Merge pull request #2 from fluencelabs/logger

Fix logger and return C example

build.gradle

@@ -2,8 +2,8 @@ group 'asmble'
 version '0.2.0'
 
 buildscript {
-    ext.kotlin_version = '1.2.51'
+    ext.kotlin_version = '1.2.61'
-    ext.asm_version = '5.2'
+    ext.asm_version = '6.2.1'
 
     repositories {
         mavenCentral()
@@ -21,7 +21,11 @@ buildscript {
 allprojects {
     apply plugin: 'java'
     group 'com.github.cretz.asmble'
-    version '0.4.0-fl-fix'
+    version '0.4.1-fl'
+
+    // skips building and running for the specified examples
+    ext.skipExamples = ['c-simple', 'go-simple', 'rust-regex']
+    // todo disabling Rust regex is temporary because some strings in wasm code exceed the size in 65353 bytes.
 
     repositories {
         mavenCentral()
@@ -54,9 +58,9 @@ project(':compiler') {
         compile "org.jetbrains.kotlin:kotlin-reflect:$kotlin_version"
         compile "org.ow2.asm:asm-tree:$asm_version"
         compile "org.ow2.asm:asm-util:$asm_version"
+        compile "org.ow2.asm:asm-commons:$asm_version"
         testCompile 'junit:junit:4.12'
         testCompile "org.jetbrains.kotlin:kotlin-test-junit:$kotlin_version"
-        testCompile "org.ow2.asm:asm-debug-all:$asm_version"
     }
 
     publishSettings(project, 'asmble-compiler', 'Asmble WASM Compiler')
@@ -67,6 +71,38 @@ project(':examples') {
         dependencies {
             compileOnly project(':compiler')
         }
+        
+        // C/C++ example helpers
+
+        task cToWasm {
+            doFirst {
+                mkdir 'build'
+                exec {
+                    def cFileName = fileTree(dir: 'src', includes: ['*.c']).files.iterator().next()
+                    commandLine 'clang', '--target=wasm32-unknown-unknown-wasm', '-O3', cFileName, '-c', '-o', 'build/lib.wasm'
+                }
+            }
+        }
+
+        task showCWast(type: JavaExec) {
+            dependsOn cToWasm
+            classpath configurations.compileClasspath
+            main = 'asmble.cli.MainKt'
+            doFirst {
+                args 'translate', 'build/lib.wasm'
+            }
+        }
+
+        task compileCWasm(type: JavaExec) {
+            dependsOn cToWasm
+            classpath configurations.compileClasspath
+            main = 'asmble.cli.MainKt'
+            doFirst {
+                def outFile = 'build/wasm-classes/' + wasmCompiledClassName.replace('.', '/') + '.class'
+                file(outFile).parentFile.mkdirs()
+                args 'compile', 'build/lib.wasm', wasmCompiledClassName, '-out', outFile
+            }
+        }
 
         // Go example helpers
 
@@ -142,45 +178,82 @@ project(':examples') {
     }
 }
 
+project(':examples:c-simple') {
+    if (project.name in skipExamples) {
+        println("[Note!] Building and runnig for ${project.name} was skipped")
+        test.onlyIf { false } // explicit skipping tests
+        compileJava.onlyIf { false } // explicit skipping compile
+        return
+    }
 
-//project(':examples:go-simple') {
+    apply plugin: 'application'
-//    apply plugin: 'application'
+    ext.wasmCompiledClassName = 'asmble.generated.CSimple'
-//    ext.wasmCompiledClassName = 'asmble.generated.GoSimple'
+    dependencies {
-//    dependencies {
+        compile files('build/wasm-classes')
-//        compile files('build/wasm-classes')
+    }
-//    }
-//    compileJava {
-//        dependsOn compileGoWasm
-//    }
-//    mainClassName = 'asmble.examples.gosimple.Main'
-//}
 
-// todo temporary disable Rust regex, because some strings in wasm code exceed the size in 65353 bytes.
+    compileJava {
+        dependsOn compileCWasm
+    }
+    mainClassName = 'asmble.examples.csimple.Main'
+}
 
-// project(':examples:rust-regex') {
+project(':examples:go-simple') {
-//     apply plugin: 'application'
+    if (project.name in skipExamples) {
-//     apply plugin: 'me.champeau.gradle.jmh'
+        println("[Note!] Building and runnig for ${project.name} was skipped")
-//     ext.wasmCompiledClassName = 'asmble.generated.RustRegex'
+        test.onlyIf { false } // explicit skipping tests
-//     dependencies {
+        compileJava.onlyIf { false } // explicit skipping compile
-//         compile files('build/wasm-classes')
+        return
-//         testCompile 'junit:junit:4.12'
+    }
-//     }
+    apply plugin: 'application'
-//     compileJava {
+    ext.wasmCompiledClassName = 'asmble.generated.GoSimple'
-//         dependsOn compileRustWasm
+    dependencies {
-//     }
+        compile files('build/wasm-classes')
-//     mainClassName = 'asmble.examples.rustregex.Main'
+    }
-//     test {
+    compileJava {
-//         testLogging.showStandardStreams = true
+        dependsOn compileGoWasm
-//         testLogging.events 'PASSED', 'SKIPPED'
+    }
-//     }
+    mainClassName = 'asmble.examples.gosimple.Main'
-//     jmh {
+}
-//         iterations = 5
+
-//         warmupIterations = 5
+ project(':examples:rust-regex') {
-//         fork = 3
+     if (project.name in skipExamples) {
-//     }
+         println("[Note!] Building and runnig for ${project.name} was skipped")
-// }
+         test.onlyIf { false } // explicit skipping tests
+         compileJava.onlyIf { false } // explicit skipping compile
+         compileTestJava.onlyIf { false } // explicit skipping compile
+         return
+     }
+     apply plugin: 'application'
+     apply plugin: 'me.champeau.gradle.jmh'
+     ext.wasmCompiledClassName = 'asmble.generated.RustRegex'
+     dependencies {
+         compile files('build/wasm-classes')
+         testCompile 'junit:junit:4.12'
+     }
+     compileJava {
+         dependsOn compileRustWasm
+     }
+     mainClassName = 'asmble.examples.rustregex.Main'
+
+     test {
+         testLogging.showStandardStreams = true
+         testLogging.events 'PASSED', 'SKIPPED'
+     }
+     jmh {
+         iterations = 5
+         warmupIterations = 5
+         fork = 3
+     }
+ }
 
 project(':examples:rust-simple') {
+    if (project.name in skipExamples) {
+        println("[Note!] Building and runnig for ${project.name} was skipped")
+        test.onlyIf { false } // explicit skipping tests
+        compileJava.onlyIf { false } // explicit skipping compile
+        return
+    }
     apply plugin: 'application'
     ext.wasmCompiledClassName = 'asmble.generated.RustSimple'
     dependencies {
@@ -193,6 +266,12 @@ project(':examples:rust-simple') {
 }
 
 project(':examples:rust-string') {
+    if (project.name in skipExamples) {
+        println("[Note!] Building and runnig for ${project.name} was skipped")
+        test.onlyIf { false } // explicit skipping tests
+        compileJava.onlyIf { false } // explicit skipping compile
+        return
+    }
     apply plugin: 'application'
     ext.wasmCompiledClassName = 'asmble.generated.RustString'
     dependencies {

compiler/src/main/java/asmble/compile/jvm/msplit/README.md

@@ -0,0 +1 @@
+Taken from https://github.com/cretz/msplit

compiler/src/main/java/asmble/compile/jvm/msplit/SplitMethod.java

@@ -0,0 +1,286 @@
+package asmble.compile.jvm.msplit;
+
+
+import org.objectweb.asm.Label;
+import org.objectweb.asm.Opcodes;
+import org.objectweb.asm.Type;
+import org.objectweb.asm.tree.*;
+
+import java.util.*;
+
+import static asmble.compile.jvm.msplit.Util.*;
+
+/** Splits a method into two */
+public class SplitMethod {
+
+  protected final int api;
+
+  /** @param api Same as for {@link org.objectweb.asm.MethodVisitor#MethodVisitor(int)} or any other ASM class */
+  public SplitMethod(int api) { this.api = api; }
+
+  /**
+   * Calls {@link #split(String, MethodNode, int, int, int)} with minSize as 20% + 1 of the original, maxSize as
+   * 70% + 1 of the original, and firstAtLeast as maxSize. The original method is never modified and the result can
+   * be null if no split points are found.
+   */
+  public Result split(String owner, MethodNode method) {
+    // Between 20% + 1 and 70% + 1 of size
+    int insnCount = method.instructions.size();
+    int minSize = (int) (insnCount * 0.2) + 1;
+    int maxSize = (int) (insnCount * 0.7) + 1;
+    return split(owner, method, minSize, maxSize, maxSize);
+  }
+
+  /**
+   * Splits the given method into two. This uses a {@link Splitter} to consistently create
+   * {@link asmble.compile.jvm.msplit.Splitter.SplitPoint}s until one reaches firstAtLeast or the largest otherwise, and then calls
+   * {@link #fromSplitPoint(String, MethodNode, Splitter.SplitPoint)}.
+   *
+   * @param owner The internal name of the owning class. Needed when splitting to call the split off method.
+   * @param method The method to split, never modified
+   * @param minSize The minimum number of instructions the split off method must have
+   * @param maxSize The maximum number of instructions the split off method can have
+   * @param firstAtLeast The number of instructions that, when first reached, will immediately be used without
+   *                     continuing. Since split points are streamed, this allows splitting without waiting to
+   *                     find the largest overall. If this is &lt= 0, it will not apply and all split points will be
+   *                     checked to find the largest before doing the split.
+   * @return The resulting split method or null if there were no split points found
+   */
+  public Result split(String owner, MethodNode method, int minSize, int maxSize, int firstAtLeast) {
+    // Get the largest split point
+    Splitter.SplitPoint largest = null;
+    for (Splitter.SplitPoint point : new Splitter(api, owner, method, minSize, maxSize)) {
+      if (largest == null || point.length > largest.length) {
+        largest = point;
+        // Early exit?
+        if (firstAtLeast > 0 && largest.length >= firstAtLeast) break;
+      }
+    }
+    if (largest == null) return null;
+    return fromSplitPoint(owner, method, largest);
+  }
+
+  /**
+   * Split the given method at the given split point. Called by {@link #split(String, MethodNode, int, int, int)}. The
+   * original method is never modified.
+   */
+  public Result fromSplitPoint(String owner, MethodNode orig, Splitter.SplitPoint splitPoint) {
+    MethodNode splitOff = createSplitOffMethod(orig, splitPoint);
+    MethodNode trimmed = createTrimmedMethod(owner, orig, splitOff, splitPoint);
+    return new Result(trimmed, splitOff);
+  }
+
+  protected MethodNode createSplitOffMethod(MethodNode orig, Splitter.SplitPoint splitPoint) {
+    // The new method is a static synthetic method named method.name + "$split" that returns an object array
+    // Key is previous local index, value is new local index
+    Map<Integer, Integer> localsMap = new HashMap<>();
+    // The new method's parameters are all stack items + all read locals
+    List<Type> args = new ArrayList<>(splitPoint.neededFromStackAtStart);
+    splitPoint.localsRead.forEach((index, type) -> {
+      args.add(type);
+      localsMap.put(index, args.size() - 1);
+    });
+    // Create the new method
+    MethodNode newMethod = new MethodNode(api,
+        Opcodes.ACC_STATIC + Opcodes.ACC_PRIVATE + Opcodes.ACC_SYNTHETIC, orig.name + "$split",
+        Type.getMethodDescriptor(Type.getType(Object[].class), args.toArray(new Type[0])), null, null);
+    // Add the written locals to the map that are not already there
+    int newLocalIndex = args.size();
+    for (Integer key : splitPoint.localsWritten.keySet()) {
+      if (!localsMap.containsKey(key)) {
+        localsMap.put(key, newLocalIndex);
+        newLocalIndex++;
+      }
+    }
+    // First set of instructions is pushing the new stack from the params
+    for (int i = 0; i < splitPoint.neededFromStackAtStart.size(); i++) {
+      Type item = splitPoint.neededFromStackAtStart.get(i);
+      newMethod.visitVarInsn(loadOpFromType(item), i);
+    }
+    // Next set of instructions comes verbatim from the original, but we have to change the local indexes
+    Set<Label> seenLabels = new HashSet<>();
+    for (int i = 0; i < splitPoint.length; i++) {
+      AbstractInsnNode insn = orig.instructions.get(i + splitPoint.start);
+      // Skip frames
+      if (insn instanceof FrameNode) continue;
+      // Store the label
+      if (insn instanceof LabelNode) seenLabels.add(((LabelNode) insn).getLabel());
+      // Change the local if needed
+      if (insn instanceof VarInsnNode) {
+        insn = insn.clone(Collections.emptyMap());
+        ((VarInsnNode) insn).var = localsMap.get(((VarInsnNode) insn).var);
+      } else if (insn instanceof IincInsnNode) {
+        insn = insn.clone(Collections.emptyMap());
+        ((VarInsnNode) insn).var = localsMap.get(((VarInsnNode) insn).var);
+      }
+      insn.accept(newMethod);
+    }
+    // Final set of instructions is an object array of stack to set and then locals written
+    // Create the object array
+    int retArrSize = splitPoint.putOnStackAtEnd.size() + splitPoint.localsWritten.size();
+    intConst(retArrSize).accept(newMethod);
+    newMethod.visitTypeInsn(Opcodes.ANEWARRAY, OBJECT_TYPE.getInternalName());
+    // So, we're going to store the arr in the next avail local
+    int retArrLocalIndex = newLocalIndex;
+    newMethod.visitVarInsn(Opcodes.ASTORE, retArrLocalIndex);
+    // Now go over each stack item and load the arr, swap w/ the stack, add the index, swap with the stack, and store
+    for (int i = splitPoint.putOnStackAtEnd.size() - 1; i >= 0; i--) {
+      Type item = splitPoint.putOnStackAtEnd.get(i);
+      // Box the item on the stack if necessary
+      boxStackIfNecessary(item, newMethod);
+      // Load the array
+      newMethod.visitVarInsn(Opcodes.ALOAD, retArrLocalIndex);
+      // Swap to put stack back on top
+      newMethod.visitInsn(Opcodes.SWAP);
+      // Add the index
+      intConst(i).accept(newMethod);
+      // Swap to put the stack value back on top
+      newMethod.visitInsn(Opcodes.SWAP);
+      // Now that we have arr, index, value, we can store in the array
+      newMethod.visitInsn(Opcodes.AASTORE);
+    }
+    // Do the same with written locals
+    int currIndex = splitPoint.putOnStackAtEnd.size();
+    for (Integer index : splitPoint.localsWritten.keySet()) {
+      Type item = splitPoint.localsWritten.get(index);
+      // Load the array
+      newMethod.visitVarInsn(Opcodes.ALOAD, retArrLocalIndex);
+      // Add the arr index
+      intConst(currIndex).accept(newMethod);
+      currIndex++;
+      // Load the var
+      newMethod.visitVarInsn(loadOpFromType(item), localsMap.get(index));
+      // Box it if necessary
+      boxStackIfNecessary(item, newMethod);
+      // Store in array
+      newMethod.visitInsn(Opcodes.AASTORE);
+    }
+    // Load the array out and return it
+    newMethod.visitVarInsn(Opcodes.ALOAD, retArrLocalIndex);
+    newMethod.visitInsn(Opcodes.ARETURN);
+    // Any try catch blocks that start in here
+    for (TryCatchBlockNode tryCatch : orig.tryCatchBlocks) {
+      if (seenLabels.contains(tryCatch.start.getLabel())) tryCatch.accept(newMethod);
+    }
+    // Reset the labels
+    newMethod.instructions.resetLabels();
+    return newMethod;
+  }
+
+  protected MethodNode createTrimmedMethod(String owner, MethodNode orig,
+      MethodNode splitOff, Splitter.SplitPoint splitPoint) {
+    // The trimmed method is the same as the original, yet the split area is replaced with a call to the split off
+    // portion. Before calling the split-off, we have to add locals to the stack part. Then afterwards, we have to
+    // replace the stack and written locals.
+    // Effectively clone the orig
+    MethodNode newMethod = new MethodNode(api, orig.access, orig.name, orig.desc,
+        orig.signature, orig.exceptions.toArray(new String[0]));
+    orig.accept(newMethod);
+    // Remove all insns, we'll re-add the ones outside the split range
+    newMethod.instructions.clear();
+    // Remove all try catch blocks and keep track of seen labels, we'll re-add them at the end
+    newMethod.tryCatchBlocks.clear();
+    Set<Label> seenLabels = new HashSet<>();
+    // Also keep track of the locals that have been stored, need to know
+    Set<Integer> seenStoredLocals = new HashSet<>();
+    // If this is an instance method, we consider "0" (i.e. "this") as seen
+    if ((orig.access & Opcodes.ACC_STATIC) == 0) seenStoredLocals.add(0);
+    // Add the insns before split
+    for (int i = 0; i < splitPoint.start; i++) {
+      AbstractInsnNode insn = orig.instructions.get(i + splitPoint.start);
+      // Skip frames
+      if (insn instanceof FrameNode) continue;
+      // Record label
+      if (insn instanceof LabelNode) seenLabels.add(((LabelNode) insn).getLabel());
+      // Check a local store has happened
+      if (insn instanceof VarInsnNode && isStoreOp(insn.getOpcode())) seenStoredLocals.add(((VarInsnNode) insn).var);
+      insn.accept(newMethod);
+    }
+    // Push all the read locals on the stack
+    splitPoint.localsRead.forEach((index, type) -> {
+      // We've seen a store for this, so just load it, otherwise use a zero val
+      // TODO: safe? if not, maybe just put at the top of the method a bunch of defaulted locals?
+      if (seenStoredLocals.contains(index)) newMethod.visitVarInsn(loadOpFromType(type), index);
+      else zeroVal(type).accept(newMethod);
+    });
+    // Invoke the split off method
+    newMethod.visitMethodInsn(Opcodes.INVOKESTATIC, owner, splitOff.name, splitOff.desc, false);
+    // Now the object array is on the stack which contains stack pieces + written locals
+    // Take off the locals
+    int localArrIndex = splitPoint.putOnStackAtEnd.size();
+    for (Integer index : splitPoint.localsWritten.keySet()) {
+      // Dupe the array
+      newMethod.visitInsn(Opcodes.DUP);
+      // Put the index on the stack
+      intConst(localArrIndex).accept(newMethod);
+      localArrIndex++;
+      // Load the written local
+      Type item = splitPoint.localsWritten.get(index);
+      newMethod.visitInsn(Opcodes.AALOAD);
+      // Cast to local type
+      if (!item.equals(OBJECT_TYPE)) {
+        newMethod.visitTypeInsn(Opcodes.CHECKCAST, boxedTypeIfNecessary(item).getInternalName());
+      }
+      // Unbox if necessary
+      unboxStackIfNecessary(item, newMethod);
+      // Store in the local
+      newMethod.visitVarInsn(storeOpFromType(item), index);
+    }
+    // Now just load up the stack
+    for (int i = 0; i < splitPoint.putOnStackAtEnd.size(); i++) {
+      boolean last = i == splitPoint.putOnStackAtEnd.size() - 1;
+      // Since the loop started with the array, we only dupe the array every time but the last
+      if (!last) newMethod.visitInsn(Opcodes.DUP);
+      // Put the index on the stack
+      intConst(i).accept(newMethod);
+      // Load the stack item
+      Type item = splitPoint.putOnStackAtEnd.get(i);
+      newMethod.visitInsn(Opcodes.AALOAD);
+      // Cast to local type
+      if (!item.equals(OBJECT_TYPE)) {
+        newMethod.visitTypeInsn(Opcodes.CHECKCAST, boxedTypeIfNecessary(item).getInternalName());
+      }
+      // Unbox if necessary
+      unboxStackIfNecessary(item, newMethod);
+      // For all but the last stack item, we need to swap with the arr ref above.
+      if (!last) {
+        // Note if the stack item takes two slots, we do a form of dup then pop since there's no swap1x2
+        if (item == Type.LONG_TYPE || item == Type.DOUBLE_TYPE) {
+          newMethod.visitInsn(Opcodes.DUP_X2);
+          newMethod.visitInsn(Opcodes.POP);
+        } else {
+          newMethod.visitInsn(Opcodes.SWAP);
+        }
+      }
+    }
+    // Now we have restored all locals and all stack...add the rest of the insns after the split
+    for (int i = splitPoint.start + splitPoint.length; i < orig.instructions.size(); i++) {
+      AbstractInsnNode insn = orig.instructions.get(i + splitPoint.start);
+      // Skip frames
+      if (insn instanceof FrameNode) continue;
+      // Record label
+      if (insn instanceof LabelNode) seenLabels.add(((LabelNode) insn).getLabel());
+      insn.accept(newMethod);
+    }
+    // Add any try catch blocks that started in here
+    for (TryCatchBlockNode tryCatch : orig.tryCatchBlocks) {
+      if (seenLabels.contains(tryCatch.start.getLabel())) tryCatch.accept(newMethod);
+    }
+    // Reset the labels
+    newMethod.instructions.resetLabels();
+    return newMethod;
+  }
+
+  /** Result of a split method */
+  public static class Result {
+    /** A copy of the original method, but changed to invoke {@link #splitOffMethod} */
+    public final MethodNode trimmedMethod;
+    /** The new method that was split off the original and is called by {@link #splitOffMethod} */
+    public final MethodNode splitOffMethod;
+
+    public Result(MethodNode trimmedMethod, MethodNode splitOffMethod) {
+      this.trimmedMethod = trimmedMethod;
+      this.splitOffMethod = splitOffMethod;
+    }
+  }
+}

compiler/src/main/java/asmble/compile/jvm/msplit/Splitter.java

@@ -0,0 +1,392 @@
+package asmble.compile.jvm.msplit;
+
+import org.objectweb.asm.Label;
+import org.objectweb.asm.Opcodes;
+import org.objectweb.asm.Type;
+import org.objectweb.asm.commons.AnalyzerAdapter;
+import org.objectweb.asm.tree.*;
+
+import java.util.*;
+
+import static asmble.compile.jvm.msplit.Util.*;
+
+/** For a given method, iterate over possible split points */
+public class Splitter implements Iterable<Splitter.SplitPoint> {
+  protected final int api;
+  protected final String owner;
+  protected final MethodNode method;
+  protected final int minSize;
+  protected final int maxSize;
+
+  /**
+   * @param api Same as for {@link org.objectweb.asm.MethodVisitor#MethodVisitor(int)} or any other ASM class
+   * @param owner Internal name of the method's owner
+   * @param method The method to find split points for
+   * @param minSize The minimum number of instructions required for the split point to be valid
+   * @param maxSize The maximum number of instructions that split points cannot exceeed
+   */
+  public Splitter(int api, String owner, MethodNode method, int minSize, int maxSize) {
+    this.api = api;
+    this.owner = owner;
+    this.method = method;
+    this.minSize = minSize;
+    this.maxSize = maxSize;
+  }
+
+  @Override
+  public Iterator<SplitPoint> iterator() { return new Iter(); }
+
+  // Types are always int, float, long, double, or ref (no other primitives)
+  /** A split point in a method that can be split off into another method */
+  public static class SplitPoint {
+    /**
+     * The locals read in this split area, keyed by index. Value type is always int, float, long, double, or object.
+     */
+    public final SortedMap<Integer, Type> localsRead;
+    /**
+     * The locals written in this split area, keyed by index. Value type is always int, float, long, double, or object.
+     */
+    public final SortedMap<Integer, Type> localsWritten;
+    /**
+     * The values of the stack needed at the start of this split area. Type is always int, float, long, double, or
+     * object.
+     */
+    public final List<Type> neededFromStackAtStart;
+    /**
+     * The values of the stack at the end of this split area that are needed to put back on the original. Type is always
+     * int, float, long, double, or object.
+     */
+    public final List<Type> putOnStackAtEnd;
+    /**
+     * The instruction index this split area begins at.
+     */
+    public final int start;
+    /**
+     * The number of instructions this split area has.
+     */
+    public final int length;
+
+    public SplitPoint(SortedMap<Integer, Type> localsRead, SortedMap<Integer, Type>localsWritten,
+        List<Type> neededFromStackAtStart, List<Type> putOnStackAtEnd, int start, int length) {
+      this.localsRead = localsRead;
+      this.localsWritten = localsWritten;
+      this.neededFromStackAtStart = neededFromStackAtStart;
+      this.putOnStackAtEnd = putOnStackAtEnd;
+      this.start = start;
+      this.length = length;
+    }
+  }
+
+  protected int compareInsnIndexes(AbstractInsnNode o1, AbstractInsnNode o2) {
+    return Integer.compare(method.instructions.indexOf(o1), method.instructions.indexOf(o2));
+  }
+
+  protected class Iter implements Iterator<SplitPoint> {
+    protected final AbstractInsnNode[] insns;
+    protected final List<TryCatchBlockNode> tryCatchBlocks;
+    protected int currIndex = -1;
+    protected boolean peeked;
+    protected SplitPoint peekedValue;
+
+    protected Iter() {
+      insns = method.instructions.toArray();
+      tryCatchBlocks = new ArrayList<>(method.tryCatchBlocks);
+      // Must be sorted by earliest starting index then earliest end index then earliest handler
+      tryCatchBlocks.sort((o1, o2) -> {
+        int cmp = compareInsnIndexes(o1.start, o2.start);
+        if (cmp == 0) compareInsnIndexes(o1.end, o2.end);
+        if (cmp == 0) compareInsnIndexes(o1.handler, o2.handler);
+        return cmp;
+      });
+    }
+
+    @Override
+    public boolean hasNext() {
+      if (!peeked) {
+        peeked = true;
+        peekedValue = nextOrNull();
+      }
+      return peekedValue != null;
+    }
+
+    @Override
+    public SplitPoint next() {
+      // If we've peeked in hasNext, use that
+      SplitPoint ret;
+      if (peeked) {
+        peeked = false;
+        ret = peekedValue;
+      } else {
+        ret = nextOrNull();
+      }
+      if (ret == null) throw new NoSuchElementException();
+      return ret;
+    }
+
+    protected SplitPoint nextOrNull() {
+      // Try for each index
+      while (++currIndex + minSize <= insns.length) {
+        SplitPoint longest = longestForCurrIndex();
+        if (longest != null) return longest;
+      }
+      return null;
+    }
+
+    protected SplitPoint longestForCurrIndex() {
+      // As a special case, if the previous insn was a line number, that was good enough
+      if (currIndex - 1 >- 0 && insns[currIndex - 1] instanceof LineNumberNode) return null;
+      // Build the info object
+      InsnTraverseInfo info = new InsnTraverseInfo();
+      info.startIndex = currIndex;
+      info.endIndex = Math.min(currIndex + maxSize - 1, insns.length - 1);
+      // Reduce the end based on try/catch blocks the start is in or that jump to
+      constrainEndByTryCatchBlocks(info);
+      // Reduce the end based on any jumps within
+      constrainEndByInternalJumps(info);
+      // Reduce the end based on any jumps into
+      constrainEndByExternalJumps(info);
+      // Make sure we didn't reduce the end too far
+      if (info.getSize() < minSize) return null;
+      // Now that we have our largest range from the start index, we can go over each updating the local refs and stack
+      // For the stack, we are going to use the
+      return splitPointFromInfo(info);
+    }
+
+    protected void constrainEndByTryCatchBlocks(InsnTraverseInfo info) {
+      // Go over all the try/catch blocks, sorted by earliest
+      for (TryCatchBlockNode block : tryCatchBlocks) {
+        int handleIndex = method.instructions.indexOf(block.handler);
+        int startIndex = method.instructions.indexOf(block.start);
+        int endIndex = method.instructions.indexOf(block.end) - 1;
+        boolean catchWithinDisallowed;
+
+        if (info.startIndex <= startIndex && info.endIndex >= endIndex) {
+          // The try block is entirely inside the range...
+          catchWithinDisallowed = false;
+          // Since it's entirely within, we need the catch handler within too
+          if (handleIndex < info.startIndex || handleIndex > info.endIndex) {
+            // Well, it's not within, so that means we can't include this try block at all
+            info.endIndex = Math.min(info.endIndex, startIndex - 1);
+          }
+        } else if (info.startIndex > startIndex && info.endIndex > endIndex) {
+          // The try block started before this range, but ends inside of it...
+          // The end has to be changed to the block's end so it doesn't go over the boundary
+          info.endIndex = Math.min(info.endIndex, endIndex);
+          // The catch can't jump in here
+          catchWithinDisallowed = true;
+        } else if (info.startIndex <= startIndex && info.endIndex < endIndex) {
+          // The try block started in this range, but ends outside of it...
+          // Can't have the block then, reduce it to before the start
+          info.endIndex = Math.min(info.endIndex, startIndex - 1);
+          // Since we don't have the block, we can't jump in here either
+          catchWithinDisallowed = true;
+        } else {
+          // The try block is completely outside, just restrict the catch from jumping in
+          catchWithinDisallowed = true;
+        }
+        // If the catch is within and not allowed to be, we have to change the end to before it
+        if (catchWithinDisallowed && info.startIndex <= handleIndex && info.endIndex >= handleIndex) {
+          info.endIndex = Math.min(info.endIndex, handleIndex - 1);
+        }
+      }
+    }
+
+    protected void constrainEndByInternalJumps(InsnTraverseInfo info) {
+      for (int i = info.startIndex; i <= info.endIndex; i++) {
+        AbstractInsnNode node = insns[i];
+        int earliestIndex;
+        int furthestIndex;
+        if (node instanceof JumpInsnNode) {
+          earliestIndex = method.instructions.indexOf(((JumpInsnNode) node).label);
+          furthestIndex = earliestIndex;
+        } else if (node instanceof TableSwitchInsnNode) {
+          earliestIndex = method.instructions.indexOf(((TableSwitchInsnNode) node).dflt);
+          furthestIndex = earliestIndex;
+          for (LabelNode label : ((TableSwitchInsnNode) node).labels) {
+            int index = method.instructions.indexOf(label);
+            earliestIndex = Math.min(earliestIndex, index);
+            furthestIndex = Math.max(furthestIndex, index);
+          }
+        } else if (node instanceof LookupSwitchInsnNode) {
+          earliestIndex = method.instructions.indexOf(((LookupSwitchInsnNode) node).dflt);
+          furthestIndex = earliestIndex;
+          for (LabelNode label : ((LookupSwitchInsnNode) node).labels) {
+            int index = method.instructions.indexOf(label);
+            earliestIndex = Math.min(earliestIndex, index);
+            furthestIndex = Math.max(furthestIndex, index);
+          }
+        } else continue;
+        // Stop here if any indexes are out of range, otherwise, change end
+        if (earliestIndex < info.startIndex || furthestIndex > info.endIndex) {
+          info.endIndex = i - 1;
+          return;
+        }
+        info.endIndex = Math.max(info.endIndex, furthestIndex);
+      }
+    }
+
+    protected void constrainEndByExternalJumps(InsnTraverseInfo info) {
+      // Basically, if any external jumps jump into our range, that can't be included in the range
+      for (int i = 0; i < insns.length; i++) {
+        if (i >= info.startIndex && i <= info.endIndex) continue;
+        AbstractInsnNode node = insns[i];
+        if (node instanceof JumpInsnNode) {
+          int index = method.instructions.indexOf(((JumpInsnNode) node).label);
+          if (index >= info.startIndex) info.endIndex = Math.min(info.endIndex, index - 1);
+        } else if (node instanceof TableSwitchInsnNode) {
+          int index = method.instructions.indexOf(((TableSwitchInsnNode) node).dflt);
+          if (index >= info.startIndex) info.endIndex = Math.min(info.endIndex, index - 1);
+          for (LabelNode label : ((TableSwitchInsnNode) node).labels) {
+            index = method.instructions.indexOf(label);
+            if (index >= info.startIndex) info.endIndex = Math.min(info.endIndex, index - 1);
+          }
+        } else if (node instanceof LookupSwitchInsnNode) {
+          int index = method.instructions.indexOf(((LookupSwitchInsnNode) node).dflt);
+          if (index >= info.startIndex) info.endIndex = Math.min(info.endIndex, index - 1);
+          for (LabelNode label : ((LookupSwitchInsnNode) node).labels) {
+            index = method.instructions.indexOf(label);
+            if (index >= info.startIndex) info.endIndex = Math.min(info.endIndex, index - 1);
+          }
+        }
+      }
+    }
+
+    protected SplitPoint splitPointFromInfo(InsnTraverseInfo info) {
+      // We're going to use the analyzer adapter and run it for the up until the end, a step at a time
+      StackAndLocalTrackingAdapter adapter = new StackAndLocalTrackingAdapter(Splitter.this);
+      // Visit all of the insns up our start.
+      // XXX: I checked the source of AnalyzerAdapter to confirm I don't need any of the surrounding stuff
+      for (int i = 0; i < info.startIndex; i++) insns[i].accept(adapter);
+      // Take the stack at the start and copy it off
+      List<Object> stackAtStart = new ArrayList<>(adapter.stack);
+      // Reset some adapter state
+      adapter.lowestStackSize = stackAtStart.size();
+      adapter.localsRead.clear();
+      adapter.localsWritten.clear();
+      // Now go over the remaining range
+      for (int i = info.startIndex; i <= info.endIndex; i++) insns[i].accept(adapter);
+      // Build the split point
+      return new SplitPoint(
+          localMapFromAdapterLocalMap(adapter.localsRead, adapter.uninitializedTypes),
+          localMapFromAdapterLocalMap(adapter.localsWritten, adapter.uninitializedTypes),
+          typesFromAdapterStackRange(stackAtStart, adapter.lowestStackSize, adapter.uninitializedTypes),
+          typesFromAdapterStackRange(adapter.stack, adapter.lowestStackSize, adapter.uninitializedTypes),
+          info.startIndex,
+          info.getSize()
+      );
+    }
+
+    protected SortedMap<Integer, Type> localMapFromAdapterLocalMap(
+        SortedMap<Integer, Object> map, Map<Object, Object> uninitializedTypes) {
+      SortedMap<Integer, Type> ret = new TreeMap<>();
+      map.forEach((k, v) -> ret.put(k, typeFromAdapterStackItem(v, uninitializedTypes)));
+      return ret;
+    }
+
+    protected List<Type> typesFromAdapterStackRange(
+        List<Object> stack, int start, Map<Object, Object> uninitializedTypes) {
+      List<Type> ret = new ArrayList<>();
+      for (int i = start; i < stack.size(); i++) {
+        Object item = stack.get(i);
+        ret.add(typeFromAdapterStackItem(item, uninitializedTypes));
+        // Jump an extra spot for longs and doubles
+        if (item == Opcodes.LONG || item == Opcodes.DOUBLE) {
+          if (stack.get(++i) != Opcodes.TOP) throw new IllegalStateException("Expected top after long/double");
+        }
+      }
+      return ret;
+    }
+
+    protected Type typeFromAdapterStackItem(Object item, Map<Object, Object> uninitializedTypes) {
+      if (item == Opcodes.INTEGER) return Type.INT_TYPE;
+      else if (item == Opcodes.FLOAT) return Type.FLOAT_TYPE;
+      else if (item == Opcodes.LONG) return Type.LONG_TYPE;
+      else if (item == Opcodes.DOUBLE) return Type.DOUBLE_TYPE;
+      else if (item == Opcodes.NULL) return OBJECT_TYPE;
+      else if (item == Opcodes.UNINITIALIZED_THIS) return Type.getObjectType(owner);
+      else if (item instanceof Label) return Type.getObjectType((String) uninitializedTypes.get(item));
+      else if (item instanceof String) return Type.getObjectType((String) item);
+      else throw new IllegalStateException("Unrecognized stack item: " + item);
+    }
+  }
+
+  protected static class StackAndLocalTrackingAdapter extends AnalyzerAdapter {
+    public int lowestStackSize;
+    public final SortedMap<Integer, Object> localsRead = new TreeMap<>();
+    public final SortedMap<Integer, Object> localsWritten = new TreeMap<>();
+
+    protected StackAndLocalTrackingAdapter(Splitter splitter) {
+      super(splitter.api, splitter.owner, splitter.method.access, splitter.method.name, splitter.method.desc, null);
+      stack = new SizeChangeNotifyList<Object>() {
+        @Override
+        protected void onSizeChanged() { lowestStackSize = Math.min(lowestStackSize, size()); }
+      };
+    }
+
+    @Override
+    public void visitVarInsn(int opcode, int var) {
+      switch (opcode) {
+        case Opcodes.ILOAD:
+        case Opcodes.LLOAD:
+        case Opcodes.FLOAD:
+        case Opcodes.DLOAD:
+        case Opcodes.ALOAD:
+          localsRead.put(var, locals.get(var));
+          break;
+        case Opcodes.ISTORE:
+        case Opcodes.FSTORE:
+        case Opcodes.ASTORE:
+          localsWritten.put(var, stack.get(stack.size() - 1));
+          break;
+        case Opcodes.LSTORE:
+        case Opcodes.DSTORE:
+          localsWritten.put(var, stack.get(stack.size() - 2));
+          break;
+      }
+      super.visitVarInsn(opcode, var);
+    }
+
+    @Override
+    public void visitIincInsn(int var, int increment) {
+      localsRead.put(var, Type.INT_TYPE);
+      localsWritten.put(var, Type.INT_TYPE);
+      super.visitIincInsn(var, increment);
+    }
+  }
+
+  protected static class SizeChangeNotifyList<T> extends AbstractList<T> {
+    protected final ArrayList<T> list = new ArrayList<>();
+
+    protected void onSizeChanged() { }
+
+    @Override
+    public T get(int index) { return list.get(index); }
+
+    @Override
+    public int size() { return list.size(); }
+
+    @Override
+    public T set(int index, T element) { return list.set(index, element); }
+
+    @Override
+    public void add(int index, T element) {
+      list.add(index, element);
+      onSizeChanged();
+    }
+
+    @Override
+    public T remove(int index) {
+      T ret = list.remove(index);
+      onSizeChanged();
+      return ret;
+    }
+  }
+
+  protected static class InsnTraverseInfo {
+    public int startIndex;
+    // Can only shrink, never increase in size
+    public int endIndex;
+
+    public int getSize() { return endIndex - startIndex + 1; }
+  }
+}

compiler/src/main/java/asmble/compile/jvm/msplit/Util.java

@@ -0,0 +1,84 @@
+package asmble.compile.jvm.msplit;
+
+import org.objectweb.asm.Opcodes;
+import org.objectweb.asm.Type;
+import org.objectweb.asm.tree.*;
+
+class Util {
+  private Util() { }
+
+  static final Type OBJECT_TYPE = Type.getType(Object.class);
+
+  static AbstractInsnNode zeroVal(Type type) {
+    if (type == Type.INT_TYPE) return new InsnNode(Opcodes.ICONST_0);
+    else if (type == Type.LONG_TYPE) return new InsnNode(Opcodes.LCONST_0);
+    else if (type == Type.FLOAT_TYPE) return new InsnNode(Opcodes.FCONST_0);
+    else if (type == Type.DOUBLE_TYPE) return new InsnNode(Opcodes.DCONST_0);
+    else return new InsnNode(Opcodes.ACONST_NULL);
+  }
+
+  static boolean isStoreOp(int opcode) {
+    return opcode == Opcodes.ISTORE || opcode == Opcodes.LSTORE || opcode == Opcodes.FSTORE ||
+        opcode == Opcodes.DSTORE || opcode == Opcodes.ASTORE;
+  }
+
+  static int storeOpFromType(Type type) {
+    if (type == Type.INT_TYPE) return Opcodes.ISTORE;
+    else if (type == Type.LONG_TYPE) return Opcodes.LSTORE;
+    else if (type == Type.FLOAT_TYPE) return Opcodes.FSTORE;
+    else if (type == Type.DOUBLE_TYPE) return Opcodes.DSTORE;
+    else return Opcodes.ASTORE;
+  }
+
+  static int loadOpFromType(Type type) {
+    if (type == Type.INT_TYPE) return Opcodes.ILOAD;
+    else if (type == Type.LONG_TYPE) return Opcodes.LLOAD;
+    else if (type == Type.FLOAT_TYPE) return Opcodes.FLOAD;
+    else if (type == Type.DOUBLE_TYPE) return Opcodes.DLOAD;
+    else return Opcodes.ALOAD;
+  }
+
+  static Type boxedTypeIfNecessary(Type type) {
+    if (type == Type.INT_TYPE) return Type.getType(Integer.class);
+    else if (type == Type.LONG_TYPE) return Type.getType(Long.class);
+    else if (type == Type.FLOAT_TYPE) return Type.getType(Float.class);
+    else if (type == Type.DOUBLE_TYPE) return Type.getType(Double.class);
+    else return type;
+  }
+
+  static void boxStackIfNecessary(Type type, MethodNode method) {
+    if (type == Type.INT_TYPE) boxCall(Integer.class, type).accept(method);
+    else if (type == Type.FLOAT_TYPE) boxCall(Float.class, type).accept(method);
+    else if (type == Type.LONG_TYPE) boxCall(Long.class, type).accept(method);
+    else if (type == Type.DOUBLE_TYPE) boxCall(Double.class, type).accept(method);
+  }
+
+  static void unboxStackIfNecessary(Type type, MethodNode method) {
+    if (type == Type.INT_TYPE) method.visitMethodInsn(Opcodes.INVOKEVIRTUAL,
+        "java/lang/Integer", "intValue", Type.getMethodDescriptor(Type.INT_TYPE), false);
+    else if (type == Type.FLOAT_TYPE) method.visitMethodInsn(Opcodes.INVOKEVIRTUAL,
+        "java/lang/Float", "floatValue", Type.getMethodDescriptor(Type.FLOAT_TYPE), false);
+    else if (type == Type.LONG_TYPE) method.visitMethodInsn(Opcodes.INVOKEVIRTUAL,
+        "java/lang/Long", "longValue", Type.getMethodDescriptor(Type.LONG_TYPE), false);
+    else if (type == Type.DOUBLE_TYPE) method.visitMethodInsn(Opcodes.INVOKEVIRTUAL,
+        "java/lang/Double", "doubleValue", Type.getMethodDescriptor(Type.DOUBLE_TYPE), false);
+  }
+
+  static AbstractInsnNode intConst(int v) {
+    switch (v) {
+      case -1: return new InsnNode(Opcodes.ICONST_M1);
+      case 0: return new InsnNode(Opcodes.ICONST_0);
+      case 1: return new InsnNode(Opcodes.ICONST_1);
+      case 2: return new InsnNode(Opcodes.ICONST_2);
+      case 3: return new InsnNode(Opcodes.ICONST_3);
+      case 4: return new InsnNode(Opcodes.ICONST_4);
+      case 5: return new InsnNode(Opcodes.ICONST_5);
+      default: return new LdcInsnNode(v);
+    }
+  }
+
+  static MethodInsnNode boxCall(Class<?> boxType, Type primType) {
+    return new MethodInsnNode(Opcodes.INVOKESTATIC, Type.getInternalName(boxType),
+        "valueOf", Type.getMethodDescriptor(Type.getType(boxType), primType), false);
+  }
+}

compiler/src/main/kotlin/asmble/ast/opt/SplitLargeFunc.kt

@@ -1,183 +0,0 @@
-package asmble.ast.opt
-
-import asmble.ast.Node
-import asmble.ast.Stack
-
-// This is a naive implementation that just grabs adjacent sets of restricted insns and breaks the one that will save
-// the most instructions off into its own function.
-open class SplitLargeFunc(
-    val minSetLength: Int = 5,
-    val maxSetLength: Int = 40,
-    val maxParamCount: Int = 30
-) {
-
-    // Null if no replacement. Second value is number of instructions saved. fnIndex must map to actual func,
-    // not imported one.
-    fun apply(mod: Node.Module, fnIndex: Int): Pair<Node.Module, Int>? {
-        // Get the func
-        val importFuncCount = mod.imports.count { it.kind is Node.Import.Kind.Func }
-        val actualFnIndex = fnIndex - importFuncCount
-        val func = mod.funcs.getOrElse(actualFnIndex) {
-            error("Unable to find non-import func at $fnIndex (actual $actualFnIndex)")
-        }
-
-        // Just take the best pattern and apply it
-        val newFuncIndex = importFuncCount + mod.funcs.size
-        return commonPatterns(mod, func).firstOrNull()?.let { pattern ->
-            // Name it as <funcname>$splitN (n is num just to disambiguate) if names are part of the mod
-            val newName = mod.names?.funcNames?.get(fnIndex)?.let {
-                "$it\$split".let { it + mod.names.funcNames.count { (_, v) -> v.startsWith(it) } }
-            }
-
-            // Go over every replacement in reverse, changing the instructions to our new set
-            val newInsns = pattern.replacements.foldRight(func.instructions) { repl, insns ->
-                insns.take(repl.range.start) +
-                    repl.preCallConsts +
-                    Node.Instr.Call(newFuncIndex) +
-                    insns.drop(repl.range.endInclusive + 1)
-            }
-
-            // Return the module w/ the new function, it's new name, and the insns saved
-            mod.copy(
-                funcs = mod.funcs.toMutableList().also {
-                    it[actualFnIndex] = func.copy(instructions = newInsns)
-                } + pattern.newFunc,
-                names = mod.names?.copy(funcNames = mod.names.funcNames.toMutableMap().also {
-                    it[newFuncIndex] = newName!!
-                })
-            ) to pattern.insnsSaved
-        }
-    }
-
-    // Results are by most insns saved. There can be overlap across patterns but never within a single pattern.
-    fun commonPatterns(mod: Node.Module, fn: Node.Func): List<CommonPattern> {
-        // Walk the stack for validation needs
-        val stack = Stack.walkStrict(mod, fn)
-
-        // Let's grab sets of insns that qualify. In this naive impl, in order to qualify the insn set needs to
-        // only have a certain set of insns that can be broken off. It can also only change the stack by 0 or 1
-        // value while never dipping below the starting stack. We also store the index they started at.
-        var insnSets = emptyList<InsnSet>()
-        // Pair in fold keyed by insn index
-        fn.instructions.foldIndexed(null as List<Pair<Int, Node.Instr>>?) { index, lastInsns, insn ->
-            if (!insn.canBeMoved) null else (lastInsns ?: emptyList()).plus(index to insn).also { fullNewInsnSet ->
-                // Get all final instructions between min and max size and with allowed param count (i.e. const count)
-                val trailingInsnSet = fullNewInsnSet.takeLast(maxSetLength)
-
-                // Get all instructions between the min and max
-                insnSets += (minSetLength..maxSetLength).
-                    asSequence().
-                    flatMap { trailingInsnSet.asSequence().windowed(it) }.
-                    filter { it.count { it.second is Node.Instr.Args.Const<*> } <= maxParamCount }.
-                    mapNotNull { newIndexedInsnSet ->
-                        // Before adding, make sure it qualifies with the stack
-                        InsnSet(
-                            startIndex = newIndexedInsnSet.first().first,
-                            insns = newIndexedInsnSet.map { it.second },
-                            valueAddedToStack = null
-                        ).withStackValueIfValid(stack)
-                    }
-            }
-        }
-        // Sort the insn sets by the ones with the most insns
-        insnSets = insnSets.sortedByDescending { it.insns.size }
-
-        // Now let's create replacements for each, keyed by the extracted func
-        val patterns = insnSets.fold(emptyMap<Node.Func, List<Replacement>>()) { map, insnSet ->
-            insnSet.extractCommonFunc().let { (func, replacement) ->
-                val existingReplacements = map.getOrDefault(func, emptyList())
-                // Ignore if there is any overlap
-                if (existingReplacements.any(replacement::overlaps)) map
-                else map + (func to existingReplacements.plus(replacement))
-            }
-        }
-
-        // Now sort the patterns by most insns saved and return
-        return patterns.map { (k, v) ->
-            CommonPattern(k, v.sortedBy { it.range.first })
-        }.sortedByDescending { it.insnsSaved }
-    }
-
-    val Node.Instr.canBeMoved get() =
-        // No blocks
-        this !is Node.Instr.Block && this !is Node.Instr.Loop && this !is Node.Instr.If &&
-        this !is Node.Instr.Else && this !is Node.Instr.End &&
-        // No breaks
-        this !is Node.Instr.Br && this !is Node.Instr.BrIf && this !is Node.Instr.BrTable &&
-        // No return
-        this !is Node.Instr.Return &&
-        // No local access
-        this !is Node.Instr.GetLocal && this !is Node.Instr.SetLocal && this !is Node.Instr.TeeLocal
-
-    fun InsnSet.withStackValueIfValid(stack: Stack): InsnSet? {
-        // This makes sure that the stack only changes by at most one item and never dips below its starting val.
-        // If it is invalid, null is returned. If it qualifies and does change 1 value, it is set.
-
-        // First, make sure the stack after the last insn is the same as the first or the same + 1 val
-        val startingStack = stack.insnApplies[startIndex].stackAtBeginning!!
-        val endingStack = stack.insnApplies.getOrNull(startIndex + insns.size)?.stackAtBeginning ?: stack.current!!
-        if (endingStack.size != startingStack.size && endingStack.size != startingStack.size + 1) return null
-        if (endingStack.take(startingStack.size) != startingStack) return null
-
-        // Now, walk the insns and make sure they never pop below the start
-        var stackCounter = 0
-        stack.insnApplies.subList(startIndex, startIndex + insns.size).forEach {
-            it.stackChanges.forEach {
-                stackCounter += if (it.pop) -1 else 1
-                if (stackCounter < 0) return null
-            }
-        }
-        // We're good, now only if the ending stack is one over the start do we have a ret val
-        return copy(
-            valueAddedToStack = endingStack.lastOrNull()?.takeIf { endingStack.size == startingStack.size + 1 }
-        )
-    }
-
-    fun InsnSet.extractCommonFunc() =
-        // This extracts a function with constants changed to parameters
-        insns.fold(Pair(
-            Node.Func(Node.Type.Func(params = emptyList(), ret = valueAddedToStack), emptyList(), emptyList()),
-            Replacement(range = startIndex until startIndex + insns.size, preCallConsts = emptyList()))
-        ) { (func, repl), insn ->
-            if (insn !is Node.Instr.Args.Const<*>) func.copy(instructions = func.instructions + insn) to repl
-            else func.copy(
-                type = func.type.copy(params = func.type.params + insn.constType),
-                instructions = func.instructions + Node.Instr.GetLocal(func.type.params.size)
-            ) to repl.copy(preCallConsts = repl.preCallConsts + insn)
-        }
-
-    protected val Node.Instr.Args.Const<*>.constType get() = when (this) {
-        is Node.Instr.I32Const -> Node.Type.Value.I32
-        is Node.Instr.I64Const -> Node.Type.Value.I64
-        is Node.Instr.F32Const -> Node.Type.Value.F32
-        is Node.Instr.F64Const -> Node.Type.Value.F64
-        else -> error("unreachable")
-    }
-
-    data class InsnSet(
-        val startIndex: Int,
-        val insns: List<Node.Instr>,
-        val valueAddedToStack: Node.Type.Value?
-    )
-
-    data class Replacement(
-        val range: IntRange,
-        val preCallConsts: List<Node.Instr>
-    ) {
-        // Subtract one because there is a call after this
-        val insnsSaved get() = (range.last + 1) - range.first - 1 - preCallConsts.size
-        fun overlaps(o: Replacement) = range.contains(o.range.first) || range.contains(o.range.last) ||
-            o.range.contains(range.first) || o.range.contains(range.last)
-    }
-
-    data class CommonPattern(
-        val newFunc: Node.Func,
-        // In order by earliest replacement first
-        val replacements: List<Replacement>
-    ) {
-        // Replacement pieces saved (with one added for the invocation) less new func instructions
-        val insnsSaved get() = replacements.sumBy { it.insnsSaved } - newFunc.instructions.size
-    }
-
-    companion object : SplitLargeFunc()
-}

compiler/src/main/kotlin/asmble/cli/Compile.kt

@@ -1,9 +1,9 @@
 package asmble.cli
 
 import asmble.ast.Script
+import asmble.compile.jvm.AsmToBinary
 import asmble.compile.jvm.AstToAsm
 import asmble.compile.jvm.ClsContext
-import asmble.compile.jvm.withComputedFramesAndMaxs
 import java.io.FileOutputStream
 
 @Suppress("NAME_SHADOWING")
@@ -69,7 +69,7 @@ open class Compile : Command<Compile.Args>() {
                 includeBinary = args.includeBinary
             )
             AstToAsm.fromModule(ctx)
-            outStream.write(ctx.cls.withComputedFramesAndMaxs())
+            outStream.write(AsmToBinary.fromClassNode(ctx.cls))
         }
     }
 

compiler/src/main/kotlin/asmble/cli/Link.kt

@@ -1,7 +1,7 @@
 package asmble.cli
 
+import asmble.compile.jvm.AsmToBinary
 import asmble.compile.jvm.Linker
-import asmble.compile.jvm.withComputedFramesAndMaxs
 import java.io.FileOutputStream
 
 open class Link : Command<Link.Args>() {
@@ -52,7 +52,7 @@ open class Link : Command<Link.Args>() {
                 defaultMaxMemPages = args.defaultMaxMem
             )
             Linker.link(ctx)
-            outStream.write(ctx.cls.withComputedFramesAndMaxs())
+            outStream.write(AsmToBinary.fromClassNode(ctx.cls))
         }
     }
 

compiler/src/main/kotlin/asmble/cli/Main.kt

@@ -3,7 +3,7 @@ package asmble.cli
 import asmble.util.Logger
 import kotlin.system.exitProcess
 
-val commands = listOf(Compile, Help, Invoke, Link, Run, SplitFunc, Translate)
+val commands = listOf(Compile, Help, Invoke, Link, Run, Translate)
 
 /**
  * Entry point of command line interface.

compiler/src/main/kotlin/asmble/cli/SplitFunc.kt

@@ -1,146 +0,0 @@
-package asmble.cli
-
-import asmble.ast.Node
-import asmble.ast.Script
-import asmble.ast.opt.SplitLargeFunc
-
-open class SplitFunc : Command<SplitFunc.Args>() {
-    override val name = "split-func"
-    override val desc = "Split a WebAssembly function into two"
-
-    override fun args(bld: Command.ArgsBuilder) = Args(
-        inFile = bld.arg(
-            name = "inFile",
-            desc = "The wast or wasm WebAssembly file name. Can be '--' to read from stdin."
-        ),
-        funcName = bld.arg(
-            name = "funcName",
-            desc = "The name (or '#' + function space index) of the function to split"
-        ),
-        inFormat = bld.arg(
-            name = "inFormat",
-            opt = "in",
-            desc = "Either 'wast' or 'wasm' to describe format.",
-            default = "<use file extension>",
-            lowPriority = true
-        ),
-        outFile = bld.arg(
-            name = "outFile",
-            opt = "outFile",
-            desc = "The wast or wasm WebAssembly file name. Can be '--' to write to stdout.",
-            default = "<inFileSansExt.split.wasm or stdout>",
-            lowPriority = true
-        ),
-        outFormat = bld.arg(
-            name = "outFormat",
-            opt = "out",
-            desc = "Either 'wast' or 'wasm' to describe format.",
-            default = "<use file extension or wast for stdout>",
-            lowPriority = true
-        ),
-        compact = bld.flag(
-            opt = "compact",
-            desc = "If set for wast out format, will be compacted.",
-            lowPriority = true
-        ),
-        minInsnSetLength = bld.arg(
-            name = "minInsnSetLength",
-            opt = "minLen",
-            desc = "The minimum number of instructions allowed for the split off function.",
-            default = "5",
-            lowPriority = true
-        ).toInt(),
-        maxInsnSetLength = bld.arg(
-            name = "maxInsnSetLength",
-            opt = "maxLen",
-            desc = "The maximum number of instructions allowed for the split off function.",
-            default = "40",
-            lowPriority = true
-        ).toInt(),
-        maxNewFuncParamCount = bld.arg(
-            name = "maxNewFuncParamCount",
-            opt = "maxParams",
-            desc = "The maximum number of params allowed for the split off function.",
-            default = "30",
-            lowPriority = true
-        ).toInt(),
-        attempts = bld.arg(
-            name = "attempts",
-            opt = "attempts",
-            desc = "The number of attempts to perform.",
-            default = "1",
-            lowPriority = true
-        ).toInt()
-    ).also { bld.done() }
-
-    override fun run(args: Args) {
-        // Load the mod
-        val translate = Translate().also { it.logger = logger }
-        val inFormat =
-            if (args.inFormat != "<use file extension>") args.inFormat
-            else args.inFile.substringAfterLast('.', "<unknown>")
-        val script = translate.inToAst(args.inFile, inFormat)
-        var mod = (script.commands.firstOrNull() as? Script.Cmd.Module)?.module ?: error("Only a single module allowed")
-
-        // Do attempts
-        val splitter = SplitLargeFunc(
-            minSetLength = args.minInsnSetLength,
-            maxSetLength = args.maxInsnSetLength,
-            maxParamCount = args.maxNewFuncParamCount
-        )
-        for (attempt in 0 until args.attempts) {
-            // Find the function
-            var index = mod.names?.funcNames?.toList()?.find { it.second == args.funcName }?.first
-            if (index == null && args.funcName.startsWith('#')) index = args.funcName.drop(1).toInt()
-            val origFunc = index?.let {
-                mod.funcs.getOrNull(it - mod.imports.count { it.kind is Node.Import.Kind.Func })
-            } ?: error("Unable to find func")
-
-            // Split it
-            val results = splitter.apply(mod, index)
-            if (results == null) {
-                logger.warn { "No instructions after attempt $attempt" }
-                break
-            }
-            val (splitMod, insnsSaved) = results
-            val newFunc = splitMod.funcs[index - mod.imports.count { it.kind is Node.Import.Kind.Func }]
-            val splitFunc = splitMod.funcs.last()
-            logger.warn {
-                "Split complete, from func with ${origFunc.instructions.size} insns to a func " +
-                    "with ${newFunc.instructions.size} insns + delegated func " +
-                    "with ${splitFunc.instructions.size} insns and ${splitFunc.type.params.size} params, " +
-                    "saved $insnsSaved insns"
-            }
-            mod = splitMod
-        }
-
-        // Write it
-        val outFile = when {
-            args.outFile != "<inFileSansExt.split.wasm or stdout>" -> args.outFile
-            args.inFile == "--" -> "--"
-            else -> args.inFile.replaceAfterLast('.', "split." + args.inFile.substringAfterLast('.'))
-        }
-        val outFormat = when {
-            args.outFormat != "<use file extension or wast for stdout>" -> args.outFormat
-            outFile == "--" -> "wast"
-            else -> outFile.substringAfterLast('.', "<unknown>")
-        }
-        translate.astToOut(outFile, outFormat, args.compact,
-            Script(listOf(Script.Cmd.Module(mod, mod.names?.moduleName))))
-    }
-
-    data class Args(
-        val inFile: String,
-        val inFormat: String,
-        val funcName: String,
-        val outFile: String,
-        val outFormat: String,
-        val compact: Boolean,
-        val minInsnSetLength: Int,
-        val maxInsnSetLength: Int,
-        val maxNewFuncParamCount: Int,
-        val attempts: Int
-    )
-
-    companion object : SplitFunc()
-}

compiler/src/main/kotlin/asmble/compile/jvm/AsmExt.kt

@@ -2,7 +2,6 @@ package asmble.compile.jvm
 
 import asmble.ast.Node
 import org.objectweb.asm.ClassReader
-import org.objectweb.asm.ClassWriter
 import org.objectweb.asm.Opcodes
 import org.objectweb.asm.Type
 import org.objectweb.asm.tree.*
@@ -189,16 +188,6 @@ fun MethodNode.toAsmString(): String {
 val Node.Type.Func.asmDesc: String get() =
     (this.ret?.typeRef ?: Void::class.ref).asMethodRetDesc(*this.params.map { it.typeRef }.toTypedArray())
 
-fun ClassNode.withComputedFramesAndMaxs(
-    cw: ClassWriter = ClassWriter(ClassWriter.COMPUTE_FRAMES + ClassWriter.COMPUTE_MAXS)
-): ByteArray {
-    // Note, compute maxs adds a bunch of NOPs for unreachable code.
-    // See $func12 of block.wast. I don't believe the extra time over the
-    // instructions to remove the NOPs is worth it.
-    this.accept(cw)
-    return cw.toByteArray()
-}
-
 fun ClassNode.toAsmString(): String {
     val stringWriter = StringWriter()
     this.accept(TraceClassVisitor(PrintWriter(stringWriter)))

compiler/src/main/kotlin/asmble/compile/jvm/AsmToBinary.kt

@@ -0,0 +1,51 @@
+package asmble.compile.jvm
+
+import asmble.compile.jvm.msplit.SplitMethod
+import org.objectweb.asm.ClassWriter
+import org.objectweb.asm.MethodTooLargeException
+import org.objectweb.asm.Opcodes
+import org.objectweb.asm.tree.ClassNode
+
+/**
+ * May mutate given class nodes on [fromClassNode] if [splitMethod] is present (the default). Uses the two-param
+ * [SplitMethod.split] call to try and split overly large methods.
+ */
+open class AsmToBinary(val splitMethod: SplitMethod? = SplitMethod(Opcodes.ASM6)) {
+    fun fromClassNode(
+        cn: ClassNode,
+        newClassWriter: () -> ClassWriter = { ClassWriter(ClassWriter.COMPUTE_FRAMES + ClassWriter.COMPUTE_MAXS) }
+    ): ByteArray {
+        while (true) {
+            val cw = newClassWriter()
+            // Note, compute maxs adds a bunch of NOPs for unreachable code.
+            // See $func12 of block.wast. I don't believe the extra time over the
+            // instructions to remove the NOPs is worth it.
+            cn.accept(cw)
+            try {
+                return cw.toByteArray()
+            } catch (e: MethodTooLargeException) {
+                if (splitMethod == null) throw e
+                // Split the offending method by removing it and replacing it with the split ones
+                require(cn.name == e.className)
+                val tooLargeIndex = cn.methods.indexOfFirst { it.name == e.methodName && it.desc == e.descriptor }
+                require(tooLargeIndex >= 0)
+                val split = splitMethod.split(cn.name, cn.methods[tooLargeIndex])
+                split ?: throw IllegalStateException("Failed to split", e)
+                // Change the split off method's name if there's already one
+                val origName = split.splitOffMethod.name
+                var foundCount = 0
+                while (cn.methods.any { it.name == split.splitOffMethod.name }) {
+                    split.splitOffMethod.name = origName + (++foundCount)
+                }
+                // Replace at the index
+                cn.methods.removeAt(tooLargeIndex)
+                cn.methods.add(tooLargeIndex, split.splitOffMethod)
+                cn.methods.add(tooLargeIndex, split.trimmedMethod)
+            }
+        }
+    }
+
+    companion object : AsmToBinary() {
+        val noSplit = AsmToBinary(null)
+    }
+}

compiler/src/main/kotlin/asmble/compile/jvm/InsnReworker.kt

@@ -150,19 +150,11 @@ open class InsnReworker {
                 // if we are at 0, add the result of said block if necessary to the count.
                 if (insideOfBlocks > 0) {
                     // If it's not a block, just ignore it
-                    val blockStackDiff = insns[insnIndex].let {
+                    (insns[insnIndex] as? Node.Instr.Args.Type)?.let {
-                        when (it) {
-                            is Node.Instr.Block -> if (it.type == null) 0 else 1
-                            is Node.Instr.Loop -> 0
-                            is Node.Instr.If -> if (it.type == null) -1 else 0
-                            else -> null
-                        }
-                    }
-                    if (blockStackDiff != null) {
                         insideOfBlocks--
                         ctx.trace { "Found block begin, number of blocks we're still inside: $insideOfBlocks" }
-                        // We're back on our block, change the count
+                        // We're back on our block, change the count if it had a result
-                        if (insideOfBlocks == 0) countSoFar += blockStackDiff
+                        if (insideOfBlocks == 0 && it.type != null) countSoFar++
                     }
                     if (insideOfBlocks > 0) continue
                 }
@@ -249,10 +241,9 @@ open class InsnReworker {
      */
     fun insnStackDiff(ctx: ClsContext, insn: Node.Instr): Int = when (insn) {
         is Node.Instr.Unreachable, is Node.Instr.Nop, is Node.Instr.Block,
-        is Node.Instr.Loop, is Node.Instr.If, is Node.Instr.Else,
+        is Node.Instr.Loop, is Node.Instr.Else, is Node.Instr.End, is Node.Instr.Br,
-        is Node.Instr.End, is Node.Instr.Br, is Node.Instr.BrIf,
         is Node.Instr.Return -> NOP
-        is Node.Instr.BrTable -> POP_PARAM
+        is Node.Instr.If, is Node.Instr.BrIf, is Node.Instr.BrTable -> POP_PARAM
         is Node.Instr.Call -> ctx.funcTypeAtIndex(insn.index).let {
             // All calls pop params and any return is a push
             (POP_PARAM * it.params.size) + (if (it.ret == null) NOP else PUSH_RESULT)

compiler/src/main/kotlin/asmble/compile/jvm/Linker.kt

@@ -201,7 +201,7 @@ open class Linker {
                 "instance" + mod.name.javaIdent.capitalize(), mod.ref.asmDesc),
             InsnNode(Opcodes.ARETURN)
         )
-        ctx.cls.methods.plusAssign(func)
+        ctx.cls.methods.plusAssign(func.toMethodNode())
     }
 
     class ModuleClass(val cls: Class<*>, overrideName: String? = null) {

compiler/src/main/kotlin/asmble/run/jvm/ScriptContext.kt

@@ -319,10 +319,15 @@ data class ScriptContext(
         bindImport(import, true, MethodType.methodType(Array<MethodHandle>::class.java)).
             invokeWithArguments()!! as Array<MethodHandle>
 
-    open class SimpleClassLoader(parent: ClassLoader, logger: Logger) : ClassLoader(parent), Logger by logger {
+    open class SimpleClassLoader(
+        parent: ClassLoader,
+        logger: Logger,
+        val splitWhenTooLarge: Boolean = true
+    ) : ClassLoader(parent), Logger by logger {
         fun fromBuiltContext(ctx: ClsContext): Class<*> {
             trace { "Computing frames for ASM class:\n" + ctx.cls.toAsmString() }
-            return ctx.cls.withComputedFramesAndMaxs().let { bytes ->
+            val writer = if (splitWhenTooLarge) AsmToBinary else AsmToBinary.noSplit
+            return writer.fromClassNode(ctx.cls).let { bytes ->
                 debug { "ASM class:\n" + bytes.asClassNode().toAsmString() }
                 defineClass("${ctx.packageName}.${ctx.className}",  bytes, 0, bytes.size)
             }

compiler/src/test/kotlin/asmble/SpecTestUnit.kt

@@ -16,7 +16,7 @@ class SpecTestUnit(name: String, wast: String, expectedOutput: String?) : BaseTe
         "nop" -> 20
         "memory_grow" -> 830
         "imports" -> 5
-        else -> 1
+        else -> 2
     }
 
     override fun warningInsteadOfErrReason(t: Throwable) = when (name) {

compiler/src/test/kotlin/asmble/run/jvm/LargeFuncTest.kt

@@ -1,29 +1,31 @@
-package asmble.ast.opt
+package asmble.run.jvm
 
 import asmble.TestBase
 import asmble.ast.Node
 import asmble.compile.jvm.AstToAsm
 import asmble.compile.jvm.ClsContext
-import asmble.run.jvm.ScriptContext
+import org.junit.Assert
 import org.junit.Test
+import org.objectweb.asm.MethodTooLargeException
 import java.nio.ByteBuffer
 import java.util.*
 import kotlin.test.assertEquals
 
-class SplitLargeFuncTest : TestBase() {
+class LargeFuncTest : TestBase() {
     @Test
-    fun testSplitLargeFunc() {
+    fun testLargeFunc() {
-        // We're going to make a large function that does some addition and then stores in mem
+        val numInsnChunks = 6001
+        // Make large func that does some math
         val ctx = ClsContext(
             packageName = "test",
             className = "Temp" + UUID.randomUUID().toString().replace("-", ""),
             logger = logger,
             mod = Node.Module(
-                memories = listOf(Node.Type.Memory(Node.ResizableLimits(initial = 2, maximum = 2))),
+                memories = listOf(Node.Type.Memory(Node.ResizableLimits(initial = 4, maximum = 4))),
                 funcs = listOf(Node.Func(
                     type = Node.Type.Func(params = emptyList(), ret = null),
                     locals = emptyList(),
-                    instructions = (0 until 501).flatMap {
+                    instructions = (0 until numInsnChunks).flatMap {
                         listOf<Node.Instr>(
                             Node.Instr.I32Const(it * 4),
                             // Let's to i * (i = 1)
@@ -47,43 +49,21 @@ class SplitLargeFuncTest : TestBase() {
         )
         // Compile it
         AstToAsm.fromModule(ctx)
+        // Confirm the method size is too large
+        try {
+            ScriptContext.SimpleClassLoader(javaClass.classLoader, logger, splitWhenTooLarge = false).
+                fromBuiltContext(ctx)
+            Assert.fail()
+        } catch (e: MethodTooLargeException) { }
+        // Try again with split
         val cls = ScriptContext.SimpleClassLoader(javaClass.classLoader, logger).fromBuiltContext(ctx)
+        // Create it and check that it still does what we expect
         val inst = cls.newInstance()
         // Run someFunc
         cls.getMethod("someFunc").invoke(inst)
         // Get the memory out
         val mem = cls.getMethod("getMemory").invoke(inst) as ByteBuffer
         // Read out the mem values
-        (0 until 501).forEach { assertEquals(it * (it - 1), mem.getInt(it * 4)) }
+        (0 until numInsnChunks).forEach { assertEquals(it * (it - 1), mem.getInt(it * 4)) }
-
-        // Now split it
-        val (splitMod, insnsSaved) = SplitLargeFunc.apply(ctx.mod, 0) ?: error("Nothing could be split")
-        // Count insns and confirm it is as expected
-        val origInsnCount = ctx.mod.funcs.sumBy { it.instructions.size }
-        val newInsnCount = splitMod.funcs.sumBy { it.instructions.size }
-        assertEquals(origInsnCount - newInsnCount, insnsSaved)
-        // Compile it
-        val splitCtx = ClsContext(
-            packageName = "test",
-            className = "Temp" + UUID.randomUUID().toString().replace("-", ""),
-            logger = logger,
-            mod = splitMod
-        )
-        AstToAsm.fromModule(splitCtx)
-        val splitCls = ScriptContext.SimpleClassLoader(javaClass.classLoader, logger).fromBuiltContext(splitCtx)
-        val splitInst = splitCls.newInstance()
-        // Run someFunc
-        splitCls.getMethod("someFunc").invoke(splitInst)
-        // Get the memory out and compare it
-        val splitMem = splitCls.getMethod("getMemory").invoke(splitInst) as ByteBuffer
-        assertEquals(mem, splitMem)
-        // Dump some info
-        logger.debug {
-            val orig = ctx.mod.funcs.first()
-            val (new, split) = splitMod.funcs.let { it.first() to it.last() }
-            "Split complete, from single func with ${orig.instructions.size} insns to func " +
-                "with ${new.instructions.size} insns + delegated func " +
-                "with ${split.instructions.size} insns and ${split.type.params.size} params"
-        }
     }
 }

examples/c-simple/README.md

@@ -0,0 +1,14 @@
+### Example: C Simple
+
+This shows a simple example of compiling C to WASM and then to the JVM. This is the C version of
+[rust-simple](../rust-simple).
+
+In order to run the C or C++ examples, the latest LLVM binaries must be on the `PATH`, built with the experimental
+WebAssembly target. This can be built by passing `-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=WebAssembly` to `cmake` when
+building WebAssembly. Or it can be downloaded from a nightly build site
+([this one](http://gsdview.appspot.com/wasm-llvm/builds/) was used for these examples at the time of writing).
+
+Everything else is basically the same as [rust-simple](../rust-simple) except with C code and using `clang`. To run
+execute the following from the root `asmble` dir:
+
+    gradlew --no-daemon :examples:c-simple:run

examples/c-simple/src/lib.c

@@ -0,0 +1,3 @@
+int addOne(int x) {
+  return x + 1;
+}

examples/c-simple/src/main/java/asmble/examples/csimple/Main.java

@@ -0,0 +1,13 @@
+package asmble.examples.csimple;
+
+import java.lang.invoke.MethodHandle;
+
+import asmble.generated.CSimple;
+
+class Main {
+    public static void main(String[] args) {
+        // Doesn't need memory or method table
+        CSimple simple = new CSimple(0, new MethodHandle[0]);
+        System.out.println("25 + 1 = " + simple.addOne(25));
+    }
+}

examples/go-simple/simple.go

@@ -2,9 +2,8 @@ package main
 
 import (
 	"fmt"
-	"os"
 )
 
 func main() {
-	fmt.Printf("Args: %v", os.Args)
+	fmt.Println("Hello, World!")
 }

settings.gradle

@@ -2,7 +2,7 @@ rootProject.name = 'asmble'
 include 'annotations',
         'compiler',
         'examples:c-simple',
-//        'examples:go-simple',
+        'examples:go-simple',
-//        'examples:rust-regex',  // todo will be enabled when the problem with string max size will be solved
+        'examples:rust-regex',
         'examples:rust-simple',
         'examples:rust-string'