSchedule() {
bool success = false;
while(!success) {
//Loop over the final node order and process each node
for(std::vector<MSchedGraphNode*>::iterator I = FinalNodeOrder.begin(),
E = FinalNodeOrder.end(); I != E; ++I) {
//CalculateEarly and Late start
int EarlyStart = -1;
int LateStart = 99999; //Set to something higher then we would ever expect (FIXME)
bool hasSucc = false;
bool hasPred = false;
if(!(*I)->isBranch()) {
//Loop over nodes in the schedule and determine if they are predecessors
//or successors of the node we are trying to schedule
for(MSSchedule::schedule_iterator nodesByCycle = schedule.begin(), nodesByCycleEnd = schedule.end();
nodesByCycle != nodesByCycleEnd; ++nodesByCycle) {
//For this cycle, get the vector of nodes schedule and loop over it
for(std::vector<MSchedGraphNode*>::iterator schedNode = nodesByCycle->second.begin(), SNE = nodesByCycle->second.end(); schedNode != SNE; ++schedNode) {
if((*I)->isPredecessor(*schedNode)) {
if(!ignoreEdge(*schedNode, *I)) {
int diff = (*I)->getInEdge(*schedNode).getIteDiff();
int ES_Temp = nodesByCycle->first + (*schedNode)->getLatency() - diff * II;
DEBUG(std::cerr << "Diff: " << diff << " Cycle: " << nodesByCycle->first << "\n");
DEBUG(std::cerr << "Temp EarlyStart: " << ES_Temp << " Prev EarlyStart: " << EarlyStart << "\n");
EarlyStart = std::max(EarlyStart, ES_Temp);
hasPred = true;
}
}
if((*I)->isSuccessor(*schedNode)) {
if(!ignoreEdge(*I,*schedNode)) {
int diff = (*schedNode)->getInEdge(*I).getIteDiff();
int LS_Temp = nodesByCycle->first - (*I)->getLatency() + diff * II;
DEBUG(std::cerr << "Diff: " << diff << " Cycle: " << nodesByCycle->first << "\n");
DEBUG(std::cerr << "Temp LateStart: " << LS_Temp << " Prev LateStart: " << LateStart << "\n");
LateStart = std::min(LateStart, LS_Temp);
hasSucc = true;
}
}
}
}
}
else {
//WARNING: HACK! FIXME!!!!
EarlyStart = II-1;
LateStart = II-1;
hasPred = 1;
hasSucc = 1;
}
DEBUG(std::cerr << "Has Successors: " << hasSucc << ", Has Pred: " << hasPred << "\n");
DEBUG(std::cerr << "EarlyStart: " << EarlyStart << ", LateStart: " << LateStart << "\n");
//Check if the node has no pred or successors and set Early Start to its ASAP
if(!hasSucc && !hasPred)
EarlyStart = nodeToAttributesMap.find(*I)->second.ASAP;
//Now, try to schedule this node depending upon its pred and successor in the schedule
//already
if(!hasSucc && hasPred)
success = scheduleNode(*I, EarlyStart, (EarlyStart + II -1));
else if(!hasPred && hasSucc)
success = scheduleNode(*I, LateStart, (LateStart - II +1));
else if(hasPred && hasSucc)
success = scheduleNode(*I, EarlyStart, std::min(LateStart, (EarlyStart + II -1)));
else
success = scheduleNode(*I, EarlyStart, EarlyStart + II - 1);
if(!success) {
++II;
schedule.clear();
break;
}
}
DEBUG(std::cerr << "Constructing Kernel\n");
success = schedule.constructKernel(II);
if(!success) {
++II;
schedule.clear();
}
}
}
bool ModuloSchedulingPass::scheduleNode(MSchedGraphNode *node,
int start, int end) {
bool success = false;
DEBUG(std::cerr << *node << " (Start Cycle: " << start << ", End Cycle: " << end << ")\n");
//Make sure start and end are not negative
if(start < 0)
start = 0;
if(end < 0)
end = 0;
bool forward = true;
if(start > end)
forward = false;
bool increaseSC = true;
int cycle = start ;
while(increaseSC) {
increaseSC = false;
increaseSC = schedule.insert(node, cycle);
if(!increaseSC)
return true;
//Increment cycle to try again
if(forward) {
++cycle;
DEBUG(std::cerr << "Increase cycle: " << cycle << "\n");
if(cycle > end)
return false;
}
else {
--cycle;
DEBUG(std::cerr << "Decrease cycle: " << cycle << "\n");
if(cycle < end)
return false;
}
}
return success;
}
/*void ModuloSchedulingPass::saveValue(const MachineInstr *inst, std::set<const Value*> &valuestoSave, std::vector<Value*> *valuesForNode) {
int numFound = 0;
Instruction *tmp;
//For each value* in this inst that is a def, we want to save a copy
//Target info
const TargetInstrInfo & mii = target.getInstrInfo();
for(unsigned i=0; i < inst->getNumOperands(); ++i) {
//get machine operand
const MachineOperand &mOp = inst->getOperand(i);
if(mOp.getType() == MachineOperand::MO_VirtualRegister && mOp.isDef()) {
//Save copy in tmpInstruction
numFound++;
tmp = TmpInstruction(mii.getMachineCodeFor(mOp.getVRegValue()),
mOp.getVRegValue());
valuesForNode->push_back(tmp);
}
}
assert(numFound == 1 && "We should have only found one def to this virtual register!");
}*/
void ModuloSchedulingPass::writePrologues(std::vector<MachineBasicBlock *> &prologues, const MachineBasicBlock *origBB, std::vector<BasicBlock*> &llvm_prologues) {
std::map<int, std::set<const MachineInstr*> > inKernel;
int maxStageCount = 0;
for(MSSchedule::kernel_iterator I = schedule.kernel_begin(), E = schedule.kernel_end(); I != E; ++I) {
maxStageCount = std::max(maxStageCount, I->second);
//Ignore the branch, we will handle this separately
if(I->first->isBranch())
continue;
//Put int the map so we know what instructions in each stage are in the kernel
if(I->second > 0) {
DEBUG(std::cerr << "Inserting instruction " << *(I->first->getInst()) << " into map at stage " << I->second << "\n");
inKernel[I->second].insert(I->first->getInst());
}
}
//Now write the prologues
for(int i = 1; i <= maxStageCount; ++i) {
BasicBlock *llvmBB = new BasicBlock();
MachineBasicBlock *machineBB = new MachineBasicBlock(llvmBB);
//Loop over original machine basic block. If we see an instruction from this
//stage that is NOT in the kernel, then it needs to be added into the prologue
//We go in order to preserve dependencies
for(MachineBasicBlock::const_iterator MI = origBB->begin(), ME = origBB->end(); ME != MI; ++MI) {
if(inKernel[i].count(&*MI)) {
inKernel[i].erase(&*MI);
if(inKernel[i].size() <= 0)
break;
else
continue;
}
else {
DEBUG(std::cerr << "Writing instruction to prologue\n");
machineBB->push_back(MI->clone());
}
}
(((MachineBasicBlock*)origBB)->getParent())->getBasicBlockList().push_back(machineBB);
prologues.push_back(machineBB);
llvm_prologues.push_back(llvmBB);
}
}
void ModuloSchedulingPass::writeEpilogues(std::vector<MachineBasicBlock *> &epilogues, const MachineBasicBlock *origBB, std::vector<BasicBlock*> &llvm_epilogues) {
std::map<int, std::set<const MachineInstr*> > inKernel;
int maxStageCount = 0;
for(MSSchedule::kernel_iterator I = schedule.kernel_begin(), E = schedule.kernel_end(); I != E; ++I) {
maxStageCount = std::max(maxStageCount, I->second);
//Ignore the branch, we will handle this separately
if(I->first->isBranch())
continue;
//Put int the map so we know what instructions in each stage are in the kernel
if(I->second > 0) {
DEBUG(std::cerr << "Inserting instruction " << *(I->first->getInst()) << " into map at stage " << I->second << "\n");
inKernel[I->second].insert(I->first->getInst());
}
}
//Now write the epilogues
for(int i = 1; i <= maxStageCount; ++i) {
BasicBlock *llvmBB = new BasicBlock();
MachineBasicBlock *machineBB = new MachineBasicBlock(llvmBB);
bool last = false;
for(MachineBasicBlock::const_iterator MI = origBB->begin(), ME = origBB->end(); ME != MI; ++MI) {
if(!last) {
if(inKernel[i].count(&*MI)) {
machineBB->push_back(MI->clone());
inKernel[i].erase(&*MI);
if(inKernel[i].size() <= 0)
last = true;
}
}
else
machineBB->push_back(MI->clone());
}
(((MachineBasicBlock*)origBB)->getParent())->getBasicBlockList().push_back(machineBB);
epilogues.push_back(machineBB);
llvm_epilogues.push_back(llvmBB);
}
}
void ModuloSchedulingPass::reconstructLoop(const MachineBasicBlock *BB) {
//The new loop will consist of an prologue, the kernel, and one or more epilogues.
std::vector<MachineBasicBlock*> prologues;
std::vector<BasicBlock*> llvm_prologues;
//Write prologue
writePrologues(prologues, BB, llvm_prologues);
//Print out prologue
for(std::vector<MachineBasicBlock*>::iterator I = prologues.begin(), E = prologues.end();
I != E; ++I) {
std::cerr << "PROLOGUE\n";
(*I)->print(std::cerr);
}
std::vector<MachineBasicBlock*> epilogues;
std::vector<BasicBlock*> llvm_epilogues;
//Write epilogues
writeEpilogues(epilogues, BB, llvm_epilogues);
//Print out prologue
for(std::vector<MachineBasicBlock*>::iterator I = epilogues.begin(), E = epilogues.end();
I != E; ++I) {
std::cerr << "EPILOGUE\n";
(*I)->print(std::cerr);
}
//create a vector of epilogues corresponding to each stage
/*std::vector<MachineBasicBlock*> epilogues;
//Create kernel
MachineBasicBlock *kernel = new MachineBasicBlock();
//keep track of stage count
int stageCount = 0;
//Target info
const TargetInstrInfo & mii = target.getInstrInfo();
//Map for creating MachinePhis
std::map<MSchedGraphNode *, std::vector<Value*> > nodeAndValueMap;
//Loop through the kernel and clone instructions that need to be put into the prologue
for(MSSchedule::kernel_iterator I = schedule.kernel_begin(), E = schedule.kernel_end(); I != E; ++I) {
//For each pair see if the stage is greater then 0
//if so, then ALL instructions before this in the original loop, need to be
//copied into the prologue
MachineBasicBlock::const_iterator actualInst;
//ignore branch
if(I->first->isBranch())
continue;
if(I->second > 0) {
assert(I->second >= stageCount && "Visiting instruction from previous stage count.\n");
//Make a set that has all the Value*'s that we read
std::set<const Value*> valuesToSave;
//For this instruction, get the Value*'s that it reads and put them into the set.
//Assert if there is an operand of another type that we need to save
const MachineInstr *inst = I->first->getInst();
for(unsigned i=0; i < inst->getNumOperands(); ++i) {
//get machine operand
const MachineOperand &mOp = inst->getOperand(i);
if(mOp.getType() == MachineOperand::MO_VirtualRegister && mOp.isUse()) {
//find the value in the map
if (const Value* srcI = mOp.getVRegValue())
valuesToSave.insert(srcI);
}
if(mOp.getType() != MachineOperand::MO_VirtualRegister && mOp.isUse()) {
assert("Our assumption is wrong. We have another type of register that needs to be saved\n");
}
}
//Check if we skipped a stage count, we need to add that stuff here
if(I->second - stageCount > 1) {
int temp = stageCount;
while(I->second - temp > 1) {
for(MachineBasicBlock::const_iterator MI = BB->begin(), ME = BB->end(); ME != MI; ++MI) {
//Check that MI is not a branch before adding, we add branches separately
if(!mii.isBranch(MI->getOpcode()) && !mii.isNop(MI->getOpcode())) {
prologue->push_back(MI->clone());
saveValue(&*MI, valuesToSave);
}
}
++temp;
}
}
if(I->second == stageCount)
continue;
stageCount = I->second;
DEBUG(std::cerr << "Found Instruction from Stage > 0\n");
//Loop over instructions in original basic block and clone them. Add to the prologue
for (MachineBasicBlock::const_iterator MI = BB->begin(), e = BB->end(); MI != e; ++MI) {
if(&*MI == I->first->getInst()) {
actualInst = MI;
break;
}
else {
//Check that MI is not a branch before adding, we add branches separately
if(!mii.isBranch(MI->getOpcode()) && !mii.isNop(MI->getOpcode()))
prologue->push_back(MI->clone());
}
}
//Now add in all instructions from this one on to its corresponding epilogue
MachineBasicBlock *epi = new MachineBasicBlock();
epilogues.push_back(epi);
for(MachineBasicBlock::const_iterator MI = actualInst, ME = BB->end(); ME != MI; ++MI) {
//Check that MI is not a branch before adding, we add branches separately
if(!mii.isBranch(MI->getOpcode()) && !mii.isNop(MI->getOpcode()))
epi->push_back(MI->clone());
}
}
}
//Create kernel
for(MSSchedule::kernel_iterator I = schedule.kernel_begin(),
E = schedule.kernel_end(); I != E; ++I) {
kernel->push_back(I->first->getInst()->clone());
}
//Debug stuff
((MachineBasicBlock*)BB)->getParent()->getBasicBlockList().push_back(prologue);
std::cerr << "PROLOGUE:\n";
prologue->print(std::cerr);
((MachineBasicBlock*)BB)->getParent()->getBasicBlockList().push_back(kernel);
std::cerr << "KERNEL: \n";
kernel->print(std::cerr);
for(std::vector<MachineBasicBlock*>::iterator MBB = epilogues.begin(), ME = epilogues.end();
MBB != ME; ++MBB) {
std::cerr << "EPILOGUE:\n";
((MachineBasicBlock*)BB)->getParent()->getBasicBlockList().push_back(*MBB);
(*MBB)->print(std::cerr);
}*/
}