Skip to content

Bound Lua inlining by register pressure #1291

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Frotty merged 9 commits into from
Sep 4, 2026
Merged

Bound Lua inlining by register pressure #1291

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
aa4891a
Bound Lua inlining by register pressure
Frotty Sep 4, 2026
d3b7927
Fix Lua register budget review findings
Frotty Sep 4, 2026
f895c2a
Fix vararg loop pressure accounting
Frotty Sep 4, 2026
52eff76
Account for Lua function-entry locals
Frotty Sep 4, 2026
45c8ef0
Use allocated pressure for Lua helper retries
Frotty Sep 4, 2026
a0d3839
Bound Lua inlining without local allocation
Frotty Sep 4, 2026
98f62cf
Account for Lua allocation classes in inlining
Frotty Sep 4, 2026
709d4fe
Budget overlapping Lua inline results
Frotty Sep 4, 2026
b2d8145
Budget Lua argument staging locals
Frotty Sep 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions de.peeeq.wurstscript/src/main/java/de/peeeq/wurstio/WurstCompilerJassImpl.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -946,6 +946,15 @@ public LuaCompilationUnit transformProgToLua() {
timeTaker.endPhase();
}

if (runArgs.isInline()) {
beginPhase(10, "inline Lua arithmetic helpers within allocated local budget");
int arithmeticHelpersInlined = optimizer.inlineLuaDivModHelpersWithinLocalBudget();
if (arithmeticHelpersInlined > 0 && runArgs.isLocalOptimizations()) {
optimizer.localOptimizations();
Comment thread
Frotty marked this conversation as resolved.
Comment thread
Frotty marked this conversation as resolved.
}
timeTaker.endPhase();
}

printDebugImProg("./test-output/lua/im " + stage++ + "_afterlocalopts.im");

boolean garbageChanged = optimizer.removeGarbage();
Expand Down
1 change: 0 additions & 1 deletion ...cript/src/main/java/de/peeeq/wurstscript/intermediatelang/optimizer/ControlFlowGraph.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,7 +163,6 @@ private Node getNode(ImStmt s) {
result.stmt = null;
} else if (s instanceof ImVarargLoop) {
result.setName("vararg loop");
result.stmt = null;
}
}
return result;
Expand Down
125 changes: 111 additions & 14 deletions ...urstscript/src/main/java/de/peeeq/wurstscript/intermediatelang/optimizer/LocalMerger.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,6 @@
import de.peeeq.wurstscript.translation.imtranslation.ImHelper;
import de.peeeq.wurstscript.translation.imtranslation.ImTranslator;
import de.peeeq.wurstscript.types.TypesHelper;
import io.vavr.collection.HashSet;
import io.vavr.collection.Set;
import it.unimi.dsi.fastutil.objects.Object2IntOpenHashMap;
import it.unimi.dsi.fastutil.objects.ObjectOpenHashSet;
Expand Down Expand Up @@ -56,10 +55,20 @@ void optimizeFunc(ImFunction func, LocalPlayerContextAnalyzer analyzer) {
private boolean canMerge(ImType a, ImType b) { return a.equalsType(b); }

private void mergeLocals(Map<ImStmt, Set<ImVar>> livenessInfo, ImFunction func) {
Map<ImVar, Set<ImVar>> interference = calculateInferenceGraph(livenessInfo);
Map<ImVar, java.util.Set<ImVar>> interference = calculateInterferenceGraph(livenessInfo, func);

Map<ImVar, Integer> declarationOrder = new IdentityHashMap<>();
int nextOrder = 0;
for (ImVar parameter : func.getParameters()) {
declarationOrder.put(parameter, nextOrder++);
}
for (ImVar local : func.getLocals()) {
declarationOrder.put(local, nextOrder++);
}

PriorityQueue<ImVar> queue = new PriorityQueue<>(
(x, y) -> interference.get(y).size() - interference.get(x).size()
Comparator.<ImVar>comparingInt(v -> interference.get(v).size()).reversed()
.thenComparingInt(declarationOrder::get)
);
queue.addAll(interference.keySet());

Expand All @@ -81,8 +90,8 @@ private void mergeLocals(Map<ImStmt, Set<ImVar>> livenessInfo, ImFunction func)
continue;
}
if (localPlayerContextAnalyzer != null
&& (localPlayerContextAnalyzer.isLocalPlayerDependent(v)
|| localPlayerContextAnalyzer.isLocalPlayerDependent(color))) {
&& localPlayerContextAnalyzer.isLocalPlayerDependent(v)
!= localPlayerContextAnalyzer.isLocalPlayerDependent(color)) {
continue;
}

Expand Down Expand Up @@ -158,17 +167,94 @@ private static int removeUnusedLocals(ImFunction f) {
return before - kept.size();
}

private Map<ImVar, Set<ImVar>> calculateInferenceGraph(Map<ImStmt, Set<ImVar>> livenessInfo) {
Map<ImVar, Set<ImVar>> g = new LinkedHashMap<>();
for (Map.Entry<ImStmt, Set<ImVar>> e : livenessInfo.entrySet()) {
Set<ImVar> live = e.getValue();
for (ImVar v1 : live) {
Set<ImVar> set = g.getOrDefault(v1, HashSet.empty());
set = set.addAll(live.filter(v2 -> canMerge(v1.getType(), v2.getType())));
g.put(v1, set);
private Map<ImVar, java.util.Set<ImVar>> calculateInterferenceGraph(
Map<ImStmt, Set<ImVar>> livenessInfo, ImFunction func) {
Map<ImVar, java.util.Set<ImVar>> graph = new LinkedHashMap<>();
for (ImVar parameter : func.getParameters()) {
graph.put(parameter, new ObjectOpenHashSet<>());
}
for (ImVar local : func.getLocals()) {
graph.put(local, new ObjectOpenHashSet<>());
}

java.util.Set<ImVar> explicitlyDefined = Collections.newSetFromMap(new IdentityHashMap<>());

// A definition interferes with every compatible value that remains live after it.
// Building only those edges is equivalent to cliquing every live set, while avoiding
// the old O(statements * liveValues^2) behavior on large inlined functions.
for (Map.Entry<ImStmt, Set<ImVar>> entry : livenessInfo.entrySet()) {
List<ImVar> defined = definedLocals(entry.getKey());
if (defined.isEmpty()) {
continue;
}
Comment thread
Frotty marked this conversation as resolved.
explicitlyDefined.addAll(defined);
for (int i = 0; i < defined.size(); i++) {
ImVar definition = defined.get(i);
java.util.Set<ImVar> neighbors = graph.computeIfAbsent(definition, ignored -> new ObjectOpenHashSet<>());
for (ImVar live : entry.getValue()) {
if (live == definition || !canMerge(definition.getType(), live.getType())) {
continue;
}
neighbors.add(live);
graph.computeIfAbsent(live, ignored -> new ObjectOpenHashSet<>()).add(definition);
}
// Vararg tuple components are assigned at the same loop boundary. They must
// occupy distinct slots even when neither component is live before the loop.
for (int j = i + 1; j < defined.size(); j++) {
ImVar other = defined.get(j);
if (canMerge(definition.getType(), other.getType())) {
neighbors.add(other);
graph.computeIfAbsent(other, ignored -> new ObjectOpenHashSet<>()).add(definition);
}
}
}
}

// Parameters and locals with no explicit assignment receive their values at function
// entry. Model that simultaneous definition so a warning-only read of an uninitialized
// local cannot be colored onto a parameter (or another implicit entry value).
List<ImVar> entryDefinitions = new ArrayList<>(func.getParameters());
for (ImVar local : func.getLocals()) {
if (!explicitlyDefined.contains(local)) {
entryDefinitions.add(local);
Comment thread
Frotty marked this conversation as resolved.
}
}
for (int i = 0; i < entryDefinitions.size(); i++) {
ImVar definition = entryDefinitions.get(i);
java.util.Set<ImVar> neighbors = graph.get(definition);
for (int j = i + 1; j < entryDefinitions.size(); j++) {
ImVar other = entryDefinitions.get(j);
if (canMerge(definition.getType(), other.getType())) {
neighbors.add(other);
graph.get(other).add(definition);
}
}
}
return graph;
}

private static List<ImVar> definedLocals(ImStmt stmt) {
if (stmt instanceof ImVarargLoop loop) {
List<ImVar> result = new ArrayList<>(loop.getLoopVars().size());
for (ImVarargLoopVar loopVar : loop.getLoopVars()) {
result.add(loopVar.getVar());
}
return result;
}
if (!(stmt instanceof ImSet set)) {
return Collections.emptyList();
}
return g;
ImLExpr left = set.getLeft();
if (left instanceof ImVarAccess access && !access.getVar().isGlobal()) {
return Collections.singletonList(access.getVar());
}
if (left instanceof ImTupleSelection selection) {
ImVar var = TypesHelper.getSimpleAndPureTupleVar(selection);
if (var != null && !var.isGlobal()) {
return Collections.singletonList(var);
}
}
return Collections.emptyList();
}

private void eliminateDeadCode(Map<ImStmt, Set<ImVar>> livenessInfo) {
Expand Down Expand Up @@ -272,6 +358,17 @@ public Map<ImStmt, Set<ImVar>> calculateLiveness(ImFunction func) {
ImStmt stmt = node.getStmt();
if (stmt == null) continue;

if (stmt instanceof ImVarargLoop loop) {
for (ImVarargLoopVar loopVar : loop.getLoopVars()) {
if (!loopVar.getVar().isGlobal()) {
def[i].add(loopVar.getVar());
}
}
// The loop body has its own CFG nodes. Visiting it here would incorrectly
// classify all body reads as uses at the loop header.
continue;
}

final int ii = i;
stmt.accept(new ImStmt.DefaultVisitor() {
@Override public void visit(ImVarAccess va) {
Expand Down
Loading
Loading