diff options
Diffstat (limited to 'chapter20/src/main/java/com')
121 files changed, 12369 insertions, 0 deletions
diff --git a/chapter20/src/main/java/com/seaofnodes/simple/ASMPrinter.java b/chapter20/src/main/java/com/seaofnodes/simple/ASMPrinter.java new file mode 100644 index 00000000..5cc13210 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/ASMPrinter.java @@ -0,0 +1,140 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.type.TypeFunPtr; +import com.seaofnodes.simple.type.TypeMem; +import com.seaofnodes.simple.type.TypeRPC; + +public abstract class ASMPrinter { + + public static SB print(SB sb, CodeGen code) { + if( code._cfg==null ) + return sb.p("Need _cfg set, run after GCM"); + + // instruction address + int iadr = 0; + for( int i=0; i<code._cfg._len; i++ ) + if( code._cfg.at(i) instanceof FunNode fun ) + iadr = print(iadr,sb,code,fun,i); + return sb; + } + + private static int print(int iadr, SB sb, CodeGen code, FunNode fun, int cfgidx) { + // Function header + sb.nl().p("---"); + fun.sig().print(sb); + sb.p("---------------------------").nl(); + + + while( !(code._cfg.at(cfgidx) instanceof ReturnNode) ) + iadr = doBlock(iadr,sb,code,fun,cfgidx++); + + // Function separator + sb.p("---"); + fun.sig().print(sb); + sb.p("---------------------------").nl(); + return iadr; + } + + static int doBlock(int iadr, SB sb, CodeGen code, FunNode fun, int cfgidx) { + CFGNode bb = code._cfg.at(cfgidx); + if( bb != fun && !(bb instanceof IfNode) && !(bb instanceof CallEndNode) && !(bb instanceof CallNode) && !(bb instanceof CProjNode && bb.in(0) instanceof CallEndNode )) + sb.p("L").p(bb._nid).p(":").nl(); + + if( !(bb instanceof CallNode) ) + for( Node n : bb._outputs ) + iadr = doInst(iadr, sb,code,fun,bb,n); + + return iadr; + } + + static int doInst(int iadr, SB sb, CodeGen code, FunNode fun, CFGNode bb, Node n) { + final int encWidth = 8; + final int opWidth = 5; + final int argWidth = 30; + + if( n instanceof CProjNode ) return iadr; + + // Phis are useful before (and during) reg alloc + if( n instanceof PhiNode phi ) { + if( phi._type instanceof TypeMem || phi._type instanceof TypeRPC ) return iadr; // Nothing for the hidden ones + sb.fix(4," ").p(" ").fix(encWidth,"").p(" ").fix(opWidth,phi._label).p(" "); + sb.p(code.reg(n)); + if( !(n instanceof ParmNode) ) { + sb.p(" = phi( "); + for( int i=1; i<n.nIns(); i++ ) + sb.p("N").p(n.in(i)._nid).p(","); + sb.unchar().p(" )"); + } + sb.nl(); + return iadr; + } + + // All blocks ending in a Region will need to either fall into or jump + // to this block. Until the post-reg-alloc block layout cleanup, we + // need to assume a jump. There's no real hardware op here, yet. + if( n instanceof RegionNode && !(n instanceof FunNode) ) { + sb.hex4(iadr++).p(" ").fix(encWidth,"??").p(" ").fix(opWidth,"JMP").p(" ").fix(argWidth,"L"+n._nid).nl(); + return iadr; + } + + // ProjNodes following a multi (e.g. Call or New results), + // get indent slightly and just print their index & node# + if( n instanceof ProjNode proj ) { + if( proj._type instanceof TypeMem ) return iadr; // Nothing for the hidden ones + sb.fix(4," ").p(" ").fix(encWidth,"").p(" ").fix(opWidth,proj._label==null ? "---" : proj._label).p(" ").p(code.reg(n)).nl(); + return iadr; + } + + // ADDR ENCODING Op--- dst = src op src // Comment + // 1234 abcdefgh ld4 RAX = [Rbase + off] // Comment + sb.hex4(iadr); + sb.p(" "); + + // Encoding + int size = 1; // TODO: Fake encoding size + iadr += size; + sb.fix(encWidth,"??"); + sb.p(" "); + + // Op; generally "ld4" or "call" + sb.fix(opWidth, n instanceof MachNode mach ? mach.op() : n.label()); + sb.p(" "); + + // General asm args + if( n instanceof MachNode mach ) { + int old = sb.len(); + mach.asm(code,sb); + sb.fix(argWidth-(sb.len()-old),""); // Pad out + + } else if( !(n._type instanceof TypeMem) ) { + // Room for some inputs + sb.fix(5, n._nid+":" ); + int off = 0; + for( int i=0; i<n.nIns(); i++ ) { + sb.fix(5, n.in(i) == null ? "___" : String.valueOf(n.in(i)._nid)); + off += 5; + if( off >= argWidth ) break; + } + if( off < argWidth ) sb.fix(argWidth-off,""); + } + sb.p(" "); + + // Comment + String comment = n.comment(); + if( comment!=null ) sb.p("// ").p(comment); + + sb.nl(); + + // MultiNodes are immediately followed by projection(s) + if( !(n instanceof CFGNode) && n instanceof MultiNode ) { + for( Node proj : n._outputs ) { + assert proj instanceof ProjNode; + doInst(iadr,sb,code,fun,bb,proj); + } + } + + return iadr; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/Ary.java b/chapter20/src/main/java/com/seaofnodes/simple/Ary.java new file mode 100644 index 00000000..73ada8e0 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/Ary.java @@ -0,0 +1,217 @@ +package com.seaofnodes.simple; + +import java.lang.StringBuilder; +import java.lang.reflect.Array; +import java.util.*; + +// ArrayList with saner syntax +public class Ary<E> extends AbstractList<E> implements List<E> { + public E[] _es; + public int _len; + + public Ary(E[] es) { this(es,es.length); } + public Ary(E[] es, int len) { if( es.length==0 ) es=Arrays.copyOf(es,1); _es=es; _len=len; } + @SuppressWarnings("unchecked") + public Ary(Class<E> clazz) { this((E[]) Array.newInstance(clazz, 1),0); } + + /** @return list is empty */ + public boolean isEmpty() { return _len==0; } + /** @return active list length */ + public int len() { return _len; } + public int size() { return len(); } + + /** @param i element index + * @return element being returned; throws if OOB + * @exception AIOOBE if !(0 <= i < _len) + */ + public E get( int i ) { return at(i); } + public E at( int i ) { + range_check(i); + return _es[i]; + } + + /** @return last element */ + public E last( ) { return at(_len-1); } + + /** Add element in amortized constant time + * @param e Element to add at end of list + **/ + public E push( E e ) { + if( _len >= _es.length ) _es = Arrays.copyOf(_es,Math.max(1,_es.length<<1)); + return (_es[_len++] = e); + } + /** @return remove and return last element */ + public E pop( ) { + return _len==0 ? null : _es[--_len]; + } + @Override public E removeLast() { return pop(); } + + + /** Add element in amortized constant time + * @param e Element to add at end of list + * @return true */ + public boolean add( E e ) { + if( _len >= _es.length ) _es = Arrays.copyOf(_es,Math.max(1,_es.length<<1)); + _es[_len++] = e; + return true; + } + + /** Fast, constant-time, element removal. Does not preserve order + * @param i element to be removed + * @return element removed */ + public E del( int i ) { + range_check(i); + E tmp = _es[i]; + _es[i]=_es[--_len]; + return tmp; + } + + /** Slow, linear-time, element removal. Preserves order. + * @param i element to be removed + * @return element removed */ + @Override public E remove( int i ) { + range_check(i); + E e = _es[i]; + System.arraycopy(_es,i+1,_es,i,(--_len)-i); + return e; + } + + /** Remove all elements */ + public void clear( ) { _len=0; } + + /** Set existing element + * @param i element to set + * @param e value to set + * @return old value + * @exception AIOOBE if !(0 <= i < _len) + */ + public E set( int i, E e ) { + range_check(i); + E old = _es[i]; + _es[i] = e; + return old; + } + + /** Set existing element + * @param i element to set + * @param e value to set + * @return old value + * @exception AIOOBE if !(0 <= i < _len) + */ + public E setX( int i, E e ) { + if( i >= _len ) { + while( i >= _es.length ) _es = Arrays.copyOf( _es, _es.length << 1 ); + _len = i+1; + } + E old = _es[i]; + _es[i] = e; + return old; + } + + public Ary<E> setLen( int len ) { + if( len > _len ) + while( len>= _es.length ) _es = Arrays.copyOf(_es,_es.length<<1); + _len = len; + while( _es.length > (len<<1) ) // Shrink if hugely too large + _es = Arrays.copyOf(_es,_es.length>>1); + Arrays.fill(_es,len,_es.length,null); + return this; + } + + public E swap( int i, int j ) { + range_check(i); + range_check(j); + E tmp = _es[i]; + if( i==j ) return tmp; + _es[i] = _es[j]; + _es[j] = tmp; + return tmp; + } + + + /** @param c Collection to be added */ + public boolean addAll( Collection<? extends E> c ) { + if( c==null ) return false; + for( E e : c ) add(e); + return true; + } + public Ary<E> addAll( Iterable<? extends E> it ) { for( E e : it ) add(e); return this; } + + /** @param es Array to be added */ + public <F extends E> Ary<E> addAll( F[] es ) { + if( es==null || es.length==0 ) return this; + while( _len+es.length > _es.length ) _es = Arrays.copyOf(_es,_es.length<<1); + System.arraycopy(es,0,_es,_len,es.length); + _len += es.length; + return this; + } + + /** @param c Collection to be added */ + public Ary<E> addAll( Ary<? extends E> c ) { + if( c==null || c._len==0 ) return this; + while( _len+c._len > _es.length ) _es = Arrays.copyOf(_es,_es.length<<1); + System.arraycopy(c._es,0,_es,_len,c._len); + _len += c._len; + return this; + } + + /** @return compact array version */ + public E[] asAry() { return Arrays.copyOf(_es,_len); } + + /** Sorts in-place + * @param c Comparator to sort by */ + public void sort_update(Comparator<? super E> c ) { Arrays.sort(_es, 0, _len, c); } + + /** Find the first matching element using ==, or -1 if none. Note that + * most del calls shuffle the list, so the first element might be random. + * @param e Element to find + * @return index of first matching element, or -1 if none */ + public int find( E e ) { + for( int i=0; i<_len; i++ ) if( _es[i]==e ) return i; + return -1; + } + + + /** @return an iterator */ + @Override public Iterator<E> iterator() { return new Iter(); } + private class Iter implements Iterator<E> { + int _i=0; + @Override public boolean hasNext() { return _i<_len; } + @Override public E next() { return _es[_i++]; } + } + + @Override public String toString() { + StringBuilder sb = new StringBuilder().append('{'); + for( int i=0; i<_len; i++ ) { + if( i>0 ) sb.append(','); + if( _es[i] != null ) sb.append(_es[i].toString()); + } + return sb.append('}').toString(); + } + + private void range_check( int i ) { + if( i < 0 || i>=_len ) + throw new ArrayIndexOutOfBoundsException(""+i+" >= "+_len); + } + + // Note that the hashCode() and equals() are not invariant to changes in the + // underlying array. If the hashCode() is used (e.g., inserting into a + // HashMap) and the then the array changes, the hashCode() will change also. + @Override public boolean equals( Object o ) { + if( this==o ) return true; + if( !(o instanceof Ary ary) ) return false; + if( _len != ary._len ) return false; + if( _es == ary._es ) return true; + for( int i=0; i<_len; i++ ) + if( !(_es[i]==null ? (ary._es[i] == null) : _es[i].equals(ary._es[i])) ) + return false; + return true; + } + @Override public int hashCode( ) { + int sum=_len; + for( int i=0; i<_len; i++ ) + sum += _es[i]==null ? 0 : _es[i].hashCode(); + return sum; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/CodeGen.java b/chapter20/src/main/java/com/seaofnodes/simple/CodeGen.java new file mode 100644 index 00000000..c48a4162 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/CodeGen.java @@ -0,0 +1,288 @@ + package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.node.cpus.x86_64_v2.MemOpX86; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; +import java.util.HashMap; +import java.util.IdentityHashMap; + +public class CodeGen { + // Last created CodeGen as a global; used all over to avoid passing about a + // "context". + public static CodeGen CODE; + + public enum Phase { + Parse, // Parse ASCII text into Sea-of-Nodes IR + Opto, // Run ideal optimizations + TypeCheck, // Last check for bad programs + InstructionSelection, // Convert to target hardware nodes + Schedule, // Global schedule (code motion) nodes + LocalSched, // Local schedule + RegAlloc; // Register allocation + } + public enum CallingConv { + Win64, + SystemV + } + public Phase _phase; + + // --------------------------- + // Compilation source code + public final String _src; + // Compile-time known initial argument type + public final TypeInteger _arg; + public CallingConv _callingConv; + + // --------------------------- + public CodeGen( String src ) { this(src, TypeInteger.BOT, 123L ); } + public CodeGen( String src, TypeInteger arg, long workListSeed ) { + CODE = this; + _phase = null; + _callingConv = null; + _start = new StartNode(arg); + _stop = new StopNode(src); + _src = src; + _arg = arg; + _iter = new IterPeeps(workListSeed); + P = new Parser(this,arg); + } + + + // --------------------------- + /** + * A counter, for unique node id generation. Starting with value 1, to + * avoid bugs confusing node ID 0 with uninitialized values. + */ + private int _uid = 1; + public int UID() { return _uid; } + public int getUID() { return _uid++; } + + // Next available memory alias number + private int _alias = 2; // 0 is for control, 1 for memory + public int getALIAS() { return _alias++; } + + + // Popular visit bitset, declared here, so it gets reused all over + public final BitSet _visit = new BitSet(); + public BitSet visit() { assert _visit.isEmpty(); return _visit; } + + // Start and stop; end points of the generated IR + public StartNode _start; + public StopNode _stop; + + // Global Value Numbering. Hash over opcode and inputs; hits in this table + // are structurally equal. + public final HashMap<Node,Node> _gvn = new HashMap<>(); + + // Compute "function indices": FIDX. + // Each new request at the same signature gets a new FIDX. + private final HashMap<TypeTuple,Integer> FIDXS = new HashMap<>(); + public TypeFunPtr makeFun( TypeTuple sig, Type ret ) { + Integer i = FIDXS.get(sig); + int fidx = i==null ? 0 : i; + FIDXS.put(sig,fidx+1); // Track count per sig + assert fidx<64; // TODO: need a larger FIDX space + return TypeFunPtr.make((byte)2,sig,ret, 1L<<fidx ); + } + // Signature for MAIN + public TypeFunPtr _main = makeFun(TypeTuple.MAIN,Type.BOTTOM); + // Reverse from a constant function pointer to the IR function being called + public FunNode link( TypeFunPtr tfp ) { + assert tfp.isConstant(); + return _linker.get(tfp.makeFrom(Type.BOTTOM)); + } + + // Insert linker mapping from constant function signature to the function + // being called. + public void link(FunNode fun) { + _linker.put(fun.sig().makeFrom(Type.BOTTOM),fun); + } + + // "Linker" mapping from constant TypeFunPtrs to heads of function. These + // TFPs all have exact single fidxs and their return is wiped to BOTTOM (so + // the return is not part of the match). + private final HashMap<TypeFunPtr,FunNode> _linker = new HashMap<>(); + + // Parser object + public final Parser P; + + // Parse ASCII text into Sea-of-Nodes IR + public CodeGen parse() { + assert _phase == null; + _phase = Phase.Parse; + + P.parse(); + JSViewer.show(); + return this; + } + + + // --------------------------- + // Iterator peepholes. + public final IterPeeps _iter; + // Stop tracking deps while assertin + public boolean _midAssert; + // Statistics on peepholes + public int _iter_cnt, _iter_nop_cnt; + // Stat gathering + public void iterCnt() { if( !_midAssert ) _iter_cnt++; } + public void iterNop() { if( !_midAssert ) _iter_nop_cnt++; } + + // Run ideal optimizations + public CodeGen opto() { + assert _phase == Phase.Parse; + _phase = Phase.Opto; + + // Pessimistic peephole optimization on a worklist + _iter.iterate(this); + + // TODO: + // Optimistic + // TODO: + // loop unroll, peel, RCE, etc + return this; + } + public <N extends Node> N add( N n ) { return (N)_iter.add(n); } + public void addAll( Ary<Node> ary ) { _iter.addAll(ary); } + + + // --------------------------- + // Last check for bad programs + public CodeGen typeCheck() { + // Demand phase Opto for cleaning up dead control flow at least, + // required for the following GCM. + assert _phase.ordinal() <= Phase.Opto.ordinal(); + _phase = Phase.TypeCheck; + + Parser.ParseException err = _stop.walk( Node::err ); + if( err != null ) + throw err; + return this; + } + + + // --------------------------- + // Code generation CPU target + public Machine _mach; + + // Convert to target hardware nodes + public CodeGen instSelect( String cpu, String callingConv ) { + assert _phase.ordinal() <= Phase.TypeCheck.ordinal(); + _phase = Phase.InstructionSelection; + + _callingConv = switch(callingConv) { + case "SystemV" -> CallingConv.SystemV; + case "Win64" -> CallingConv.Win64; + default -> throw Utils.TODO(); // Unknown calling convention + }; + + // Look for CPU in fixed named place: + // com.seaofnodes.simple.node.cpus."cpu"."cpu.class" + String clzFile = "com.seaofnodes.simple.node.cpus."+cpu+"."+cpu; + try { _mach = ((Class<Machine>) Class.forName( clzFile )).getDeclaredConstructor().newInstance(); } + catch( Exception e ) { throw new RuntimeException(e); } + + // Convert to machine ops + _uid = 1; // All new machine nodes reset numbering + var map = new IdentityHashMap<Node,Node>(); + _instSelect( _stop, map ); + _stop = ( StopNode)map.get(_stop ); + _start = (StartNode)map.get(_start); + _instOuts(_stop,visit()); + _visit.clear(); + return this; + } + + // Walk all ideal nodes, recursively mapping ideal to machine nodes, then + // make a machine node for "this". + private Node _instSelect( Node n, IdentityHashMap<Node,Node> map ) { + if( n==null ) return null; + Node x = map.get(n); + if( x !=null ) return x; // Been there, done that + + // If n is a MachConcrete, then its part of a multi-node expansion. + // It does not need instruction selection (already selected!) + // but it does need its inputs walked. + if( n instanceof MachConcreteNode || n instanceof MemOpX86 ) { + for( int i=0; i < n.nIns(); i++ ) + n._inputs.set(i, _instSelect(n.in(i),map) ); + return n; + } + + // Produce a machine node from n; map it to flag as done so stops cycles. + map.put(n, x=_mach.instSelect(n) ); + // Walk machine op and replace inputs with mapped inputs + for( int i=0; i < x.nIns(); i++ ) + x._inputs.set(i, _instSelect(x.in(i),map) ); + if( x instanceof MachNode mach ) + mach.postSelect(); // Post selection action + + // Updates forward edges only. + n._outputs.clear(); + return x; + } + + // Walk all machine Nodes, and set their output edges + private void _instOuts( Node n, BitSet visit ) { + if( visit.get(n._nid) ) return; + visit.set(n._nid); + for( Node in : n._inputs ) + if( in!=null ) { + in._outputs.push(n); + _instOuts(in,visit); + } + } + + + // --------------------------- + // Control Flow Graph in RPO order. + Ary<CFGNode> _cfg = new Ary<>(CFGNode.class); + + // Global schedule (code motion) nodes + public CodeGen GCM() { return GCM(false); } + public CodeGen GCM( boolean show) { + assert _phase.ordinal() <= Phase.InstructionSelection.ordinal(); + _phase = Phase.Schedule; + + // Build the loop tree, fix never-exit loops + _start.buildLoopTree(_stop); + GlobalCodeMotion.buildCFG(this); + if( show ) + System.out.println(new GraphVisualizer().generateDotOutput(_stop,null,null)); + return this; + } + + // --------------------------- + // Local (basic block) scheduler phase, a classic list scheduler + public CodeGen localSched() { + assert _phase == Phase.Schedule; + _phase = Phase.LocalSched; + ListScheduler.sched(this); + return this; + } + + + // --------------------------- + public String reg(Node n) { + if( _phase == Phase.RegAlloc ) throw Utils.TODO(); + return "N"+ n._nid; + } + + + // --------------------------- + SB asm(SB sb) { return ASMPrinter.print(sb,this); } + String asm() { return asm(new SB()).toString(); } + + + // Testing shortcuts + Node ctrl() { return _stop.ret().ctrl(); } + Node expr() { return _stop.ret().expr(); } + String print() { return _stop.print(); } + + // Debugging helper + @Override public String toString() { return _stop.p(9999); } + + // Debugging helper + public Node f(int idx) { return _stop.find(idx); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/GlobalCodeMotion.java b/chapter20/src/main/java/com/seaofnodes/simple/GlobalCodeMotion.java new file mode 100644 index 00000000..12dd9243 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/GlobalCodeMotion.java @@ -0,0 +1,241 @@ + package com.seaofnodes.simple; + +import com.seaofnodes.simple.IterPeeps.WorkList; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; +import java.util.*; + +public abstract class GlobalCodeMotion { + + // Arrange that the existing isCFG() Nodes form a valid CFG. The + // Node.use(0) is always a block tail (either IfNode or head of the + // following block). There are no unreachable infinite loops. + public static void buildCFG( CodeGen code ) { + Ary<CFGNode> rpo = new Ary<>(CFGNode.class); + _rpo_cfg(null, code._start, code.visit(), rpo); + // Reverse in-place + for( int i=0; i< rpo.size()>>1; i++ ) { + int j = rpo.size()-1-i; + CFGNode tmp = rpo.get(i); + rpo.set(i, rpo.get(j)); + rpo.set(j, tmp); + } + // Set global CFG + code._cfg = rpo; + + schedEarly(code); + code._visit.clear(); + schedLate (code); + } + + // Post-Order of CFG + private static void _rpo_cfg(CFGNode def, Node use, BitSet visit, Ary<CFGNode> rpo) { + if( !(use instanceof CFGNode cfg) || visit.get(cfg._nid) ) + return; // Been there, done that + if( def instanceof CallNode && cfg instanceof FunNode ) + return; // Ignore linked function calls + visit.set(cfg._nid); + for( Node useuse : cfg._outputs ) + _rpo_cfg(cfg,useuse,visit,rpo); + rpo.add(cfg); + } + + // ------------------------------------------------------------------------ + // Visit all nodes in CFG Reverse Post-Order, essentially defs before uses + // (except at loops). Since defs are visited first - and hoisted as early + // as possible, when we come to a use we place it just after its deepest + // input. + private static void schedEarly(CodeGen code) { + // Reverse Post-Order on CFG + for( CFGNode cfg : code._cfg ) { + cfg.loopDepth(); + for( Node n : cfg._inputs ) + _schedEarly(n,code._visit ); + if( cfg instanceof RegionNode ) + for( Node phi : cfg._outputs ) + if( phi instanceof PhiNode ) + _schedEarly(phi,code._visit ); + } + } + + private static void _schedEarly(Node n, BitSet visit) { + if( n==null || visit.get(n._nid) ) return; // Been there, done that + assert !(n instanceof CFGNode); + visit.set(n._nid); + // Schedule not-pinned not-CFG inputs before self. Since skipping + // Pinned, this never walks the backedge of Phis (and thus spins around + // a data-only loop), eventually attempting relying on some pre-visited- + // not-post-visited data op with no scheduled control. + for( Node def : n._inputs ) + if( def!=null && !(def instanceof PhiNode) ) + _schedEarly(def,visit); + + // If not-pinned (e.g. constants, projections, phi) and not-CFG + if( !n.isPinned() ) { + // Schedule at deepest input + CFGNode early = CodeGen.CODE._start; // Maximally early, lowest idepth + if( n.in(0) instanceof CFGNode cfg ) early = cfg; + for( int i=1; i<n.nIns(); i++ ) + if( n.in(i)!=null && n.in(i).cfg0().idepth() > early.idepth() ) + early = n.in(i).cfg0(); // Latest/deepest input + n.setDef(0,early); + } + } + + // ------------------------------------------------------------------------ + private static void schedLate( CodeGen code ) { + CFGNode[] late = new CFGNode[code.UID()]; + Node[] ns = new Node[code.UID()]; + // Breadth-first scheduling + breadth(code._stop,ns,late); + + // Copy the best placement choice into the control slot + for( int i=0; i<late.length; i++ ) + if( ns[i] != null && !(ns[i] instanceof ProjNode) ) + ns[i].setDef(0,late[i]); + } + + private static void breadth(Node stop, Node[] ns, CFGNode[] late) { + // Things on the worklist have some (but perhaps not all) uses done. + WorkList<Node> work = new WorkList<>(); + work.push(stop); + Node n; + outer: + while( (n = work.pop()) != null ) { + assert late[n._nid]==null; // No double visit + // These I know the late schedule of, and need to set early for loops + if( n instanceof CFGNode cfg ) late[n._nid] = cfg.blockHead() ? cfg : cfg.cfg(0); + else if( n instanceof PhiNode phi ) late[n._nid] = phi.region(); + else if( n instanceof ProjNode && n.in(0) instanceof CFGNode cfg ) late[n._nid] = cfg; + else { + + // All uses done? + for( Node use : n._outputs ) + if( use!=null && late[use._nid]==null ) + continue outer; // Nope, await all uses done + + // Loads need their memory inputs' uses also done + if( n instanceof LoadNode ld ) + for( Node memuse : ld.mem()._outputs ) + if( late[memuse._nid]==null && + // New makes new memory, never crushes load memory + !(memuse instanceof NewNode) && + // Load-use directly defines memory + (memuse._type instanceof TypeMem || + // Load-use indirectly defines memory + (memuse._type instanceof TypeTuple tt && tt._types[ld._alias] instanceof TypeMem)) ) + continue outer; + + // All uses done, schedule + _doSchedLate(n,ns,late); + } + + // Walk all inputs and put on worklist, as their last-use might now be done + for( Node def : n._inputs ) + if( def!=null && late[def._nid]==null ) { + work.push(def); + // if the def has a load use, maybe the load can fire + for( Node ld : def._outputs ) + if( ld instanceof LoadNode && late[ld._nid]==null ) + work.push(ld); + } + } + } + + private static void _doSchedLate(Node n, Node[] ns, CFGNode[] late) { + // Walk uses, gathering the LCA (Least Common Ancestor) of uses + CFGNode early = n.in(0) instanceof CFGNode cfg ? cfg : n.in(0).cfg0(); + assert early != null; + CFGNode lca = null; + for( Node use : n._outputs ) + if( use != null ) + lca = use_block(n,use, late)._idom(lca,null); + + // Loads may need anti-dependencies, raising their LCA + if( n instanceof LoadNode load ) + lca = find_anti_dep(lca,load,early,late); + + // Walk up from the LCA to the early, looking for best place. This is + // the lowest execution frequency, approximated by least loop depth and + // deepest control flow. + CFGNode best = lca; + lca = lca.idom(); // Already found best for starting LCA + for( ; lca != early.idom(); lca = lca.idom() ) + if( better(lca,best) ) + best = lca; + assert !(best instanceof IfNode); + ns [n._nid] = n; + late[n._nid] = best; + } + + // Block of use. Normally from late[] schedule, except for Phis, which go + // to the matching Region input. + private static CFGNode use_block(Node n, Node use, CFGNode[] late) { + if( !(use instanceof PhiNode phi) ) + return late[use._nid]; + CFGNode found=null; + for( int i=1; i<phi.nIns(); i++ ) + if( phi.in(i)==n ) + found = phi.region().cfg(i)._idom(found,null); + + assert found!=null; + return found; + } + + + // Least loop depth first, then largest idepth + private static boolean better( CFGNode lca, CFGNode best ) { + return lca.loopDepth() < best.loopDepth() || + (lca.idepth() > best.idepth() || best instanceof IfNode); + } + + private static CFGNode find_anti_dep(CFGNode lca, LoadNode load, CFGNode early, CFGNode[] late) { + // We could skip final-field loads here. + // Walk LCA->early, flagging Load's block location choices + for( CFGNode cfg=lca; early!=null && cfg!=early.idom(); cfg = cfg.idom() ) + cfg._anti = load._nid; + // Walk load->mem uses, looking for Stores causing an anti-dep + for( Node mem : load.mem()._outputs ) { + switch( mem ) { + case StoreNode st: + assert late[st._nid]!=null; + lca = anti_dep(load,late[st._nid],st.cfg0(),lca,st); + break; + case CallNode st: + assert late[st._nid]!=null; + lca = anti_dep(load,late[st._nid],st.cfg0(),lca,st); + break; + case PhiNode phi: + // Repeat anti-dep for matching Phi inputs. + // No anti-dep edges but may raise the LCA. + for( int i=1; i<phi.nIns(); i++ ) + if( phi.in(i)==load.mem() ) + lca = anti_dep(load,phi.region().cfg(i),load.mem().cfg0(),lca,null); + break; + case NewNode st: break; + case LoadNode ld: break; // Loads do not cause anti-deps on other loads + case ReturnNode ret: break; // Load must already be ahead of Return + case MemMergeNode ret: break; // Mem uses now on ScopeMin + case NeverNode never: break; + default: throw Utils.TODO(); + } + } + return lca; + } + + // + private static CFGNode anti_dep( LoadNode load, CFGNode stblk, CFGNode defblk, CFGNode lca, Node st ) { + // Walk store blocks "reach" from its scheduled location to its earliest + for( ; stblk != defblk.idom(); stblk = stblk.idom() ) { + // Store and Load overlap, need anti-dependence + if( stblk._anti==load._nid ) { + lca = stblk._idom(lca,null); // Raise Loads LCA + if( lca == stblk && st != null && st._inputs.find(load) == -1 ) // And if something moved, + st.addDef(load); // Add anti-dep as well + return lca; // Cap this stores' anti-dep to here + } + } + return lca; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/GraphVisualizer.java b/chapter20/src/main/java/com/seaofnodes/simple/GraphVisualizer.java new file mode 100644 index 00000000..30e25916 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/GraphVisualizer.java @@ -0,0 +1,288 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; + +import java.util.*; + +import static com.seaofnodes.simple.Utils.TODO; + +/** + * Simple visualizer that outputs GraphViz dot format. + * The dot output must be saved to a file and run manually via dot to generate the SVG output. + * Currently, this is done manually. + */ +public class GraphVisualizer { + + /** + * If set to true we put the control nodes in a separate cluster from + * data nodes. + */ + boolean _separateControlCluster = false; + + public GraphVisualizer(boolean separateControlCluster) { this._separateControlCluster = separateControlCluster; } + public GraphVisualizer() { this(false); } + + public String generateDotOutput(Parser parse) { return generateDotOutput(parse._code._stop,parse._scope,parse._xScopes); } + public String generateDotOutput(StopNode stop, Node scope, Stack<ScopeNode> xScopes) { + + // Since the graph has cycles, we need to create a flat list of all the + // nodes in the graph. + Collection<Node> all = findAll(xScopes, stop, scope); + StringBuilder sb = new StringBuilder(); + sb.append("digraph chapter18 {\n"); + sb.append("/*\n"); + sb.append(stop._src); + sb.append("\n*/\n"); + + // To keep the Scopes below the graph and pointing up into the graph we + // need to group the Nodes in a subgraph cluster, and the scopes into a + // different subgraph cluster. THEN we can draw edges between the + // scopes and nodes. If we try to cross subgraph cluster borders while + // still making the subgraphs DOT gets confused. + sb.append("\trankdir=BT;\n"); // Force Nodes before Scopes + + // CNC - turned off Apr/8/2024, gives more flex in the layout and + // removes some of the more ludicrous layout choices. + // Preserve node input order + //sb.append("\tordering=\"in\";\n"); + + // Merge multiple edges hitting the same node. Makes common shared + // nodes much prettier to look at. + sb.append("\tconcentrate=\"true\";\n"); + + // Force nested scopes to order + sb.append("\tcompound=\"true\";\n"); + + // Just the Nodes first, in a cluster no edges + nodes(sb, all); + + // Now the scopes, in a cluster no edges + if( xScopes != null ) + for( ScopeNode xscope : xScopes ) + if( !xscope.isDead() ) + scope( sb, xscope ); + + // Walk the Node edges + nodeEdges(sb, all); + + // Walk the active Scope edges + if( xScopes != null ) + for( ScopeNode xscope : xScopes ) + if( !xscope.isDead() ) + scopeEdges( sb, xscope ); + + sb.append("}\n"); + return sb.toString(); + } + + private void nodesByCluster(StringBuilder sb, boolean doCtrl, Collection<Node> all) { + if (!_separateControlCluster && doCtrl) // all nodes in 1 cluster + return; + // Just the Nodes first, in a cluster no edges + sb.append(doCtrl ? "\tsubgraph cluster_Controls {\n" : "\tsubgraph cluster_Nodes {\n"); // Magic "cluster_" in the subgraph name + for (Node n : all) { + if (n instanceof ProjNode || n instanceof CProjNode || n instanceof MemMergeNode || n==Parser.XCTRL) + continue; // Do not emit, rolled into MultiNode or Scope cluster already + if( _separateControlCluster && doCtrl && !(n instanceof CFGNode) ) continue; + if( _separateControlCluster && !doCtrl && (n instanceof CFGNode) ) continue; + sb.append("\t\t").append(n.uniqueName()).append(" [ "); + String lab = n.glabel(); + if (n instanceof MultiNode) { + // Make a box with the MultiNode on top, and all the projections on the bottom + sb.append("shape=plaintext label=<\n"); + sb.append("\t\t\t<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\" CELLPADDING=\"4\">\n"); + sb.append("\t\t\t<TR><TD BGCOLOR=\"yellow\">").append(lab).append("</TD></TR>\n"); + sb.append("\t\t\t<TR>"); + boolean doProjTable = false; + n._outputs.sort((x,y) -> x instanceof ProjNode xp && y instanceof ProjNode yp ? (xp._idx - yp._idx) : (x._nid - y._nid)); + for (Node use : n._outputs) { + if (use instanceof ProjNode proj) { + if (!doProjTable) { + doProjTable = true; + sb.append("<TD>").append("\n"); + sb.append("\t\t\t\t<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\">").append("\n"); + sb.append("\t\t\t\t<TR>"); + } + sb.append("<TD PORT=\"p").append(proj._idx).append("\">").append(proj.glabel()).append("</TD>"); + } + if (use instanceof CProjNode proj) { + if (!doProjTable) { + doProjTable = true; + sb.append("<TD>").append("\n"); + sb.append("\t\t\t\t<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\">").append("\n"); + sb.append("\t\t\t\t<TR>"); + } + sb.append("<TD PORT=\"p").append(proj._idx).append("\" BGCOLOR=\"yellow\">").append(proj.glabel()).append("</TD>"); + } + } + if (doProjTable) { + sb.append("</TR>").append("\n"); + sb.append("\t\t\t\t</TABLE>").append("\n"); + sb.append("\t\t\t</TD>"); + } + sb.append("</TR>\n"); + sb.append("\t\t\t</TABLE>>\n\t\t"); + + } else { + // control nodes have box shape + // other nodes are ellipses, i.e. default shape + if( n instanceof CFGNode ) sb.append("shape=box style=filled fillcolor=yellow "); + else if (n instanceof PhiNode) sb.append("style=filled fillcolor=lightyellow "); + sb.append("label=\"").append(lab).append("\" "); + } + sb.append("];\n"); + } + if (!_separateControlCluster) { + // Force Region & Phis to line up + for (Node n : all) { + if (n instanceof RegionNode region) { + sb.append("\t\t{ rank=same; "); + sb.append(region.uniqueName()).append(";"); + for (Node phi : region._outputs) + if (phi instanceof PhiNode) sb.append(phi.uniqueName()).append(";"); + sb.append("}\n"); + } + } + } + sb.append("\t}\n"); // End Node cluster + } + + private void nodes(StringBuilder sb, Collection<Node> all) { + nodesByCluster(sb, true, all); + nodesByCluster(sb, false, all); + } + + // Build a nested scope display, walking the _prev edge + private void scope( StringBuilder sb, ScopeNode scope ) { + sb.append("\tnode [shape=plaintext];\n"); + int last = scope.nIns(); + int max = scope._lexSize.size(); + for( int i = 0; i < max; i++ ) { + int level = max-i-1; + String scopeName = makeScopeName(scope, level); + sb.append("\tsubgraph cluster_").append(scopeName).append(" {\n"); // Magic "cluster_" in the subgraph name + // Special for memory ScopeMinNode + if( level==0 ) { + MemMergeNode n = scope.mem(); + sb.append("\t\t").append(n.uniqueName()).append(" [label=\"").append(n.glabel()).append("\"];\n"); + } + sb.append("\t\t").append(scopeName).append(" [label=<\n"); + sb.append("\t\t\t<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\">\n"); + // Add the scope level + sb.append("\t\t\t<TR><TD BGCOLOR=\"cyan\">").append(level).append("</TD>"); + int lexStart=scope._lexSize.at(level); + for( int j=lexStart; j<last; j++ ) { + var v = scope._vars.at(j); + sb.append("<TD PORT=\"").append(makePortName(scopeName, v._name)).append("\">").append(v._name).append("</TD>"); + } + last = lexStart; + sb.append("</TR>\n"); + sb.append("\t\t\t</TABLE>>];\n"); + } + // Scope clusters nest, so the graphics shows the nested scopes, so + // they are not closed as they are printed; so they just keep nesting. + // We close them all at once here. + sb.append( "\t}\n".repeat( max ) ); // End all Scope clusters + } + + private String makeScopeName(ScopeNode sn, int level) { return sn.uniqueName() + "_" + level; } + private String makePortName(String scopeName, String varName) { return scopeName + "_" + varName; } + + // Walk the node edges + private void nodeEdges(StringBuilder sb, Collection<Node> all) { + // All them edge labels + sb.append("\tedge [ fontname=Helvetica, fontsize=8 ];\n"); + for( Node n : all ) { + // Do not display the Constant->Start edge; + // ProjNodes handled by Multi; + // ScopeNodes are done separately + if( n instanceof ConstantNode || n instanceof ProjNode || n instanceof CProjNode || n instanceof ScopeNode || n==Parser.XCTRL ) + continue; + for( int i=0; i<n.nIns(); i++ ) { + Node def = n.in(i); + if( n instanceof PhiNode && def instanceof RegionNode ) { + // Draw a dotted use->def edge from Phi to Region. + sb.append('\t').append(n.uniqueName()); + sb.append(" -> "); + sb.append(def.uniqueName()); + sb.append(" [style=dotted taillabel=").append(i).append("];\n"); + } else if( def != null ) { + // Most edges land here use->def + sb.append('\t').append(n.uniqueName()).append(" -> "); + if( def instanceof CProjNode proj ) { + String mname = proj.ctrl().uniqueName(); + sb.append(mname).append(":p").append(proj._idx); + } else if( def instanceof ProjNode proj ) { + String mname = proj.in(0).uniqueName(); + sb.append(mname).append(":p").append(proj._idx); + } else sb.append(def.uniqueName()); + // Number edges, so we can see how they track + sb.append("[taillabel=").append(i); + // The edge from New to ctrl is just for anchoring the New + if ( n instanceof NewNode ) + sb.append(" color=green"); + // control edges are colored red + else if( def instanceof CFGNode ) + sb.append(" color=red"); + else if( def.isMem() ) + sb.append(" color=blue"); + // Backedges do not add a ranking constraint + if( i==2 && (n instanceof PhiNode || n instanceof LoopNode) ) + sb.append(" constraint=false"); + sb.append("];\n"); + } + } + } + } + + // Walk the scope edges + private void scopeEdges( StringBuilder sb, ScopeNode scope ) { + sb.append("\tedge [style=dashed color=cornflowerblue];\n"); + int level=0; + for( var v : scope._vars ) { + Node def = scope.in(v._idx); + while( def instanceof ScopeNode lazy ) + def = lazy.in(v._idx); + if( def==null ) continue; + while( level < scope._lexSize.size() && v._idx >= scope._lexSize.at(level) ) + level++; + String scopeName = makeScopeName(scope, level-1); + sb.append("\t") + .append(scopeName).append(":") + .append('"').append(makePortName(scopeName, v._name)).append('"') // wrap port name with quotes because $ctrl is not valid unquoted + .append(" -> "); + if( def instanceof CProjNode proj ) { + sb.append(def.in(0).uniqueName()).append(":p").append(proj._idx); + } else if( def instanceof ProjNode proj ) { + sb.append(def.in(0).uniqueName()).append(":p").append(proj._idx); + } else sb.append(def.uniqueName()); + sb.append(";\n"); + } + } + + /** + * Finds all nodes in the graph. + */ + public static Collection<Node> findAll(Stack<ScopeNode> ss, Node... ns) { + final HashMap<Integer, Node> all = new HashMap<>(); + for( Node n : ns ) + walk(all, n); + if( ss != null ) + for( Node n : ss ) + walk(all, n); + return all.values(); + } + + /** + * Walk a subgraph and populate distinct nodes in the all list. + */ + private static void walk(HashMap<Integer,Node> all, Node n) { + if(n == null ) return; + if (all.get(n._nid) != null) return; // Been there, done that + all.put(n._nid, n); + for (Node c : n._inputs) + walk(all, c); + for( Node c : n._outputs ) + walk(all, c); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/IRPrinter.java b/chapter20/src/main/java/com/seaofnodes/simple/IRPrinter.java new file mode 100644 index 00000000..3cddacfd --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/IRPrinter.java @@ -0,0 +1,279 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.util.ArrayList; +import java.util.BitSet; +import java.util.HashMap; + +public abstract class IRPrinter { + + // Print a node on 1 line, columnar aligned, as: + // NNID NNAME DDEF DDEF [[ UUSE UUSE ]] TYPE + // 1234 sssss 1234 1234 1234 1234 1234 1234 tttttt + public static SB printLine( Node n, SB sb ) { + sb.p("%4d %-7.7s ".formatted(n._nid,n.label())); + if( n._inputs==null ) + return sb.p("DEAD\n"); + for( Node def : n._inputs ) + sb.p(def==null ? "____" : "%4d".formatted(def._nid)) + // Lazy Phi indicator + .p(n instanceof MemMergeNode && def instanceof MemMergeNode ? "^" : " "); + for( int i = n._inputs.size(); i<4; i++ ) + sb.p(" "); + sb.p(" [[ "); + for( Node use : n._outputs ) + sb.p(use==null ? "____ " : "%4d ".formatted(use._nid)); + int lim = 6 - Math.max(n._inputs.size(),4); + for( int i = n._outputs.size(); i<lim; i++ ) + sb.p(" "); + sb.p(" ]] "); + if( n._type!= null ) sb.p(n._type.str()); + return sb.p("\n"); + } + + + public static String prettyPrint(Node node, int depth) { + return CodeGen.CODE._phase.ordinal() > CodeGen.Phase.Schedule.ordinal() + ? _prettyPrintScheduled( node, depth ) + : _prettyPrint( node, depth ); + } + + // Another bulk pretty-printer. Makes more effort at basic-block grouping. + private static String _prettyPrint( Node node, int depth ) { + // First, a Breadth First Search at a fixed depth. + BFS bfs = new BFS(node,depth); + // Convert just that set to a post-order + ArrayList<Node> rpos = new ArrayList<>(); + BitSet visit = new BitSet(); + for( int i=bfs._lim; i< bfs._bfs.size(); i++ ) + postOrd( bfs._bfs.get(i), null, rpos, visit, bfs._bs); + // Reverse the post-order walk + SB sb = new SB(); + boolean gap=false; + for( int i=rpos.size()-1; i>=0; i-- ) { + Node n = rpos.get(i); + if( n instanceof CFGNode || n.isMultiHead() ) { + if( !gap ) sb.p("\n"); // Blank before multihead + if( n instanceof FunNode fun ) + fun.sig().print(sb.p("--- ").p(fun._name==null ? "" : fun._name).p(" "),false).p("----------------------\n"); + printLine( n, sb ); // Print head + while( --i >= 0 ) { + Node t = rpos.get(i); + if( !t.isMultiTail() ) { i++; break; } + printLine( t, sb ); + } + if( n instanceof ReturnNode ret ) { + FunNode fun = ret.fun(); + fun.sig().print(sb.p("--- ").p(fun._name==null ? "" : fun._name).p(" "),false).p("----------------------\n"); + } + if( !(n instanceof CallNode) ) { + sb.p("\n"); // Blank after multitail + gap = true; + } + } else { + printLine( n, sb ); + gap = false; + } + } + return sb.toString(); + } + + private static void postOrd(Node n, Node prior, ArrayList<Node> rpos, BitSet visit, BitSet bfs) { + if( !bfs.get(n._nid) ) + return; // Not in the BFS visit + if( n instanceof FunNode && !(prior instanceof StartNode) ) + return; // Only visit Fun from Start + if( visit.get(n._nid) ) return; // Already post-order walked + visit.set(n._nid); + // First walk the CFG, then everything + if( n instanceof CFGNode ) { + for( Node use : n._outputs ) + // Follow CFG, not across call/function borders, and not around backedges + if( use instanceof CFGNode && !(n instanceof CallNode && use instanceof FunNode) && + use.nOuts() >= 1 && !(use._outputs.get(0) instanceof LoopNode) ) + postOrd(use, n, rpos,visit,bfs); + for( Node use : n._outputs ) + // Follow CFG, not across call/function borders + if( use instanceof CFGNode && !(n instanceof CallNode && use instanceof FunNode) ) + postOrd(use,n,rpos,visit,bfs); + } + // Follow all outputs + for( Node use : n._outputs ) + if( use != null && + !(n instanceof CallNode && use instanceof FunNode) && + (n instanceof FunNode || !(use instanceof ParmNode)) ) + postOrd(use, n, rpos,visit,bfs); + // Post-order + rpos.add(n); + } + + // Breadth-first search, broken out in a class to keep in more independent. + // Maintains a root-set of Nodes at the limit (or past by 1 if MultiHead). + public static class BFS { + // A breadth first search, plus MultiHeads for any MultiTails + public final ArrayList<Node> _bfs; + public final BitSet _bs; // Visited members by node id + public final int _depth; // Depth limit + public final int _lim; // From here to _bfs._len can be roots for a reverse search + public BFS( Node base, int d ) { + _depth = d; + _bfs = new ArrayList<>(); + _bs = new BitSet(); + + add(base); // Prime the pump + int idx=0, lim=1; // Limit is where depth counter changes + while( idx < _bfs.size() ) { // Ran out of nodes below depth + Node n = _bfs.get(idx++); + for( Node def : n._inputs ) + if( def!=null && !_bs.get(def._nid) ) + add(def); + if( idx==lim ) { // Depth counter changes at limit + if( --d < 0 ) + break; // Ran out of depth + lim = _bfs.size(); // New depth limit + } + } + // Toss things past the limit except multi-heads + while( idx < _bfs.size() ) { + Node n = _bfs.get(idx); + if( n.isMultiHead() ) idx++; + else del(idx); + } + // Root set is any node with no inputs in the visited set + lim = _bfs.size(); + for( int i=_bfs.size()-1; i>=0; i-- ) + if( !any_visited(_bfs.get(i)) ) + swap( i,--lim); + _lim = lim; + } + void swap( int x, int y ) { + if( x==y ) return; + Node tx = _bfs.get(x); + Node ty = _bfs.get(y); + _bfs.set(x,ty); + _bfs.set(y,tx); + } + void add(Node n) { + _bfs.add(n); + _bs.set(n._nid); + } + void del(int idx) { + _bs.clear(_bfs.get(idx)._nid); + Utils.del(_bfs, idx); + } + boolean any_visited( Node n ) { + for( Node def : n._inputs ) + if( def!=null && _bs.get(def._nid) ) + return true; + return false; + } + } + + // Bulk pretty printer, knowing scheduling information is available + private static String _prettyPrintScheduled( Node node, int depth ) { + // Backwards DFS walk to depth. + HashMap<Integer,Integer> ds = new HashMap<>(); + Ary<Node> ns = new Ary<>(Node.class); + _walk(ds,ns,node,depth); + // Remove data projections, these are force-printed behind their multinode head + for( int i=0; i<ns.size(); i++ ) { + if( ns.get(i) instanceof ProjNode proj && !(proj.in(0) instanceof CFGNode) ) { + ns.del(i--); + ds.remove(proj._nid); + } + } + // Print by block with least idepth + SB sb = new SB(); + Ary<Node> bns = new Ary<>(Node.class); + while( !ds.isEmpty() ) { + CFGNode blk = null; + for( Node n : ns ) { + CFGNode cfg = n instanceof CFGNode cfg0 && cfg0.blockHead() ? cfg0 : n.cfg0(); + if( blk==null || cfg.idepth() < blk.idepth() ) + blk = cfg; + } + Integer d = ds.remove(blk._nid); + ns.del(ns.find(blk)); + + // Print block header + sb.p("%-13.13s".formatted(label(blk)+":")); + sb.p( " ".repeat(4) ).p(" [[ "); + if( blk instanceof StartNode ) ; + else if( blk instanceof RegionNode || blk instanceof StopNode ) + for( int i=(blk instanceof StopNode ? 3 : 1); i<blk.nIns(); i++ ) + label(sb,blk.cfg(i)); + else + label(sb,blk.cfg(0)); + sb.p(" ]] \n"); + printLine(blk,sb); + + // Collect block contents that are in the depth limit + bns.clear(); + int xd = Integer.MAX_VALUE; + for( Node use : blk._outputs ) { + Integer i = ds.get(use._nid); + if( i!=null && !(use instanceof CFGNode cfg && cfg.blockHead()) ) { + if( bns.find(use)==-1 ) + bns.add(use); + xd = Math.min(xd,i); + } + } + // Print Phis up front, if any + for( int i=0; i<bns.size(); i++ ) + if( bns.get(i) instanceof PhiNode phi ) + printLine( phi, sb,bns,i--,ds,ns); + + // Print block contents in depth order, bumping depth until whole block printed + for( ; !bns.isEmpty(); xd++ ) + for( int i=0; i<bns.size(); i++ ) { + Node n = bns.get(i); + if( ds.get(n._nid)==xd ) { + printLine( n, sb, bns, i--, ds,ns ); + if( n instanceof MultiNode && !(n instanceof CFGNode) ) { + for( Node use : n._outputs ) { + printLine(use,sb,bns,bns.indexOf(use),ds,ns); + } + } + } + } + sb.p("\n"); + } + return sb.toString(); + } + + private static void _walk( HashMap<Integer,Integer> ds, Ary<Node> ns, Node node, int d ) { + Integer nd = ds.get(node._nid); + if( nd!=null && d <= nd ) return; // Been there, done that + Integer old = ds.put(node._nid,d) ; + if( old == null ) + ns.add(node); + if( d == 0 ) return; // Depth cutoff + for( Node def : node._inputs ) + if( def != null && + !(node instanceof LoopNode loop && loop.back()==def) && + // Don't walk into or out of functions + !(node instanceof CallEndNode && def instanceof ReturnNode) && + !(node instanceof FunNode && def instanceof CallNode) && + !(node instanceof ParmNode && !(def instanceof FunNode)) + ) + _walk(ds,ns,def,d-1); + } + + static String label( CFGNode blk ) { + if( blk instanceof StartNode ) return "START"; + return (blk instanceof LoopNode ? "LOOP" : "L")+blk._nid; + } + static void label( SB sb, CFGNode blk ) { + if( !blk.blockHead() ) blk = blk.cfg(0); + sb.p( "%-9.9s ".formatted( label( blk ) ) ); + } + static void printLine( Node n, SB sb, Ary<Node> bns, int i, HashMap<Integer,Integer> ds, Ary<Node> ns ) { + printLine( n, sb ); + if( i != -1 ) bns.del(i); + ds.remove(n._nid); + int idx = ns.find(n); + if( idx!=-1 ) ns.del(idx); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/IterPeeps.java b/chapter20/src/main/java/com/seaofnodes/simple/IterPeeps.java new file mode 100644 index 00000000..265e0b21 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/IterPeeps.java @@ -0,0 +1,196 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; + +import java.util.Arrays; +import java.util.BitSet; +import java.util.Random; + +/** + * The IterPeeps runs after parsing. It iterates the peepholes to a fixed point + * so that no more peepholes apply. This should be linear because peepholes rarely + * (never?) increase code size. The graph should monotonically reduce in some + * dimension, which is usually size. It might also reduce in e.g. number of + * MulNodes or Load/Store nodes, swapping out more "expensive" Nodes for cheaper + * ones. + * <br> + * The theoretical overall worklist is mindless just grabbing the next thing and + * doing it. If the graph changes, put the neighbors on the worklist. + * <br> + * Lather, Rinse, Repeat until the worklist runs dry. + * <p> + * The main issues we have to deal with: + * + * <ul> + * <li>Nodes have uses; replacing some set of Nodes with another requires more graph + * reworking. Not rocket science, but it can be fiddly. Its helpful to have a + * small set of graph munging utilities, and the strong invariant that the graph + * is stable and correct between peepholes. In our case `Node.subsume` does + * most of the munging, building on our prior stable Node utilities.</li> + * + * <li>Changing a Node also changes the graph "neighborhood". The neighbors need to + * be checked to see if THEY can also peephole, and so on. After any peephole + * or graph update we put a Nodes uses and defs on the worklist.</li> + * + * <li>Our strong invariant is that for all Nodes, either they are on the worklist + * OR no peephole applies. This invariant is easy to check, although expensive. + * Basically the normal "iterate peepholes to a fixed point" is linear, and this + * check is linear at each peephole step... so quadratic overall. Its a useful + * assert, but one we can disable once the overall algorithm is stable - and + * then turn it back on again when some new set of peepholes is misbehaving. + * The code for this is turned on in `IterPeeps.iterate` as `assert + * progressOnList(stop);`</li> + * </ul> + */ +public class IterPeeps { + + private final WorkList<Node> _work; + + IterPeeps( long seed ) { _work = new WorkList<>(seed); } + + public <N extends Node> N add( N n ) { return (N)_work.push(n); } + + public void addAll( Ary<Node> ary ) { _work.addAll(ary); } + + /** + * Iterate peepholes to a fixed point + */ + public void iterate( CodeGen code ) { + assert progressOnList(code); + int cnt=0; + + Node n; + while( (n=_work.pop()) != null ) { + if( n.isDead() ) continue; + cnt++; // Useful for debugging, searching which peephole broke things + Node x = n.peepholeOpt(); + if( x != null ) { + if( x.isDead() ) continue; + // peepholeOpt can return brand-new nodes, needing an initial type set + if( x._type==null ) x.setType(x.compute()); + // Changes require neighbors onto the worklist + if( x != n || !(x instanceof ConstantNode) ) { + // All outputs of n (changing node) not x (prior existing node). + for( Node z : n._outputs ) _work.push(z); + // Everybody gets a free "go again" in case they didn't get + // made in their final form. + _work.push(x); + // If the result is not self, revisit all inputs (because + // there's a new user), and replace in the graph. + if( x != n ) { + for( Node z : n. _inputs ) _work.push(z); + n.subsume(x); + } + } + // If there are distant neighbors, move to worklist + n.moveDepsToWorklist(); + JSViewer.show(); // Show again + assert progressOnList(code); // Very expensive assert + } + if( n.isUnused() && !(n instanceof StopNode) ) + n.kill(); // Just plain dead + } + + } + + // Visit ALL nodes and confirm the invariant: + // Either you are on the _work worklist OR running `iter()` makes no progress. + + // This invariant ensures that no progress is missed, i.e., when the + // worklist is empty we have indeed done all that can be done. To help + // with debugging, the {@code assert} is broken out in a place where it is easy to + // stop if a change is found. + + // Also, the normal usage of `iter()` may attempt peepholes with distance + // neighbors and these should fail, but will then try to add dependencies + // {@link #Node.addDep} which is a side effect in an assert. The {@link + // #midAssert} is used to stop this side effect. + private boolean progressOnList(CodeGen code) { + code._midAssert = true; + Node changed = code._stop.walk( n -> { + Node m = n; + if( n.compute().isa(n._type) && (!n.iskeep() || n._nid<=6) ) { // Types must be forwards, even if on worklist + if( _work.on(n) ) return null; + m = n.peepholeOpt(); + if( m==null ) return null; + } + System.err.println("BREAK HERE FOR BUG"); + return m; + }); + code._midAssert = false; + return changed==null; + } + + /** + * Classic WorkList, with a fast add/remove, dup removal, random pull. + * The Node's nid is used to check membership in the worklist. + */ + public static class WorkList<E extends Node> { + + private Node[] _es; + private int _len; + private final BitSet _on; // Bit set if Node._nid is on WorkList + private final Random _R; // For randomizing pull from the WorkList + private final long _seed; + + /* Useful stat - how many nodes are processed in the post parse iterative opt */ + private long _totalWork = 0; + + WorkList() { this(123); } + WorkList(long seed) { + _es = new Node[1]; + _len=0; + _on = new BitSet(); + _seed = seed; + _R = new Random(); + _R.setSeed(_seed); + } + + /** + * Pushes a Node on the WorkList, ensuring no duplicates + * If Node is null it will not be added. + */ + public E push( E x ) { + if( x==null ) return null; + int idx = x._nid; + if( !_on.get(idx) ) { + _on.set(idx); + if( _len==_es.length ) + _es = Arrays.copyOf(_es,_len<<1); + _es[_len++] = x; + _totalWork++; + } + return x; + } + + public void addAll( Ary<E> ary ) { + for( E n : ary ) + push(n); + } + + /** + * True if Node is on the WorkList + */ + boolean on( E x ) { return _on.get(x._nid); } + boolean isEmpty() { return _len==0; } + + /** + * Removes a random Node from the WorkList; null if WorkList is empty + */ + E pop() { + if( _len == 0 ) return null; + int idx = _R.nextInt(_len); + E x = (E)_es[idx]; + _es[idx] = _es[--_len]; // Compress array + _on.clear(x._nid); + return x; + } + + public void clear() { + _len = 0; + _on.clear(); + _R.setSeed(_seed); + _totalWork = 0; + } + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/JSViewer.java b/chapter20/src/main/java/com/seaofnodes/simple/JSViewer.java new file mode 100644 index 00000000..fcb0c28a --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/JSViewer.java @@ -0,0 +1,402 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; +import java.io.IOException; +import java.util.Stack; +import java.nio.file.Paths; +import java.util.Collection; +import static com.seaofnodes.simple.CodeGen.CODE; + +public class JSViewer implements AutoCloseable { + // Display programs in an endless loop + public static void main( String[] args ) throws Exception { + try(var js = new JSViewer()) { + js.run(); + } + } + + static boolean SHOW; + + // WebSocket to the browser for display + static SimpleWebSocket SERVER; + + static int N; // Dot frames + + JSViewer() throws Exception { + // Launch server; handshake + SERVER = new SimpleWebSocket(Paths.get("docs/index.html").toUri(),12345) ; + } + + void run( ) throws Exception { + SERVER.put("!"); + while( true ) { + String src = SERVER.get(); + switch( src ) { + case null: return; + case "null": return; + case "+": break; // Client requests more frames, but we send them all anyway + default: + System.out.println(src); + try { + N=0; + // Use parser scope, xscope when building views + CodeGen code = new CodeGen(src); + SHOW = true; + show(); + // Parse program, generating views at every parse point + code.parse(); + // No longer user parse internal state when building views + code.opto(); + + // Catch and ignore Parser errors + } catch(RuntimeException re) { + System.err.println(re); + } finally { + SHOW = false; + SERVER.put("#"); // Final frame + } + break; + } + } + } + + @Override public void close() throws IOException { + SERVER.close(); + SERVER=null; + } + + public static void show() { if( SERVER!=null && SHOW ) _show(); } + private static void _show() { + // Skip util we at least get the Parse object made + if( CODE._phase==null || CODE._phase.ordinal() < CodeGen.Phase.Parse.ordinal() ) + return; + boolean midParse = CODE._phase == CodeGen.Phase.Parse; + Stack<ScopeNode> xScopes = midParse ? CODE.P._xScopes : null; + + Collection<Node> all = GraphVisualizer.findAll(xScopes, CODE._stop, midParse ? CODE.P._scope: null); + SB sb = new SB(); + sb.p("digraph view_").p(N++).p(" {\n").ii(); + if( midParse ) + sb.i().p("// POS: ").p(CODE.P.pos()).nl(); + + // To keep the Scopes below the graph and pointing up into the graph we + // need to group the Nodes in a subgraph cluster, and the scopes into a + // different subgraph cluster. THEN we can draw edges between the + // scopes and nodes. If we try to cross subgraph cluster borders while + // still making the subgraphs DOT gets confused. + sb.i().p("rankdir=BT;\n"); // Force Nodes before Scopes + + // CNC - turned off Apr/8/2024, gives more flex in the layout and + // removes some of the more ludicrous layout choices. + // Preserve node input order + //sb.append("ordering=\"in\";\n"); + + // Merge multiple edges hitting the same node. Makes common shared + // nodes much prettier to look at. + sb.i().p("concentrate=\"true\";\n"); + + // Force nested scopes to order + sb.i().p("compound=\"true\";\n"); + + // Just the Nodes first, in a cluster no edges + nodes(sb, all); + + // Now the scopes, in a cluster no edges + if( midParse ) + for( ScopeNode xscope : xScopes ) + if( !xscope.isDead() ) + scope( sb, xscope ); + + // Walk the Node edges + nodeEdges(sb, midParse && !xScopes.empty() ? xScopes.peek() : null, all); + + // Walk the active Scope edges + if( midParse ) + for( ScopeNode xscope : xScopes ) + if( !xscope.isDead() ) + scopeEdges( sb, xscope ); + + + sb.p("}\n").di(); + // Tell client another DOT frame + String dot = sb.toString(); + System.out.println(dot); + try { + SERVER.put(dot); + } catch( IOException ioe ) { + try { SERVER.close();} catch( IOException ignored ){} + } + } + + + private static void nodes(SB sb, Collection<Node> all) { + // Just the Nodes first, in a cluster no edges + sb.i().p("subgraph cluster_Nodes {\n").ii(); // Magic "cluster_" in the subgraph name + for( Node n : all ) { + if( n instanceof ProjNode || n instanceof CProjNode || n instanceof MemMergeNode ) + continue; // Do not emit, rolled into MultiNode or Scope cluster already + sb.i().p(n.uniqueName()).p(" [ "); + if( n instanceof MultiNode && !(n instanceof StartNode) ) { + // Make a box with the MultiNode on top, and all the projections on the bottom + sb. p("shape=plaintext label=<\n").ii(); + sb.i().p("<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\" CELLPADDING=\"4\">\n"); + // Row over single cell for node label + cell(sb.i().p("<TR>"), n.glabel(), n, null).p("</TR>\n"); + // Row over single cell, for nested table + sb.i().p("<TR><TD>\n").ii(); + sb.i().p("<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\">").p("\n"); + sb.i().p("<TR>"); + n._outputs.sort((x,y) -> x instanceof ProjNode xp && y instanceof ProjNode yp ? (xp._idx - yp._idx) : ((x==null ? 99999 : x._nid) - (y==null ? 99999 : y._nid))); + boolean empty_row=true; + for( Node use : n._outputs ) { + int idx = idx(n); + if( idx != -1 ) { + cell(sb,use.glabel(),use,"p"+idx); + empty_row=false; + } + } + // At least one cell on row + if( empty_row ) sb.p("<TD></TD>"); + sb. p("</TR>").p("\n"); + sb.i().p("</TABLE>").p("\n").di(); + sb.i().p("</TD></TR>\n"); + sb.i().p("</TABLE>>\n").di(); + sb.i().p("];\n"); + + } else { + // control nodes have box shape + // other nodes are ellipses, i.e. default shape + //else if (n instanceof PhiNode) sb.p("style=filled fillcolor=lightyellow "); + node(sb,n.glabel(),n); + sb.p("];\n"); + } + } + // Force Region & Phis to line up + for (Node n : all) { + if (n instanceof RegionNode region) { + sb.i().p("{ rank=same; "); + sb.p(region.uniqueName()).p(";"); + for (Node phi : region._outputs) + if (phi instanceof PhiNode) sb.p(phi.uniqueName()).p(";"); + sb.p("}\n"); + } + } + sb.di().i().p("}").nl(); // End Node cluster + } + + // Build a nested scope display, walking the _prev edge + private static void scope( SB sb, ScopeNode scope ) { + sb.i().p("node [shape=plaintext];\n"); + int last = scope.nIns(); + int max = scope._lexSize.size()+1; // One more than lexsize + for( int i = 0; i < max; i++ ) { + int level = max-i-1; + String scopeName = makeScopeName(scope, level); + sb.i().p("subgraph cluster_").p(scopeName).p(" {\n").ii(); // Magic "cluster_" in the subgraph name + sb.i().p(scopeName).p(" [label=<\n").ii(); + sb.i().p("<TABLE BORDER=\"0\" CELLBORDER=\"1\" CELLSPACING=\"0\">\n"); + int lexStart = level==0 ? 0 : scope._lexSize.at(level-1); + + // Special for memory ScopeMinNode + MemMergeNode n = scope.nIns()>1 ? scope.mem() : null; + if( level==0 && n!=null && n.nIns()>2 ) { + sb.i().p("<TR>"); + for( int m=2; m<n.nIns(); m++ ) + cell(sb,"#"+m,n.alias(m),"m"+m); + sb.p("</TR>\n"); // End scope level + } + + // Scope variables, empty if none + if( lexStart<last ) { + sb.i().p("<TR>\n"); + for( int j=lexStart; j<last; j++ ) { + if( scope.in(j)==null ) continue; + var v = scope._vars.at(j); + cell(sb.i(),v._name,scope.in(j),makePortName(scopeName, v._name)).nl(); + } + last = lexStart; + sb.i().p("</TR>\n"); + } + sb.i().p("</TABLE>>];\n").di(); + // Scope label + sb.i().p(scopeName).p("A [label=\"").p(level).p("\" shape=doublecircle style=filled fillcolor=aqua fontsize=10 margin=0 width=0.2 ];\n"); + } + // Scope clusters nest, so the graphics shows the nested scopes, so + // they are not closed as they are printed; so they just keep nesting. + // We close them all at once here. + for( int i=0; i<max; i++ ) + sb.di().i().p("}\n"); + } + + // Append a cell, with color + private static SB cell(SB sb, String text, Node n, String port) { + sb.p("<TD"); + String color = color(n); + sb.p(" BGCOLOR=\"").p(color).p("\""); + if( port!=null ) sb.p(" PORT=\"").p(port).p("\""); + return colorcell(sb.p(">"),text,n,dark(color)).p("</TD>"); + } + private static SB node(SB sb, String text, Node n) { + String color = color(n); + sb.p("style=filled fillcolor=").p(color).p(" "); + String shape = switch( n ) { + case FunNode fun -> "shape=box peripheries=2"; + case CFGNode cfg -> "shape=box"; + default -> ""; + }; + sb.p(shape).p(" "); + return colorcell(sb.p("label=<"),text,n,dark(color)).p(">"); + } + + private static String color(Node n) { + return switch( n ) { + case CFGNode cfg -> "yellow"; + case NewNode nnn -> "lightgreen"; + case StructNode s -> "lightgreen"; + default -> typeColor(n._type); + }; + } + private static String typeColor(Type t) { + return t==null ? "white" : switch( t ) { + case TypeMem mem -> "blue"; + case TypeMemPtr tmp -> "green"; + case TypeFunPtr tfp -> "orange"; + case TypeRPC rpc -> "antiquewhite"; + case TypeInteger ti -> "lightblue"; + case TypeFloat tf -> "aqua"; + case Type tn -> tn==Type.NIL ? "lightgray" : "red"; + }; + } + private static final String[] DARKS = new String[]{"blue","green","black"}; + private static boolean dark(String color) { return Utils.find(DARKS,color) != -1; } + + + + // Called with an open div + private static SB colorcell(SB sb, String text, Node n, boolean dark) { + if( dark ) sb.p("<font color=\"white\">"); + sb.p(text); + if( n._type!=null ) { + sb.p("<br /><font point-size=\"10\">"); + n._type.gprint(sb); + sb.p("</font>"); + } + if( dark ) sb.p("</font>"); + return sb; + } + + + private static String makeScopeName(ScopeNode sn, int level) { return sn.uniqueName() + "_" + level; } + private static String makePortName(String scopeName, String varName) { return scopeName + "_" + varName; } + + // Walk the node edges + private static void nodeEdges(SB sb, ScopeNode scope, Collection<Node> all) { + // All them edge labels + String scopeName = scope!=null ? makeScopeName(scope, 0) : null; + sb.i().p("edge [ fontname=Helvetica, fontsize=8 ];\n"); + for( Node n : all ) { + // Do not display the Constant->Start edge; + if( n instanceof ConstantNode || + n instanceof XCtrlNode || + // ProjNodes handled by Multi; + n instanceof ProjNode || + n instanceof CProjNode || + // ScopeNodes are done separately + n instanceof ScopeNode || + n instanceof MemMergeNode + ) + continue; + if( n.isDead() ) + throw Utils.TODO(); + // Normal edges + for( int i=0; i<n.nIns(); i++ ) { + Node def = n.in(i); + if( def==null ) continue; + sb.i().p(n.uniqueName()).p(" -> "); + if( n instanceof PhiNode && def instanceof RegionNode ) { + // Draw a dotted use->def edge from Phi to Region. + sb.p(def.uniqueName()); + sb.p(" [style=dotted taillabel=").p(i).p("];\n"); + } else { + // Most edges land here use->def + if( def instanceof CProjNode proj ) { + String mname = proj.ctrl().uniqueName(); + sb.p(mname).p(":p").p(proj._idx); + } else if( def instanceof ProjNode proj ) { + String mname = proj.in(0).uniqueName(); + sb.p(mname).p(":p").p(proj._idx); + } else sb.p(def.uniqueName()); + // Number edges, so we can see how they track + sb.p("[taillabel=").p(i); + // Color the edge + nodeEdgeColor(sb,def._type); + // Backedges do not add a ranking constraint + if( i==2 && (n instanceof PhiNode || n instanceof LoopNode) ) + sb.p(" constraint=false"); + sb.p("];\n"); + } + } + + // Bonus edge if hooked by parser + if( (n.iskeep() || n.isUnused()) && scopeName != null ) { + sb.i().p(scopeName).p(" -> ").p(n.uniqueName()).p(" [ style=dashed color=grey];\n"); + } + // Force functions above return + if( n instanceof ReturnNode ret && ret.inProgress() && !ret.fun().isDead() ) + sb.i().p(ret.ctrl().uniqueName()).p(" -> ").p(ret.fun().uniqueName()).p("[style=invis]\n"); + } + } + + private static void nodeEdgeColor(SB sb, Type t) { + sb.p(" color=").p(typeColor(t)); + } + + // Walk the scope edges + private static void scopeEdges( SB sb, ScopeNode scope ) { + sb.i().p("edge [style=dashed color=cornflowerblue];\n"); + int level=0; + for( var v : scope._vars ) { + if( v._idx==1 ) continue; // Memory done separately + Node def = scope.in(v._idx); + while( def instanceof ScopeNode lazy ) + def = lazy.in(v._idx); + if( def==null ) continue; + while( level < scope._lexSize.size() && v._idx >= scope._lexSize.at(level) ) + level++; + String scopeName = makeScopeName(scope, level); + sb.i() + .p(scopeName).p(":") + .p('"').p(makePortName(scopeName, v._name)).p('"') // wrap port name with quotes because $ctrl is not valid unquoted + .p(" -> "); + defPort(sb,def).p(";\n"); + } + + // Memory + if( scope.nIns()>1 ) { + MemMergeNode n = scope.mem(); + for( int i=2; i<n.nIns(); i++ ) { + Node def = n.in(i); + while( def instanceof ScopeNode lazy ) + def = lazy.in(i); + if( def==null ) continue; + String scopeName = makeScopeName(scope, 0); + sb.i().p(scopeName).p(":m").p(i).p(" -> "); + defPort(sb,def).p(";\n"); + } + } + + } + + // (C)Projection index or -1 + private static int idx(Node n) { + return n instanceof ProjNode p ? p._idx : (n instanceof CProjNode cp ? cp._idx : -1); + } + + private static SB defPort(SB sb, Node def) { + int idx = idx(def); + return idx != -1 + ? sb.p(def.in(0).uniqueName()).p(":p").p(idx) + : sb.p(def.uniqueName()); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/ListScheduler.java b/chapter20/src/main/java/com/seaofnodes/simple/ListScheduler.java new file mode 100644 index 00000000..18055e19 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/ListScheduler.java @@ -0,0 +1,231 @@ + package com.seaofnodes.simple; + +import com.seaofnodes.simple.IterPeeps.WorkList; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; +import java.util.*; + +public abstract class ListScheduler { + + // eXtra per-node stuff for scheduling. + private static final IdentityHashMap<Node,XSched> XS = new IdentityHashMap<>(); + private static class XSched { + static final Ary<XSched> FREE = new Ary<>(XSched.class); + static void alloc(CFGNode bb, Node n) { + XSched x = FREE.pop(); + XS.put(n, (x==null ? new XSched() : x).init(bb,n)); + } + void free() { _n = null; FREE.push(this); } + + static XSched get(Node n) { return n==null ? null : XS.get(n); } + + private XSched init(CFGNode bb, Node n) { + _n = n; + _bcnt = 0; + _ruse = _rdef = _single = false; + + // Count block-locals and remote inputs + if( !(n instanceof PhiNode) ) + for( Node def : n._inputs ) + if( def != null && def != bb ) { + if( def.cfg0()==bb ) _bcnt++; // Raise block-local input count + else _ruse = true; + } + // Count remote outputs and allowed registers + if( n instanceof MultiNode ) + for( Node use : n._inputs ) + computeSingleRDef(bb,use); + else + computeSingleRDef(bb,n); + return this; + } + + private void computeSingleRDef(CFGNode bb, Node n) { + RegMaskRW rmask = n instanceof MachNode mach ? mach.outregmap().copy() : null; + for( Node use : n._outputs ) { + // Remote use, so this is a remote def + if( use!=null && use.cfg0()!=bb ) _rdef = true; + // And mask between n's def mask and all n's users masks' + if( use instanceof MachNode mach && rmask != null ) + for( int i=1; i<use.nIns(); i++ ) + if( use.in(i)==n ) + rmask.and(mach.regmap(i)); + } + // Defines in a single register + _single = rmask!=null && rmask.size1(); + } + + // If _bcnt==0, declare ready; move user bcnts into rcnts. + boolean isReady() { + if( _bcnt > 0 || _rcnt > 0 ) return false; + if( _n instanceof MultiNode ) + for( Node use : _n._outputs ) + if( use instanceof ProjNode ) readyUp(use); + // anti-dep from a multireg directly + else _readyUp(use); + else + readyUp(_n); + return true; + } + private static void readyUp(Node n) { + for( Node use : n._outputs ) + _readyUp(use); + } + private static void _readyUp(Node use) { + XSched xs = get(use); + if( xs!=null && !(use instanceof PhiNode) ) + { assert xs._bcnt>0; xs._bcnt--; xs._rcnt++; } + } + + boolean decIsReady() { + assert _rcnt > 0; + _rcnt--; + return isReady(); + } + + Node _n; // Node this extra is for + int _bcnt; // Not-ready not-scheduled inputs + int _rcnt; // Ready not-scheduled inputs + boolean _ruse; // Node IS a "remote use", uses a other-block value + boolean _rdef; // Node IS a "remote def" of a value used in other block + boolean _single; // Defines a single register, likely to conflict + } + + + // List schedule every block + public static void sched( CodeGen code ) { + XS.clear(); + for( CFGNode bb : code._cfg ) + if( bb.blockHead() ) + local(bb); + } + + // Classic list scheduler + private static void local( CFGNode bb ) { + assert XS.isEmpty(); + int len = bb.nOuts(); + + // Count block-locals, remote inputs and outputs + for( int i=0; i<len; i++ ) + XSched.alloc(bb,bb.out(i)); + + // Nodes are ready if they are not used by other nodes in this block. + // Move ready up front. + int ready = 0; + for( int i=0; i<len; i++ ) + if( XSched.get(bb.out(i)).isReady() ) // No block-local inputs? + bb._outputs.swap(ready++,i); // Move into ready set + assert len==0 || ready > 0; + + // Classic list scheduler. Behind sched is scheduled. Between sched + // and ready have zero counts and are ready to schedule. Ahead of + // ready is undefined. + // + // As nodes are scheduled, the nodes they use decrement their bcnts[]. Once + // that gets to 0, all the nodes using it are scheduled, so it's ready. + int sched = 0; + while( sched < len ) { + assert sched < ready; + int pick = best(bb._outputs,sched,ready); // Pick best + Node best = bb._outputs.swap(pick,sched++); // Swap best to head of ready and move into the scheduled set + // Lower ready count of users, and since projections are ready as + // soon as their multi is ready, lower those now too. + for( Node use : best._outputs ) + if( use instanceof ProjNode ) + for( Node useuse : use._outputs ) + ready = ready(bb,useuse,ready); + else + ready = ready(bb,use,ready); + } + + // Reset for next block + for( XSched x : XS.values() ) + x.free(); + XS.clear(); + } + + private static int ready(CFGNode bb, Node use, int ready) { + if( use!=null && !(use instanceof PhiNode) && use.in(0)==bb && XSched.get(use).decIsReady() ) + bb._outputs.set(ready++, use); // Became ready, move into ready set + //bb._outputs.swap(ready++,use); // Became ready, move into ready set + return ready; + } + + + // Pick best between sched and ready. + private static int best( Ary<Node> blk, int sched, int ready ) { + int pick = sched; + int score = score(blk.at(pick)); + for( int i=sched+1; i<ready; i++ ) { + int nscore = score(blk.at(i)); + if( nscore > score ) + { score = nscore; pick = i; } + } + return pick; + } + + // Highest score wins. Max at 1000, min at 10, except specials. + private static final int[] CNT = new int[3]; + static int score( Node n ) { + if( n instanceof ProjNode ) return 1001; // Pinned at block entry + if( n instanceof CProjNode ) return 1001; // Pinned at block entry + if( n instanceof PhiNode ) return 1000; + if( n instanceof CFGNode ) return 1; // Pinned at block exit + + int score = 500; + // Ideal nodes ignore register pressure scheduling + if( !(n instanceof MachNode) ) + return score; + + // Register pressure local scheduling: avoid overlapping specialized + // registers usages. + + // Subtract 10 if this op forces a live range to exist that cannot be + // resolved immediately; i.e., live count goes up. Scale to 20 if + // multi-def. + + // Subtract 100 if this op forces a single-def live range to exist + // which might conflict on the same register with other live ranges. + // Defines a single register based on def & uses, and the output is not + // ready. Scale to 200 for multi-def. + CNT[1]=CNT[2]=0; + XSched xn = XSched.get(n); + if( xn._rdef ) score = 200; // If defining a remote value, just generically stall alot + if( n instanceof MultiNode ) { + for( Node use : n._outputs ) + singleUseNotReady( use, xn._single ); + } else if( n.nOuts() == 1 ) + singleUseNotReady( n, xn._single ); + score += -10 * Math.min( CNT[1], 2 ); + score += -100 * Math.min( CNT[2], 2 ); + + // Add 10 if this op ends a single-def live range, add 20 for 2 or more. + // Scale to 100,200 if the single-def is also single-register. + CNT[1]=CNT[2]=0; + for( int i=1; i<n.nIns(); i++ ) { + XSched xd = XSched.get( n.in(i) ); + if( xd != null && xd._n.nOuts()==1 ) + CNT[xd._single || ((MachNode)n).regmap(i).size1() ? 2 : 1]++; + } + score += 10 * Math.min( CNT[1], 2 ); + score += 100 * Math.min( CNT[2], 2 ); + + //// Nothing special + //if( n instanceof Pe ) return 500; // RISC want to go after PE + //if( n instanceof Risc ) return 400; // RISC want to go after PE + assert 10 <= score && score <= 990; + return score; + } + + // Single use, and that use is not ready to fire once this one does. + // If true, stalling 'n' might reduce 'n's lifetime. + // Return 0 for false, 1 if true, 10 if also single register + private static void singleUseNotReady( Node n, boolean single ) { + if( n.nOuts() != 1 ) return; + XSched xu = XSched.get(n.out(0)); + if( xu !=null && xu._bcnt==0 ) + return; + CNT[single ? 2 : 1]++; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/Machine.java b/chapter20/src/main/java/com/seaofnodes/simple/Machine.java new file mode 100644 index 00000000..b29e9aba --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/Machine.java @@ -0,0 +1,21 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.node.CFGNode; +import com.seaofnodes.simple.node.IfNode; +import com.seaofnodes.simple.node.Node; + +abstract public class Machine { + // Human readable machine name. Something like "x86-64" or "arm" or "risc5" + public abstract String name(); + // Human-readable name for a register number, e.g. "RAX" or "R0" + public abstract String reg( int reg ); + // Create a split op; any register to any register, including stack slots + public abstract Node split(); + // Return a MachNode unconditional branch + public abstract CFGNode jump(); + // Break an infinite loop + public abstract IfNode never( CFGNode ctrl ); + // Instruction select from ideal nodes + public abstract Node instSelect( Node n ); +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/Parser.java b/chapter20/src/main/java/com/seaofnodes/simple/Parser.java new file mode 100644 index 00000000..61a53d68 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/Parser.java @@ -0,0 +1,1729 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; +import java.text.ParseException; +import java.util.*; + +/** + * The Parser converts a Simple source program to the Sea of Nodes intermediate + * representation directly in one pass. There is no intermediate Abstract + * Syntax Tree structure. + * <p> + * This is a simple recursive descent parser. All lexical analysis is done here as well. + */ +public class Parser { + + public static ConstantNode ZERO; // Very common node, cached here + public static ConstantNode NIL; // Very common node, cached here + public static XCtrlNode XCTRL; // Very common node, cached here + + // Compile driver + public final CodeGen _code; + + // The Lexer. Thin wrapper over a byte[] buffer with a cursor. + private final Lexer _lexer; + + /** + * Current ScopeNode - ScopeNodes change as we parse code, but at any point of time + * there is one current ScopeNode. The reason the current ScopeNode can change is to do with how + * we handle branching. See {@link #parseIf()}. + * <p> + * Each ScopeNode contains a stack of lexical scopes, each scope is a symbol table that binds + * variable names to Nodes. The top of this stack represents current scope. + * <p> + * We keep a list of all ScopeNodes so that we can show them in graphs. + * @see #parseIf() + * @see #_xScopes + */ + public ScopeNode _scope; + + /** + * List of keywords disallowed as identifiers + */ + public static final HashSet<String> KEYWORDS = new HashSet<>(){{ + add("bool"); + add("break"); + add("byte"); + add("continue"); + add("else"); + add("f32"); + add("f64"); + add("false"); + add("flt"); + add("i16"); + add("i32"); + add("i64"); + add("i8"); + add("if"); + add("int"); + add("new"); + add("null"); + add("return"); + add("struct"); + add("true"); + add("u1"); + add("u16"); + add("u32"); + add("u8"); + add("while"); + }}; + + /** + * We clone ScopeNodes when control flows branch; it is useful to have + * a list of all active ScopeNodes for purposes of visualization of the SoN graph + */ + public final Stack<ScopeNode> _xScopes = new Stack<>(); + + ScopeNode _continueScope; + ScopeNode _breakScope; // Merge all the while-breaks here + FunNode _fun; // Current function being parsed + + // Mapping from a type name to a Type. The string name matches + // `type.str()` call. No TypeMemPtrs are in here, because Simple does not + // have C-style '*ptr' references. + public static HashMap<String, Type> TYPES; + + // Mapping from a type name to the constructor for a Type. + public final HashMap<String, StructNode> INITS; + + + public Parser(CodeGen code, TypeInteger arg) { + _code = code; + _lexer = new Lexer(code._src); + _scope = new ScopeNode(); + _continueScope = _breakScope = null; + ZERO = con(TypeInteger.ZERO).keep(); + NIL = con(Type.NIL).keep(); + XCTRL= new XCtrlNode().peephole().keep(); + TYPES = defaultTypes(); + INITS = new HashMap<>(); + } + + @Override + public String toString() { return _lexer.toString(); } + + public static HashMap<String, Type> defaultTypes() { + return new HashMap<>() {{ + put("bool",TypeInteger.U1 ); + put("byte",TypeInteger.U8 ); + put("f32" ,TypeFloat .F32); + put("f64" ,TypeFloat .F64); + put("flt" ,TypeFloat .F64); + put("i16" ,TypeInteger.I16); + put("i32" ,TypeInteger.I32); + put("i64" ,TypeInteger.BOT); + put("i8" ,TypeInteger.I8 ); + put("int" ,TypeInteger.BOT); + put("u1" ,TypeInteger.U1 ); + put("u16" ,TypeInteger.U16); + put("u32" ,TypeInteger.U32); + put("u8" ,TypeInteger.U8 ); + put("val" ,Type.TOP); // Marker type, indicates type inference + put("var" ,Type.BOTTOM); // Marker type, indicates type inference + }}; + } + + // Debugging utility to find a Node by index + public Node f(int nid) { return _code.f(nid); } + + private Node ctrl() { return _scope.ctrl(); } + private <N extends Node> N ctrl(N n) { return _scope.ctrl(n); } + + public void parse() { parse(false); } + public void parse(boolean show) { + _xScopes.push(_scope); + _scope.define(ScopeNode.CTRL, Type.CONTROL , false, null, _lexer); + _scope.define(ScopeNode.MEM0, TypeMem.BOT , false, null, _lexer); + _scope.define(ScopeNode.ARG0, TypeInteger.BOT, false, null, _lexer); + + ctrl(XCTRL); + _scope.mem(new MemMergeNode(false)); + + // Parse whole program, as-if function header "{ int arg -> body }" + parseFunctionBody(_code._main,loc(),"arg"); + _code.link(_code._main)._name = "main"; + + if( !_lexer.isEOF() ) throw _errorSyntax("unexpected"); + + // Clean up and reset + _xScopes.pop(); + _scope.kill(); + for( StructNode init : INITS.values() ) + init.unkeep().kill(); + INITS.clear(); + _code._stop.peephole(); + if( show ) showGraph(); + } + + /** + * Parses a function body, assuming the header is parsed. + */ + private ReturnNode parseFunctionBody(TypeFunPtr sig, Lexer loc, String... ids) { + // Stack parser state on the local Java stack, and unstack it later + Node oldctrl = ctrl().keep(); + Node oldmem = _scope.mem().keep(); + FunNode oldfun = _fun; + ScopeNode breakScope = _breakScope; _breakScope = null; + ScopeNode continueScope = _continueScope; _continueScope = null; + + FunNode fun = _fun = (FunNode)peep(new FunNode(loc(),sig,null,_code._start)); + // Once the function header is available, install in linker table - + // allowing recursive functions. Linker matches on declared args and + // exact fidx, and ignores the return (because the fidx will only match + // the exact single function). + _code.link(fun); + + Node rpc = new ParmNode("$rpc",0,TypeRPC.BOT,fun,con(TypeRPC.BOT)).peephole(); + + // Build a multi-exit return point for all function returns + RegionNode r = new RegionNode((Lexer)null,null,null).init(); + assert r.inProgress(); + PhiNode rmem = new PhiNode(ScopeNode.MEM0,TypeMem.BOT,r,null).init(); + PhiNode rrez = new PhiNode(ScopeNode.ARG0,Type.BOTTOM,r,null).init(); + ReturnNode ret = new ReturnNode(r, rmem, rrez, rpc, fun).init(); + fun.setRet(ret); + assert ret.inProgress(); + _code._stop.addDef(ret); + + // Pre-call the function from Start, with worse-case arguments. This + // represents all the future, yet-to-be-parsed functions calls and + // external calls. + _scope.push(ScopeNode.Kind.Function); + ctrl(fun); // Scope control from function + // Private mem alias tracking per function + MemMergeNode mem = new MemMergeNode(true); + mem.addDef(null); // Alias#0 + mem.addDef(new ParmNode(ScopeNode.MEM0,1,TypeMem.BOT,fun,con(TypeMem.BOT)).peephole()); // All aliases + _scope.mem(mem); + // All args, "as-if" called externally + for( int i=0; i<ids.length; i++ ) { + Type t = sig.arg(i); + _scope.define(ids[i], t, false, new ParmNode(ids[i],i+2,t,fun,con(t)).peephole(), loc); + } + + // Parse the body + Node last=ZERO; + while (!peek('}') && !_lexer.isEOF()) + last = parseStatement(); + + // Last expression is the return + if( ctrl()._type==Type.CONTROL ) + fun.addReturn(ctrl(), _scope.mem().merge(), last); + + // Pop off the inProgress node on the multi-exit Region merge + assert r.inProgress(); + r ._inputs.pop(); + rmem._inputs.pop(); + rrez._inputs.pop(); + r._loc = loc(); // Final position + assert !r.inProgress(); + + // Force peeps, which have been avoided due to inProgress + ret.setDef(1,rmem.peephole()); + ret.setDef(2,rrez.peephole()); + ret.setDef(0,r.peephole()); + ret = (ReturnNode)ret.peephole(); + + // Function scope ends + _scope.pop(); + _fun = oldfun; + _breakScope = breakScope; + _continueScope = continueScope; + // Reset control and memory to pre-function parsing days + ctrl(oldctrl.unkeep()); + _scope.mem(oldmem.unkeep()); + + return ret; + } + + /** + * Parses a block + * + * <pre> + * '{' statements '}' + * </pre> + * Does not parse the opening or closing '{}' + * @return a {@link Node} or {@code null} + */ + private Node parseBlock(ScopeNode.Kind kind) { + // Enter a new scope + _scope.push(kind); + Node last = ZERO; + while (!peek('}') && !_lexer.isEOF()) + last = parseStatement(); + // Exit scope + last.keep(); + _scope.pop(); + return last.unkeep(); + } + + /** + * Parses a statement + * + * <pre> + * returnStatement | declStatement | blockStatement | ifStatement | expressionStatement + * </pre> + * @return a {@link Node} or {@code null} + */ + private Node parseStatement() { + if( false ) return null; + else if (matchx("return") ) return parseReturn(); + else if (matchx("if") ) return parseIf(); + else if (matchx("while") ) return parseWhile(); + else if (matchx("for") ) return parseFor(); + else if (matchx("break") ) return parseBreak(); + else if (matchx("continue")) return parseContinue(); + else if (matchx("struct") ) return parseStruct(); + else if (matchx("#showGraph")) return require(showGraph(),";"); + else if (matchx(";") ) return null; // Empty statement + // Break ambiguity around leading function types and starting a block + else if (peek('{') && !isTypeFun() ) { + match("{"); + return require(parseBlock(ScopeNode.Kind.Block),"}"); + } + // Declaration or normal assignment/expression + else return parseDeclarationStatement(); + } + + /** + * Parses a while statement + * + * <pre> + * while ( expression ) statement + * </pre> + * @return a {@link Node}, never {@code null} + */ + private Node parseWhile() { + require("("); + return parseLooping(false); + } + + + /** + * Parses a for statement + * + * <pre> + * for( var x=init; test; incr ) body + * </pre> + * @return a {@link Node}, never {@code null} + */ + private Node parseFor() { + // { var x=init,y=init,...; + // while( test ) { + // body; + // next; + // } + // } + require("("); + _scope.push(ScopeNode.Kind.Block); // Scope for the index variables + if( !match(";") ) // Can be empty init "for(;test;next) body" + parseDeclarationStatement(); // Non-empty init + Node rez = parseLooping(true); + _scope.pop(); // Exit index variable scope + return rez; + } + + // Shared by `for` and `while` + private Node parseLooping( boolean doFor ) { + + var savedContinueScope = _continueScope; + var savedBreakScope = _breakScope; + + // Loop region has two control inputs, the first is the entry + // point, and second is back edge that is set after loop is parsed + // (see end_loop() call below). Note that the absence of back edge is + // used as an indicator to switch off peepholes of the region and + // associated phis; see {@code inProgress()}. + + ctrl(new LoopNode(loc(),ctrl()).peephole()); // Note we set back edge to null here + + // At loop head, we clone the current Scope (this includes all + // names in every nesting level within the Scope). + // We create phis eagerly for all the names we find, see dup(). + + // Save the current scope as the loop head + ScopeNode head = _scope.keep(); + // Clone the head Scope to create a new Scope for the body. + // Create phis eagerly as part of cloning + _xScopes.push(_scope = _scope.dup(true)); // The true argument triggers creating phis + + // Parse predicate + var pred = peek(';') ? con(1) : parseAsgn(); + require( doFor ? ";" : ")" ); + + // IfNode takes current control and predicate + Node ifNode = new IfNode(ctrl(), pred.keep()).peephole(); + // Setup projection nodes + Node ifT = new CProjNode(ifNode. keep(), 0, "True" ).peephole().keep(); + Node ifF = new CProjNode(ifNode.unkeep(), 1, "False").peephole(); + + // for( ;;next ) body + int nextPos = -1, nextEnd = -1; + if( doFor ) { + // Skip the next expression and parse it later + nextPos = pos(); + skipAsgn(); + nextEnd = pos(); + require(")"); + } + + // Clone the body Scope to create the break/exit Scope which accounts for any + // side effects in the predicate. The break/exit Scope will be the final + // scope after the loop, and its control input is the False branch of + // the loop predicate. Note that body Scope is still our current scope. + ctrl(ifF); + _xScopes.push(_breakScope = _scope.dup()); + _breakScope.addGuards(ifF,pred,true); // Up-cast predicate + + // No continues yet + _continueScope = null; + + // Parse the true side, which corresponds to loop body + // Our current scope is the body Scope + ctrl(ifT.unkeep()); // set ctrl token to ifTrue projection + _scope.addGuards(ifT,pred.unkeep(),false); // Up-cast predicate + parseStatement(); // Parse loop body + _scope.removeGuards(ifT); + + // Merge the loop bottom into other continue statements + if (_continueScope != null) { + _continueScope = jumpTo(_continueScope); + _scope.kill(); + _scope = _continueScope; + } + + // Now append the next code onto the body code + if( doFor ) { + int old = pos(nextPos); + if( !peek(')') ) + parseAsgn(); + if( pos() != nextEnd ) + throw errorSyntax( "Unexpected code after expression" ); + pos(old); + } + + // The true branch loops back, so whatever is current _scope.ctrl gets + // added to head loop as input. endLoop() updates the head scope, and + // goes through all the phis that were created earlier. For each phi, + // it sets the second input to the corresponding input from the back + // edge. If the phi is redundant, it is replaced by its sole input. + var exit = _breakScope; + head.endLoop(_scope, exit); + head.unkeep().kill(); + + _xScopes.pop(); // Cleanup + _xScopes.pop(); // Cleanup + + _continueScope = savedContinueScope; + _breakScope = savedBreakScope; + + // At exit the false control is the current control, and + // the scope is the exit scope after the exit test. + _xScopes.pop(); + _xScopes.push(exit); + _scope = exit; + return ZERO; + } + + private ScopeNode jumpTo(ScopeNode toScope) { + ScopeNode cur = _scope.dup(); + ctrl(XCTRL); // Kill current scope + // Prune nested lexical scopes that have depth > than the loop head + // We use _breakScope as a proxy for the loop head scope to obtain the depth + while( cur._lexSize.size() > _breakScope._lexSize.size() ) + cur.pop(); + // If this is a continue then first time the target is null + // So we just use the pruned current scope as the base for the + // "continue" + if( toScope == null ) + return cur; + // toScope is either the break scope, or a scope that was created here + assert toScope._lexSize.size() <= _breakScope._lexSize.size(); + toScope.ctrl(toScope.mergeScopes(cur,loc()).peephole()); + return toScope; + } + + private void checkLoopActive() { if (_breakScope == null) throw error("No active loop for a break or continue"); } + + private Node parseContinue() { checkLoopActive(); _continueScope = require(jumpTo( _continueScope ),";"); return ZERO; } + private Node parseBreak () { + checkLoopActive(); + // At the time of the break, and loop-exit conditions are only valid if + // they are ALSO valid at the break. It is the intersection of + // conditions here, not the union. + _breakScope.removeGuards(_breakScope.ctrl()); + _breakScope = require(jumpTo(_breakScope ),";"); + _breakScope.addGuards(_breakScope.ctrl(), null, false); + return ZERO; + } + + // Look for an unbalanced `)`, skipping balanced + private void skipAsgn() { + int paren=0; + while( true ) + // Next X char handles skipping complex comments + switch( _lexer.nextXChar() ) { + case Character.MAX_VALUE: + throw Utils.TODO(); + case ')': + if( --paren<0 ) { + posT( pos() - 1 ); + return; // Leave the `)` behind + } + break; + case '(': paren ++; break; + default: break; + } + } + + + /** + * Parses a statement + * + * <pre> + * if ( expression ) statement [else statement] + * </pre> + * @return a {@link Node}, never {@code null} + */ + private Node parseIf() { + // Parse predicate + require("("); + var pred = require(parseAsgn(), ")"); + return parseTrinary(pred,true,"else"); + } + + // Parse a conditional expression, merging results. + private Node parseTrinary( Node pred, boolean stmt, String fside ) { + pred.keep(); + + // IfNode takes current control and predicate + Node ifNode = new IfNode(ctrl(), pred).peephole(); + // Setup projection nodes + Node ifT = new CProjNode(ifNode. keep(), 0, "True" ).peephole().keep(); + Node ifF = new CProjNode(ifNode.unkeep(), 1, "False").peephole().keep(); + // In if true branch, the ifT proj node becomes the ctrl + // But first clone the scope and set it as current + ScopeNode fScope = _scope.dup(); // Duplicate current scope + _xScopes.push(fScope); // For graph visualization we need all scopes + + // Parse the true side + ctrl(ifT.unkeep()); // set ctrl token to ifTrue projection + _scope.addGuards(ifT,pred,false); // Up-cast predicate + Node lhs = (stmt ? parseStatement() : parseAsgn()).keep(); // Parse true-side + _scope.removeGuards(ifT); + + // See if a one-sided def was made: "if(pred) int x = 1;" and throw. + // See if any forward-refs were made, and copy them to the other side: + // "pred ? n*fact(n-1) : 1" + fScope.balanceIf(_scope); + + ScopeNode tScope = _scope; + + // Parse the false side + _scope = fScope; // Restore scope, then parse else block if any + ctrl(ifF.unkeep()); // Ctrl token is now set to ifFalse projection + // Up-cast predicate, even if not else clause, because predicate can + // remain true if the true clause exits: `if( !ptr ) return 0; return ptr.fld;` + _scope.addGuards(ifF,pred,true); + boolean doRHS = match(fside); + Node rhs = (doRHS + ? (stmt ? parseStatement() : parseAsgn()) + : con(lhs._type.makeZero())).keep(); + _scope.removeGuards(ifF); + if( doRHS ) + fScope = _scope; + pred.unkeep(); + + // Check the trinary widening int/flt + if( !stmt ) { + rhs = widenInt( rhs.unkeep(), lhs._type ).keep(); + lhs = widenInt( lhs.unkeep(), rhs._type ).keep(); + } + + _scope = tScope; + _xScopes.pop(); // Discard pushed from graph display + // See if a one-sided def was made: "if(pred) int x = 1;" and throw. + // See if any forward-refs were made, and copy them to the other side: + // "pred ? n*fact(n-1) : 1" + tScope.balanceIf(fScope); + + // Merge results + RegionNode r = ctrl(tScope.mergeScopes(fScope,loc())); + Node ret = peep(new PhiNode("",lhs._type.meet(rhs._type),r,lhs.unkeep(),rhs.unkeep())); + // Immediately fail e.g. `arg ? 7 : ptr` + ParseException err; + if( !stmt && (err=ret.err()) !=null ) throw err; + r.peephole(); + return ret; + } + + /** + * Parses a return statement; "return" already parsed. + * The $ctrl edge is killed. + * + * <pre> + * 'return' expr ; + * </pre> + * @return an expression {@link Node}, never {@code null} + */ + private Node parseReturn() { + var expr = require(parseAsgn(), ";"); + // Need default memory, since it can be lazy, need to force + // a non-lazy Phi + _fun.addReturn(ctrl(), _scope.mem().merge(), expr); + ctrl(XCTRL); // Kill control + return expr; + } + + /** + * Dumps out the node graph + * @return {@code null} + */ + Node showGraph() { + System.out.println(new GraphVisualizer().generateDotOutput(_code._stop,_scope,_xScopes)); + return null; + } + + /** Parse: [name '='] expr + */ + private Node parseAsgn() { + int old = pos(); + String name = _lexer.matchId(); + // Just a plain expression, no assignment. + // Distinguish `var==expr` from `var=expr` + if( name==null || KEYWORDS.contains(name) || !matchOpx('=','=') ) + { pos(old); return parseExpression(); } + + // Parse assignment expression + Node expr = parseAsgn(); + + // Final variable to update + Var def = _scope.lookup(name); + if( def==null ) + throw error("Undefined name '" + name + "'"); + + // TOP fields are for late-initialized fields; these have never + // been written to, and this must be the final write. Other writes + // outside the constructor need to check the final bit. + if( _scope.in(def._idx)._type!=Type.TOP && def._final && + // Inside a constructor, final assign is OK, outside nope + !(_scope.inCon() && def._idx >= _scope._lexSize.last()) ) + throw error("Cannot reassign final '"+name+"'"); + + // Lift expression, based on type + Node lift = liftExpr(expr.keep(), def.type(), def._final); + // Update + _scope.update(name,lift); + // Return un-lifted expr + return expr.unkeep(); + } + + // Make finals deep; widen ints to floats; narrow wide int types. + // Early error if types do not match variable. + private Node liftExpr( Node expr, Type t, boolean xfinal ) { + assert !(expr._type instanceof TypeMemPtr tmp) || !tmp.isFRef(); + // Final is deep on ptrs + if( xfinal && t instanceof TypeMemPtr tmp ) { + t = tmp.makeRO(); + expr = peep(new ReadOnlyNode(expr)); + } + // Auto-widen int to float + expr = widenInt( expr, t ); + // Auto-narrow wide ints to narrow ints + expr = zsMask(expr,t); + // Type is sane + if( expr._type!=Type.BOTTOM && !expr._type.shallowISA(t) ) + throw error("Type " + expr._type.str() + " is not of declared type " + t.str()); + return expr; + } + + private Node widenInt( Node expr, Type t ) { + return (expr._type instanceof TypeInteger || expr._type==Type.NIL) && t instanceof TypeFloat + ? peep(new ToFloatNode(expr)) : expr; + } + + /** + * Parse declaration or expression statement + * declStmt = type var['=' exprAsgn][, var['=' exprAsgn]]* ';' | exprAsgn ';' + * <p> + * exprAsgn = var '=' exprAsgn | expr + */ + private Node parseDeclarationStatement() { + Type t = type(); + if( t == null ) + return require(parseAsgn(),";"); + + // now parse var['=' asgnexpr] in a loop + Node n = parseDeclaration(t); + while( match(",") ) + n = parseDeclaration(t); + return require(n,";"); + } + + /** Parse final: [!]var['=' asgn] + */ + private Node parseDeclaration(Type t) { + assert t!=null; + // Has var/val instead of a user-declared type + boolean inferType = t==Type.TOP || t==Type.BOTTOM; + boolean hasBang = match("!"); + Lexer loc = loc(); + String name = requireId(); + // Optional initializing expression follows + boolean xfinal = false; + Node expr; + if( match("=") ) { + expr = parseAsgn(); + // TOP means val and val is always final + xfinal = (t==Type.TOP) || + // BOTTOM is var and var is always not-final + (t!=Type.BOTTOM && + // no Bang AND + !hasBang && + // not-null (expecting null to be set to not-null) + expr._type != Type.NIL && + // Pointers are final by default; int/flt are not-final by default. + (t instanceof TypeNil)); + // var/val, then type comes from expression + if( inferType ) { + if( expr._type==Type.NIL ) + throw error("a not-null/non-zero expression"); + t = expr._type; + if( !xfinal ) t = t.glb(); // Widen if not final + } + if( t instanceof TypeFunPtr && expr._type instanceof TypeFunPtr tfp && tfp.isConstant() ) + _code.link(tfp)._name = name; + + } else { + // Need an expression to infer the type. + // Also, if not-null then need an initializing expression. + // Allowed in a class def, because type/init will happen in the constructor. + if( (inferType || (t instanceof TypeNil tn && !tn.nullable() )) && !_scope.inCon() ) + throw errorSyntax("=expression"); + // Initial value for uninitialized struct fields. + expr = switch( t ) { + // Nullable pointers get a NIL; not-null get a TOP which + // signals that they *must* be initialized in the constructor. + case TypeNil tn -> tn.nullable() ? NIL : con(Type.TOP); + case TypeInteger ti -> ZERO; + case TypeFloat tf -> con(TypeFloat.FZERO); + // Bottom signals type inference: they must be initialized in + // the constructor and that's when we'll discover the type. + case Type tt -> { assert tt==Type.BOTTOM; yield con(tt); } + }; + } + + // Lift expression, based on type + Node lift = liftExpr(expr, t, xfinal); + + if( xfinal && t instanceof TypeMemPtr tmp ) + t = tmp.makeRO(); + + // Lift type to the declaration. This will report as an error later if + // we cannot lift the type. + if( !lift._type.isa(t) ) + lift = peep(new CastNode(t,null,lift)); + // Define a new name + if( !_scope.define(name,t,xfinal,lift, loc) ) + throw error("Redefining name '" + name + "'", loc); + return lift; + } + + + + /** + * Parse a struct declaration, and return the following statement. + * Only allowed in top level scope. + * Structs cannot be redefined. + * + * @return The statement following the struct + */ + private Node parseStruct() { + if (_xScopes.size() > 1) throw error("struct declarations can only appear in top level scope"); + String typeName = requireId(); + Type t = TYPES.get(typeName); + if( t!=null && !(t instanceof TypeMemPtr tmp && tmp.isFRef() ) ) + throw error("struct '" + typeName + "' cannot be redefined"); + + // A Block scope parse, and inspect the scope afterward for fields. + _scope.push(ScopeNode.Kind.Constructor); + require("{"); + while (!peek('}') && !_lexer.isEOF()) + parseStatement(); + + // Grab the declarations and build fields and a Struct + int lexlen = _scope._lexSize.last(); + int varlen = _scope._vars._len; + StructNode s = new StructNode(); + Ary<Field> fs = new Ary<>(Field.class); + for( int i=lexlen; i<varlen; i++ ) { + s.addDef(_scope.in(i)); + Var v = _scope._vars.at(i); + if( !v.isFRef() ) // Promote to outer scope, not defined here + fs.push(Field.make(v._name,v.type(),_code.getALIAS(),v._final)); + } + TypeStruct ts = s._ts = TypeStruct.make(typeName, fs.asAry()); + TYPES.put(typeName, TypeMemPtr.make(ts)); + INITS.put(typeName,s.peephole().keep()); + // Done with struct/block scope + require("}"); + require(";"); + _scope.pop(); + return ZERO; + } + + + // Parse and return a type or null. Valid types always are followed by an + // 'id' which the caller must parse. This lets us distinguish forward ref + // types (which ARE valid here) from local vars in an (optional) forward + // ref type position. + + // t = int|i8|i16|i32|i64|u8|u16|u32|u64|byte|bool | flt|f32|f64 | val | var | struct[?] + private Type type() { + int old1 = pos(); + // Only type with a leading `{` is a function pointer... + if( peek('{') ) return typeFunPtr(); + + // Otherwise you get a type name + String tname = _lexer.matchId(); + if( tname==null ) return null; + + // Convert the type name to a type. + Type t0 = TYPES.get(tname); + // No new types as keywords + if( t0 == null && KEYWORDS.contains(tname) ) + return posT(old1); + if( t0 == Type.BOTTOM || t0 == Type.TOP ) return t0; // var/val type inference + Type t1 = t0 == null ? TypeMemPtr.make(TypeStruct.makeFRef(tname)) :t0; // Null: assume a forward ref type + // Nest arrays and '?' as needed + Type t2 = t1; + while( true ) { + if( match("?") ) { + if( !(t2 instanceof TypeMemPtr tmp) ) + throw error("Type "+t0+" cannot be null"); + if( tmp.nullable() ) throw error("Type "+t2+" already allows null"); + t2 = tmp.makeNullable(); + } else if( match("[]") ) { + t2 = typeAry(t2); + } else + break; + } + + // Check no forward ref + if( t0 != null ) return t2; + // Check valid forward ref, after parsing all the type extra bits. + // Cannot check earlier, because cannot find required 'id' until after "[]?" syntax + int old2 = pos(); + match("!"); + String id = _lexer.matchId(); + if( !(peek(',') || peek(';') || match("->")) ) + return posT(old1); + pos(old2); // Reset lexer to reparse + if( id==null ) + return posT(old1); // Reset lexer to reparse + // Yes a forward ref, so declare it + TYPES.put(tname,t1); + return t2; + } + + // Make an array type of t + private TypeMemPtr typeAry( Type t ) { + if( t instanceof TypeMemPtr tmp && tmp.notNull() ) + throw error("Arrays of reference types must always be nullable"); + String tname = "["+t.str()+"]"; + Type ta = TYPES.get(tname); + if( ta != null ) return (TypeMemPtr)ta; + // Need make an array type. + TypeStruct ts = TypeStruct.makeAry(TypeInteger.U32,_code.getALIAS(),t,_code.getALIAS()); + assert ts.str().equals(tname); + TypeMemPtr tary = TypeMemPtr.make(ts); + TYPES.put(tname,tary); + return tary; + } + + // A function type is `{ type... -> type }` or `{ type }`. + private Type typeFunPtr() { + int old = pos(); // Record lexer position + match("{"); // Skip already-peeked '{' + Type t0 = type(); // Either return or first arg + if( t0==null ) return posT(old); // Not a function + if( match("}") ) // No-arg function { -> type } + return TypeFunPtr.make(match("?"),TypeTuple.BOT,t0); + Ary<Type> ts = new Ary<>(Type.class); + ts.push(t0); // First argument + while( true ) { + if( match("->") ) { // End of arguments, parse return + Type ret = type(); + if( ret==null || !match("}") ) + return posT(old); // Not a function + return TypeFunPtr.make(match("?"),TypeTuple.make(ts.asAry()),ret); + } + Type t1 = type(); + if( t1==null ) return posT(old); // Not a function + ts.push(t1); + } + } + + // True if a TypeFunPtr, without advancing parser + private boolean isTypeFun() { + int old = pos(); + if( typeFunPtr()==null ) return false; + pos(old); + return true; + } + + /** + * Parse an expression of the form: + * + * <pre> + * expr : bitwise [? expr [: expr]] + * </pre> + * @return an expression {@link Node}, never {@code null} + */ + private Node parseExpression() { + Node expr = parseBitwise(); + return match("?") ? parseTrinary(expr,false,":") : expr; + } + + /** + * Parse an bitwise expression + * + * <pre> + * bitwise : compareExpr (('&' | '|' | '^') compareExpr)* + * </pre> + * @return a bitwise expression {@link Node}, never {@code null} + */ + private Node parseBitwise() { + Node lhs = parseComparison(); + while( true ) { + if( false ) ; + else if( match("&") ) lhs = new AndNode(loc(),lhs,null); + else if( match("|") ) lhs = new OrNode(loc(),lhs,null); + else if( match("^") ) lhs = new XorNode(loc(),lhs,null); + else break; + lhs.setDef(2,parseComparison()); + lhs = peep(lhs); + } + return lhs; + } + + /** + * Parse an expression of the form: + * + * <pre> + * expr : additiveExpr op additiveExpr + * </pre> + * @return an comparator expression {@link Node}, never {@code null} + */ + private Node parseComparison() { + var lhs = parseShift(); + while( true ) { + int idx=0; boolean negate=false; + // Test for any local nodes made, and "keep" lhs during peepholes + if( false ) ; + else if( match("==") ) { idx=2; lhs = new BoolNode.EQ(lhs, null); } + else if( match("!=") ) { idx=2; lhs = new BoolNode.EQ(lhs, null); negate=true; } + else if( match("<=") ) { idx=2; lhs = new BoolNode.LE(lhs, null); } + else if( match("<" ) ) { idx=2; lhs = new BoolNode.LT(lhs, null); } + else if( match(">=") ) { idx=1; lhs = new BoolNode.LE(null, lhs); } + else if( match(">" ) ) { idx=1; lhs = new BoolNode.LT(null, lhs); } + else break; + // Peepholes can fire, but lhs is already "hooked", kept alive + lhs.setDef(idx,parseShift()); + lhs = peep(lhs.widen()); + if( negate ) // Extra negate for != + lhs = peep(new NotNode(lhs)); + } + return lhs; + } + + /** + * Parse an additive expression + * + * <pre> + * shiftExpr : additiveExpr (('<<' | '>>' | '>>>') additiveExpr)* + * </pre> + * @return a shift expression {@link Node}, never {@code null} + */ + private Node parseShift() { + Node lhs = parseAddition(); + while( true ) { + if( false ) ; + else if( match("<<") ) lhs = new ShlNode(loc(),lhs,null); + else if( match(">>>")) lhs = new ShrNode(loc(),lhs,null); + else if( match(">>") ) lhs = new SarNode(loc(),lhs,null); + else break; + lhs.setDef(2,parseAddition()); + ParseException err; + if( (err=lhs.err()) != null ) + throw err; + lhs = peep(lhs.widen()); + } + return lhs; + } + + /** + * Parse an additive expression + * + * <pre> + * additiveExpr : multiplicativeExpr (('+' | '-') multiplicativeExpr)* + * </pre> + * @return an add expression {@link Node}, never {@code null} + */ + private Node parseAddition() { + Node lhs = parseMultiplication(); + while( true ) { + if( false ) ; + else if( match("+") ) lhs = new AddNode(lhs,null); + else if( match("-") ) lhs = new SubNode(lhs,null); + else break; + lhs.setDef(2,parseMultiplication()); + lhs = peep(lhs.widen()); + } + return lhs; + } + + /** + * Parse an multiplicativeExpr expression + * + * <pre> + * multiplicativeExpr : unaryExpr (('*' | '/') unaryExpr)* + * </pre> + * @return a multiply expression {@link Node}, never {@code null} + */ + private Node parseMultiplication() { + var lhs = parseUnary(); + while( true ) { + if( false ) ; + else if( match("*") ) lhs = new MulNode(lhs,null); + else if( match("/") ) lhs = new DivNode(lhs,null); + else break; + lhs.setDef(2,parseUnary()); + lhs = peep(lhs.widen()); + } + return lhs; + } + + /** + * Parse a unary minus expression. + * + * <pre> + * unaryExpr : ('-') unaryExpr | '!') unaryExpr | postfixExpr | primaryExpr | '--' Id | '++' Id + * </pre> + * @return a unary expression {@link Node}, never {@code null} + */ + private Node parseUnary() { + // Pre-dec/pre-inc + int old = pos(); + if( match("--") || match("++") ) { + int delta = _lexer.peek(-1)=='+' ? 1 : -1; // Pre vs post + String name = _lexer.matchId(); + if( name!=null ) { + Var n = _scope.lookup(name); + if( n != null && !(n.type() instanceof TypeMemPtr) ) { + if( n._final ) + throw error("Cannot reassign final '"+n._name+"'"); + Node expr = n.type() instanceof TypeFloat + ? peep(new AddFNode(_scope.in(n),con(TypeFloat.constant(delta)))) + : zsMask(peep(new AddNode(_scope.in(n),con(delta))),n.type()); + _scope.update(n,expr); + return expr; + } + } + // Reset, try again + pos(old); + } + if (match("-")) return peep(new MinusNode(parseUnary()).widen()); + if (match("!")) return peep(new NotNode(parseUnary())); + return parsePostfix(parsePrimary()); + } + + /** + * Parse a primary expression: + * + * <pre> + * primaryExpr : integerLiteral | true | false | null | new Type | '(' expression ')' | Id['++','--'] + * </pre> + * @return a primary {@link Node}, never {@code null} + */ + private Node parsePrimary() { + if( _lexer.isNumber(_lexer.peek()) ) return parseLiteral(); + if( matchx("true" ) ) return con(1); + if( matchx("false") ) return ZERO; + if( matchx("null" ) ) return NIL; + if( match ("(" ) ) return require(parseAsgn(), ")"); + if( matchx("new" ) ) return alloc(); + if( match ("{" ) ) return require(func(),"}"); + // Expect an identifier now + Var n = requireLookupId(); + Node rvalue = _scope.in(n); + if( rvalue._type == Type.BOTTOM ) + if( rvalue instanceof FRefNode ) return rvalue; + else throw error("Cannot read uninitialized field '"+n._name+"'"); + // Check for assign-update, x += e0; + char ch = _lexer.matchOperAssign(); + if( ch==0 ) return rvalue; + Node op = opAssign(ch,rvalue,n._final,n.type(),n._name); + _scope.update(n,op); + return postfix(ch) ? rvalue.unkeep() : op; + } + + // Check for assign-update, "x += e0". + // Returns "x + e0"; the caller assigns value. + // if ch is postfix, then lhs is kept and caller will unkeep and return it. + private Node opAssign(char ch, Node lhs, boolean xfinal, Type t, String name ) { + if( xfinal ) + throw error("Cannot reassign final '"+name+"'"); + // RHS of the update. + lhs.keep(); // Alive across parseAsgn + Node rhs = + (byte)ch == 1 ? con( 1) : // var++ + (byte)ch == -1 ? con(-1) : // var-- + parseAsgn(); // var op= rhs + if( !postfix(ch) ) lhs.unkeep(); // Allow to die in next peep + // 4 cases: + // int + int ==>> narrow int + // int + flt ==>> error, caller must fail assigning flt into int + // flt + int ==>> use float op, wrap toFloat() + // flt + flt ==>> use float op + Node op = switch(ch) { + case 1, (char)-1, + '+' -> new AddNode(lhs,rhs); + case '-' -> new SubNode(lhs,rhs); + case '*' -> new MulNode(lhs,rhs); + case '/' -> new DivNode(lhs,rhs); + default -> throw Utils.TODO(); + }; + // Convert to float ops, or narrow int types; error if not declared type. + // Also, if postfix LHS is still keep() + return liftExpr(peep(op.widen()),t,false); + } + + + // Expect an ID here. If not found, assume a forward reference function + Var requireLookupId() { + String id = _lexer.matchId(); + if( id == null || KEYWORDS.contains(id) ) + throw errorSyntax("an identifier or expression"); + Var n = _scope.lookup(id); + // If missing, assume a forward reference + if( n==null ) { + _scope.define(id, FRefNode.FREF_TYPE, true, XCTRL, loc()); + n = _scope.lookup(id); + } else { + // Lookup worked on an out-of-function value. + if( _scope.outOfFunction(n) ) { + // No closures, so this has to be a final constant (which + // includes forward refs) + Node def = _scope.in(n._idx); + if( !(def instanceof FRefNode) && !(n._final && def._type.isConstant()) ) + throw error("Variable '"+n._name+"' is out of function scope and must be a final constant"); + } + } + return n; + } + + /** + Parse an allocation + */ + private Node alloc() { + Type t = type(); + if( t==null ) throw error("Expected a type"); + // Parse ary[ length_expr ] + if( match("[") ) { + if( !t.makeZero().isa(t) ) + throw error("Cannot allocate a non-nullable, since arrays always zero/null fill"); + Node len = parseAsgn(); + if( !(len._type instanceof TypeInteger) ) + throw error("Cannot allocate an array with length "+len._type); + require("]"); + TypeMemPtr tmp = typeAry(t); + return newArray(tmp._obj,len); + } + + if( !(t instanceof TypeMemPtr tmp) ) + throw error("Cannot allocate a "+t.str()); + + // Parse new struct { default_initialization } + StructNode s = INITS.get(tmp._obj._name); + if( s==null ) throw error("Unknown struct type '" + tmp._obj._name + "'"); + + Field[] fs = s._ts._fields; + // if the object is fully initialized, we can skip a block here. + // Check for constructor block: + boolean hasConstructor = match("{"); + Ary<Node> init=s._inputs; int idx=0; + if( hasConstructor ) { + idx = _scope.nIns(); + // Push a scope, and pre-assign all struct fields. + _scope.push(ScopeNode.Kind.Block); + Lexer loc = loc(); + for( int i=0; i<fs.length; i++ ) + _scope.define(fs[i]._fname, fs[i]._type, fs[i]._final, s.in(i)._type==Type.TOP ? con(Type.BOTTOM) : s.in(i), loc); + // Parse the constructor body + require(parseBlock(ScopeNode.Kind.Constructor),"}"); + init = _scope._inputs; + } + // Check that all fields are initialized + for( int i=idx; i<init.size(); i++ ) + if( init.at(i)._type == Type.TOP || init.at(i)._type == Type.BOTTOM ) + throw error("'"+tmp._obj._name+"' is not fully initialized, field '" + fs[i-idx]._fname + "' needs to be set in a constructor"); + Node ptr = newStruct(tmp._obj, con(tmp._obj.offset(fs.length)), idx, init ); + if( hasConstructor ) + _scope.pop(); + return ptr; + } + + + /** + * Return a NewNode initialized memory. + * @param obj is the declared type, with GLB fields + * @param init is a collection of initialized fields + */ + private Node newStruct( TypeStruct obj, Node size, int idx, Ary<Node> init ) { + Field[] fs = obj._fields; + if( fs==null ) + throw error("Unknown struct type '" + obj._name + "'"); + int len = fs.length; + Node[] ns = new Node[2+len]; + ns[0] = ctrl(); // Control in slot 0 + // Total allocated length in bytes + ns[1] = size; + // Memory aliases for every field + for( int i = 0; i < len; i++ ) + ns[2+i] = memAlias(fs[i]._alias); + Node nnn = new NewNode(TypeMemPtr.make(obj), ns).peephole().keep(); + for( int i = 0; i < len; i++ ) + memAlias(fs[i]._alias, new ProjNode(nnn,i+2,memName(fs[i]._alias)).peephole()); + Node ptr = new ProjNode(nnn.unkeep(),1,obj._name).peephole().keep(); + + // Initial nonzero values for every field + for( int i = 0; i < len; i++ ) { + Node val = init.get( i + idx ); + if( val._type != val._type.makeZero() ) { + Node mem = memAlias(fs[i]._alias); + Node st = new StoreNode(loc(),fs[i]._fname,fs[i]._alias,fs[i]._type,mem,ptr,con(obj.offset(i)),val,true).peephole(); + memAlias(fs[i]._alias,st); + } + } + + return ptr.unkeep(); + } + + private static final Ary<Node> ALTMP = new Ary<>(Node.class); + private Node newArray(TypeStruct ary, Node len) { + int base = ary.aryBase (); + int scale= ary.aryScale(); + Node size = peep(new AddNode(con(base),peep(new ShlNode(null,len.keep(),con(scale))))); + ALTMP.clear(); ALTMP.add(len.unkeep()); ALTMP.add(con(ary._fields[1]._type.makeZero())); + return newStruct(ary,size,0,ALTMP); + } + + // We set up memory aliases by inserting special vars in the scope these + // variables are prefixed by $ so they cannot be referenced in Simple code. + // Using vars has the benefit that all the existing machinery of scoping + // and phis work as expected + private Node memAlias(int alias ) { return _scope.mem(alias ); } + private void memAlias(int alias, Node st) { _scope.mem(alias, st); } + public static String memName(int alias) { return ("$"+alias).intern(); } + + /** + * Parse postfix expression; this can be a field expression, an array + * lookup or a postfix operator like '#' + * + * <pre> + * expr ('.' FIELD)* [ = expr ] // Field reference + * expr '#' // Postfix unary read operator + * expr ['++' | '--' ] // Postfix unary write operator + * expr ('[' expr ']')* = [ = expr ] // Array reference + * expr '(' [args,]* ')' // Function call + * </pre> + */ + private Node parsePostfix(Node expr) { + String name; + if( match(".") ) name = requireId(); + else if( match("#") ) name = "#"; + else if( match("[") ) name = "[]"; + else if( match("(") ) return parsePostfix(require(functionCall(expr),")")); + else return expr; // No postfix + + if( expr._type==Type.NIL ) + throw error("Accessing unknown field '" + name + "' from 'null'"); + + // Sanity check expr for being a reference + if( !(expr._type instanceof TypeMemPtr ptr) ) { + throw error( "Expected "+(name=="#" || name=="[]" ? "array" : "reference")+" but found " + expr._type.str() ); + } + + // Sanity check field name for existing + TypeMemPtr tmp = (TypeMemPtr)TYPES.get(ptr._obj._name); + if( tmp == null ) throw error("Accessing unknown field '" + name + "' from '" + ptr + "'"); + TypeStruct base = tmp._obj; + int fidx = base.find(name); + if( fidx == -1 ) throw error("Accessing unknown field '" + name + "' from '" + ptr.str() + "'"); + + // Get field type and layout offset from base type and field index fidx + Field f = base._fields[fidx]; // Field from field index + Type tf = f._type; + if( tf instanceof TypeMemPtr ftmp && ftmp.isFRef() ) + tf = ftmp.makeFrom(((TypeMemPtr)(TYPES.get(ftmp._obj._name)))._obj); + + // Field offset; fixed for structs, computed for arrays + Node off = (name.equals("[]") // If field is an array body + // Array index math + ? peep(new AddNode(con(base.aryBase()),peep(new ShlNode(null,require(parseAsgn(),"]"),con(base.aryScale()))))) + // Struct field offsets are hardwired + : con(base.offset(fidx))).keep(); + + // Disambiguate "obj.fld==x" boolean test from "obj.fld=x" field assignment + if( matchOpx('=','=') ) { + Node val = parseAsgn().keep(); + Node lift = liftExpr( val, tf, f._final ); + + Node st = new StoreNode(loc(), name, f._alias, tf, memAlias(f._alias), expr, off.unkeep(), lift, false); + // Arrays include control, as a proxy for a safety range check. + // Structs don't need this; they only need a NPE check which is + // done via the type system. + if( base.isAry() ) st.setDef(0,ctrl()); + memAlias(f._alias, st.peephole()); + return val.unkeep(); // "obj.a = expr" returns the expression while updating memory + } + + Node load = new LoadNode(loc(),name, f._alias, tf, memAlias(f._alias), expr.keep(), off); + // Arrays include control, as a proxy for a safety range check + // Structs don't need this; they only need a NPE check which is + // done via the type system. + if( base.isAry() && !name.equals("#") ) load.setDef(0,ctrl()); + load = peep(load); + + // Check for assign-update, "ptr.fld += expr" or "ary[idx]++" + char ch = _lexer.matchOperAssign(); + if( ch!=0 ) { + Node op = opAssign(ch,load,f._final,tf,name); + Node st = new StoreNode(loc(), name, f._alias, tf, memAlias(f._alias), expr.unkeep(), off, op, false); + // Arrays include control, as a proxy for a safety range check. + // Structs don't need this; they only need a NPE check which is + // done via the type system. + if( base.isAry() ) st.setDef(0,ctrl()); + memAlias(f._alias, peep(st)); + load = postfix(ch) ? load.unkeep() : op; + // And use the original loaded value as the result + } else expr.unkill(); + off.unkill(); + + return parsePostfix(load); + } + + + // zero/sign extend. "i" is limited to either classic unsigned (min==0) or + // classic signed (min=minus-power-of-2); max=power-of-2-minus-1. + private Node zsMask(Node val, Type t ) { + if( !(val._type instanceof TypeInteger tval && t instanceof TypeInteger t0 && !tval.isa(t0)) ) { + if( !(val._type instanceof TypeFloat tval && t instanceof TypeFloat t0 && !tval.isa(t0)) ) + return val; + // Float rounding + return peep(new RoundF32Node(val)); + } + if( t0._min==0 ) // Unsigned + return peep(new AndNode(null,val,con(t0._max))); + // Signed extension + int shift = Long.numberOfLeadingZeros(t0._max)-1; + Node shf = con(shift); + if( shf._type==TypeInteger.ZERO ) + return val; + return peep(new SarNode(null,peep(new ShlNode(null,val,shf.keep())),shf.unkeep())); + } + + /** + * Parse a function body; the caller will parse the surrounding "{}" + * + * <pre> + * { [type arg,]* -> expr } + * { expr } // The no-argument function + * </pre> + */ + private Node func() { + Ary<Type> ts = new Ary<>(Type.class); + Ary<String> ids = new Ary<>(String.class); + _lexer.skipWhiteSpace(); + Lexer loc = loc(); // First argument location + while( true ) { + Type t = type(); // Arg type + if( t==null ) break; + String id = requireId(); + ts .push(t ); // Push type/arg pairs + ids.push(id); + match(","); + } + require("->"); + // Make a concrete function type, with a fidx + TypeFunPtr tfp = _code.makeFun(TypeTuple.make(ts.asAry()),Type.BOTTOM); + ReturnNode ret = parseFunctionBody(tfp,loc,ids.asAry()); + return con(ret._fun.sig()); + } + + /** + * Parse function call arguments; caller will parse the surrounding "()" + * <pre> + * ( arg* ) + * </pre> + */ + private Node functionCall(Node expr) { + if( expr._type == Type.NIL ) + throw error("Calling a null function pointer"); + if( !(expr instanceof FRefNode) && !expr._type.isa(TypeFunPtr.BOT) ) + throw error("Expected a function but got "+expr._type.glb().str()); + expr.keep(); // Keep while parsing args + + Ary<Node> args = new Ary<Node>(Node.class); + args.push(null); // Space for ctrl,mem + args.push(null); + while( !peek(')') ) { + Node arg = parseAsgn(); + if( arg==null ) break; + args.push(arg.keep()); + if( !match(",") ) break; + } + // Control & memory after parsing args + args.set(0,ctrl().keep()); + args.set(1,_scope.mem().merge().keep()); + args.push(expr); // Function pointer + // Unkeep them all + for( Node arg : args ) + arg.unkeep(); + // Dead into the call? Skip all the node gen + if( ctrl()._type == Type.XCONTROL ) { + for( Node arg : args ) + if( arg.isUnused() ) + arg.kill(); + return con(Type.TOP); + } + + // Into the call + CallNode call = (CallNode)new CallNode(loc(), args.asAry()).peephole(); + + // Post-call setup + CallEndNode cend = (CallEndNode)new CallEndNode(call).peephole(); + // Control from CallEnd + ctrl(new CProjNode(cend,0,ScopeNode.CTRL).peephole()); + // Memory from CallEnd + MemMergeNode mem = new MemMergeNode(true); + mem.addDef(null); // Alias#0 + mem.addDef(new ProjNode(cend,1,ScopeNode.MEM0).peephole()); + _scope.mem(mem); + // Call result + return new ProjNode(cend,2,null).peephole(); + } + + /** + * Parse integer literal + * + * <pre> + * integerLiteral: [1-9][0-9]* | [0] + * floatLiteral: [digits].[digits]?[e [digits]]? + * </pre> + */ + private ConstantNode parseLiteral() { return con(_lexer.parseNumber()); } + public static Node con( long con ) { return con==0 ? ZERO : con(TypeInteger.constant(con)); } + public static ConstantNode con( Type t ) { return (ConstantNode)new ConstantNode(t).peephole(); } + public Node peep( Node n ) { + // Peephole, then improve with lexically scoped guards + return _scope.upcastGuard(n.peephole()); + } + + ////////////////////////////////// + // Utilities for lexical analysis + + // Return true and skip if "syntax" is next in the stream. + private boolean match (String syntax) { return _lexer.match (syntax); } + // Match must be "exact", not be followed by more id letters + private boolean matchx(String syntax) { return _lexer.matchx(syntax); } + private boolean matchOpx(char c0, char c1) { return _lexer.matchOpx(c0,c1); } + // Return true and do NOT skip if 'ch' is next + private boolean peek(char ch) { return _lexer.peek(ch); } + private boolean peekIsId() { return _lexer.peekIsId(); } + + // ch is +/- 1, means oper++ or oper-- means postfix + private static boolean postfix(char ch) { + return (byte)ch== 1 || (byte)ch== -1; + } + + public int pos() { return _lexer._position; } + private int pos(int pos) { + int old = _lexer._position; + _lexer._position = pos; + return old; + } + private Type posT(int pos) { _lexer._position = pos; return null; } + // Source code location + Lexer loc() { return new Lexer(_lexer); } + + + // Require and return an identifier + private String requireId() { + String id = _lexer.matchId(); + if (id != null && !KEYWORDS.contains(id) ) return id.intern(); + throw error("Expected an identifier, found '"+id+"'"); + } + + private String matchId() { + int old = pos(); + String id = _lexer.matchId(); + if( id==null ) return null; + if( !KEYWORDS.contains(id) ) return id; + pos(old); + return null; + } + + // Require an exact match + private Parser require(String syntax) { require(null, syntax); return this; } + private <N> N require(N n, String syntax) { + if (match(syntax)) return n; + throw errorSyntax(syntax); + } + + //////////////////////////////////// + // Lexer components + + // Lexer provides low level access to the raw file bytes, peeks and matches + // short strings, parses numbers, skips comments and whitespace, tracks + // line numbers, allows the parse position to be saved and restored, and + // serves as a location indicator for errors. + + public static class Lexer { + + // Input buffer; an array of text bytes read from a file or a string + private final byte[] _input; + // Tracks current position in input buffer + private int _position = 0; + + // + private int _line_number = 1; + // Start of current line + private int _line_start = 0; + + /** + * Record the source text for lexing + */ + public Lexer(String source) { + this(source.getBytes()); + } + + /** + * Direct from disk file source + */ + public Lexer(byte[] buf) { + _input = buf; + } + + /** + * Copy a lexer from a lexer + */ + private Lexer(Lexer l) { + _input = l._input; + _position = l._position; + _line_number = l._line_number; + _line_start = l._line_start; + } + + // Very handy in the debugger, shows the unparsed program + @Override + public String toString() { + return new String(_input, _position, _input.length - _position); + } + + // True if at EOF + private boolean isEOF() { + return _position >= _input.length; + } + + // Peek next character, or report EOF + private char peek() { + return isEOF() ? Character.MAX_VALUE // Special value that causes parsing to terminate + : (char) _input[_position]; + } + // Just crash if misused + public byte peek(int off) { return _input[_position+off]; } + + private void inc() { + if( _position++ < _input.length && _input[_position-1] == '\n' ) { + _line_number++; + _line_start = _position; + } + } + // Does not honor LF, so caller must roll back position on a LF + private char nextChar() { + char ch = peek(); + inc(); + return ch; + } + + // True if a white space + private boolean isWhiteSpace() { + return peek() <= ' '; // Includes all the use space, tab, newline, CR + } + + /** + * Return the next non-white-space character + */ + private void skipWhiteSpace() { + while( true ) { + if( isWhiteSpace() ) inc(); + // Skip // to end of line + else if( _position+2 < _input.length && + _input[_position ] == '/' && + _input[_position+1] == '/') { + inc(); inc(); + while( !isEOF() && _input[_position] != '\n' ) inc(); + } else break; + } + } + + // Next non-white-space character, or EOF + public char nextXChar() { skipWhiteSpace(); return nextChar(); } + + // Return true, if we find "syntax" after skipping white space; also + // then advance the cursor past syntax. + // Return false otherwise, and do not advance the cursor. + boolean match(String syntax) { + assert syntax.indexOf('\n')==-1; // No newlines in match + skipWhiteSpace(); + int len = syntax.length(); + if (_position + len > _input.length) return false; + for (int i = 0; i < len; i++) + if ((char) _input[_position + i] != syntax.charAt(i)) + return false; + _position += len; + return true; + } + + // Match must be exact and not followed by more ID characters. + // Prevents identifier "ifxy" from matching an "if" statement. + boolean matchx(String syntax) { + if( !match(syntax) ) return false; + if( !isIdLetter(peek()) ) return true; + _position -= syntax.length(); + return false; + } + boolean matchOpx(char c0, char c1) { + skipWhiteSpace(); + if( _position+1 >= _input.length || _input[_position]!=c0 || _input[_position+1]==c1 ) + return false; + inc(); + return true; + } + + private boolean peek(char ch) { + skipWhiteSpace(); + return peek()==ch; + } + + boolean peekIsId() { + skipWhiteSpace(); + return isIdStart(peek()); + } + + // Return an identifier or null + String matchId() { + return peekIsId() ? parseId() : null; + } + + // Used for errors + String getAnyNextToken() { + if (isEOF()) return ""; + if (isIdStart(peek())) return parseId(); + if (isNumber(peek())) return parseNumberString(); + if (isPunctuation(peek())) return parsePunctuation(); + return String.valueOf(peek()); + } + + + boolean isNumber(char ch) {return Character.isDigit(ch);} + + // Return a constant Type, either TypeInteger or TypeFloat + private Type parseNumber() { + int old = _position; + int len = isLongOrDouble(); + if( len > 0 ) { + if( len > 1 && _input[old]=='0' ) + throw error("Syntax error: integer values cannot start with '0'",this); + long i = Long.parseLong(new String(_input,old,len)); + return TypeInteger.constant(i); + } + return TypeFloat.constant(Double.parseDouble(new String(_input,old,-len))); + } + private String parseNumberString() { + int old = _position; + int len = Math.abs(isLongOrDouble()); + return new String(_input,old,len); + } + + // Return +len that ends a long + // Return -len that ends a double + private int isLongOrDouble() { + int old = _position; + char c; + while( Character.isDigit(c=nextChar()) ) ; + if( !(c=='e' || c=='.') ) + return --_position - old; + while( Character.isDigit(c=nextChar()) || c=='e' || c=='.' ) ; + return -(--_position - old); + } + + // First letter of an identifier + private boolean isIdStart(char ch) { + return Character.isAlphabetic(ch) || ch == '_'; + } + + // All characters of an identifier, e.g. "_x123" + private boolean isIdLetter(char ch) { + return Character.isLetterOrDigit(ch) || ch == '_'; + } + + private String parseId() { + int start = _position; + while (isIdLetter(nextChar())) ; + return new String(_input, start, --_position - start); + } + + // + private boolean isPunctuation(char ch) { + return "=;[]<>()+-/*&|^".indexOf(ch) != -1; + } + + private String parsePunctuation() { + int start = _position; + return new String(_input, start, 1); + } + + // Next oper= character, or 0. + // As a convenience, mark "++" as a char 1 and "--" as char -1 (65535) + public char matchOperAssign() { + skipWhiteSpace(); + if( _position+2 >= _input.length ) return 0; + char ch0 = (char)_input[_position]; + if( "+-/*&|^".indexOf(ch0) == -1 ) return 0; + char ch1 = (char)_input[_position+1]; + if( ch1 == '=' ) { _position += 2; return ch0; } + if( isIdLetter((char)_input[_position+2]) ) return 0; + if( ch0 == '+' && ch1 == '+' ) { _position += 2; return (char) 1; } + if( ch0 == '-' && ch1 == '-' ) { _position += 2; return (char)-1; } + return 0; + } + } + + ParseException errorSyntax(String syntax) { return _errorSyntax("expected "+syntax); } + private ParseException _errorSyntax(String msg) { + return error("Syntax error, "+msg+": " + _lexer.getAnyNextToken()); + } + ParseException error(String msg) { return error(msg,_lexer); } + public static ParseException error(String msg, Lexer loc) { return new ParseException(msg,loc); } + + public static class ParseException extends RuntimeException { + public final Lexer _loc; + // file:line:charoff err + //String msg = "src:"+_line_number+":"+(_position-_line_start)+" "+errorMessage; + ParseException( String msg, Lexer loc ) { super(msg); _loc = loc; assert loc!=null; } + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/RegMask.java b/chapter20/src/main/java/com/seaofnodes/simple/RegMask.java new file mode 100644 index 00000000..20be7157 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/RegMask.java @@ -0,0 +1,39 @@ +package com.seaofnodes.simple; + +import java.util.BitSet; + +// A "register mask" - 1 bit set for each allowed register. In addition "stack +// slot" registers may be allowed, effectively making the set infinite. +public class RegMask { + + public static final RegMask EMPTY = new RegMask(); + + final BitSet _bits; + + public RegMask(long x ) { _bits = BitSet.valueOf(new long[]{x}); } + public RegMask(long[] xs) { _bits = BitSet.valueOf( xs); } + // Internal constructor + RegMask() { _bits = new BitSet(); } + RegMask(BitSet bs) { _bits = bs; } + + public boolean get(int r) { return _bits.get(r); } + + // Defensive writable copy + public RegMaskRW copy() { return new RegMaskRW( (BitSet)_bits.clone() ); } + + boolean size1() { return _bits.cardinality()==1; } +} + +class RegMaskRW extends RegMask { + public RegMaskRW() { super(); } + public RegMaskRW(long x) { super(new long[]{x}); } + public RegMaskRW(long[] xs) { super(xs); } + public RegMaskRW(BitSet bs) { super(bs); } + public void set(int r) { _bits.set(r); } + public void clr(int r) { _bits.clear(r); } + public void set(int r, boolean b) { _bits.set(r,b); } + public RegMaskRW and( RegMask r ) { + _bits.and(r._bits); + return this; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/SB.java b/chapter20/src/main/java/com/seaofnodes/simple/SB.java new file mode 100644 index 00000000..7d580c21 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/SB.java @@ -0,0 +1,91 @@ +package com.seaofnodes.simple; + +/** Tight/tiny StringBuilder wrapper. + * Short short names on purpose; so they don't obscure the printing. + * Can't believe this wasn't done long, long ago. */ +public final class SB { + private final StringBuilder _sb; + private int _indent = 0; + public SB( ) { _sb = new StringBuilder( ); } + public SB(String s) { _sb = new StringBuilder(s); } + public SB p( String s ) { _sb.append(s); return this; } + public SB p( float s ) { + if( Float.isNaN(s) ) + _sb.append( "Float.NaN"); + else if( Float.isInfinite(s) ) { + _sb.append(s > 0 ? "Float.POSITIVE_INFINITY" : "Float.NEGATIVE_INFINITY"); + } else _sb.append(s); + return this; + } + public SB p( double s ) { + if( Double.isNaN(s) ) + _sb.append("Double.NaN"); + else if( Double.isInfinite(s) ) { + _sb.append(s > 0 ? "Double.POSITIVE_INFINITY" : "Double.NEGATIVE_INFINITY"); + } else _sb.append(s); + return this; + } + public SB p( char s ) { _sb.append(s); return this; } + public SB p( int s ) { _sb.append(s); return this; } + public SB p( long s ) { _sb.append(s); return this; } + public SB p( boolean s) { _sb.append(s); return this; } + // 4 hex digits + public SB hex4(int s) { + assert (s>>4*4)==0; // Fits in 16 bits + for( int i=0; i<4; i++ ) { + int digit = (s>>((3-i)*4)) & 0xf; + _sb.append((char)((digit <= 9 ? '0' : ('A'-10))+digit)); + } + return this; + } + // Fixed width field + public SB fix( int sz, String s ) { + assert sz >=0; + for( int i=0; i<sz; i++ ) + _sb.append( i < s.length() ? s.charAt(i) : ' '); + return this; + } + + // Not spelled "p" on purpose: too easy to accidentally say "p(1.0)" and + // suddenly call the autoboxed version. + public SB pobj( Object s ) { _sb.append(s.toString()); return this; } + public SB i( int d ) { for( int i=0; i<d+_indent; i++ ) p(" "); return this; } + public SB i( ) { return i(0); } + public SB ip(String s) { return i().p(s); } + public SB s() { _sb.append(' '); return this; } + + // Increase indentation + public SB ii( int i) { _indent += i; return this; } + public SB ii() { return ii(1); } + // Decrease indentation + public SB di( int i) { _indent -= i; return this; } + public SB di() { return di(1); } + + public SB nl( ) { return p(System.lineSeparator()); } + // Last printed a nl + public boolean was_nl() { + int len = _sb.length(); + String nl = System.lineSeparator(); + int nlen = nl.length(); + if( len < nlen ) return false; + for( int i=0; i<nlen; i++ ) + if( _sb.charAt(len-nlen+i)!=nl.charAt(i) ) + return false; + return true; + } + + // Delete last char. Useful when doing string-joins and JSON printing and an + // extra separater char needs to be removed: + // + // sb.p('['); + // for( Foo foo : foos ) + // sb.p(foo).p(','); + // sb.unchar().p(']'); // remove extra trailing comma + // + public SB unchar() { return unchar(1); } + public SB unchar(int x) { _sb.setLength(_sb.length()-x); return this; } + + public SB clear() { _sb.setLength(0); return this; } + public int len() { return _sb.length(); } + @Override public String toString() { return _sb.toString(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/SimpleWebSocket.java b/chapter20/src/main/java/com/seaofnodes/simple/SimpleWebSocket.java new file mode 100644 index 00000000..1ce766ff --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/SimpleWebSocket.java @@ -0,0 +1,148 @@ +package com.seaofnodes.simple; + +import java.io.*; +import java.net.ServerSocket; +import java.net.Socket; +import java.net.URI; +import java.nio.charset.StandardCharsets; +import java.security.MessageDigest; +import java.security.NoSuchAlgorithmException; +import java.util.Base64; + +// Why-Oh-Why is this needed? + +// Pass in a URI and a port. +// Opens the client; client has to open a websocket at the given port. +// Connects, handshakes. +// Thereafter you can call get/put to swap frames with client. + +class SimpleWebSocket extends ServerSocket { + final InputStream _in; + final OutputStream _out; + + SimpleWebSocket(URI uri, int port) throws IOException, NoSuchAlgorithmException { + super(port); + // Launch client + java.awt.Desktop.getDesktop().browse(uri); + // Look for client; get in/out streams + Socket sock = accept(); + _in = sock. getInputStream(); + _out = sock.getOutputStream(); + + + // Establish web socket with handshake; easier with a buffered reader + BufferedReader br = new BufferedReader(new InputStreamReader(_in)); + String key = null; + while( br.ready() ) { + String line = br.readLine(); + if( line.startsWith("Sec-WebSocket-Key: ") ) + key = line.substring(19); + } + assert key!=null; + // Magic handshake + byte[] response = ("HTTP/1.1 101 Switching Protocols\r\n" + + "Connection: Upgrade\r\n" + + "Upgrade: websocket\r\n" + + "Sec-WebSocket-Accept: " + + Base64.getEncoder().encodeToString(MessageDigest.getInstance("SHA-1").digest((key + "258EAFA5-E914-47DA-95CA-C5AB0DC85B11").getBytes( StandardCharsets.UTF_8 ))) + + "\r\n\r\n").getBytes( StandardCharsets.UTF_8 ); + _out.write(response,0,response.length); + } + + final byte[] _ins = new byte[1024]; + final byte[] _str = new byte[1024]; + final byte[] _key = new byte[4]; + int _x, _len; + private int rawget() { assert _x<_len; return _ins[_x++]&0xFF; } + + // TODO: Flip to Direct Byte Buffer, this is just a poor-mans version + private void flipRead() throws IOException { + // Move _x to _len to offset 0 + System.arraycopy(_ins,_x,_ins,0,(_len-_x)); + _len = _len-_x; + _x=0; + // Read more + int len = _in.read(_ins,_len,_ins.length-_len); + if( len == -1 ) + throw Utils.TODO(); // EOF + _len += len; + } + // Block until you read at least len + private void atLeast(int len) throws IOException { + if( _x+len <= _len ) return; + flipRead(); + } + + + // Classic read from browser client. Encoded, so decoded here + public String get() throws IOException { + atLeast(1); // Need 1 byte for opcode + int op = rawget(); + if( (op&0x80)!=0x80 ) // FIN bit not set? + throw Utils.TODO( "Multi-part from client: " + op ); // Multipart message + op &= 0x7F; // Strip FIN bit + switch( op&15) { + case 1: { + atLeast(5); // Min message length + // Message length + int len = rawget()-128; + if( len > 125 ) { + if( len > 126 ) + throw Utils.TODO(); // Multi-byte length + len = (rawget()<<8) | rawget(); + } + atLeast(len+4); // Rest of message + // Encoding key + _key[0] = (byte)rawget(); + _key[1] = (byte)rawget(); + _key[2] = (byte)rawget(); + _key[3] = (byte)rawget(); + + // Decode bytes + for( int i=0; i<len; i++ ) + _str[i] = (byte)(rawget() ^ _key[i & 3]); + + return new String(_str,0,len); + } + // Client closes + case 8: return null; + // Ping + case 9: + // Send a PONG back. TODO: echo PING data in the PONG + _out.write(0x8A); // FIN + PONG + _out.write(0x00); // LENGTH==0 + return get(); + // Not (yet) implemented: continuation frames, ping/pong + default: + throw Utils.TODO("Unexpected op from client: "+op); // Multi-part message + } + } + + // Classic put to browser client. No encoding. + public void put(String msg) throws IOException { + int len = msg.length(); + if( len <= 125 ) { + _out.write(129); // FIN, opcode 1 - whole text message + _out.write(len); + for( int i=0; i<msg.length(); i++ ) + _out.write(msg.charAt(i)); + return; + } + if( len > 65535 ) + throw Utils.TODO("MOAR LENGTH"); + _out.write(129); // FIN, opcode 1 - whole text message + _out.write(126); // Length is 16b + _out.write(len>>>8); + _out.write(len&255); + for( int i=0; i<msg.length(); i++ ) + _out.write(msg.charAt(i)); + } + + @Override public void close() throws IOException { + if( _out != null ) { + _out.write(0b10001000); + _out.flush(); + } + super.close(); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/Utils.java b/chapter20/src/main/java/com/seaofnodes/simple/Utils.java new file mode 100644 index 00000000..1346c3ea --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/Utils.java @@ -0,0 +1,53 @@ +package com.seaofnodes.simple; + +import java.util.ArrayList; + +public class Utils { + public static RuntimeException TODO() { return TODO("Not yet implemented"); } + public static RuntimeException TODO(String msg) { return new RuntimeException(msg); } + + /** + * Fast, constant-time, element removal. Does not preserve order + * + * @param array ArrayList to modify + * @param i element to be removed + * @return element removed + */ + public static <E> void del(ArrayList<E> array, int i) { + E last = array.removeLast(); + if (i < array.size()) array.set(i, last); + } + + /** + * Search a list for an element by reference + * + * @param ary List to search in + * @param x Object to be searched + * @return >= 0 on success, -1 on failure + */ + public static <E> int find( ArrayList<E> ary, E x ) { + for( int i=0; i<ary.size(); i++ ) + if( ary.get(i)==x ) + return i; + return -1; + } + /** + * Search a list for an element by reference + * + * @param ary List to search in + * @param x Object to be searched + * @return >= 0 on success, -1 on failure + */ + public static <E> int find( E[] ary, E x ) { + for( int i=0; i<ary.length; i++ ) + if( ary[i]==x ) + return i; + return -1; + } + + // Rotate a long, nice for hashes + public static long rot( long x, int n ) { return (x<<n) | (x>>>n); } + + // Fold a 64bit hash into 32 bits + public static int fold( long x ) { return (int)((x>>32) ^ x); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/Var.java b/chapter20/src/main/java/com/seaofnodes/simple/Var.java new file mode 100644 index 00000000..e71d7ef2 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/Var.java @@ -0,0 +1,55 @@ +package com.seaofnodes.simple; + +import com.seaofnodes.simple.type.*; + + +/** + * The tracked fields are now complex enough to deserve a array-of-structs layout + */ +public class Var { + + public final String _name; // Declared name + public int _idx; // index in containing scope + private Type _type; // Declared type + public boolean _final; // Final field + public Parser.Lexer _loc; // Source location + public boolean _fref; // Forward ref + + public Var(int idx, String name, Type type, boolean xfinal, Parser.Lexer loc ) { + this(idx,name,type,xfinal,loc,false); + } + public Var(int idx, String name, Type type, boolean xfinal, Parser.Lexer loc, boolean fref) { + _idx = idx; + _name = name; + _type = type; + _final = xfinal; + _loc = loc; + _fref = fref; + } + public Type type() { + if( !_type.isFRef() ) return _type; + // Update self to no longer use the forward ref type + Type def = Parser.TYPES.get(((TypeMemPtr)_type)._obj._name); + return (_type=_type.meet(def)); + } + public Type lazyGLB() { + Type t = type(); + return t instanceof TypeMemPtr ? t : t.glb(); + } + + // Forward reference variables (not types) must be BOTTOM and + // distinct from inferred variables + public boolean isFRef() { return _fref; } + + public void defFRef( Type type, boolean xfinal, Parser.Lexer loc ) { + assert isFRef() && xfinal; + _type = type; + _final = true; + _loc = loc; + _fref = false; + } + + @Override public String toString() { + return _type.toString()+(_final ? " ": " !")+_name; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/AddFNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/AddFNode.java new file mode 100644 index 00000000..7d4e2910 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/AddFNode.java @@ -0,0 +1,44 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class AddFNode extends Node { + public AddFNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "AddF"; } + + @Override public String glabel() { return "+"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("+"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0 && + in(2)._type instanceof TypeFloat i1) { + if (i0.isConstant() && i1.isConstant()) + return TypeFloat.constant(i0.value()+i1.value()); + } + return in(1)._type.meet(in(2)._type); + } + + @Override + public Node idealize() { + Node lhs = in(1); + Type t2 = in(2)._type; + + // Add of 0. + if ( t2.isConstant() && t2 instanceof TypeFloat i && i.value()==0 ) + return lhs; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new AddFNode(lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/AddNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/AddNode.java new file mode 100644 index 00000000..6b7f2418 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/AddNode.java @@ -0,0 +1,190 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.*; + +import java.util.BitSet; + +import static com.seaofnodes.simple.Parser.con; + +public class AddNode extends Node { + public AddNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "Add"; } + + @Override public String glabel() { return "+"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("+"), visited); + return sb.append(")"); + } + + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i1 && + t2 instanceof TypeInteger i2 ) { + if (i1.isConstant() && i2.isConstant()) + return TypeInteger.constant(i1.value()+i2.value()); + // Fold ranges like {0-1} + {2-3} into {2-4}. + if( !overflow(i1._min,i2._min) && + !overflow(i1._max,i2._max) ) + return TypeInteger.make(i1._min+i2._min,i1._max+i2._max); + } + return TypeInteger.BOT; + } + + static boolean overflow( long x, long y ) { + if( (x ^ y ) < 0 ) return false; // unequal signs, never overflow + return (x ^ (x + y)) < 0; // sum has unequal signs, so overflow + } + @Override + public Node idealize () { + Node lhs = in(1); + Node rhs = in(2); + Type t2 = rhs._type; + + // Add of 0. We do not check for (0+x) because this will already + // canonicalize to (x+0) + if( t2 == TypeInteger.ZERO ) + return lhs; + + // Add of same to a multiply by 2 + if( lhs==rhs ) + return new ShlNode(null,lhs,con(1)); + + // Goal: a left-spine set of adds, with constants on the rhs (which then fold). + + // Move non-adds to RHS + if( !(lhs instanceof AddNode) && rhs instanceof AddNode ) + return swap12(); + + // x+(-y) becomes x-y + if( rhs instanceof MinusNode minus ) + return new SubNode(lhs,minus.in(1)); + + // Now we might see (add add non) or (add non non) or (add add add) but never (add non add) + + // Do we have x + (y + z) ? + // Swap to (x + y) + z + // Rotate (add add add) to remove the add on RHS + if( rhs instanceof AddNode add ) + return new AddNode(new AddNode(lhs,add.in(1)).peephole(), add.in(2)); + + // Now we might see (add add non) or (add non non) but never (add non add) nor (add add add) + if( !(lhs instanceof AddNode) ) + // Rotate; look for (add (phi cons) con/(phi cons)) + return spine_cmp(lhs,rhs,this) ? swap12() : phiCon(this,true); + + // Now we only see (add add non) + + // Dead data cycle; comes about from dead infinite loops. Do nothing, + // the loop will peep as dead after a bit. + if( lhs.in(1) == lhs ) + return null; + + // Do we have (x + con1) + con2? + // Replace with (x + (con1+con2) which then fold the constants + // lhs.in(2) is con1 here + // If lhs.in(2) is not a constant, we add ourselves as a dependency + // because if it later became a constant then we could make this + // transformation. + if( lhs.in(2).addDep(this)._type.isConstant() && rhs._type.isConstant() ) + return new AddNode(lhs.in(1),new AddNode(lhs.in(2),rhs).peephole()); + + + // Do we have ((x + (phi cons)) + con) ? + // Do we have ((x + (phi cons)) + (phi cons)) ? + // Push constant up through the phi: x + (phi con0+con0 con1+con1...) + Node phicon = phiCon(this,true); + if( phicon!=null ) return phicon; + + // Now we sort along the spine via rotates, to gather similar things together. + + // Do we rotate (x + y) + z + // into (x + z) + y ? + if( spine_cmp(lhs.in(2),rhs,this) ) + return new AddNode(new AddNode(lhs.in(1),rhs).peephole(),lhs.in(2)); + + return null; + } + + // Rotation is only valid for associative ops, e.g. Add, Mul, And, Or, Xor. + // Do we have ((phi cons)|(x + (phi cons)) + con|(phi cons)) ? + // Push constant up through the phi: x + (phi con0+con0 con1+con1...) + static Node phiCon(Node op, boolean rotate) { + Node lhs = op.in(1); + Node rhs = op.in(2); + if( rhs._type==TypeInteger.TOP ) return null; + // LHS is either a Phi of constants, or another op with Phi of constants + PhiNode lphi = pcon(lhs,op); + if( rotate && lphi==null && lhs.nIns() > 2 ) { + // Only valid to rotate constants if both are same associative ops + if( lhs.getClass() != op.getClass() ) return null; + lphi = pcon(lhs.in(2),op); // Will rotate with the Phi push + } + if( lphi==null ) return null; + if( lphi.region().nIns() <=2 ) return null; // Phi is collapsing + + // RHS is a constant or a Phi of constants + if( !(rhs instanceof ConstantNode) && pcon(rhs,op)==null ) + return null; + + // If both are Phis, must be same Region + if( rhs instanceof PhiNode && lphi.in(0) != rhs.in(0) ) + return null; + + // Note that this is the exact reverse of Phi pulling a common op down + // to reduce total op-count. We don't get in an endless push-up + // push-down peephole cycle because the constants all fold first. + Node[] ns = new Node[lphi.nIns()]; + ns[0] = lphi.in(0); + // Push constant up through the phi: x + (phi con0+con0 con1+con1...) + for( int i=1; i<ns.length; i++ ) + ns[i] = op.copy(lphi.in(i), rhs instanceof PhiNode ? rhs.in(i) : rhs).peephole(); + String label = lphi._label + (rhs instanceof PhiNode rphi ? rphi._label : ""); + Node phi = new PhiNode(label,lphi._declaredType,ns).peephole(); + // Rotate needs another op, otherwise just the phi + return lhs==lphi ? phi : op.copy(lhs.in(1),phi); + } + + /** + * Tests if the op is a phi and has all constant inputs. + * If not, returns null. + * If op is a phi, but its inputs are not all constants, then dep is added as + * a dependency to the phi's non-const input, because if later the input turn into a constant + * dep can make progress. + */ + static PhiNode pcon(Node op, Node dep) { + return op instanceof PhiNode phi && phi.allCons(dep) ? phi : null; + } + + // Compare two off-spine nodes and decide what order they should be in. + // Do we rotate ((x + hi) + lo) into ((x + lo) + hi) ? + // Generally constants always go right, then Phi-of-constants, then muls, then others. + // Ties with in a category sort by node ID. + // TRUE if swapping hi and lo. + static boolean spine_cmp( Node hi, Node lo, Node dep ) { + if( lo._type.isConstant() ) return false; + if( hi._type.isConstant() ) return true ; + + if( lo instanceof PhiNode lphi && lphi.region()._type==Type.XCONTROL ) return false; + if( hi instanceof PhiNode hphi && hphi.region()._type==Type.XCONTROL ) return false; + + if( lo instanceof PhiNode && lo.allCons(dep) ) return false; + if( hi instanceof PhiNode && hi.allCons(dep) ) return true ; + + if( lo instanceof PhiNode && !(hi instanceof PhiNode) ) return true; + if( hi instanceof PhiNode && !(lo instanceof PhiNode) ) return false; + + // Same category of "others" + return lo._nid > hi._nid; + } + + @Override Node copy(Node lhs, Node rhs) { return new AddNode(lhs,rhs); } + @Override Node copyF() { return new AddFNode(null,null); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/AndNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/AndNode.java new file mode 100644 index 00000000..36e3a949 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/AndNode.java @@ -0,0 +1,58 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; + +import java.util.BitSet; + +public class AndNode extends LogicalNode { + public AndNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "And"; } + @Override public String op() { return "&"; } + + @Override public String glabel() { return "&"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1 ) { + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()&i1.value()); + // Sharpen allowed bits if either value is narrowed + long mask = i0.mask() & i1.mask(); + return mask < 0 ? TypeInteger.BOT : TypeInteger.make(0,mask); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // And of -1. We do not check for (-1&x) because this will already + // canonicalize to (x&-1) + if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==-1 ) + return lhs; + + // Move constants to RHS: con*arg becomes arg*con + if ( t1.isConstant() && !t2.isConstant() ) + return swap12(); + + // Do we have ((x & (phi cons)) & con) ? + // Do we have ((x & (phi cons)) & (phi cons)) ? + // Push constant up through the phi: x & (phi con0&con0 con1&con1...) + Node phicon = AddNode.phiCon(this,true); + if( phicon!=null ) return phicon; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new AndNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/BoolNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/BoolNode.java new file mode 100644 index 00000000..84be11a9 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/BoolNode.java @@ -0,0 +1,134 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; +import static com.seaofnodes.simple.type.TypeInteger.*; + +abstract public class BoolNode extends Node { + + public BoolNode(Node lhs, Node rhs) { + super(null, lhs, rhs); + } + + abstract public String op(); // String opcode name + + @Override + public String label() { return getClass().getSimpleName(); } + + @Override + public String glabel() { return op(); } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append(op()), visited); + return sb.append(")"); + } + + @Override + public TypeInteger compute() { + Type t1 = in(1)._type; + Type t2 = in(2)._type; + // Exactly equals? + if( t1.isHigh() || t2.isHigh() ) + return BOOL.dual(); + if( in(1)==in(2) ) + // LT fails, both EQ and LE succeed + return this instanceof LT ? FALSE : TRUE; + if( t1 instanceof TypeInteger i1 && + t2 instanceof TypeInteger i2 ) + return doOp(i1,i2); + if( t1 instanceof TypeFloat f1 && + t2 instanceof TypeFloat f2 && + f1.isConstant() && f2.isConstant() ) + return doOp(f1.value(), f2.value()) ? TRUE : FALSE; + return BOOL; + } + + TypeInteger doOp(TypeInteger t1, TypeInteger t2) { throw Utils.TODO(); } + boolean doOp(double lhs, double rhs) { throw Utils.TODO(); } + Node copyF(Node lhs, Node rhs) { return null; } + + @Override + public Node idealize() { + // Compare of same + if( in(1)==in(2) ) + return this instanceof LT ? Parser.ZERO : new ConstantNode(TRUE); + + // Equals pushes constant to the right; 5==X becomes X==5. + if( this instanceof EQ ) { + if( !(in(2) instanceof ConstantNode) ) { + // con==noncon becomes noncon==con + if( in(1) instanceof ConstantNode || in(1)._nid > in(2)._nid ) + // Equals sorts by NID otherwise: non.high == non.low becomes non.low == non.high + return in(1)._type instanceof TypeFloat ? new EQF(in(2),in(1)) : new EQ(in(2),in(1)); + } + // Equals X==0 becomes a !X + if( (in(2)._type == ZERO || in(2)._type == Type.NIL) ) + return new NotNode(in(1)); + } + + // Do we have ((x * (phi cons)) * con) ? + // Do we have ((x * (phi cons)) * (phi cons)) ? + // Push constant up through the phi: x * (phi con0*con0 con1*con1...) + Node phicon = AddNode.phiCon(this,this instanceof EQ); + if( phicon!=null ) return phicon; + + return null; + } + + public static class EQ extends BoolNode { + public EQ(Node lhs, Node rhs) { super(lhs,rhs); } + public String op() { return "=="; } + TypeInteger doOp(TypeInteger i1, TypeInteger i2) { + if( i1==i2 && i1.isConstant() ) return TRUE; + if( i1._max < i2._min || i1._min > i2._max ) return FALSE; + return BOOL; + } + Node copy(Node lhs, Node rhs) { return new EQ(lhs,rhs); } + Node copyF() { return new EQF(null,null); } + } + public static class LT extends BoolNode { + public LT(Node lhs, Node rhs) { super(lhs,rhs); } + public String op() { return "<" ; } + public String glabel() { return "<"; } + TypeInteger doOp(TypeInteger i1, TypeInteger i2) { + if( i1._max < i2._min ) return TRUE; + if( i1._min >= i2._max ) return FALSE; + return BOOL; + } + Node copy(Node lhs, Node rhs) { return new LT(lhs,rhs); } + Node copyF() { return new LTF(null,null); } + } + public static class LE extends BoolNode { + public LE(Node lhs, Node rhs) { super(lhs,rhs); } + public String op() { return "<="; } + public String glabel() { return "<="; } + TypeInteger doOp(TypeInteger i1, TypeInteger i2) { + if( i1._max <= i2._min ) return TRUE; + if( i1._min > i2._max ) return FALSE; + return BOOL; + } + Node copy(Node lhs, Node rhs) { return new LE(lhs,rhs); } + Node copyF() { return new LEF(null,null); } + } + + public static class EQF extends EQ { public EQF(Node lhs, Node rhs) { super(lhs,rhs); } boolean doOp(double lhs, double rhs) { return lhs == rhs; } } + public static class LTF extends LT { public LTF(Node lhs, Node rhs) { super(lhs,rhs); } boolean doOp(double lhs, double rhs) { return lhs < rhs; } } + public static class LEF extends LE { public LEF(Node lhs, Node rhs) { super(lhs,rhs); } boolean doOp(double lhs, double rhs) { return lhs <= rhs; } } + + // Unsigned less that, for range checks. Not directly user writable. + public static class ULT extends BoolNode { + public ULT(Node lhs, Node rhs) { super(lhs,rhs); } + public String op() { return "u<" ; } + public String glabel() { return "u<"; } + TypeInteger doOp(TypeInteger i1, TypeInteger i2) { + if( Long.compareUnsigned(i1._max,i2._min) < 0 ) return TRUE; + if( Long.compareUnsigned(i1._min,i2._max) >= 0 ) return FALSE; + return BOOL; + } + Node copy(Node lhs, Node rhs) { return new ULT(lhs,rhs); } + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CFGNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CFGNode.java new file mode 100644 index 00000000..174725f5 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CFGNode.java @@ -0,0 +1,162 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Ary; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; +import java.util.HashMap; +import java.util.HashSet; + +/** Control Flow Graph Nodes + * <p> + * CFG nodes have a immediate dominator depth (idepth) and a loop nesting + * depth(loop_depth). + * <p> + * idepth is computed lazily upon first request, and is valid even in the + * Parser, and is used by peepholes during parsing and afterward. + * <p> + * loop_depth is computed after optimization as part of scheduling. + * + */ +public abstract class CFGNode extends Node { + + public CFGNode(Node... nodes) { super(nodes); } + public CFGNode(CFGNode cfg) { + super(cfg); + _idepth = cfg._idepth; + _ltree = cfg._ltree; + _pre = cfg._pre; + } + + public CFGNode cfg(int idx) { return (CFGNode)in(idx); } + + // Block head is Start, Region, CProj, but not e.g. If, Return, Stop + public boolean blockHead() { return false; } + + // Get the one control following; error to call with more than one e.g. an + // IfNode or other multi-way branch. + public CFGNode uctrl() { + CFGNode c = null; + for( Node n : _outputs ) + if( n instanceof CFGNode cfg ) + { assert c==null; c = cfg; } + return c; + } + + + // ------------------------------------------------------------------------ + /** + * Immediate dominator tree depth, used to approximate a real IDOM during + * parsing where we do not have the whole program, and also peepholes + * change the CFG incrementally. + * <p> + * See {@link <a href="https://en.wikipedia.org/wiki/Dominator_(graph_theory)">...</a>} + */ + public int _idepth; + public int idepth() { return _idepth==0 ? (_idepth=idom().idepth()+1) : _idepth; } + + // Return the immediate dominator of this Node and compute dom tree depth. + public CFGNode idom(Node dep) { return cfg(0); } + public final CFGNode idom() { return idom(null); } + + // Return the LCA of two idoms + public CFGNode _idom(CFGNode rhs, Node dep) { + if( rhs==null ) return this; + CFGNode lhs = this; + while( lhs != rhs ) { + var comp = lhs.idepth() - rhs.idepth(); + if( comp >= 0 ) lhs = ((CFGNode)lhs.addDep(dep)).idom(); + if( comp <= 0 ) rhs = ((CFGNode)rhs.addDep(dep)).idom(); + } + return lhs; + } + + // Anti-dependence field support + public int _anti; // Per-CFG field to help find anti-deps + + // ------------------------------------------------------------------------ + // Loop nesting + public LoopNode loop() { return _ltree._head; } + public int loopDepth() { return _ltree==null ? 0 : _ltree.depth(); } + + LoopTree _ltree; + int _pre; // Pre-order numbers for loop tree finding + private static class LoopTree { + LoopTree _par; + final LoopNode _head; + int _depth; + LoopTree(LoopNode head) { _head = head; } + @Override public String toString() { return "LOOP"+_head._nid; } + int depth() { + return _depth==0 ? (_par==null ? 0 : (_depth = _par.depth()+1)) : _depth; + } + } + + // ------------------------------------------------------------------------ + // Tag all CFG Nodes with their containing LoopNode; LoopNodes themselves + // also refer to *their* containing LoopNode, as well as have their depth. + // Start is a LoopNode which contains all at depth 1. + public void buildLoopTree(StopNode stop) { + _ltree = stop._ltree = Parser.XCTRL._ltree = new LoopTree((StartNode)this); + _bltWalk(2,null,stop, new BitSet()); + } + int _bltWalk( int pre, FunNode fun, StopNode stop, BitSet post ) { + // Pre-walked? + if( _pre!=0 ) return pre; + _pre = pre++; + // Pre-walk + for( Node use : _outputs ) + if( use instanceof CFGNode usecfg && !skip( usecfg ) ) + pre = usecfg._bltWalk( pre, use instanceof FunNode fuse ? fuse : fun, stop, post ); + + // Post-order work: find innermost loop + LoopTree inner = null, ltree; + for( Node use : _outputs ) { + if( !(use instanceof CFGNode usecfg) ) continue; + if( skip(usecfg) ) continue; + if( usecfg._type == Type.XCONTROL || // Do not walk dead control + usecfg._type == TypeTuple.IF_NEITHER ) // Nor dead IFs + continue; + // Child visited but not post-visited? + if( !post.get(usecfg._nid) ) { + // Must be a backedge to a LoopNode then + ltree = usecfg._ltree = new LoopTree((LoopNode)usecfg); + } else { + // Take child's loop choice, which must exist + ltree = usecfg._ltree; + // If falling into a loop, use the target loop's parent instead + if( ltree._head == usecfg ) { + if( ltree._par == null ) + // This loop never had an If test choose to take its + // exit, i.e. it is a no-exit infinite loop. + ltree._par = ltree._head.forceExit(fun,stop)._ltree; + ltree = ltree._par; + } + } + // Sort inner loops. The decision point is some branch far removed + // from either loop head OR either backedge so requires pre-order + // numbers to figure out innermost. + if( inner == null ) { inner = ltree; continue; } + if( inner == ltree ) continue; // No change + LoopTree outer = ltree._head._pre > inner._head._pre ? inner : ltree; + inner = ltree._head._pre > inner._head._pre ? ltree : inner; + inner._par = outer; + } + // Set selected loop + if( inner!=null ) + _ltree = inner; + // Tag as post-walked + post.set(_nid); + return pre; + } + + boolean skip(CFGNode usecfg) { + // Only walk control users that are alive. + // Do not walk from a Call to linked Fun's. + return usecfg instanceof XCtrlNode || + (this instanceof CallNode && usecfg instanceof FunNode) || + (this instanceof ReturnNode && usecfg instanceof CallEndNode); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CProjNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CProjNode.java new file mode 100644 index 00000000..690ef590 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CProjNode.java @@ -0,0 +1,62 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeMem; +import com.seaofnodes.simple.type.TypeTuple; +import java.util.BitSet; + +public class CProjNode extends CFGNode { + + // Which slice of the incoming multipart value + public final int _idx; + + // Debugging label + public final String _label; + + public CProjNode(Node ctrl, int idx, String label) { + super(ctrl); + _idx = idx; + _label = label; + } + public CProjNode(CProjNode c) { super(c); _idx = c._idx; _label = c._label; } + + @Override public String label() { return _label; } + + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { return sb.append(_label); } + + @Override public boolean isMultiTail() { return in(0).isMultiHead(); } + @Override public boolean blockHead() { return true; } + + public CFGNode ctrl() { return cfg(0); } + + @Override + public Type compute() { + Type t = ctrl()._type; + return t instanceof TypeTuple tt ? tt._types[_idx] : Type.BOTTOM; + } + + @Override + public Node idealize() { + if( ctrl()._type instanceof TypeTuple tt ) { + if( tt._types[_idx]==Type.XCONTROL ) + return Parser.XCTRL; // We are dead + if( ctrl() instanceof IfNode && tt._types[1-_idx]==Type.XCONTROL ) // Only true for IfNodes + return ctrl().in(0); // We become our input control + } + + // Flip a negating if-test, to remove the not + if( ctrl() instanceof IfNode iff && iff.pred().addDep(this) instanceof NotNode not ) + return new CProjNode(new IfNode(iff.ctrl(),not.in(1)).peephole(),1-_idx,_idx==0 ? "False" : "True"); + + // Copy of some other input + return ((MultiNode)ctrl()).pcopy(_idx); + } + + @Override + boolean eq( Node n ) { return _idx == ((CProjNode)n)._idx; } + + @Override + int hash() { return _idx; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CallEndNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CallEndNode.java new file mode 100644 index 00000000..4e50c1fc --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CallEndNode.java @@ -0,0 +1,90 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +/** + * CallEnd + */ +public class CallEndNode extends CFGNode implements MultiNode { + + // When set true, this Call/CallEnd/Fun/Return is being trivially inlined + private boolean _folding; + + public CallEndNode(CallNode call) { super(new Node[]{call}); } + public CallEndNode(CallEndNode cend) { super(cend); } + + @Override + public String label() { return "CallEnd"; } + @Override public boolean isMultiHead() { return true; } + @Override public boolean blockHead() { return true; } + + CallNode call() { return (CallNode)in(0); } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("cend( "); + sb.append( in(0) instanceof CallNode ? "Call, " : "----, "); + for( int i=1; i<nIns()-1; i++ ) + in(i)._print0(sb,visited).append(","); + sb.setLength(sb.length()-1); + return sb.append(")"); + } + + @Override + public Type compute() { + if( !(in(0) instanceof CallNode call) ) + return TypeTuple.RET.dual(); + Type ret = call.fptr().addDep(this)._type instanceof TypeFunPtr tfp ? tfp.ret() : Type.BOTTOM; + return TypeTuple.make(call._type,TypeMem.BOT,ret); + } + + @Override + public Node idealize() { + + // Trivial inlining: call site calls a single function; single function + // is only called by this call site. + if( !_folding && nIns()==2 && in(0) instanceof CallNode call ) { + Node fptr = call.fptr(); + if( fptr.nOuts() == 1 && // Only user is this call + fptr instanceof ConstantNode && // We have an immediate call + // Function is being called, and its not-null + fptr._type instanceof TypeFunPtr tfp && tfp.notNull() && + // Arguments are correct + call.err()==null ) { + ReturnNode ret = (ReturnNode)in(1); + FunNode fun = ret.fun(); + // Expecting Start, and the Call + if( fun.nIns()==3 ) { + assert fun.in(1) instanceof StartNode && fun.in(2)==call; + // Disallow self-recursive inlining (loop unrolling by another name) + CFGNode idom = call; + while( !(idom instanceof FunNode fun2) ) + idom = idom.idom(); + if( idom != fun ) { + // Trivial inline: rewrite + _folding = true; + // Rewrite Fun so the normal RegionNode ideal collapses + fun._folding = true; + fun.setDef(1,Parser.XCTRL); // No default/unknown StartNode caller + fun.setDef(2,call.ctrl()); // Bypass the Call; + fun.ret().setDef(3,null); // Return is folding also + CodeGen.CODE.addAll(fun._outputs); + return this; + } + } else { + fun.addDep(this); + } + } else { // Function ptr has multiple users (so maybe multiple call sites) + fptr.addDep(this); + } + } + + return null; + } + + @Override public Node pcopy(int idx) { + return _folding ? in(1).in(idx) : null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CallNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CallNode.java new file mode 100644 index 00000000..a9ca852c --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CallNode.java @@ -0,0 +1,146 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.IterPeeps; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.util.BitSet; + +/** + * Call + */ +public class CallNode extends CFGNode { + + // Source location for late reported errors + public final Parser.Lexer _loc; + + public CallNode(Parser.Lexer loc, Node... nodes) { super(nodes); _loc = loc; } + public CallNode(CallNode call) { super(call); _loc = call._loc; } + + @Override + public String label() { return "Call"; } + + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { + String fname = name(); + if( fname == null ) fptr()._print0(sb,visited); + else sb.append(fname); + sb.append("( "); + for( int i=2; i<nIns()-1; i++ ) + in(i)._print0(sb,visited).append(","); + sb.setLength(sb.length()-1); + return sb.append(")"); + } + public String name() { + if( fptr()._type instanceof TypeFunPtr tfp && tfp.isConstant() ) + return CodeGen.CODE.link(tfp)._name; + return null; + } + + + + Node ctrl() { return in(0); } + Node mem () { return in(1); } + // Args mapped 1-to-1 on inputs, so conceptually start at 2 + Node arg(int idx) { return in(idx); } + // Same arg accounting as TFPs, although the numbering starts at 2 + int nargs() { return nIns()-3; } // Minus control, memory, fptr + // args from input 2 to last; last is function input + public Node fptr() { return _inputs.last(); } + + // Find the Call End from the Call + public CallEndNode cend() { + // Always in use slot 0 + if( nOuts()>0 && out(0) instanceof CallEndNode cend ) { + assert _cend()==cend; + return cend; + } else { + assert _cend()==null; + return null; + } + } + private CallEndNode _cend() { + CallEndNode cend=null; + for( Node n : _outputs ) + if( n instanceof CallEndNode cend0 ) + { assert cend == null; cend = cend0; } + return cend; + } + + // Get the one control following; error to call with more than one e.g. an + // IfNode or other multi-way branch. + @Override public CFGNode uctrl() { return cend(); } + + + @Override + public Type compute() { + return ctrl()._type; + } + + @Override + public Node idealize() { + CallEndNode cend = cend(); + if( cend==null ) return null; // Still building + + // Link: call calls target function. Linking makes the target FunNode + // point to this Call, and all his Parms point to the call arguments; + // also the CallEnd points to the Return. + if( fptr()._type instanceof TypeFunPtr tfp && tfp.nargs() == nargs() ) { + // If fidxs is negative, then infinite unknown functions + long fidxs = tfp.fidxs(); + if( fidxs > 0 ) { + // Wipe out the return which matching in the linker table + // Walk the (63 max) bits and link + for( ; fidxs!=0; fidxs = TypeFunPtr.nextFIDX(fidxs) ) { + int fidx = Long.numberOfTrailingZeros(fidxs); + TypeFunPtr tfp0 = tfp.makeFrom(fidx); + FunNode fun = CodeGen.CODE.link(tfp0); + if( fun!=null && !fun._folding && !linked(fun) ) + link(fun); + } + } + } + + return null; + } + + // True if Fun is linked to this Call + boolean linked( FunNode fun ) { + for( Node n : fun._inputs ) + if( n == this ) + return true; + return false; + } + + + // Link so this calls fun + private void link( FunNode fun ) { + assert !linked(fun); + fun.addDef(this); + for( Node use : fun._outputs ) + if( use instanceof ParmNode parm ) + parm.addDef(parm._idx==0 ? cend() : arg(parm._idx)); + // Call end points to function return + CodeGen.CODE.add(cend()).addDef(fun.ret()); + assert linked(fun); + } + + @Override + public Parser.ParseException err() { + if( !(fptr()._type instanceof TypeFunPtr tfp) ) + throw Utils.TODO(); + if( !tfp.notNull() ) + return Parser.error( "Might be null calling "+tfp, _loc); + if( nargs() != tfp.nargs() ) + return Parser.error( "Expecting "+tfp.nargs()+" arguments, but found "+nargs(), _loc); + + // Check for args + for( int i=0; i<tfp.nargs(); i++ ) + if( !arg(i+2)._type.isa(tfp.arg(i)) ) + return Parser.error( "Argument #"+i+" isa "+arg(i+2)._type+", but must be a "+tfp.arg(i), _loc); + + return null; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CastNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CastNode.java new file mode 100644 index 00000000..d8a968c1 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CastNode.java @@ -0,0 +1,58 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +// Upcast (join) the input to a t. Used after guard test to lift an input. +// Can also be used to make a type-assertion if ctrl is null. +public class CastNode extends Node { + private final Type _t; + public CastNode(Type t, Node ctrl, Node in) { + super(ctrl, in); + _t = t; + setType(compute()); + } + + @Override public String label() { return "("+_t.str()+")"; } + + @Override + public String uniqueName() { return "Cast_" + _nid; } + + @Override public boolean isPinned() { return true; } + @Override public boolean isConst() { return true; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return in(1)._print0(sb.append(label()), visited); + } + + @Override + public Type compute() { + return in(1)._type.join(_t); + } + + @Override + public Node idealize() { + return in(1)._type.isa(_t) ? in(1) : null; + } + + @Override + boolean eq(Node n) { + CastNode cast = (CastNode)n; // Contract + return _t==cast._t; + } + + @Override + int hash() { return _t.hashCode(); } + + @Override + public Parser.ParseException err() { + // Has a condition to test, so OK + if( in(0) != null ) return null; + // No condition to test, so this must optimize away + throw Utils.TODO(); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ConstantNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ConstantNode.java new file mode 100644 index 00000000..d788cff7 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ConstantNode.java @@ -0,0 +1,70 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.util.BitSet; + +/** + * A Constant node represents a constant value. At present, the only constants + * that we allow are integer literals; therefore Constants contain an integer + * value. As we add other types of constants, we will refactor how we represent + * Constants. + * <p> + * Constants have no semantic inputs. However, we set Start as an input to + * Constants to enable a forward graph walk. This edge carries no semantic + * meaning, and it is present <em>solely</em> to allow visitation. + * <p> + * The Constant's value is the value stored in it. + */ + +public class ConstantNode extends Node { + public final Type _con; + public ConstantNode( Type type ) { + super(new Node[]{CodeGen.CODE._start}); + _con = _type = type; + } + public ConstantNode( ConstantNode con ) { super(con); _con = con._type; } + + public static Node make( Type type ) { + if( type==Type. CONTROL ) return new CtrlNode(); + if( type==Type.XCONTROL ) return new XCtrlNode(); + return new ConstantNode(type); + } + + @Override public String label() { return "#"+_con; } + @Override public String glabel() { return _con.gprint(new SB().p("#")).toString(); } + + @Override + public String uniqueName() { return "Con_" + _nid; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + if( _con instanceof TypeFunPtr tfp ) { + FunNode fun = CodeGen.CODE.link(tfp); + if( fun._name != null ) + return sb.append("{ ").append(fun._name).append("}"); + } + return sb.append(_con.print(new SB())); + } + + @Override public boolean isMultiTail() { return true; } + @Override public boolean isConst() { return true; } + + @Override + public Type compute() { return _con; } + + @Override + public Node idealize() { return null; } + + @Override + boolean eq(Node n) { + ConstantNode con = (ConstantNode)n; // Contract + return _con==con._con; + } + + @Override + int hash() { return _con.hashCode(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/CtrlNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/CtrlNode.java new file mode 100644 index 00000000..0ea060bd --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/CtrlNode.java @@ -0,0 +1,15 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import java.util.BitSet; + +public class CtrlNode extends CFGNode { + public CtrlNode() { super(CodeGen.CODE._start); } + @Override public String label() { return "Ctrl"; } + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { return sb.append("Cctrl"); } + @Override public boolean isConst() { return true; } + @Override public Type compute() { return Type.CONTROL; } + @Override public Node idealize() { return null; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/DivFNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/DivFNode.java new file mode 100644 index 00000000..36fd4524 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/DivFNode.java @@ -0,0 +1,46 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class DivFNode extends Node { + public DivFNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "DivF"; } + + @Override public String glabel() { return "/"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("/"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0 && + in(2)._type instanceof TypeFloat i1) { + if (i0.isConstant() && i1.isConstant()) + return TypeFloat.constant(i0.value()/i1.value()); + } + return in(1)._type.meet(in(2)._type); + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // Div of 1. + if ( t2.isConstant() && t2 instanceof TypeFloat i && i.value()==1 ) + return lhs; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new DivFNode(lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/DivNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/DivNode.java new file mode 100644 index 00000000..0d5ed297 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/DivNode.java @@ -0,0 +1,46 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class DivNode extends Node { + public DivNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "Div"; } + + @Override public String glabel() { return "//"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("/"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i1 && + t2 instanceof TypeInteger i2 ) { + if (i1.isConstant() && i2.isConstant()) + return i2.value() == 0 + ? TypeInteger.ZERO + : TypeInteger.constant(i1.value()/i2.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + // Div of 1. + if( in(2)._type == TypeInteger.TRUE ) + return in(1); + return null; + } + + @Override Node copy(Node lhs, Node rhs) { return new DivNode(lhs,rhs); } + @Override Node copyF() { return new DivFNode(null,null); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/FRefNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/FRefNode.java new file mode 100644 index 00000000..f0c4045d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/FRefNode.java @@ -0,0 +1,35 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Var; +import com.seaofnodes.simple.type.Type; +import java.util.BitSet; + +/** + * A Forward Reference. Its any final constant, including functions. When + * the Def finally appears its plugged into the forward reference, which then + * peepholes to the Def. + */ +public class FRefNode extends ConstantNode { + public static final Type FREF_TYPE = Type.BOTTOM; + public final Var _n; + public FRefNode( Var n ) { super(FREF_TYPE); _n = n; } + + @Override public String label() { return "FRef"+_n; } + + @Override public String uniqueName() { return "FRef_" + _nid; } + + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { + return sb.append("FRef_").append(_n); + } + + @Override public Node idealize() { + // When FRef finds its definition, idealize to it + return nIns()==1 ? null : in(1); + } + + public Parser.ParseException err() { return Parser.error("Undefined name '"+_n._name+"'",_n._loc); } + + @Override boolean eq(Node n) { return this==n; } + @Override int hash() { return _n._idx; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/FunNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/FunNode.java new file mode 100644 index 00000000..506e7758 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/FunNode.java @@ -0,0 +1,149 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.util.BitSet; +import static com.seaofnodes.simple.CodeGen.CODE; + +public class FunNode extends RegionNode { + + // When set true, this Call/CallEnd/Fun/Return is being trivially inlined + boolean _folding; + + private TypeFunPtr _sig; // Initial signature + private ReturnNode _ret; // Return pointer + + public String _name; // Debug name + + public FunNode( Parser.Lexer loc, TypeFunPtr sig, Node... nodes ) { super(loc,nodes); _sig = sig; } + public FunNode( FunNode fun ) { super( fun, fun._loc ); _sig = fun.sig(); _name = fun._name; } + + @Override + public String label() { return _name == null ? "$fun" : _name; } + + // Find the one CFG user from Fun. It's not always the Return, but always + // the Return *is* a CFG user of Fun. + @Override public CFGNode uctrl() { + for( Node n : _outputs ) + if( n instanceof CFGNode cfg && + (cfg instanceof RegionNode || cfg.cfg0()==this) ) + return cfg; + return null; + } + + public ParmNode rpc() { + ParmNode rpc = null; + for( Node n : _outputs ) + if( n instanceof ParmNode parm && parm._idx==0 ) + { assert rpc==null; rpc=parm; } + return rpc; + } + + // Cannot create the Return and Fun at the same time; one has to be first. + // So setting the return requires a second step. + public void setRet(ReturnNode ret) { _ret=ret; } + ReturnNode ret() { assert _ret!=null; return _ret; } + + // Signature can improve over time + public TypeFunPtr sig() { return _sig; } + void setSig( TypeFunPtr sig ) { + assert sig.isa(_sig); + if( _sig != sig ) { + CODE.add(this); + _sig = sig; + } + } + + @Override + public Type compute() { + // Only dead if no callers after SCCP + return Type.CONTROL; + } + + @Override + public Node idealize() { + + // Some linked path dies + Node progress = deadPath(); + if( progress!=null ) { + if( nIns()==3 && in(2) instanceof CallNode call ) + CODE.add(call.cend()); // If Start and one call, check for inline + return progress; + } + + // When can we assume no callers? Or no other callers (except main)? + // In a partial compilation, we assume Start gets access to any/all + // top-level public structures and recursively what they point to. + // This in turn is valid arguments to every callable function. + // + // In a total compilation, we can start from Start and keep things + // more contained. + + // If no default/unknown caller, use the normal RegionNode ideal rules + // to collapse + if( unknownCallers() ) return null; + + // If down to a single input, become that input + if( nIns()==2 && !hasPhi() ) { + CODE.add( CODE._stop ); // Stop will remove dead path + CODE.add( _ret ); // Return will compute to TOP control + return in(1); // Collapse if no Phis; 1-input Phis will collapse on their own + } + + return null; + } + + // Bypass Region idom, always assume depth == 1, one more than Start + @Override public int idepth() { return (_idepth=1); } + // Bypass Region idom, always assume idom is Start + @Override public CFGNode idom(Node dep) { return cfg(1); } + + // Always in-progress until we run out of unknown callers + public boolean unknownCallers() { return in(1) instanceof StartNode; } + + @Override public boolean inProgress() { return unknownCallers(); } + + // Add a new function exit point. + public void addReturn(Node ctrl, Node mem, Node rez) { _ret.addReturn(ctrl,mem,rez); } + + // Build the function body + public BitSet body() { + + // Reverse up (stop to start) CFG only, collect bitmap. + BitSet cfgs = new BitSet(); + cfgs.set(_nid); + walkUp(ret(),cfgs ); + + // Top down (start to stop) all flavors. CFG limit to bitmap. + // If data use bottoms out in wrong CFG, returns false - but tries all outputs. + // If any output hits an in-CFG use (e.g. phi), then keep node. + BitSet body = new BitSet(); + walkDown(this, cfgs, body, new BitSet()); + return body; + } + + private static void walkUp(CFGNode n, BitSet cfgs) { + if( cfgs.get(n._nid) ) return; + cfgs.set(n._nid); + if( n instanceof RegionNode r ) + for( int i=1; i<n.nIns(); i++ ) + walkUp(n.cfg(i),cfgs); + else walkUp(n.cfg0(),cfgs); + } + + private static boolean walkDown( Node n, BitSet cfgs, BitSet body, BitSet visit ) { + if( visit.get(n._nid) ) return body.get(n._nid); + visit.set(n._nid); + + if( n instanceof CFGNode && !cfgs.get(n._nid) ) + return false; + if( n.in(0)!=null && !cfgs.get(n.in(0)._nid) ) + return false; + boolean in = n.in(0)!=null || n instanceof CFGNode; + for( Node use : n._outputs ) + in |= walkDown(use,cfgs,body,visit); + if( in ) body.set(n._nid); + return in; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/IfNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/IfNode.java new file mode 100644 index 00000000..9e6d2dff --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/IfNode.java @@ -0,0 +1,69 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.IterPeeps; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; +import java.util.HashSet; + +public class IfNode extends CFGNode implements MultiNode { + + public IfNode(Node ctrl, Node pred) { + super(ctrl, pred); + CodeGen.CODE.add(this); // Because idoms are complex, just add it + } + public IfNode(IfNode iff) { super(iff); } + + @Override + public String label() { return "If"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("if( "); + return in(1)._print0(sb, visited).append(" )"); + } + + @Override public boolean isMultiHead() { return true; } + + public Node ctrl() { return in(0); } + public Node pred() { return in(1); } + + // No one unique control follows + @Override public CFGNode uctrl() { return null; } + + @Override + public Type compute() { + // If the If node is not reachable then neither is any following Proj + if (ctrl()._type != Type.CONTROL && ctrl()._type != Type.BOTTOM ) + return TypeTuple.IF_NEITHER; + Node pred = pred(); + Type t = pred._type; + // High types mean NEITHER side is reachable. + // Wait until the type falls to decide which way to go. + if( t.isHigh() ) + return TypeTuple.IF_NEITHER; + // If constant is 0 then false branch is reachable + // Else true branch is reachable + if( t.isConstant() ) + return (t==Type.NIL || t==TypeInteger.ZERO || (t instanceof TypeFunPtr tfp && tfp._fidxs==0) ) ? TypeTuple.IF_FALSE : TypeTuple.IF_TRUE; + // If adding a zero makes a difference, the predicate must not have a zero/null + if( !t.makeZero().isa(t) ) + return TypeTuple.IF_TRUE; + + return TypeTuple.IF_BOTH; + } + + @Override + public Node idealize() { + // Hunt up the immediate dominator tree. If we find an identical if + // test on either the true or false branch, that side wins. + if( !pred()._type.isHighOrConst() ) + for( CFGNode dom = idom(), prior=this; dom!=null; prior = dom, dom = dom.idom() ) + if( dom.addDep(this) instanceof IfNode iff && iff.pred().addDep(this)==pred() && prior instanceof CProjNode prj ) { + setDef(1,con( prj._idx==0 ? 1 : 0 )); + return this; + } + return null; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/LoadNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/LoadNode.java new file mode 100644 index 00000000..6c74042b --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/LoadNode.java @@ -0,0 +1,210 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +/** + * Load represents extracting a value from inside a memory object, + * in chapter 10 this means Struct fields. + */ +public class LoadNode extends MemOpNode { + + /** + * Load a value from a ptr.field. + * + * @param name The field we are loading + * @param mem The memory alias node - this is updated after a Store + * @param ptr The ptr to the struct base from where we load a field + * @param off The offset inside the struct base + */ + public LoadNode(Parser.Lexer loc, String name, int alias, Type glb, Node mem, Node ptr, Node off) { + super(loc, name, alias, glb, mem, ptr, off); + } + + // GraphVis DOT code (must be valid Java identifiers) and debugger labels + @Override public String label() { return "ld_"+mlabel(); } + // GraphVis node-internal labels + @Override public String glabel() { return "." +_name; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { return sb.append(".").append(_name); } + + @Override + public Type compute() { + if( mem()._type instanceof TypeMem mem ) { + // Update declared forward ref to the actual + if( _declaredType.isFRef() && mem._t instanceof TypeMemPtr tmp && !tmp.isFRef() ) + _declaredType = tmp; + // No lifting if ptr might null-check + if( err()==null ) + return _declaredType.join(mem._t); + } + return _declaredType; + } + + @Override + public Node idealize() { + Node ptr = ptr(); + Node mem = mem(); + + // Simple Load-after-Store on same address. + if( mem instanceof StoreNode st && + ptr == st.ptr() && off() == st.off() ) { // Must check same object + assert _name.equals(st._name); // Equiv class aliasing is perfect + return extend(st.val()); + } + + // Simple Load-after-New on same address. + if( mem instanceof ProjNode p && p.in(0) instanceof NewNode nnn && + ptr == nnn.proj(1) ) // Must check same object + return zero(nnn); // Load zero from new + + // Uplift control to a prior dominating load. + for( Node memuse : mem._outputs ) + // Find a prior load, has same mem,ptr,off but higher ctrl + if( memuse != this && memuse instanceof LoadNode ld && ptr==ld.ptr() && off()==ld.off() && + cfg0()!=null && cfg0()._idom(ld.cfg0(),this) == ld.cfg0() ) // Higher control means load is legal earlier + return ld; + + // Load-after-Store on same address, but bypassing provably unrelated + // stores. This is a more complex superset of the above two peeps. + // "Provably unrelated" is really weak. + if( ptr instanceof ReadOnlyNode ro ) + ptr = ro.in(1); + outer: + while( true ) { + switch( mem ) { + case StoreNode st: + if( ptr == st.ptr().addDep(this) && off() == st.off() ) + return extend(castRO(st.val())); // Proved equal + // Can we prove unequal? Offsets do not overlap? + if( !off()._type.join(st.off()._type).isHigh() && // Offsets overlap + !neverAlias(ptr,st.ptr()) ) // And might alias + break outer; // Cannot tell, stop trying + // Pointers cannot overlap + mem = st.mem(); // Proved never equal + break; + case PhiNode phi: + // Assume related + phi.addDep(this); + break outer; + case ConstantNode top: break outer; // Assume shortly dead + case ProjNode mproj: // Memory projection + switch( mproj.in(0) ) { + case NewNode nnn1: + if( ptr instanceof ProjNode pproj && pproj.in(0) == mproj.in(0) ) + return zero(nnn1); + if( !(ptr instanceof ProjNode pproj && pproj.in(0) instanceof NewNode) ) + break outer; // Cannot tell, ptr not related to New + mem = nnn1.in(nnn1.findAlias(_alias));// Bypass unrelated New + break; + case StartNode start: break outer; + case CallEndNode cend: break outer; // TODO: Bypass no-alias call + default: throw Utils.TODO(); + } + break; + case MemMergeNode merge: mem = merge.alias(_alias); break; + + default: + throw Utils.TODO(); + } + } + + // Push a Load up through a Phi, as long as it collapses on at least + // one arm. If at a Loop, the backedge MUST collapse - else we risk + // spinning the same transform around the loop indefinitely. + // BEFORE (2 Sts, 1 Ld): AFTER (1 St, 0 Ld): + // if( pred ) ptr.x = e0; val = pred ? e0 + // else ptr.x = e1; : e1; + // val = ptr.x; ptr.x = val; + if( mem() instanceof PhiNode memphi && memphi.region()._type == Type.CONTROL && memphi.nIns()== 3 && + // Offset can be hoisted + off() instanceof ConstantNode && + // Pointer can be hoisted + hoistPtr(ptr,memphi) ) { + + // Profit on RHS/Loop backedge + if( profit(memphi,2) || + // Else must not be a loop to count profit on LHS. + (!(memphi.region() instanceof LoopNode) && profit(memphi,1)) ) { + Node ld1 = ld(1); + Node ld2 = ld(2); + return new PhiNode(_name,_declaredType,memphi.region(),ld1,ld2); + } + } + + return null; + } + + // Load a flavored zero from a New + private Node zero(NewNode nnn) { + TypeStruct ts = nnn._ptr._obj; + Type zero = ts._fields[ts.findAlias(_alias)]._type.makeZero(); + return castRO(new ConstantNode(zero).peephole()); + } + + private Node ld( int idx ) { + Node mem = mem(), ptr = ptr(); + return new LoadNode(_loc,_name,_alias,_declaredType,mem.in(idx),ptr instanceof PhiNode && ptr.in(0)==mem.in(0) ? ptr.in(idx) : ptr, off()).peephole(); + } + + private static boolean neverAlias( Node ptr1, Node ptr2 ) { + return ptr1.in(0) != ptr2.in(0) && + // Unrelated allocations + ptr1 instanceof ProjNode && ptr1.in(0) instanceof NewNode && + ptr2 instanceof ProjNode && ptr2.in(0) instanceof NewNode; + } + + private static boolean hoistPtr(Node ptr, PhiNode memphi ) { + // Can I hoist ptr above the Region? + if( !(memphi.region() instanceof RegionNode r) ) + return false; // Dead or dying Region/Phi + // If ptr from same Region, then yes, just use hoisted split pointers + if( ptr instanceof PhiNode pphi && pphi.region() == r ) + return true; + + // No, so can we lift this ptr? + CFGNode cptr = ptr.cfg0(); + if( cptr != null ) + // Pointer is controlled high + // TODO: Really needs to be the LCA of all inputs is high + return cptr.idepth() <= r.idepth(); + + // Dunno without a longer walk + return false; + } + + // Profitable if we find a matching Store on this Phi arm. + private boolean profit(PhiNode phi, int idx) { + Node px = phi.in(idx); + if( px==null ) return false; + if( px._type instanceof TypeMem mem && mem._t.isHighOrConst() ) return true; + if( px instanceof StoreNode st1 && ptr()==st1.ptr() && off()==st1.off() ) return true; + px.addDep(this); + return false; + } + + // Read-Only is a deep property, and cannot be cast-away + private Node castRO(Node rez) { + if( ptr()._type.isFinal() && !rez._type.isFinal() ) + return new ReadOnlyNode(rez).peephole(); + return rez; + } + + // When a load bypasses a store, the store might truncate bits - and the + // load will need to zero/sign-extend. + private Node extend(Node val) { + if( !(_declaredType instanceof TypeInteger ti) ) return val; + if( ti._min==0 ) // Unsigned + return new AndNode(null,val,con(ti._max)); + // Signed extension + int shift = Long.numberOfLeadingZeros(ti._max)-1; + Node shf = con(shift); + if( shf._type==TypeInteger.ZERO ) + return val; + //return peep(new SarNode(null,peep(new ShlNode(null,val,shf.keep())),shf.unkeep())); + throw Utils.TODO(); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/LogicalNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/LogicalNode.java new file mode 100644 index 00000000..ccc1f6b5 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/LogicalNode.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public abstract class LogicalNode extends Node { + // Source location for late reported errors + Parser.Lexer _loc; + + public LogicalNode(Parser.Lexer loc, Node lhs, Node rhs) { super(null, lhs, rhs); _loc = loc; } + abstract String op(); + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append(op()), visited); + return sb.append(")"); + } + + + @Override public Parser.ParseException err() { + if( in(1)._type.isHigh() || in(2)._type.isHigh() ) return null; + if( !(in(1)._type instanceof TypeInteger) ) return Parser.error("Cannot '"+op()+"' " + in(1)._type.glb(),_loc); + if( !(in(2)._type instanceof TypeInteger) ) return Parser.error("Cannot '"+op()+"' " + in(2)._type.glb(), _loc); + return null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/LoopNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/LoopNode.java new file mode 100644 index 00000000..d2b0e36c --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/LoopNode.java @@ -0,0 +1,86 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeMem; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; +import java.util.HashSet; + +public class LoopNode extends RegionNode { + public LoopNode( Parser.Lexer loc, Node entry ) { super(loc,null,entry,null); } + public LoopNode( LoopNode loop ) { super(loop); } + + public CFGNode entry() { return cfg(1); } + public CFGNode back () { return cfg(2); } + + @Override + public String label() { return "Loop"; } + + @Override + public Type compute() { + if( inProgress() ) return Type.CONTROL; + return entry()._type; + } + + // Bypass Region idom, same as the default idom() using use in(1) instead of in(0) + @Override public int idepth() { return _idepth==0 ? (_idepth=idom().idepth()+1) : _idepth; } + // Bypass Region idom, same as the default idom() using use in(1) instead of in(0) + @Override public CFGNode idom(Node dep) { return entry(); } + + // If this is an unreachable loop, it may not have an exit. If it does not + // (i.e., infinite loop), force an exit to make it reachable. + public StopNode forceExit( FunNode fun, StopNode stop ) { + // Walk the backedge, then immediate dominator tree util we hit this + // Loop again. If we ever hit a CProj from an If (as opposed to + // directly on the If) we found our exit. + CFGNode x = back(); + while( x != this ) { + if( x instanceof CProjNode exit && exit.in(0) instanceof IfNode iff ) { + CFGNode other = iff.cproj(1-exit._idx); + if( other!=null && other.loopDepth() < loopDepth() ) + return stop; // Found an exit, not an infinite loop + } + x = x.idom(); + } + // Found a no-exit loop. Insert an exit + NeverNode iff = new NeverNode(back()); + for( Node use : _outputs ) + if( use instanceof PhiNode ) + iff.addDef(use); + CProjNode t = new CProjNode(iff,0,"True" ).init(); + CProjNode f = new CProjNode(iff,1,"False").init(); + setDef(2,f); + + // Now fold control into the exit. Might have 1 valid exit, or an + // XCtrl or a bunch of prior NeverNode exits. + Node top = new ConstantNode(Type.TOP).peephole(); + ReturnNode ret = fun.ret(); + Node ctrl = ret.ctrl(), mem = ret.mem(), expr = ret.expr(); + if( ctrl._type != Type.XCONTROL ) { + // Perfect aligned exit? + if( !(ctrl instanceof RegionNode r && + mem instanceof PhiNode pmem && pmem.region()==r && + expr instanceof PhiNode prez && prez.region()==r ) ) { + // Nope, insert an aligned exit layer + ctrl = new RegionNode(_loc,null,ctrl).init(); + mem = new PhiNode((RegionNode)ctrl,mem ).init(); + expr = new PhiNode((RegionNode)ctrl,expr).init(); + } + // Append new Never exit + ctrl.addDef(t ); + mem .addDef(top); + expr.addDef(top); + } else { + ctrl = t; + mem = top; + expr = top; + } + ret.setDef(0,ctrl); + ret.setDef(1,mem ); + ret.setDef(2,expr); + + return stop; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MachConcreteNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MachConcreteNode.java new file mode 100644 index 00000000..0c49fc64 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MachConcreteNode.java @@ -0,0 +1,27 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.Utils; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; + +// Generic machine-specific class, has a few Node implementations that have to +// exist (abstract) but are not useful past the optimizer. +public abstract class MachConcreteNode extends Node implements MachNode{ + + public MachConcreteNode(Node node) { super(node); } + public MachConcreteNode(Node[]nodes) { super(nodes); } + + @Override public String label() { return op(); } + @Override public Type compute () { throw Utils.TODO(); } + @Override public Node idealize() { throw Utils.TODO(); } + + @Override public StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("(").append(op()).append(","); + for( int i=1; i<nIns(); i++ ) + (in(i)==null ? sb.append(" ---") : in(i)._print0(sb, visited)).append(","); + sb.setLength(sb.length()-1); + return sb.append(")"); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MachNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MachNode.java new file mode 100644 index 00000000..9f2fee85 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MachNode.java @@ -0,0 +1,49 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import java.io.ByteArrayOutputStream; + +public interface MachNode { + + // Easy access to the abstract Machine + default Machine machine() { return CodeGen.CODE._mach; } + + // Run a post-instruction-selection action + default void postSelect() { } + + // Register mask allowed on input i. 0 for no register. + abstract public RegMask regmap(int i); + // Register mask allowed as a result. 0 for no register. + abstract public RegMask outregmap(); + // Multi-reg-defining machine op; idx comes from Proj. + // Sometimes these Projs have no uses, and just exist to kill a register. + default RegMask outregmap(int idx) { throw Utils.TODO(); } + + // If this op updates the same register as input 'idx', e.g. many X86 "R1 += + // R2" style ops. This forces the output live range to equal the input live + // range, so they both get the same register. + + // This can hold true for projections, if the instruction updates many + // registers that it also takes in. Note that ops also use the same + // registers for both inputs and outputs but these are pinned and without any + // choices (e.g. always RCX for both an input and output) and are better + // served with just declaring as a fixed register. + + // Returns either 0 for not-two-adr or the updated input; for projections + // this is their Multi's updated input. + default int twoAddress( ) { return 0; } + + // Instructions cheaper to recreate than to spill, such as loading small constants + default boolean isClone() { return false; } + // Encoding is appended into the byte array; size is returned + abstract public int encoding(ByteArrayOutputStream bytes); + + // Human-readable form appended to the SB. Things like the encoding, + // indentation, leading address or block labels not printed here. + // Just something like "R17=[R18+12] // Load array base". + // General form: "dst=src+src". + default void asm(CodeGen code, SB sb) { throw Utils.TODO(); } + + // Human readable opcode; e.g. "ld4" or "call". + default String op() { throw Utils.TODO(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java new file mode 100644 index 00000000..5363bb6f --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java @@ -0,0 +1,154 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.*; +import java.util.*; + +/** + * Memory Merge - a merge of many aliases into a "fat memory". All aliases + * are here, but most will be lazy - take the default fat memory. + */ +public class MemMergeNode extends Node { + + /* + * In-Progress means this is being used by the Parser to track memory + * aliases. No optimizations are allowed. When no longer "in progress" + * normal peeps work. + */ + public final boolean _inProgress; + + public MemMergeNode( boolean inProgress) { _type = TypeMem.BOT; _inProgress = inProgress; } + public MemMergeNode(MemMergeNode mem) { super(mem); _inProgress = false; } + + + // If being used by a Scope, this is "in progress" from the Parser. + // Otherwise, it's a memory merge + boolean inProgress() { return _inProgress; } + + @Override public String label() { return "ALLMEM"; } + @Override public boolean isMem() { return true; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append(_inProgress ? "Merge[" : "MEM[ "); + for( int j=2; j<nIns(); j++ ) { + sb.append(j); + sb.append(":"); + Node n = in(j); + while( n instanceof ScopeNode loop ) { + sb.append("Lazy_"); + n = loop.mem(j); + } + if( n==null ) sb.append("___ "); + else n._print0(sb, visited).append(" "); + } + sb.setLength(sb.length()-1); + return sb.append("]"); + } + + // Make a memory merge: no longer a Scope really, tracking memory state but + // not related to the parser in any way. + public Node merge() { + // Force default memory to not be lazy + MemMergeNode merge = new MemMergeNode(false); + for( Node n : _inputs ) + merge.addDef(n); + merge._mem(1,null); + return merge.peephole(); + } + + + @Override public Type compute() { return TypeMem.BOT; } + + @Override public Node idealize() { + if( inProgress() ) return null; + + // If not merging any memory (all memory is just the default) + if( nIns()==2 ) + return in(1); // Become default memory + + return null; + } + + public Node in( Var v ) { return in(v._idx); } + + public Node alias( int alias ) { + return in(alias<nIns() && in(alias)!=null ? alias : 1); + } + + Node alias( int alias, Node st ) { + while( alias >= nIns() ) addDef(null); + return setDef(alias,st); + } + + // Read or update from memory. + // A shared implementation allows us to create lazy phis both during + // lookups and updates; the lazy phi creation is part of chapter 8. + Node _mem( int alias, Node st ) { + // Memory projections are made lazily; if one does not exist + // then it must be START.proj(1) + Node old = alias(alias); + if( old instanceof ScopeNode loop ) { + MemMergeNode loopmem = loop.mem(); + Node memdef = loopmem.alias(alias); + // Lazy phi! + old = memdef instanceof PhiNode phi && loop.ctrl()==phi.region() + // Loop already has a real Phi, use it + ? memdef + // Set real Phi in the loop head + // The phi takes its one input (no backedge yet) from a recursive + // lookup, which might have insert a Phi in every loop nest. + : loopmem.alias(alias, new PhiNode(Parser.memName(alias), TypeMem.BOT,loop.ctrl(),loopmem._mem(alias,null),null).peephole() ); + alias(alias,old); + } + // Memory projections are made lazily; expand as needed + return st==null ? old : alias(alias,st); // Not lazy, so this is the answer + } + + + void _merge( MemMergeNode that, RegionNode r) { + int len = Math.max(nIns(),that.nIns()); + for( int i = 2; i < len; i++) + if( alias(i) != that.alias(i) ) { // No need for redundant Phis + // If we are in lazy phi mode we need to a lookup + // by alias as it will trigger a phi creation + Node lhs = this._mem(i,null); + Node rhs = that._mem(i,null); + alias(i, new PhiNode(Parser.memName(i), lhs._type.glb().meet(rhs._type.glb()), r, lhs, rhs).peephole()); + } + } + + // Fill in the backedge of any inserted Phis + void _endLoopMem( ScopeNode scope, MemMergeNode back, MemMergeNode exit ) { + Node exit_def = exit.alias(1); + for( int i=1; i<nIns(); i++ ) { + if( in(i) instanceof PhiNode phi && phi.region()==scope.ctrl() ) { + assert phi.in(2)==null; + phi.setDef(2,back.in(i)==scope ? phi : back.in(i)); // Fill backedge + } + if( exit_def == scope ) // Replace a lazy-phi on the exit path also + exit.alias(i,in(i)); + } + } + + // Now one-time do a useless-phi removal + void _useless( ) { + for( int i=2; i<nIns(); i++ ) { + if( in(i) instanceof PhiNode phi ) { + // Do an eager useless-phi removal + Node in = phi.peephole(); + CodeGen.CODE.addAll(phi._outputs); + phi.moveDepsToWorklist(); + if( in != phi ) { + if( !phi.iskeep() ) // Keeping phi around for parser elsewhere + phi.subsume(in); + setDef(i,in); // Set the update back into Scope + } + } + } + } + + @Override boolean eq( Node n ) { + return this==n || !_inProgress; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MemOpNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MemOpNode.java new file mode 100644 index 00000000..d9ab6b89 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MemOpNode.java @@ -0,0 +1,74 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeMemPtr; +import java.lang.StringBuilder; +import java.util.BitSet; + +/** + * Convenience common base for Load and Store. + */ +public abstract class MemOpNode extends Node { + + public final String _name; + public final int _alias; + // Declared type; not final because it might be a forward-reference + // which will be lazily improved when the reference is declared. + public Type _declaredType; + // Source location for late reported errors + public final Parser.Lexer _loc; + + public MemOpNode(Parser.Lexer loc, String name, int alias, Type glb, Node mem, Node ptr, Node off) { + super(null, mem, ptr, off); + _name = name; + _alias = alias; + _declaredType = glb; + _loc = loc; + } + public MemOpNode(Parser.Lexer loc, String name, int alias, Type glb, Node mem, Node ptr, Node off, Node value) { + this(loc,name, alias, glb, mem, ptr, off); + addDef(value); + } + public MemOpNode( Node mach, MemOpNode mop ) { + super(mach); + _name = mop._name; + _alias = mop._alias; + _declaredType = mop._declaredType; + _loc = mop._loc; + } + + // + static String mlabel(String name) { return name.equals("[]") ? "ary" : (name.equals("#") ? "len" : name); } + String mlabel() { return mlabel(_name); } + + public Node mem() { return in(1); } + public Node ptr() { return in(2); } + public Node off() { return in(3); } + + @Override StringBuilder _print1( StringBuilder sb, BitSet visited ) { return _printMach(sb,visited); } + public StringBuilder _printMach( StringBuilder sb, BitSet visited ) { throw Utils.TODO(); } + + + @Override + boolean eq(Node n) { + MemOpNode mem = (MemOpNode)n; // Invariant + return _alias==mem._alias; // When comparing types error to use "equals"; always use "==" + } + + @Override + int hash() { return _alias; } + + @Override + public Parser.ParseException err() { + Type ptr = ptr()._type; + // Already an error, but better error messages come from elsewhere + if( ptr == Type.BOTTOM ) return null; + if( ptr.isHigh() ) return null; // Assume it will fall to not-null + // Better be a not-nil TMP + if( ptr instanceof TypeMemPtr tmp && tmp.notNull() ) + return null; + return Parser.error( "Might be null accessing '" + _name + "'",_loc); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MinusFNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MinusFNode.java new file mode 100644 index 00000000..2b749ba3 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MinusFNode.java @@ -0,0 +1,36 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class MinusFNode extends Node { + public MinusFNode(Node in) { super(null, in); } + + @Override public String label() { return "MinusF"; } + + @Override public String glabel() { return "-"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("(-"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0) + return i0.isConstant() ? TypeFloat.constant(-i0.value()) : i0; + return TypeFloat.F64; + } + + @Override + public Node idealize() { + // -(-x) is x + if( in(1) instanceof MinusFNode minus ) + return minus.in(1); + + return null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MinusNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MinusNode.java new file mode 100644 index 00000000..dd2cbd26 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MinusNode.java @@ -0,0 +1,40 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import com.seaofnodes.simple.Utils; + +import java.util.BitSet; + +public class MinusNode extends Node { + public MinusNode(Node in) { super(null, in); } + + @Override public String label() { return "Minus"; } + + @Override public String glabel() { return "-"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("(-"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if( in(1)._type.isHigh() ) return TypeInteger.TOP; + if( in(1)._type instanceof TypeInteger i0 && + !(i0==TypeInteger.BOT || i0._min == Long.MIN_VALUE || i0._max == Long.MIN_VALUE) ) + return TypeInteger.make(-i0._max,-i0._min); + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + // -(-x) is x + if( in(1) instanceof MinusNode minus ) + return minus.in(1); + + return null; + } + @Override Node copyF() { return new MinusFNode(null); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MulFNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MulFNode.java new file mode 100644 index 00000000..2d44a9fb --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MulFNode.java @@ -0,0 +1,51 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class MulFNode extends Node { + public MulFNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "MulF"; } + + @Override public String glabel() { return "*"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("*"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0 && + in(2)._type instanceof TypeFloat i1) { + if (i0.isConstant() && i1.isConstant()) + return TypeFloat.constant(i0.value()*i1.value()); + } + return in(1)._type.meet(in(2)._type); + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // Mul of 1. We do not check for (1*x) because this will already + // canonicalize to (x*1) + if ( t2.isConstant() && t2 instanceof TypeFloat i && i.value()==1 ) + return lhs; + + // Move constants to RHS: con*arg becomes arg*con + if ( t1.isConstant() && !t2.isConstant() ) + return swap12(); + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new MulFNode(lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MulNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MulNode.java new file mode 100644 index 00000000..4fdf987d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MulNode.java @@ -0,0 +1,81 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class MulNode extends Node { + public MulNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "Mul"; } + + @Override public String glabel() { return "*"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("*"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i1 && + t2 instanceof TypeInteger i2 ) { + if( i1==TypeInteger.ZERO || i2==TypeInteger.ZERO) + return TypeInteger.ZERO; + if (i1.isConstant() && i2.isConstant()) + return TypeInteger.constant(i1.value()*i2.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // Move constants to RHS: con*arg becomes arg*con + if ( t1.isConstant() && !t2.isConstant() ) + return swap12(); + + // Multiply by constant + if ( t2.isConstant() && t2 instanceof TypeInteger i ) { + // Mul of 1. We do not check for (1*x) because this will already + // canonicalize to (x*1) + long c = i.value(); + if( c==1 ) return lhs; + if( c==0 ) return Parser.ZERO; + // Mul by a power of 2, +/-1. Bit patterns more complex than this + // are unlikely to win on an X86 vs the normal "imul", and so + // become machine-specific. + if( (c & (c-1)) == 0 ) + return new ShlNode(null,lhs,con(Long.numberOfTrailingZeros(c))); + // 2**n + 1, e.g. 9 + long d = c-1; + if( (d & (d-1)) == 0 ) + return new AddNode(new ShlNode(null,lhs,con(Long.numberOfTrailingZeros(d))).peephole(),lhs); + // 2**n - 1, e.g. 31 + long e = c+1; + if( (e & (e-1)) == 0 ) + return new SubNode(new ShlNode(null,lhs,con(Long.numberOfTrailingZeros(e))).peephole(),lhs); + + } + + // Do we have ((x * (phi cons)) * con) ? + // Do we have ((x * (phi cons)) * (phi cons)) ? + // Push constant up through the phi: x * (phi con0*con0 con1*con1...) + Node phicon = AddNode.phiCon(this,true); + if( phicon!=null ) return phicon; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new MulNode(lhs,rhs); } + @Override Node copyF() { return new MulFNode(null,null); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/MultiNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/MultiNode.java new file mode 100644 index 00000000..17d00881 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/MultiNode.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Ary; + + +public interface MultiNode { + + abstract Ary<Node> outs(); + + // Find a projection by index + default ProjNode proj( int idx ) { + for( Node out : outs() ) + if( out instanceof ProjNode prj && prj._idx==idx ) + return prj; + return null; + } + + // Find a projection by index + default CProjNode cproj( int idx ) { + for( Node out : outs() ) + if( out instanceof CProjNode prj && prj._idx==idx ) + return prj; + return null; + } + + // Return not-null if this projection index is a ideal copy. + // Called by ProjNode ideal and used to collapse Multis. + default Node pcopy(int idx) { return null; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/NeverNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/NeverNode.java new file mode 100644 index 00000000..e77ed653 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/NeverNode.java @@ -0,0 +1,21 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.TypeInteger; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeTuple; + +import java.util.BitSet; + +// "Never true" for infinite loop exits +public class NeverNode extends IfNode { + public NeverNode(Node ctrl) { super(ctrl,Parser.ZERO); } + + @Override public String label() { return "Never"; } + + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { return sb.append("Never"); } + + @Override public Type compute() { return TypeTuple.IF_BOTH; } + + @Override public Node idealize() { return null; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/NewNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/NewNode.java new file mode 100644 index 00000000..80545d66 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/NewNode.java @@ -0,0 +1,83 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.*; +import com.seaofnodes.simple.Utils; +import java.util.BitSet; + +/** + * Allocation! Allocate a chunk of memory, and pre-zero it. + * The inputs include control and size, and ALL aliases being set. + * The output is large tuple, one for every alias plus the created pointer. + * New is expected to be followed by projections for every alias. + */ +public class NewNode extends Node implements MultiNode { + + public final TypeMemPtr _ptr; + public final int _len; + + public NewNode(TypeMemPtr ptr, Node... nodes) { + super(nodes); + assert !ptr.nullable(); + _ptr = ptr; + _len = ptr._obj._fields.length; + // Control in slot 0 + assert nodes[0]._type==Type.CONTROL || nodes[0]._type == Type.XCONTROL; + // Malloc-length in slot 1 + assert nodes[1]._type instanceof TypeInteger || nodes[1]._type==Type.NIL; + for( int i=0; i<_len; i++ ) + // Memory slices for all fields. + assert nodes[2+i]._type.isa(TypeMem.BOT); + } + + public NewNode(NewNode nnn) { super(nnn); _ptr = nnn._ptr; _len = nnn._len; } + + public Node mem (int idx) { return in(idx+2); } + + @Override public String label() { return "new_"+glabel(); } + @Override public String glabel() { + return _ptr._obj.isAry() ? "ary_"+_ptr._obj._fields[1]._type.str() : _ptr._obj.str(); + } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("new "); + return sb.append(_ptr._obj.str()); + } + + int findAlias(int alias) { + int idx = _ptr._obj.findAlias(alias); + assert idx!= -1; // Error, caller should be calling + return idx+2; // Skip ctrl, size + } + + // 0 - ctrl + // 1 - byte size + // 2-len+2 - aliases, one per field + // len+2 - 2*len+2 - initial values, one per field + public Node size() { return in(1); } + + @Override + public TypeTuple compute() { + Field[] fs = _ptr._obj._fields; + Type[] ts = new Type[fs.length+2]; + ts[0] = Type.CONTROL; + ts[1] = _ptr; + for( int i=0; i<fs.length; i++ ) { + Type mt = in(i+2)._type; + TypeMem mem = mt==Type.TOP ? TypeMem.TOP : (TypeMem)mt; + Type tfld = mem._t.meet(mem._t.makeZero()); + Type tfld2 = tfld.join(fs[i]._type); + ts[i+2] = TypeMem.make(fs[i]._alias,tfld2); + } + return TypeTuple.make(ts); + } + + @Override + public Node idealize() { return null; } + + @Override + boolean eq(Node n) { return this == n; } + + @Override + int hash() { return _ptr.hashCode(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/Node.java b/chapter20/src/main/java/com/seaofnodes/simple/node/Node.java new file mode 100644 index 00000000..bbf1ebe9 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/Node.java @@ -0,0 +1,640 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.*; +import java.util.function.Function; +import static com.seaofnodes.simple.CodeGen.CODE; + +/** + * All Nodes in the Sea of Nodes IR inherit from the Node class. + * The Node class provides common functionality used by all subtypes. + * Subtypes of Node specialize by overriding methods. + */ +public abstract class Node { + + /** + * Each node has a unique dense Node ID within a compilation context + * The ID is useful for debugging, for using as an offset in a bitvector, + * as well as for computing equality of nodes (to be implemented later). + */ + public final int _nid; + + /** + * Inputs to the node. These are use-def references to Nodes. + * <p> + * Generally fixed length, ordered, nulls allowed, no unused trailing space. + * Ordering is required because e.g. "a/b" is different from "b/a". + * The first input (offset 0) is often a {@link CFGNode} node. + */ + public final Ary<Node> _inputs; + + /** + * Outputs reference Nodes that are not null and have this Node as an + * input. These nodes are users of this node, thus these are def-use + * references to Nodes. + * <p> + * Outputs directly match inputs, making a directed graph that can be + * walked in either direction. These outputs are typically used for + * efficient optimizations but otherwise have no semantics meaning. + */ + public final Ary<Node> _outputs; + + + /** + * Current computed type for this Node. This value changes as the graph + * changes and more knowledge is gained about the program. + */ + public Type _type; + + + Node(Node... inputs) { + _nid = CODE.getUID(); // allocate unique dense ID + _inputs = new Ary<>(Node.class); + Collections.addAll(_inputs,inputs); + _outputs = new Ary<>(Node.class); + for( Node n : _inputs ) + if( n != null ) + n.addUse( this ); + } + + // Make a Node using the existing arrays of nodes. + // Used by any pass rewriting all Node classes but not the edges. + Node( Node n ) { + assert CodeGen.CODE._phase == CodeGen.Phase.InstructionSelection; + _nid = CODE.getUID(); // allocate unique dense ID + _inputs = new Ary<>(n._inputs.asAry()); + _outputs = new Ary<>(Node.class); + _type = n._type; + _deps = null; + _hash = 0; + } + + // Easy reading label for debugger, e.g. "Add" or "Region" or "EQ" + public abstract String label(); + + // Unique label for graph visualization, e.g. "Add12" or "Region30" or "EQ99" + public String uniqueName() { + // Get rid of $ as graphviz doesn't like it + String label = label().replaceAll("\\$", ""); + return label + _nid; + } + + // Graphical label, e.g. "+" or "Region" or "==" + public String glabel() { return label(); } + + // Extra fun stuff, for assembly printing. Jump labels, parser locations, + // variable types, etc. + public String comment() { return null; } + + // ------------------------------------------------------------------------ + + // Debugger Printing. + + // {@code toString} is what you get in the debugger. It has to print 1 + // line (because this is what a debugger typically displays by default) and + // has to be robust with broken graph/nodes. + @Override + public final String toString() { return print(); } + + // This is a *deep* print. We print with a mutually recursive tik-tok + // style; the common _print0 calls the per-Node _print1, which calls back + // to _print0; + public final String print() { + return _print0(new StringBuilder(), new BitSet()).toString(); + } + + // This is the common print: check for repeats, check for DEAD and print + // "DEAD" else call the per-Node print1. + final StringBuilder _print0(StringBuilder sb, BitSet visited) { + if (visited.get(_nid) && !(this instanceof ConstantNode) ) + return sb.append(label()); + visited.set(_nid); + return isDead() + ? sb.append(uniqueName()).append(":DEAD") + : _print1(sb, visited); + } + // Every Node implements this; a partial-line recursive print + abstract StringBuilder _print1(StringBuilder sb, BitSet visited); + + public String p(int depth) { return IRPrinter.prettyPrint(this,depth); } + + public boolean isMultiHead() { return false; } + public boolean isMultiTail() { return false; } + public boolean isConst () { return false; } + + // ------------------------------------------------------------------------ + // Graph Node & Edge manipulation + + /** + * Gets the ith input node + * @param i Offset of the input node + * @return Input node or null + */ + public Node in(int i) { return _inputs.get(i); } + public Node out(int i) { return _outputs.get(i); } + public final Ary<Node> outs() { return _outputs; } + + public int nIns() { return _inputs.size(); } + + public int nOuts() { return _outputs.size(); } + + public boolean isUnused() { return nOuts() == 0; } + + public CFGNode cfg0() { return (CFGNode)in(0); } + + /** + * Change a <em>def</em> into a Node. Keeps the edges correct, by removing + * the corresponding <em>use->def</em> edge. This may make the original + * <em>def</em> go dead. This function is co-recursive with {@link #kill}. + * <p> + + * This method is the normal path for altering a Node, because it does the + * proper default edge maintenance. It also <em>immediately</em> kills + * Nodes that lose their last use; at times care must be taken to avoid + * killing Nodes that are being used without having an output Node. This + * definitely happens in the middle of recursive {@link #peephole} calls. + * + * @param idx which def to set + * @param new_def the new definition + * @return new_def for flow coding + */ + public <N extends Node> N setDef(int idx, N new_def ) { + unlock(); + Node old_def = in(idx); + if( old_def == new_def ) return new_def; // No change + // If new def is not null, add the corresponding def->use edge + // This needs to happen before removing the old node's def->use edge as + // the new_def might get killed if the old node kills it recursively. + if( new_def != null ) + new_def.addUse(this); + // Set the new_def over the old (killed) edge + _inputs.set(idx,new_def); + if( old_def != null ) { // If the old def exists, remove a def->use edge + if( old_def.delUse(this) ) // If we removed the last use, the old def is now dead + old_def.kill(); // Kill old def + else CODE.add(old_def); // Else old lost a use, so onto worklist + } + moveDepsToWorklist(); + // Return new_def for easy flow-coding + return new_def; + } + + // Remove the numbered input, compressing the inputs in-place. This + // shuffles the order deterministically - which is suitable for Region and + // Phi, but not for every Node. If the def goes dead, it is recursively + // killed, which may include 'this' Node. + Node delDef(int idx) { + unlock(); + Node old_def = in(idx); + _inputs.del(idx); + if( old_def.delUse(this) ) // If we removed the last use, the old def is now dead + old_def.kill(); // Kill old def + old_def.moveDepsToWorklist(); + return this; + } + + // Insert the numbered input, sliding other inputs to the right + Node insertDef(int idx, Node new_def) { + _inputs.add(idx,null); + return setDef(idx,new_def); + } + + + /** + * Add a new def to an existing Node. Keep the edges correct by + * adding the corresponding <em>def->use</em> edge. + * + * @param new_def the new definition, appended to the end of existing definitions + * @return new_def for flow coding + */ + public Node addDef(Node new_def) { + unlock(); + // Add use->def edge + _inputs.add(new_def); + // If new def is not null, add the corresponding def->use edge + if( new_def != null ) + new_def.addUse(this); + return new_def; + } + + // Breaks the edge invariants, used temporarily + protected <N extends Node> N addUse(Node n) { _outputs.add(n); return (N)this; } + + // Remove node 'use' from 'def's (i.e. our) output list, by compressing the list in-place. + // Return true if the output list is empty afterward. + // Error is 'use' does not exist; ok for 'use' to be null. + protected boolean delUse( Node use ) { + _outputs.del(_outputs.find(use)); + return _outputs.isEmpty(); + } + + // Shortcut for "popping" until n nodes. A "pop" is basically a + // setDef(last,null) followed by lowering the nIns() count. + void popUntil(int n) { + unlock(); + while( nIns() > n ) { + Node old_def = _inputs.pop(); + if( old_def != null && // If it exists and + old_def.delUse(this) ) // If we removed the last use, the old def is now dead + old_def.kill(); // Kill old def + } + } + + /** + * Kill a Node with no <em>uses</em>, by setting all of its <em>defs</em> + * to null. This may recursively kill more Nodes, and is basically dead + * code elimination. + */ + public void kill( ) { + unlock(); + moveDepsToWorklist(); + assert isUnused(); // Has no uses, so it is dead + _type=null; // Flag as dead + while( nIns()>0 ) { // Set all inputs to null, recursively killing unused Nodes + Node old_def = _inputs.removeLast(); + // Revisit neighbor because removed use + if( old_def != null && CODE.add(old_def).delUse(this) ) + old_def.kill(); // If we removed the last use, the old def is now dead + } + assert isDead(); // Really dead now + } + + // Mostly used for asserts and printing. + public boolean isDead() { return isUnused() && nIns()==0 && _type==null; } + + // Shortcuts to stop DCE mid-parse + // Add bogus null use to keep node alive + public <N extends Node> N keep() { return addUse(null); } + // Remove bogus null. + public <N extends Node> N unkeep() { + delUse(null); + return (N)this; + } + // Test "keep" status + public boolean iskeep() { return _outputs.find(null) != -1; } + public void unkill() { + if( unkeep().isUnused() ) + kill(); + } + + + // Replace self with nnn in the graph, making 'this' go dead + public void subsume( Node nnn ) { + assert nnn!=this; + while( nOuts() > 0 ) { + Node n = _outputs.removeLast(); + n.unlock(); + int idx = n._inputs.find(this); + n._inputs.set(idx,nnn); + nnn.addUse(n); + CODE.addAll(n._outputs); + } + kill(); + } + + // ------------------------------------------------------------------------ + // Graph-based optimizations + + /** + * Try to peephole at this node and return a better replacement Node. + * Always returns some not-null Node (often this). + */ + public final Node peephole( ) { + if( _type==null ) // Brand-new node, never peeped before + JSViewer.show(); + Node n = peepholeOpt(); + if( n!=null ) // Made progress? + JSViewer.show(); // Show again + return n==null ? this : deadCodeElim(n._nid >= _nid ? n.peephole() : n); // Cannot return null for no-progress + } + + /** + * Try to peephole at this node and return a better replacement Node if + * possible. We compute a {@link Type} and then check and replace: + * <ul> + * <li>if the Type {@link Type#isConstant}, we replace with a {@link ConstantNode}</li> + * <li>in a future chapter we will look for a + * <a href="https://en.wikipedia.org/wiki/Common_subexpression_elimination">Common Subexpression</a> + * to eliminate.</li> + * <li>we ask the Node for a better replacement. The "better replacement" + * is things like {@code (1+2)} becomes {@code 3} and {@code (1+(x+2))} becomes + * {@code (x+(1+2))}. By canonicalizing expressions we fold common addressing + * math constants, remove algebraic identities and generally simplify the + * code. </li> + * + * Unlike peephole above, this explicitly returns null for no-change, or not-null + * for a better replacement (which can be this). + * </ul> + */ + public final Node peepholeOpt( ) { + CODE.iterCnt(); + // Compute initial or improved Type + Type old = setType(compute()); + + // Replace constant computations from non-constants with a constant + // node. If peeps are disabled, still allow high Phis to collapse; + // they typically come from dead Regions, and we want the Region to + // collapse, which requires the Phis to die first. + if( _type.isHighOrConst() && !isConst() ) + return ConstantNode.make(_type).peephole(); + + // Global Value Numbering + if( _hash==0 ) { + Node n = CODE._gvn.get(this); // Will set _hash as a side effect + if( n==null ) + CODE._gvn.put(this,this); // Put in table now + else { + // Because of random worklist ordering, the two equal nodes + // might have different types. Because of monotonicity, both + // types are valid. To preserve monotonicity, the resulting + // shared Node has to have the best of both types. + n.setType(n._type.join(_type)); + _hash = 0; // Clear, since it never went in the table + return deadCodeElim(n);// Return previous; does Common Subexpression Elimination + } + } + + // Ask each node for a better replacement + Node n = idealize(); + if( n != null ) // Something changed + return n; // Report progress + + if( old!=_type ) return this; // Report progress; + CODE.iterNop(); + return null; // No progress + } + + // m is the new Node, self is the old. + // Return 'm', which may have zero uses but is alive nonetheless. + // If self has zero uses (and is not 'm'), {@link #kill} self. + private Node deadCodeElim(Node m) { + // If self is going dead and not being returned here (Nodes returned + // from peephole commonly have no uses (yet)), then kill self. + if( m != this && isUnused() && !isDead() ) { + // Killing self - and since self recursively kills self's inputs we + // might end up killing 'm', which we are returning as a live Node. + // So we add a bogus extra null output edge to stop kill(). + m.keep(); // Keep m alive + kill(); // Kill self because replacing with 'm' + m.unkeep(); // Okay to peephole m + } + return m; + } + + /** + * This function needs to be + * <a href="https://en.wikipedia.org/wiki/Monotonic_function">Monotonic</a> + * as it is part of a Monotone Analysis Framework. + * <a href="https://www.cse.psu.edu/~gxt29/teaching/cse597s21/slides/08monotoneFramework.pdf">See for example this set of slides</a>. + * <p> + * For Chapter 2, all our Types are really integer constants, and so all + * the needed properties are trivially true, and we can ignore the high + * theory. Much later on, this will become important and allow us to do + * many fancy complex optimizations trivially... because theory. + * <p> + * {@link #compute()} needs to be stand-alone, and cannot recursively call compute + * on its inputs, because programs are cyclic (have loops!) and this will just + * infinitely recurse until stack overflow. Instead, compute typically + * computes a new type from the {@link #_type} field of its inputs. + */ + public abstract Type compute(); + + // Set the type. Assert monotonic progress. + // If changing, add users to worklist. + public Type setType(Type type) { + Type old = _type; + assert old==null || type.isa(old); // Since _type not set, can just re-run this in assert in the debugger + if( old == type ) return old; + _type = type; // Set _type late for easier assert debugging + CODE.addAll(_outputs); + moveDepsToWorklist(); + return old; + } + + public <N extends Node> N init() { _type = compute(); return (N)this; } + + /** + * This function rewrites the current Node into a more "idealized" form. + * This is the bulk of our peephole rewrite rules, and we use this to + * e.g. turn arbitrary collections of adds and multiplies with mixed + * constants into a normal form that's easy for hardware to implement. + * Example: An array addressing expression: + * <pre> ary[idx+1]</pre> + * might turn into Sea-of-Nodes IR: + * <pre> (ary+12)+((idx+1) * 4)</pre> + * This expression can then be idealized into: + * <pre> ary + ((idx*4) + (12 + (1*4)))</pre> + * And more folding: + * <pre> ary + ((idx<<2) + 16)</pre> + * And during code-gen: + * <pre> MOV4 Rary,[Ridx<<2 +16] // or some such hardware-specific notation </pre> + * <p> + * {@link #idealize} has a very specific calling convention: + * <ul> + * <li>If NO change is made, return {@code null} + * <li>If ANY change is made, return not-null; this can be {@code this} + * <li>The returned Node does NOT call {@link #peephole} on itself; the {@link #peephole} call will recursively peephole it. + * <li>Any NEW nodes that are not directly returned DO call {@link #peephole}. + * </ul> + * <p> + * Examples: + * <table border="3"> + * <tr><th> before </th><th> after </th><th>return </th><th>comment </th></tr> + * <tr><td>{@code (x+5) }</td><td>{@code (x+5) }</td><td>{@code null }</td><td>No change</td></tr> + * <tr><td>{@code (5+x) }</td><td>{@code (x+5) }</td><td>{@code this }</td><td>Swapped arguments</td></tr> + * <tr><td>{@code ((x+1)+2)}</td><td>{@code (x+(1+2))}</td><td>{@code (x+_) }</td><td>Returns 2 new Nodes</td></tr> + * </table> + * + * The last entry deserves more discussion. The new Node {@code (1+2)} + * created in {@link #idealize} calls {@link #peephole} (which then folds + * into a constant). The other new Node {@code (x+3)} does not call + * peephole, because it is returned and peephole itself will recursively + * call peephole. + * <p> + * Since idealize calls peephole and peephole calls idealize, you must be + * careful that all idealizations are <em>monotonic</em>: all transforms + * remove some feature, so that the set of available transforms always + * shrinks. If you don't, you risk an infinite peephole loop! + * + * @return Either a new or changed node, or null for no changes. + */ + public abstract Node idealize(); + + + // Some of the peephole rules get complex, and search further afield than + // just the nearest neighbor. These peepholes can fail the pattern match + // on a node some distance away, and if that node ever changes we should + // retry the peephole. Track a set of Nodes dependent on `this`, and + // revisit them if `this` changes. + Ary<Node> _deps; + + /** + * Add a node to the list of dependencies. Only add it if its not an input + * or output of this node, that is, it is at least one step away. The node + * being added must benefit from this node being peepholed. + */ + Node addDep( Node dep ) { + // Running peepholes during the big assert cannot have side effects + // like adding dependencies. + if( CODE._midAssert ) return this; + if( dep == null ) return this; + if( _deps==null ) _deps = new Ary<>(Node.class); + if( _deps .find(dep) != -1 ) return this; // Already on list + if( _inputs .find(dep) != -1 ) return this; // No need for deps on immediate neighbors + if( _outputs.find(dep) != -1 ) return this; + _deps.add(dep); + return this; + } + + // Move the dependents onto a worklist, and clear for future dependents. + public void moveDepsToWorklist( ) { + if( _deps==null ) return; + CODE.addAll(_deps); + _deps.clear(); + } + + + // Two nodes are equal if they have the same inputs and the same "opcode" + // which means the same Java class, plus same internal parts. + @Override public final boolean equals( Object o ) { + if( o==this ) return true; + if( o.getClass() != getClass() ) return false; + Node n = (Node)o; + int len = _inputs.size(); + if( len != n._inputs.size() ) return false; + for( int i=0; i<len; i++ ) + if( in(i) != n.in(i) ) + return false; + return eq(n); + } + // Subclasses add extra checks (such as ConstantNodes have same constant), + // and can assume "this!=n" and has the same Java class. + boolean eq( Node n ) { return true; } + + + // Cached hash. If zero, then not computed AND this Node is NOT in the GVN + // table - and can have its edges hacked (which will change his hash + // anyway). If Non-Zero then this Node is IN the GVN table, or is being + // probed to see if it can be inserted. His edges are "locked", because + // hacking his edges will change his hash. + int _hash; + + // If the _hash is set, then the Node is in the GVN table; remove it. + void unlock() { + if( _hash==0 ) return; + Node old = CODE._gvn.remove(this); // Pull from table + assert old==this; + _hash=0; // Out of table now + } + + + // Hash of opcode and inputs + @Override public final int hashCode() { + if( _hash != 0 ) return _hash; + int hash = hash(); + for( Node n : _inputs ) + if( n != null ) + hash = hash ^ (hash<<17) ^ (hash>>13) ^ n._nid; + if( hash==0 ) hash = 0xDEADBEEF; // Bad hash, so use some junky thing + return (_hash=hash); + } + // Subclasses add extra hash info (such as ConstantNodes constant) + int hash() { return 0; } + + // ------------------------------------------------------------------------ + // + + /** Is this Node Memory related */ + public boolean isMem() { return false; } + + /** Pinned in the schedule; these are data nodes whose input#0 is not allowed to change */ + public boolean isPinned() { return false; } + + // Semantic change to the graph (so NOT a peephole), used by the Parser. + // If any input is a float, flip to a float-flavored opcode and widen any + // non-float input. + public final Node widen() { + if( !hasFloatInput() ) return this; + Node flt = copyF(); + if( flt==null ) return this; + for( int i=1; i<nIns(); i++ ) + flt.setDef(i, in(i)._type instanceof TypeFloat ? in(i) : new ToFloatNode(in(i)).peephole()); + kill(); + return flt; + } + private boolean hasFloatInput() { + for( int i=1; i<nIns(); i++ ) + if( in(i)._type instanceof TypeFloat ) + return true; + return false; + } + Node copyF() { return null; } + + + // ------------------------------------------------------------------------ + // Peephole utilities + + // Swap inputs without letting either input go dead during the swap. + Node swap12() { + unlock(); // Hash is order dependent + _inputs.swap(1,2); + return this; + } + + /** + * Does this node contain all constants? Ignores in(0), as is usually + * control. In an input is not a constant, we add dep as a dependency to + * it because dep can make progress if the input becomes a constant later. + * It is sufficient for one of the non-const inputs to have the dependency, + * so we don't bother checking the rest. + */ + boolean allCons(Node dep) { + for( int i=1; i<nIns(); i++ ) + if( !(in(i)._type.isConstant()) ) { + in(i).addDep(dep); // If in(i) becomes a constant later, will trigger some peephole + return false; + } + return true; + } + + + // Make a shallow copy (same class) of this Node, with given inputs and + // empty outputs and a new Node ID. The original inputs are ignored. + // Does not need to be implemented in isCFG() nodes. + Node copy(Node lhs, Node rhs) { throw Utils.TODO("Binary ops need to implement copy"); } + + // Report any post-optimize errors + public Parser.ParseException err() { return null; } + + // Common integer constants + public static ConstantNode con(long x) { return (ConstantNode)(new ConstantNode(TypeInteger.constant(x)).peephole()); } + + // Utility to walk the entire graph applying a function; return the first + // not-null result. + final public <E> E walk( Function<Node,E> pred ) { + assert CODE._visit.isEmpty(); + E rez = _walk(pred); + CODE._visit.clear(); + return rez; + } + + private <E> E _walk( Function<Node,E> pred ) { + if( CODE._visit.get(_nid) ) return null; // Been there, done that + CODE._visit.set(_nid); + E x = pred.apply(this); + if( x != null ) return x; + for( Node def : _inputs ) if( def != null && (x = def._walk(pred)) != null ) return x; + for( Node use : _outputs ) if( use != null && (x = use._walk(pred)) != null ) return x; + return null; + } + + /** + * Debugging utility to find a Node by index + */ + public Node find(int nid) { + return walk( n -> n._nid==nid ? n : null ); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/NotNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/NotNode.java new file mode 100644 index 00000000..b454de66 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/NotNode.java @@ -0,0 +1,32 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +public class NotNode extends Node { + public NotNode(Node in) { super(null, in); } + + @Override public String label() { return "Not"; } + + @Override public String glabel() { return "!"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("(!"), visited); + return sb.append(")"); + } + + @Override + public TypeInteger compute() { + Type t0 = in(1)._type; + if( t0.isHigh() ) return TypeInteger.BOOL.dual(); + if( t0 == Type.NIL || t0 == TypeInteger.ZERO ) return TypeInteger.TRUE; + if( t0 instanceof TypeNil tn && tn.notNull() ) return TypeInteger.FALSE; + if( t0 instanceof TypeInteger i && (i._min > 0 || i._max < 0) ) return TypeInteger.FALSE; + return TypeInteger.BOOL; + } + + @Override + public Node idealize() { return null; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/OrNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/OrNode.java new file mode 100644 index 00000000..9b03dc5d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/OrNode.java @@ -0,0 +1,55 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; + +import java.util.BitSet; + +public class OrNode extends LogicalNode { + public OrNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "Or"; } + @Override public String op() { return "|"; } + + @Override public String glabel() { return "|"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1 ) { + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()|i1.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // Or of 0. We do not check for (0|x) because this will already + // canonicalize to (x|0) + if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==0 ) + return lhs; + + // Move constants to RHS: con*arg becomes arg*con + if ( t1.isConstant() && !t2.isConstant() ) + return swap12(); + + // Do we have ((x | (phi cons)) | con) ? + // Do we have ((x | (phi cons)) | (phi cons)) ? + // Push constant up through the phi: x | (phi con0|con0 con1|con1...) + Node phicon = AddNode.phiCon(this,true); + if( phicon!=null ) return phicon; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new OrNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ParmNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ParmNode.java new file mode 100644 index 00000000..8d7c4205 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ParmNode.java @@ -0,0 +1,48 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import java.util.BitSet; + +public class ParmNode extends PhiNode { + + // Argument indices are mapped one-to-one on CallNode inputs + public final int _idx; // Argument index + + public ParmNode(String label, int idx, Type declaredType, Node... inputs) { + super(label,declaredType,inputs); + _idx = idx; + } + public ParmNode(ParmNode parm) { super(parm, parm._label, parm._declaredType); _idx = parm._idx; } + + @Override public String label() { return MemOpNode.mlabel(_label); } + + @Override public String glabel() { return _label; } + + public FunNode fun() { return (FunNode)in(0); } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + if( "main".equals(fun()._name) && _label.equals("arg") ) + return sb.append("arg"); + sb.append("Parm_").append(_label).append("("); + for( Node in : _inputs ) { + if (in == null) sb.append("____"); + else in._print0(sb, visited); + sb.append(","); + } + sb.setLength(sb.length()-1); + sb.append(")"); + return sb; + } + + + @Override + public Node idealize() { + if( inProgress() ) return null; + return super.idealize(); + } + + // Always in-progress until we run out of unknown callers + @Override public boolean inProgress() { return fun().inProgress(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/PhiNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/PhiNode.java new file mode 100644 index 00000000..17748afe --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/PhiNode.java @@ -0,0 +1,201 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +public class PhiNode extends Node { + + public final String _label; + + // The Phi type we compute must stay within the domain of the Phi. + // Example Int stays Int, Ptr stays Ptr, Control stays Control, Mem stays Mem. + final Type _declaredType; + + public PhiNode(String label, Type declaredType, Node... inputs) { super(inputs); _label = label; assert declaredType!=null; _declaredType = declaredType; } + public PhiNode(PhiNode phi, String label, Type declaredType) { super(phi); _label = label; _declaredType = declaredType; } + public PhiNode(PhiNode phi) { super(phi); _label = phi._label; _declaredType = phi._declaredType; } + + public PhiNode(RegionNode r, Node sample) { + super(new Node[]{r}); + _label = ""; + _declaredType = sample._type; + while( nIns() < r.nIns() ) + addDef(sample); + } + + @Override public String label() { return "Phi_"+MemOpNode.mlabel(_label); } + + @Override public String glabel() { return "φ_"+_label; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + if( !(region() instanceof RegionNode r) || r.inProgress() ) + sb.append("Z"); + sb.append("Phi("); + for( Node in : _inputs ) { + if (in == null) sb.append("____"); + else in._print0(sb, visited); + sb.append(","); + } + sb.setLength(sb.length()-1); + sb.append(")"); + return sb; + } + + public CFGNode region() { return (CFGNode)in(0); } + @Override public boolean isMultiTail() { return true; } + @Override public boolean isMem() { return _declaredType instanceof TypeMem; } + @Override public boolean isPinned() { return true; } + + @Override + public Type compute() { + if( !(region() instanceof RegionNode r) ) + return region()._type==Type.XCONTROL ? (_type instanceof TypeMem ? TypeMem.TOP : Type.TOP) : _type; + // During parsing Phis have to be computed type pessimistically. + if( r.inProgress() ) return _declaredType; + // Set type to local top of the starting type + Type t = _declaredType.glb().dual();//Type.TOP; + for (int i = 1; i < nIns(); i++) + // If the region's control input is live, add this as a dependency + // to the control because we can be peeped should it become dead. + if( r.in(i).addDep(this)._type != Type.XCONTROL ) + t = t.meet(in(i)._type); + return t; + } + + @Override + public Node idealize() { + if( !(region() instanceof RegionNode r ) ) + return in(1); // Input has collapse to e.g. starting control. + if( r.inProgress() || r.nIns()<=1 ) + return null; // Input is in-progress + + // If we have only a single unique input, become it. + Node live = singleUniqueInput(); + if (live != null) + return live; + + // No bother if region is going to fold dead paths soon + for( int i=1; i<nIns(); i++ ) + if( r.in(i)._type == Type.XCONTROL ) + return null; + + // Pull "down" a common data op. One less op in the world. One more + // Phi, but Phis do not make code. + // Phi(op(A,B),op(Q,R),op(X,Y)) becomes + // op(Phi(A,Q,X), Phi(B,R,Y)). + Node op = in(1); + if( !isMem() && op.nIns()==3 && op.in(0)==null && same_op() ) { + assert !(op instanceof CFGNode); + Node[] lhss = new Node[nIns()]; + Node[] rhss = new Node[nIns()]; + lhss[0] = rhss[0] = in(0); // Set Region + for( int i=1; i<nIns(); i++ ) { + lhss[i] = in(i).in(1); + rhss[i] = in(i).in(2); + } + Node phi_lhs = new PhiNode(_label, _declaredType,lhss).peephole(); + Node phi_rhs = new PhiNode(_label, _declaredType,rhss).peephole(); + Node down = op.copy(phi_lhs,phi_rhs); + // Test not running backwards, which can happen for e.g. And's + if( down.compute().isa(compute()) ) + return down; + down.kill(); + } + + // If merging Phi(N, cast(N)) - we are losing the cast JOIN effects, so just remove. + if( nIns()==3 ) { + if( in(1) instanceof CastNode cast && cast.in(1).addDep(this)==in(2) ) return in(2); + if( in(2) instanceof CastNode cast && cast.in(1).addDep(this)==in(1) ) return in(1); + } + // If merging a null-checked null and the checked value, just use the value. + // if( val ) ..; phi(Region,False=0/null,True=val); + // then replace with plain val. + if( nIns()==3 ) { + int nullx = -1; + if( in(1)._type == in(1)._type.makeZero() ) nullx = 1; + if( in(2)._type == in(2)._type.makeZero() ) nullx = 2; + if( nullx != -1 ) { + Node val = in(3-nullx); + if( val instanceof CastNode cast ) + val = cast.in(1); + if( r.idom(this).addDep(this) instanceof IfNode iff && iff.pred().addDep(this)==val ) { + // Must walk the idom on the null side to make sure we hit False. + CFGNode idom = (CFGNode)r.in(nullx); + while( idom != null && idom.nIns() > 0 && idom.in(0) != iff ) idom = idom.idom(); + if( idom instanceof CProjNode proj && proj._idx==1 ) + return val; + } + } + } + + return null; + } + + private boolean same_op() { + for( int i=2; i<nIns(); i++ ) + if( in(1).getClass() != in(i).getClass() ) + return false; + return true; + } + + /** + * If only single unique input, return it + */ + private Node singleUniqueInput() { + if( region() instanceof LoopNode loop && loop.entry()._type == Type.XCONTROL ) + return null; // Dead entry loops just ignore and let the loop collapse + Node live = null; + for( int i=1; i<nIns(); i++ ) { + // If the region's control input is live, add this as a dependency + // to the control because we can be peeped should it become dead. + if( region().in(i).addDep(this)._type != Type.XCONTROL && in(i) != this ) + if( live == null || live == in(i) ) live = in(i); + else return null; + } + return live; + } + + @Override + boolean allCons(Node dep) { + if( !(region() instanceof RegionNode r) ) return false; + // When the region completes (is no longer in progress) the Phi can + // become a "all constants" Phi, and the "dep" might make progress. + addDep(dep); + if( r.inProgress() ) return false; + return super.allCons(dep); + } + + // True if last input is null + public boolean inProgress() { + return in(nIns()-1) == null; + } + + // Never equal if inProgress + @Override boolean eq( Node n ) { + return !inProgress(); + } + + @Override + public Parser.ParseException err() { + if( _type != Type.BOTTOM ) return null; + + // BOTTOM means we mixed e.g. int and ptr + for( int i=1; i<nIns(); i++ ) + // Already an error, but better error messages come from elsewhere + if( in(i)._type == Type.BOTTOM ) + return null; + + // Gather a minimal set of types that "cover" all the rest + boolean ti=false, tf=false, tp=false, tn=false; + for( int i=1; i<nIns(); i++ ) { + Type t = in(i)._type; + ti |= t instanceof TypeInteger x; + tf |= t instanceof TypeFloat x; + tp |= t instanceof TypeMemPtr x; + tn |= t==Type.NIL; + } + return ReturnNode.mixerr(ti,tf,tp,tn, ((RegionNode)region())._loc); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ProjNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ProjNode.java new file mode 100644 index 00000000..ef7b17e4 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ProjNode.java @@ -0,0 +1,55 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeMem; +import com.seaofnodes.simple.type.TypeTuple; + +import java.util.BitSet; + +public class ProjNode extends Node { + + // Which slice of the incoming multipart value + public final int _idx; + + // Debugging label + public final String _label; + + public ProjNode(Node ctrl, int idx, String label) { + super(new Node[]{ctrl}); + _idx = idx; + _label = label; + } + public ProjNode(ProjNode p) { super(p); _idx = p._idx; _label = p._label; } + + @Override public String label() { return _label; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + if( _label != null ) return sb.append(_label); + if( in(0) instanceof CallEndNode cend && cend.call()!=null ) + return cend.call()._print0(sb,visited); + return sb.append("LONELY PROJ"); + } + + @Override public CFGNode cfg0() { + return in(0) instanceof CFGNode cfg ? cfg : in(0).cfg0(); + } + + @Override public boolean isMultiTail() { return in(0).isMultiHead(); } + @Override public boolean isMem() { return _type instanceof TypeMem; } + @Override public boolean isPinned() { return true; } + + @Override + public Type compute() { + Type t = in(0)._type; + return t instanceof TypeTuple tt ? tt._types[_idx] : Type.BOTTOM; + } + + @Override public Node idealize() { return ((MultiNode)in(0)).pcopy(_idx); } + + @Override + boolean eq( Node n ) { return _idx == ((ProjNode)n)._idx; } + + @Override + int hash() { return _idx; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ReadOnlyNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ReadOnlyNode.java new file mode 100644 index 00000000..b755af3f --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ReadOnlyNode.java @@ -0,0 +1,30 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeMemPtr; + +import java.util.BitSet; + +/** + * Cast a pointer to read-only + */ +public class ReadOnlyNode extends Node { + public ReadOnlyNode( Node n ) { super(null,n); } + public ReadOnlyNode( ReadOnlyNode n ) { super(n); } + @Override public String label() { return "ReadOnly"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return in(1)._print0(sb.append("(const)"),visited); + } + @Override public Type compute() { + Type t = in(1)._type; + return t instanceof TypeMemPtr tmp ? tmp.makeRO() : t; + } + + @Override public Node idealize() { + if( in(1)._type instanceof TypeMemPtr tmp && tmp.isFinal() ) + return in(1); + return null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/RegionNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/RegionNode.java new file mode 100644 index 00000000..5d3db5c6 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/RegionNode.java @@ -0,0 +1,126 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.Type; +import java.util.*; + +public class RegionNode extends CFGNode { + // Source location for late discovered errors + public Parser.Lexer _loc; + + public RegionNode(Parser.Lexer loc, Node... nodes) { super(nodes); _loc = loc; } + public RegionNode(RegionNode r, Parser.Lexer loc) { super(r); _loc = loc; } + public RegionNode(RegionNode r) { super(r); _loc = r._loc; } + + @Override + public String label() { return "Region"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return sb.append(label()); + } + + @Override public boolean isMultiHead() { return true; } + @Override public boolean blockHead() { return true; } + + @Override + public Type compute() { + if( inProgress() ) return Type.CONTROL; + Type t = Type.XCONTROL; + for (int i = 1; i < nIns(); i++) + t = t.meet(in(i)._type); + return t; + } + + @Override + public Node idealize() { + if( inProgress() ) return null; + + Node progress = deadPath(); + if( progress!=null ) return progress; + + // If down to a single input, become that input + if( nIns()==2 && !hasPhi() ) + return in(1); // Collapse if no Phis; 1-input Phis will collapse on their own + + // If a CFG diamond with no merging, delete: "if( pred ) {} else {};" + if( !hasPhi() && // No Phi users, just a control user + in(1) instanceof CProjNode p1 && + in(2) instanceof CProjNode p2 && + p1.in(0).addDep(this)==p2.in(0).addDep(this) && + p1.in(0) instanceof IfNode iff ) { + // Replace with the iff.ctrl directly + if( nIns()==3 ) return iff.ctrl(); + // Just delete the path for fat Regions + setDef(1,iff.ctrl()); + return delDef(2); + } + + return null; + } + + Node deadPath() { + // Delete dead paths into a Region + int path = findDeadInput(); + if( path==0 ) return null; + // Do not delete the entry path of a loop (ok to remove the back edge + // and make the loop a single-entry Region which folds away the Loop). + // Folding the entry path confused the loop structure, moving the + // backedge to the entry point. + if( this instanceof LoopNode loop && loop.entry()==in(path) ) + return null; + // Cannot use the obvious output iterator here, because a Phi deleting + // an input might recursively delete *itself*. This shuffles the + // output array, and we might miss iterating an unrelated Phi. So on + // rare occasions we repeat the loop to get all the Phis. + int nouts = 0; + while( nouts != nOuts() ) { + nouts = nOuts(); + for( int i=0; i<nOuts(); i++ ) + if( out(i) instanceof PhiNode phi && phi.nIns()==nIns() ) + CodeGen.CODE.addAll(phi.delDef(path)._outputs); + } + return isDead() ? Parser.XCTRL : delDef(path); + } + + + private int findDeadInput() { + for( int i=1; i<nIns(); i++ ) + if( in(i)._type==Type.XCONTROL ) + return i; + return 0; // All inputs alive + } + + boolean hasPhi() { + for( Node phi : _outputs ) + if( phi instanceof PhiNode ) + return true; + return false; + } + + // Immediate dominator of Region is a little more complicated. + @Override public int idepth() { + if( _idepth!=0 ) return _idepth; + int d=0; + for( Node n : _inputs ) + if( n!=null ) + d = Math.max(d,((CFGNode)n).idepth()+1); + return _idepth=d; + } + + @Override public CFGNode idom(Node dep) { + CFGNode lca = null; + // Walk the LHS & RHS idom trees in parallel until they match, or either fails. + // Because this does not cache, it can be linear in the size of the program. + for( int i=1; i<nIns(); i++ ) + lca = cfg(i)._idom(lca,dep); + return lca; + } + + // True if last input is null + public boolean inProgress() { return nIns()>1 && in(nIns()-1) == null; } + + // Never equal if inProgress + @Override boolean eq( Node n ) { return !inProgress(); } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ReturnNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ReturnNode.java new file mode 100644 index 00000000..20b6c7b9 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ReturnNode.java @@ -0,0 +1,129 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +/** + * The Return node has two inputs. The first input is a control node and the + * second is the data node that supplies the return value. + * <p> + * In this presentation, Return functions as a Stop node, since multiple <code>return</code> statements are not possible. + * The Stop node will be introduced in Chapter 6 when we implement <code>if</code> statements. + * <p> + * The Return's output is the value from the data node. + */ +public class ReturnNode extends CFGNode { + + public FunNode _fun; + + public ReturnNode(Node ctrl, Node mem, Node data, Node rpc, FunNode fun ) { + super(ctrl, mem, data, rpc); + _fun = fun; + } + public ReturnNode( ReturnNode ret, FunNode fun ) { super(ret); _fun = fun; } + + public Node ctrl() { return in(0); } + public Node mem () { return in(1); } + public Node expr() { return in(2); } + public Node rpc () { return in(3); } + public FunNode fun() { return _fun; } + + @Override + public String label() { return "Return"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("return "); + expr()._print0(sb, visited); + return sb.append(";"); + } + + // No one unique control follows; can be many call end sites + @Override public CFGNode uctrl() { return null; } + + @Override + public Type compute() { + if( inProgress () ) return TypeTuple.RET; // In progress + if( _fun.isDead() ) return TypeTuple.RET.dual(); // Dead another way + return TypeTuple.make(ctrl()._type,mem()._type,expr()._type); + } + + @Override public Node idealize() { + if( inProgress () ) return null; + if( _fun.isDead() ) return null; + + // Upgrade signature based on return type + Type ret = expr()._type; + TypeFunPtr fcn = _fun.sig(); + assert ret.isa(fcn.ret()); + if( ret != fcn.ret() ) + _fun.setSig(fcn.makeFrom(ret)); + + // If dead (cant be reached; infinite loop), kill the exit values + if( ctrl()._type==Type.XCONTROL && + !(mem() instanceof ConstantNode && expr() instanceof ConstantNode) ) { + Node top = new ConstantNode(Type.TOP).peephole(); + setDef(1,top); + setDef(2,top); + return this; + } + + return null; + } + + public boolean inProgress() { + return ctrl().getClass() == RegionNode.class && ((RegionNode)ctrl()).inProgress(); + } + + // Gather parse-time return types for error reporting + private Type mt = Type.TOP; + private boolean ti=false, tf=false, tp=false, tn=false; + + // Add a return exit to the current parsing function + void addReturn( Node ctrl, Node rmem, Node expr ) { + assert inProgress(); + + // Gather parse-time return types for error reporting + Type t = expr._type; + mt = mt.meet(t); + ti |= t instanceof TypeInteger x; + tf |= t instanceof TypeFloat x; + tp |= t instanceof TypeMemPtr x; + tn |= t==Type.NIL; + + // Merge path into the One True Return + RegionNode r = (RegionNode)ctrl(); + // Assert that the Phis are in particular outputs; not reordered or shuffled + PhiNode mem = (PhiNode)r.out(0); assert mem._declaredType == TypeMem.BOT; + PhiNode rez = (PhiNode)r.out(1); assert rez._declaredType == Type.BOTTOM; + // Pop "inProgress" null off + r ._inputs.pop(); + mem._inputs.pop(); + rez._inputs.pop(); + // Add new return point + r .addDef(ctrl); + mem.addDef(rmem); + rez.addDef(expr); + // Back to being inProgress + r .addDef(null); + mem.addDef(null); + rez.addDef(null); + } + + @Override public Parser.ParseException err() { + return expr()._type/*mt*/==Type.BOTTOM ? mixerr(ti,tf,tp,tn,_fun._loc) : null; + } + + static Parser.ParseException mixerr( boolean ti, boolean tf, boolean tp, boolean tn, Parser.Lexer loc ) { + if( !ti && !tf && !tp && !tn ) + return Parser.error("No defined return type", loc); + SB sb = new SB().p("No common type amongst "); + if( ti ) sb.p("int and "); + if( tf ) sb.p("f64 and "); + if( tp || tn ) sb.p("reference and "); + return Parser.error(sb.unchar(5).toString(),loc); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/RoundF32Node.java b/chapter20/src/main/java/com/seaofnodes/simple/node/RoundF32Node.java new file mode 100644 index 00000000..44d3a587 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/RoundF32Node.java @@ -0,0 +1,38 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class RoundF32Node extends Node { + public RoundF32Node(Node lhs) { super(null, lhs); } + + @Override public String label() { return "RoundF32"; } + + @Override public String glabel() { return "(f32)"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return in(1)._print0(sb.append("((f32)"), visited).append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0 && i0.isConstant() ) + return TypeFloat.constant((float)i0.value()); + return in(1)._type; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Type t1 = lhs._type; + + // RoundF32 of float + if( t1 instanceof TypeFloat tf && tf._sz==32 ) + return lhs; + + return null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/SarNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/SarNode.java new file mode 100644 index 00000000..4a71ec06 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/SarNode.java @@ -0,0 +1,50 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class SarNode extends LogicalNode { + public SarNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "Sar"; } + @Override public String op() { return ">>"; } + + @Override public String glabel() { return ">>"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if (t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1) { + if( i0 == TypeInteger.ZERO ) + return TypeInteger.ZERO; + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()>>i1.value()); + if( i1.isConstant() ) { + int log = (int)i1.value(); + return TypeInteger.make(-1L<<(63-log),(1L<<(63-log))-1); + } + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t2 = rhs._type; + + // Sar of 0. + if( t2.isConstant() && t2 instanceof TypeInteger i && (i.value()&63)==0 ) + return lhs; + + // TODO: x >> 3 >> (y ? 1 : 2) ==> x >> (y ? 4 : 5) + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new SarNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ScopeNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ScopeNode.java new file mode 100644 index 00000000..1208ff63 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ScopeNode.java @@ -0,0 +1,428 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.*; +import java.util.*; + +/** + * The Scope node is purely a parser helper - it tracks names to nodes with a + * stack of variables. + */ +public class ScopeNode extends MemMergeNode { + + /** + * The control is a name that binds to the currently active control + * node in the graph + */ + public static final String CTRL = "$ctrl"; + public static final String ARG0 = "arg"; + public static final String MEM0 = "$mem"; + + // All active/live variables in all nested scopes, all run together + public final Ary<Var> _vars; + + // Size of each nested lexical scope + public final Ary<Integer> _lexSize; + + // Lexical scope is one of normal Block, constructor or function + public enum Kind { Block, Constructor, Function }; + public final Ary<Kind> _kinds; + + // Extra guards; tested predicates and casted results + private final Ary<Node> _guards; + + // A new ScopeNode + public ScopeNode() { + super(true); + _vars = new Ary<>(Var .class); + _lexSize= new Ary<>(Integer.class); + _kinds = new Ary<>(Kind .class); + _guards = new Ary<>(Node .class); + } + + @Override public String label() { return "Scope"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + sb.append("Scope[ "); + int j=1; + for( int i=0; i<nIns(); i++ ) { + if( j < _lexSize._len && i == _lexSize.at(j) ) { sb.append("| "); j++; } + Var v = _vars.get(i); + sb.append(v.type().print(new SB())); + sb.append(" "); + if( v._final ) sb.append("!"); + sb.append(v._name); + sb.append("="); + Node n = in(i); + while( n instanceof ScopeNode loop ) { + sb.append("Lazy_"); + n = loop.in(i); + } + if( n==null ) sb.append("___"); + else n._print0(sb, visited); + sb.append(", "); + } + sb.setLength(sb.length()-2); + return sb.append("]"); + } + + + public Node ctrl() { return in(0); } + public MemMergeNode mem() { return (MemMergeNode)in(1); } + + /** + * The ctrl of a ScopeNode is always bound to the currently active + * control node in the graph, via a special name '$ctrl' that is not + * a valid identifier in the language grammar and hence cannot be + * referenced in Simple code. + * + * @param n The node to be bound to '$ctrl' + * + * @return Node that was bound + */ + public <N extends Node> N ctrl(N n) { return setDef(0,n); } + public Node mem(Node n) { return setDef(1,n); } + + public void push(Kind kind) { + assert _lexSize._len==_kinds._len; + _lexSize.push(_vars.size()); + _kinds .push(kind); + } + + // Pop a lexical scope + public void pop() { + assert _lexSize._len==_kinds._len; + promote(); + int n = _lexSize.pop(); + _kinds.pop(); + popUntil(n); + _vars.setLen(n); + } + + + // Look for forward references in the last lexical scope and promote to the + // next outer lexical scope. At the last scope declare them an error. + public void promote() { + int n = _lexSize.last(); + for( int i=n; i<nIns(); i++ ) { + Var v = _vars.at(i); + if( !v.isFRef() ) continue; + if( _lexSize._len==1 ) + throw Parser.error("Undefined name '" + v._name + "'",v._loc); + _vars.swap(n,i); + _inputs.swap(n,i); + v._idx = n; + n++; + _lexSize.set(_lexSize._len-1,n); + } + } + + + public boolean inCon() { return _kinds.last() == Kind.Constructor; } + + // Is v outside any current function scope? + public boolean outOfFunction( Var v ) { + for( int i=_lexSize._len-1; i>=0 && v._idx<_lexSize.at(i); i-- ) + if( _kinds.at(i)==Kind.Function ) + return true; + return false; + } + + + // Find name in reverse, return an index into _vars or -1. Linear scan + // instead of hashtable, but probably doesn't matter until the scan + // typically hits many dozens of variables. + int find( String name ) { + for( int i=_vars.size()-1; i>=0; i-- ) + if( _vars.at(i)._name.equals(name) ) + return i; + return -1; + } + + /** + * Create a new variable name in the current scope + */ + public boolean define( String name, Type declaredType, boolean xfinal, Node init, Parser.Lexer loc ) { + assert _lexSize.isEmpty() || name.charAt(0)!='$' ; // Later scopes do not define memory + if( _lexSize._len > 0 ) + for( int i=_vars.size()-1; i>=_lexSize.last(); i-- ) { + Var n = _vars.at(i); + if( n._name.equals(name) ) { + if( !n.isFRef() ) return false; // Double define + FRefNode fref = (FRefNode)in(n._idx); // Get forward ref + if( !xfinal || !declaredType.isConstant() ) throw fref.err(); // Must be a final constant + n.defFRef(declaredType,xfinal,loc); // Declare full correct type, final, source location + setDef(n._idx,fref.addDef(init)); // Set FRef to defined; tell parser also + } + } + Var v = new Var(nIns(),name,declaredType,xfinal,loc,init==Parser.XCTRL); + _vars.add(v); + // Creating a forward reference + if( init==Parser.XCTRL ) + init = new FRefNode(v).init(); + addDef(init); + return true; + } + + // Read from memory + public Node mem( int alias ) { return mem()._mem(alias,null); } + // Write to memory + public void mem( int alias, Node st ) { mem()._mem(alias,st); } + + + /** + * Lookup a name in all scopes starting from most deeply nested. + * + * @param name Name to be looked up + * @return null if not found, or the implementing Node + */ + public Var lookup( String name ) { + int idx = find(name); + // -1 is missed in all scopes, not found + return idx == -1 ? null : update(_vars.at(idx),null); + } + + /** + * If the name is present in any scope, then redefine else null + * + * @param name Name being redefined + * @param n The node to bind to the name + */ + public void update( String name, Node n ) { + int idx = find(name); + assert idx>=0; + update(_vars.at(idx),n); + } + + public Var update( Var v, Node st ) { + Node old = in(v._idx); + if( old instanceof ScopeNode loop ) { + // Lazy Phi! + Node def = loop.in(v._idx); + old = def instanceof PhiNode phi && loop.ctrl()==phi.region() + // Loop already has a real Phi, use it + ? def + // Set real Phi in the loop head + // The phi takes its one input (no backedge yet) from a recursive + // lookup, which might have insert a Phi in every loop nest. + : loop.setDef(v._idx,new PhiNode(v._name, v.lazyGLB(), loop.ctrl(), loop.in(loop.update(v,null)._idx),null).peephole()); + setDef(v._idx,old); + } + assert !v._final || st==null; + if( st!=null ) setDef(v._idx,st); // Set new value + return v; + } + + /** + * Duplicate a ScopeNode; including all levels, up to Nodes. So this is + * neither shallow (would dup the Scope but not the internal HashMap + * tables), nor deep (would dup the Scope, the HashMap tables, but then + * also the program Nodes). + * <p> + * If the {@code loop} flag is set, the edges are filled in as the original + * Scope, as an indication of Lazy Phis at loop heads. The goal here is to + * not make Phis at loop heads for variables which are never touched in the + * loop body. + * <p> + * The new Scope is a full-fledged Node with proper use<->def edges. + */ + public ScopeNode dup() { return dup(false); } + public ScopeNode dup(boolean loop) { + ScopeNode dup = new ScopeNode(); + // Our goals are: + // 1) duplicate the name bindings of the ScopeNode across all stack levels + // 2) Make the new ScopeNode a user of all the nodes bound + // 3) Ensure that the order of defs is the same to allow easy merging + dup._vars .addAll(_vars ); + dup._lexSize.addAll(_lexSize); + dup._kinds .addAll(_kinds ); + dup._guards .addAll(_guards ); + // The dup'd guards all need dup'd keepers, to keep proper accounting + // when later removing all guards + for( Node n : _guards ) + if( !(n instanceof CFGNode) ) + n.keep(); + dup.addDef(ctrl()); // Control input is just copied + + // Memory input is a shallow copy + MemMergeNode memdup = new MemMergeNode(true), mem = mem(); + memdup.addDef(null); + memdup.addDef(loop ? this : mem.in(1)); + for( int i=2; i<mem.nIns(); i++ ) + // For lazy phis on loops we use a sentinel + // that will trigger phi creation on update + memdup.addDef(loop ? this : mem.in(i)); + dup.addDef(memdup); + + // Copy of other inputs + for( int i=2; i<nIns(); i++ ) + // For lazy phis on loops we use a sentinel + // that will trigger phi creation on update + dup.addDef(loop && !_vars.at(i)._final ? this : in(i)); + return dup; + } + + /** + * Merges the names whose node bindings differ, by creating Phi node for such names + * The names could occur at all stack levels, but a given name can only differ in the + * innermost stack level where the name is bound. + * + * @param that The ScopeNode to be merged into this + * @return A new node representing the merge point + */ + public RegionNode mergeScopes(ScopeNode that, Parser.Lexer loc) { + RegionNode r = ctrl(new RegionNode(loc,null,ctrl(), that.ctrl()).keep()); + mem()._merge(that.mem(),r); + this ._merge(that ,r); + that.kill(); // Kill merged scope + CodeGen.CODE.add(r); + return r.unkeep(); + } + + private void _merge(ScopeNode that, RegionNode r) { + for( int i = 2; i < nIns(); i++) + if( in(i) != that.in(i) ) { // No need for redundant Phis + // If we are in lazy phi mode we need to a lookup + // by name as it will trigger a phi creation + Var v = _vars.at(i); + Node lhs = this.in(this.update(v,null)); + Node rhs = that.in(that.update(v,null)); + setDef(i, new PhiNode(v._name, v.type(), r, lhs, rhs).peephole()); + } + } + + // Balance arms of an IF. Extra lonely defs are thrown: "if(pred) int x;". + // Forward refs are copied to the other side, "as if" they were there all along. + public void balanceIf( ScopeNode scope ) { + for( int i = nIns(); i < scope.nIns(); i++ ) { + Var n = scope._vars.at(i); + if( n.isFRef() ) { // RHS has forward refs + _vars.add(n); // Copy to LHS + addDef(scope.in(i)); + } else + throw Parser.error("Cannot define a '"+n._name+"' on one arm of an if",n._loc); + } + } + + + // peephole the backedge scope into this loop head scope + // We set the second input to the phi from the back edge (i.e. loop body) + public void endLoop(ScopeNode back, ScopeNode exit ) { + Node ctrl = ctrl(); + assert ctrl instanceof LoopNode loop && loop.inProgress(); + ctrl.setDef(2,back.ctrl()); + + mem()._endLoopMem( this, back.mem(), exit.mem() ); + this ._endLoop ( this, back , exit ); + back.kill(); // Loop backedge is dead + // Now one-time do a useless-phi removal + mem()._useless(); + this ._useless(); + + // The exit mem's lazy default value had been the loop top, + // now it goes back to predating the loop. + exit.mem().setDef(1,mem().in(1)); + } + + // Fill in the backedge of any inserted Phis + void _endLoop( ScopeNode scope, Node back, Node exit ) { + for( int i=2; i<nIns(); i++ ) { + if( _vars.at(i)._final ) continue; // Final vars did not get modified in the loop + if( _vars.at(i).type().isHighOrConst() ) continue; // Cannot lift higher than a constant, so no Phi + if( back.in(i) != scope ) { + PhiNode phi = (PhiNode)in(i); + assert phi.region()==scope.ctrl() && phi.in(2)==null; + phi.setDef(2,back.in(i)); // Fill backedge + } + if( exit.in(i) == scope ) // Replace a lazy-phi on the exit path also + exit.setDef(i,in(i)); + } + } + + + // Up-casting: using the results of an If to improve a value. + // E.g. "if( ptr ) ptr.field;" is legal because ptr is known not-null. + public void addGuards( Node ctrl, Node pred, boolean invert ) { + assert ctrl instanceof CFGNode; + _guards.add(ctrl); // Marker to distinguish 0,1,2 guards + // add pred & its cast to the normal input list, with special Vars + if( pred==null || pred.isDead() ) + return; // Dead, do not add any guards + // Invert the If conditional + if( invert ) + pred = pred instanceof NotNode not ? not.in(1) : CodeGen.CODE.add(new NotNode(pred).peephole()); + // This is a zero/null test. + // Compute the positive test type. + Type tnz = pred._type.nonZero(); + if( tnz!=null ) + _addGuard(tnz,ctrl,pred); + + // Compute the negative test type. + if( pred instanceof NotNode not ) { + Node npred = not.in(1); + Type tzero = npred._type.makeZero(); + _addGuard(tzero,ctrl,npred); + } + } + + private void _addGuard(Type guard, Node ctrl, Node pred) { + Type tcast = guard.join(pred._type); + if( tcast != pred._type && !tcast.isHigh() ) { + Node cast = new CastNode(tcast,ctrl,pred.keep()).peephole().keep(); + _guards.add(pred); + _guards.add(cast); + replace(pred,cast); + } + } + + + // Remove matching pred/cast pairs from this guarded region. + public ScopeNode removeGuards( Node ctrl ) { + assert ctrl instanceof CFGNode; + // 0,1 or 2 guards + while( true ) { + Node g = _guards.pop(); + if( g == ctrl ) break; + if( g instanceof CFGNode ) continue; + g .unkill(); // Pop/kill cast + _guards.pop().unkill(); // Pop/kill pred + } + return this; + } + + // If we find a guarded instance of pred, replace with the upcasted version + public Node upcastGuard( Node pred ) { + // If finding an instanceof pred, replace with cast. + // Otherwise, just pred itself. + int i = _guards._len; + while( i > 0 ) { + Node cast = _guards.at(--i); + if( cast instanceof CFGNode ) continue; // Marker between guard sets + Node xpred = _guards.at(--i); + if( xpred == pred ) + return cast; + } + return pred; + } + + // Kill guards also + @Override public void kill() { + for( Node n : _guards ) + if( !(n instanceof CFGNode) ) + n.unkill(); + _guards.clear(); + // Can have lazy uses remaining + if( isUnused() ) + super.kill(); + } + + + private void replace( Node old, Node cast ) { + assert old!=null && old!=cast; + for( int i=0; i<nIns(); i++ ) + if( in(i)==old ) + setDef(i,cast); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ShlNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ShlNode.java new file mode 100644 index 00000000..21dbd503 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ShlNode.java @@ -0,0 +1,53 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class ShlNode extends LogicalNode { + public ShlNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "Shl"; } + @Override public String op() { return "<<"; } + + @Override public String glabel() { return "<<"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if (t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1 ) { + if( i0 == TypeInteger.ZERO ) + return TypeInteger.ZERO; + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()<<i1.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + + if( rhs._type instanceof TypeInteger shl && shl.isConstant() ) { + // Shl of 0. + if( (shl.value()&63)==0 ) + return lhs; + // (x + c) << i => (x << i) + (c << i) + if( lhs instanceof AddNode add && add.addDep(this).in(2)._type instanceof TypeInteger c && c.isConstant() ) { + long sum = c.value() << shl.value(); + if( Integer.MIN_VALUE <= sum && sum <= Integer.MAX_VALUE ) + return new AddNode(new ShlNode(_loc,add.in(1),rhs).peephole(), Parser.con(sum) ); + } + } + + // TODO: x << 3 << (y ? 1 : 2) ==> x << (y ? 4 : 5) + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new ShlNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ShrNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ShrNode.java new file mode 100644 index 00000000..789fb8a8 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ShrNode.java @@ -0,0 +1,46 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class ShrNode extends LogicalNode { + public ShrNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "Shr"; } + @Override public String op() { return ">>>"; } + + @Override public String glabel() { return ">>>"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if (t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1 ) { + if( i0 == TypeInteger.ZERO ) + return TypeInteger.ZERO; + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()>>>i1.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t2 = rhs._type; + + // Shr of 0. + if( t2.isConstant() && t2 instanceof TypeInteger i && (i.value()&63)==0 ) + return lhs; + + // TODO: x >>> 3 >>> (y ? 1 : 2) ==> x >>> (y ? 4 : 5) + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new ShrNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/StartNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/StartNode.java new file mode 100644 index 00000000..26e08535 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/StartNode.java @@ -0,0 +1,58 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.*; +import java.util.Arrays; +import java.util.BitSet; +import java.util.HashSet; +import static com.seaofnodes.simple.Utils.TODO; + +/** + * The Start node represents the start of the function. + * + * Start initially has 1 input (arg) from outside and the initial control. + * In ch10 we also add mem aliases as structs get defined; each field in struct + * adds a distinct alias to Start's tuple. + */ +public class StartNode extends LoopNode implements MultiNode { + + final Type _arg; + + public StartNode(Type arg) { super((Parser.Lexer)null,null); _arg = arg; _type = compute(); } + public StartNode(StartNode start) { super(start); _arg = start._arg; } + + @Override public String label() { return "Start"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return sb.append(label()); + } + + @Override public boolean isMultiHead() { return true; } + @Override public boolean blockHead() { return true; } + @Override public CFGNode cfg0() { return null; } + + // Get the one control following; error to call with more than one e.g. an + // IfNode or other multi-way branch. For Start, its "main" + @Override public CFGNode uctrl() { + // Find "main", its the start. + CFGNode C = null; + for( Node use : _outputs ) + if( use instanceof FunNode fun && fun.sig().isa(CodeGen.CODE._main) ) + { assert C==null; C = fun; } + return C; + } + + + @Override public TypeTuple compute() { + return TypeTuple.make(Type.CONTROL,TypeMem.TOP,_arg); + } + + @Override public Node idealize() { return null; } + + // No immediate dominator, and idepth==0 + @Override public int idepth() { return 0; } + @Override public CFGNode idom(Node dep) { return null; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/StopNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/StopNode.java new file mode 100644 index 00000000..6aaba3b3 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/StopNode.java @@ -0,0 +1,64 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.type.Type; +import java.util.BitSet; + +public class StopNode extends CFGNode { + + public final String _src; + + public StopNode(String src) { + super(); + _src = src; + _type = compute(); + } + public StopNode(StopNode stop) { super(stop); _src = stop._src; } + + @Override + public String label() { + return "Stop"; + } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + // For the sake of many old tests, and single value prints as "return val" + if( ret()!=null ) return ret()._print0(sb,visited); + sb.append("Stop[ "); + for( Node ret : _inputs ) + ret._print0(sb, visited).append(" "); + return sb.append("]"); + } + + @Override public boolean blockHead() { return true; } + + + // If a single Return, return it. + // Otherwise, null because ambiguous. + public ReturnNode ret() { + return nIns()==1 && in(0) instanceof ReturnNode ret ? ret : null; + } + + @Override + public Type compute() { + return Type.BOTTOM; + } + + @Override + public Node idealize() { + int len = nIns(); + for( int i=0; i<nIns(); i++ ) + if( ((ReturnNode)in(i)).fun().isDead() ) + delDef(i--); + if( len != nIns() ) return this; + return null; + } + + @Override public int idepth() { + if( _idepth!=0 ) return _idepth; + int d=0; + for( Node ret : _inputs ) + d = Math.max(d,((ReturnNode)ret).idepth()+1); + return _idepth=d; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/StoreNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/StoreNode.java new file mode 100644 index 00000000..93294dba --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/StoreNode.java @@ -0,0 +1,113 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +/** + * Store represents setting a value to a memory based object, in chapter 10 + * this means a field inside a struct. + */ +public class StoreNode extends MemOpNode { + + private final boolean _init; // Initializing writes are allowed to write null + + /** + * @param name The struct field we are assigning to + * @param mem The memory alias node - this is updated after a Store + * @param ptr The ptr to the struct base where we will store a value + * @param off The offset inside the struct base + * @param value Value to be stored + */ + public StoreNode(Parser.Lexer loc, String name, int alias, Type glb, Node mem, Node ptr, Node off, Node value, boolean init) { + super(loc, name, alias, glb, mem, ptr, off, value); + _init = init; + } + + // GraphVis DOT code and debugger labels + @Override public String label() { return "st_"+mlabel(); } + // GraphVis node-internal labels + @Override public String glabel() { return "." +_name+"="; } + @Override public boolean isMem() { return true; } + + public Node val() { return in(4); } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return sb.append(".").append(_name).append("=").append( val()).append(";"); + } + + @Override + public Type compute() { + Type val = val()._type; + TypeMem mem = (TypeMem)mem()._type; // Invariant + if( mem == TypeMem.TOP ) return TypeMem.TOP; + Type t = Type.BOTTOM; // No idea on field contents + // Same alias, lift val to the declared type and then meet into other fields + if( mem._alias == _alias ) { + // Update declared forward ref to the actual + if( _declaredType.isFRef() && val instanceof TypeMemPtr tmp && !tmp.isFRef() ) + _declaredType = tmp; + val = val.join(_declaredType); + t = val.meet(mem._t); + } + return TypeMem.make(_alias,t); + } + + @Override + public Node idealize() { + + // Simple store-after-store on same address. Should pick up the + // required init-store being stomped by a first user store. + if( mem() instanceof StoreNode st && + ptr()==st.ptr() && // Must check same object + off()==st.off() && // And same offset (could be "same alias" but this handles arrays to same index) + ptr()._type instanceof TypeMemPtr && // No bother if weird dead pointers + // Must have exactly one use of "this" or you get weird + // non-serializable memory effects in the worse case. + checkOnlyUse(st) ) { + assert _name.equals(st._name); // Equiv class aliasing is perfect + setDef(1,st.mem()); + return this; + } + + // Value is automatically truncated by narrow store + if( val() instanceof AndNode and && and.in(2)._type.isConstant() ) { + int log = _declaredType.log_size(); + if( log<3 ) { // And-mask vs narrow store + long mask = ((TypeInteger)and.in(2)._type).value(); + long bits = (1L<<(8<<log))-1; + // Mask does not mask any of the stored bits + if( (bits&mask)==bits ) + // So and-mask is already covered by the store + { setDef(4,and.in(1)); return this; } + } + } + + + return null; + } + + // Check that "mem" has no uses except "this" + private boolean checkOnlyUse(Node mem) { + if( mem.nOuts()==1 ) return true; + // Add deps on the other uses (can be e.g. ScopeNode mid-parse) so that + // when the other uses go away we can retry. + for( Node use : mem._outputs ) + if( use != this ) + use.addDep(this); + return false; + } + + @Override + public Parser.ParseException err() { + Parser.ParseException err = super.err(); + if( err != null ) return err; + TypeMemPtr tmp = (TypeMemPtr)ptr()._type; + if( tmp._obj.field(_name)._final && !_init ) + return Parser.error("Cannot modify final field '"+_name+"'",_loc); + Type t = val()._type; + //return _init || t.isa(_declaredType) ? null : Parser.error("Cannot store "+t+" into field "+_declaredType+" "+_name,_loc); + return null; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/StructNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/StructNode.java new file mode 100644 index 00000000..076b342a --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/StructNode.java @@ -0,0 +1,46 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.type.*; +import java.util.BitSet; + +/** + * Build a compound object + */ +public class StructNode extends Node { + + public TypeStruct _ts; + + @Override public String label() { return _ts==null ? "STRUCT?" : _ts.str(); } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + if( _ts==null ) return sb.append("STRUCT?"); + sb.append(_ts._name).append(" {"); + for( int i=0; i<nIns(); i++ ) { + sb.append(_ts._fields[i]._fname).append(":"); + sb.append((in(i)==null ? Type.BOTTOM : in(i)._type).print(new SB())); + sb.append("; "); + } + sb.setLength(sb.length()-2); + return sb.append("}"); + } + + @Override + public TypeStruct compute() { + if( _ts==null ) return TypeStruct.BOT; + Field[] fs = new Field[_ts._fields.length]; + for( int i=0; i<fs.length; i++ ) + fs[i] = _ts._fields[i].makeFrom(in(i)==null ? Type.TOP : in(i)._type); + return TypeStruct.make(_ts._name,fs); + } + + @Override + public Node idealize() { return null; } + + @Override + boolean eq(Node n) { return _ts == ((StructNode)n)._ts; } + + @Override + int hash() { return _ts==null ? 0 : _ts.hashCode(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/SubFNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/SubFNode.java new file mode 100644 index 00000000..b591718d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/SubFNode.java @@ -0,0 +1,44 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; + +import java.util.BitSet; + +public class SubFNode extends Node { + public SubFNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "SubF"; } + + @Override public String glabel() { return "-"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("-"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + if (in(1)._type instanceof TypeFloat i0 && + in(2)._type instanceof TypeFloat i1) { + if (i0.isConstant() && i1.isConstant()) + return TypeFloat.constant(i0.value()-i1.value()); + } + return in(1)._type.meet(in(2)._type); + } + + @Override + public Node idealize() { + Node lhs = in(1); + Type t2 = in(2)._type; + + // Sub of 0. + if ( t2.isConstant() && t2 instanceof TypeFloat i && i.value()==0 ) + return lhs; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new SubFNode(lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/SubNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/SubNode.java new file mode 100644 index 00000000..339c65ca --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/SubNode.java @@ -0,0 +1,59 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class SubNode extends Node { + public SubNode(Node lhs, Node rhs) { super(null, lhs, rhs); } + + @Override public String label() { return "Sub"; } + + @Override public String glabel() { return "-"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + in(1)._print0(sb.append("("), visited); + in(2)._print0(sb.append("-"), visited); + return sb.append(")"); + } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + // Sub of same is 0 + if( in(1)==in(2) ) + return TypeInteger.ZERO; + if( t1 instanceof TypeInteger i1 && + t2 instanceof TypeInteger i2 ) { + if (i1.isConstant() && i2.isConstant()) + return TypeInteger.constant(i1.value()-i2.value()); + // Fold ranges like {2-3} - {0-1} into {1-3}. + if( !AddNode.overflow(i1._min,-i2._max) && + !AddNode.overflow(i1._max,-i2._min) && + i2._min != Long.MIN_VALUE ) + return TypeInteger.make(i1._min-i2._max,i1._max-i2._min); + } + + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + // x - (-y) is x+y + if( in(2) instanceof MinusNode minus ) + return new AddNode(in(1),minus.in(1)); + + // (-x) - y is -(x+y) + if( in(1) instanceof MinusNode minus ) + return new MinusNode(new AddNode(minus.in(1),in(2)).peephole()); + + return null; + } + + @Override Node copy(Node lhs, Node rhs) { return new SubNode(lhs,rhs); } + @Override Node copyF() { return new SubFNode(null,null); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/ToFloatNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/ToFloatNode.java new file mode 100644 index 00000000..1f4d67e0 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/ToFloatNode.java @@ -0,0 +1,34 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFloat; +import com.seaofnodes.simple.type.TypeInteger; + +import java.util.BitSet; + +public class ToFloatNode extends Node { + public ToFloatNode(Node lhs) { super(null, lhs); } + + @Override public String label() { return "ToFloat"; } + + @Override public String glabel() { return "(flt)"; } + + @Override + StringBuilder _print1(StringBuilder sb, BitSet visited) { + return in(1)._print0(sb.append("(flt)"), visited); + } + + @Override + public Type compute() { + if( in(1)._type == Type.NIL ) return TypeFloat.FZERO; + if (in(1)._type instanceof TypeInteger i0 ) { + if( i0.isHigh() ) return TypeFloat.F64.dual(); + if( i0.isConstant() ) + return TypeFloat.constant(i0.value()); + } + return TypeFloat.F64; + } + + @Override public Node idealize() { return null; } + @Override Node copy(Node lhs, Node rhs) { return new ToFloatNode(lhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/XCtrlNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/XCtrlNode.java new file mode 100644 index 00000000..9b4245c7 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/XCtrlNode.java @@ -0,0 +1,16 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import java.util.BitSet; + +public class XCtrlNode extends CFGNode { + public XCtrlNode() { super(new Node[]{CodeGen.CODE._start}); } + @Override public String label() { return "Xctrl"; } + @Override StringBuilder _print1(StringBuilder sb, BitSet visited) { return sb.append("Xctrl"); } + @Override public boolean isConst() { return true; } + @Override public boolean isMultiTail() { return true; } + @Override public Type compute() { return Type.XCONTROL; } + @Override public Node idealize() { return null; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/XorNode.java b/chapter20/src/main/java/com/seaofnodes/simple/node/XorNode.java new file mode 100644 index 00000000..37d7147d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/XorNode.java @@ -0,0 +1,54 @@ +package com.seaofnodes.simple.node; + +import com.seaofnodes.simple.Parser; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.util.BitSet; + +public class XorNode extends LogicalNode { + public XorNode(Parser.Lexer loc, Node lhs, Node rhs) { super(loc, lhs, rhs); } + + @Override public String label() { return "Xor"; } + @Override public String op() { return "^"; } + + @Override public String glabel() { return "^"; } + + @Override + public Type compute() { + Type t1 = in(1)._type, t2 = in(2)._type; + if( t1.isHigh() || t2.isHigh() ) + return TypeInteger.TOP; + if( t1 instanceof TypeInteger i0 && + t2 instanceof TypeInteger i1 ) { + if( i0.isConstant() && i1.isConstant() ) + return TypeInteger.constant(i0.value()^i1.value()); + } + return TypeInteger.BOT; + } + + @Override + public Node idealize() { + Node lhs = in(1); + Node rhs = in(2); + Type t1 = lhs._type; + Type t2 = rhs._type; + + // Xor of 0. We do not check for (0^x) because this will already + // canonicalize to (x^0) + if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==0 ) + return lhs; + + // Move constants to RHS: con*arg becomes arg*con + if ( t1.isConstant() && !t2.isConstant() ) + return swap12(); + + // Do we have ((x ^ (phi cons)) ^ con) ? + // Do we have ((x ^ (phi cons)) ^ (phi cons)) ? + // Push constant up through the phi: x ^ (phi con0^con0 con1^con1...) + Node phicon = AddNode.phiCon(this,true); + if( phicon!=null ) return phicon; + + return null; + } + @Override Node copy(Node lhs, Node rhs) { return new XorNode(_loc,lhs,rhs); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFMemX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFMemX86.java new file mode 100644 index 00000000..4f75edcd --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFMemX86.java @@ -0,0 +1,33 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.*; + +public class AddFMemX86 extends MemOpX86 { + AddFMemX86( AddFNode add, LoadNode ld , Node base, Node idx, int off, int scale, Node val ) { + super(add,ld, base, idx, off, scale, 0, val ); + } + + @Override public RegMask regmap(int i) { + if( i==1 ) return RegMask.EMPTY; // Memory + if( i==2 ) return x86_64_v2.RMASK; // base + if( i==3 ) return x86_64_v2.RMASK; // index + if( i==4 ) return x86_64_v2.XMASK; // value + throw Utils.TODO(); + } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + + + // General form: "add dst = src + [base + idx<<2 + 12]" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = "); + sb.p(val()==null ? "#"+_imm : code.reg(val())).p(" + "); + asm_address(code,sb); + } + @Override public String op() { return "addf"+_sz; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFX86.java new file mode 100644 index 00000000..a3f5c7b3 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddFX86.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class AddFX86 extends MachConcreteNode implements MachNode { + AddFX86( Node addf ) { super(addf); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.XMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "addf dst += src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" + ").p(code.reg(in(2))); + } + + @Override public String op() { return "addf"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddIX86.java new file mode 100644 index 00000000..6a332ff5 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddIX86.java @@ -0,0 +1,39 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class AddIX86 extends MachConcreteNode implements MachNode { + final TypeInteger _ti; + AddIX86( Node add, TypeInteger ti ) { + super(add); + _inputs.pop(); + _ti = ti; + } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { + // assert i== i; + return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "addi dst += #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" + #"); + _ti.print(sb); + } + + @Override public String op() { + return _ti.value() == 1 ? "inc" : (_ti.value() == -1 ? "dec" : "addi"); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddMemX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddMemX86.java new file mode 100644 index 00000000..7d45d757 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddMemX86.java @@ -0,0 +1,25 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.*; + +public class AddMemX86 extends MemOpX86 { + AddMemX86( AddNode add, LoadNode ld , Node base, Node idx, int off, int scale, int imm, Node val ) { + super(add,ld, base, idx, off, scale, imm, val ); + } + + @Override public String op() { return "add"+_sz; } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // General form: "add dst = src + [base + idx<<2 + 12]" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = "); + sb.p(val()==null ? "#"+_imm : code.reg(val())).p(" + "); + asm_address(code,sb); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddX86.java new file mode 100644 index 00000000..61aa5a9e --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AddX86.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class AddX86 extends MachConcreteNode implements MachNode { + AddX86( Node add ) { super(add); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "add dst += src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" + ").p(code.reg(in(2))); + } + + @Override public String op() { return "add"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AndIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AndIX86.java new file mode 100644 index 00000000..f2ee1c66 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/AndIX86.java @@ -0,0 +1,42 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; +import java.lang.StringBuilder; + +public class AndIX86 extends MachConcreteNode implements MachNode { + final TypeInteger _ti; + AndIX86(Node and, TypeInteger ti) { + super(and); + _inputs.pop(); + _ti = ti; + } + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "andi dst & #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" & #"); + _ti.print(sb); + } + + @Override public String op() { return "andi"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallRX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallRX86.java new file mode 100644 index 00000000..31d303ed --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallRX86.java @@ -0,0 +1,31 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.io.ByteArrayOutputStream; + +public class CallRX86 extends CallNode implements MachNode { + CallRX86( CallNode call ) { super(call); } + + @Override public String label() { return op(); } + @Override public RegMask regmap(int i) { + return i==_inputs._len + ? x86_64_v2.WMASK // Function call target + : x86_64_v2.callInMask(i); // Normal argument + } + @Override public RegMask outregmap() { return x86_64_v2.RET_MASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(fptr())); + } + + @Override public String op() { return "callr"; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallX86.java new file mode 100644 index 00000000..9c246c24 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CallX86.java @@ -0,0 +1,40 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; + +public class CallX86 extends CallNode implements MachNode { + final TypeFunPtr _tfp; + final String _name; + CallX86( CallNode call, TypeFunPtr tfp ) { + super(call); + _inputs.pop(); // Pop constant target + assert tfp.isConstant(); + _tfp = tfp; + _name = CodeGen.CODE.link(tfp)._name; + } + + @Override public String label() { return op(); } + @Override public RegMask regmap(int i) { + return x86_64_v2.callInMask(i); // Normal argument + } + @Override public RegMask outregmap() { return x86_64_v2.RET_MASK; } + + @Override public String name() { return _name; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + @Override public void asm(CodeGen code, SB sb) { + sb.p(_name); + } + + @Override public String op() { return "call"; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpIX86.java new file mode 100644 index 00000000..2fa7fbb1 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpIX86.java @@ -0,0 +1,43 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +// Compare immediate. Sets flags. +public class CmpIX86 extends MachConcreteNode implements MachNode { + final int _imm; + final String _bop; + CmpIX86( BoolNode bool, TypeInteger ti ) { + super(bool); + _inputs.pop(); // Toss ideal ConstantNode away, embedded into machine op + _bop = bool.op(); // One of <,<=,== + _imm = (int)ti.value(); + assert _imm == ti.value(); + } + CmpIX86( Node cmp, double ignore ) { + super(cmp); + _bop = "=="; + _imm = 0; + } + + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.RMASK; } + @Override public RegMask outregmap() { return x86_64_v2.FLAGS_MASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + @Override public void asm(CodeGen code, SB sb) { + String dst = code.reg(this); + if( dst!="FLAGS" ) sb.p(dst).p(" = "); + sb.p(code.reg(in(1))); + if( _imm != 0 ) sb.p(", #").p(_imm); + } + + @Override public String op() { return _imm==0 ? "test" : "cmp"; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpMemX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpMemX86.java new file mode 100644 index 00000000..873ca9df --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpMemX86.java @@ -0,0 +1,33 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.*; + +public class CmpMemX86 extends MemOpX86 { + final boolean _invert; // Op switched LHS, RHS + CmpMemX86( BoolNode bool, LoadNode ld , Node base, Node idx, int off, int scale, int imm, Node val, boolean invert ) { + super(bool,ld, base, idx, off, scale, imm, val ); + _invert = invert; + } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.FLAGS_MASK; } + + // General form: "cmp dst = src, [base + idx<<2 + 12]" + @Override public void asm(CodeGen code, SB sb) { + String dst = code.reg(this); + if( dst!="FLAGS" ) sb.p(dst).p(" = "); + if( _invert ) asm_address(code,sb).p(","); + if( val()==null ) { + if( _imm!=0 ) sb.p("#").p(_imm); + } else { + sb.p(code.reg(val())); + } + if( !_invert ) asm_address(code,sb.p(", ")); + } + + @Override public String op() { return ((val()==null && _imm==0) ? "test" : "cmp") + _sz; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpX86.java new file mode 100644 index 00000000..6b502249 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/CmpX86.java @@ -0,0 +1,27 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class CmpX86 extends MachConcreteNode implements MachNode { + CmpX86( Node cmp ) { super(cmp); } + + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.RMASK; } + @Override public RegMask outregmap() { return x86_64_v2.FLAGS_MASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "add dst += src" + @Override public void asm(CodeGen code, SB sb) { + String dst = code.reg(this); + if( dst!="FLAGS" ) sb.p(dst).p(" = "); + sb.p(code.reg(in(1))).p(", ").p(code.reg(in(2))); + } + + @Override public String op() { return "cmp"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivFX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivFX86.java new file mode 100644 index 00000000..d73bfd5b --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivFX86.java @@ -0,0 +1,33 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class DivFX86 extends MachConcreteNode implements MachNode { + DivFX86( Node divf) { super(divf); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.XMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + + // General form: "divf dst /= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" / ").p(code.reg(in(2))); + } + + @Override public String op() { + // 8 bytes = 64 bits + return "divf"; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivX86.java new file mode 100644 index 00000000..1dc20017 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/DivX86.java @@ -0,0 +1,31 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class DivX86 extends MachConcreteNode implements MachNode { + DivX86( Node div ) { super(div); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { + assert i==1 || i==2; + return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "div dst /= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" / ").p(code.reg(in(2))); + } + + @Override public String op() { return "div"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FltX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FltX86.java new file mode 100644 index 00000000..21529f7b --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FltX86.java @@ -0,0 +1,36 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.ConstantNode; +import com.seaofnodes.simple.node.MachNode; + +import java.io.ByteArrayOutputStream; + +public class FltX86 extends ConstantNode implements MachNode { + FltX86(ConstantNode con ) { super(con); } + + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // General int registers + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // Human-readable form appended to the SB. Things like the encoding, + // indentation, leading address or block labels not printed here. + // Just something like "ld4\tR17=[R18+12] // Load array base". + // General form: "op\tdst=src+src" + @Override public void asm(CodeGen code, SB sb) { + _con.print(sb.p(code.reg(this)).p(" #")); + } + + @Override public String op() { + return "fld"; // Some fancier encoding + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java new file mode 100644 index 00000000..f969fae4 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java @@ -0,0 +1,21 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.FunNode; +import com.seaofnodes.simple.node.MachNode; +import java.io.ByteArrayOutputStream; + +public class FunX86 extends FunNode implements MachNode { + FunX86( FunNode fun ) { super(fun); } + + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/I2f8X86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/I2f8X86.java new file mode 100644 index 00000000..983f817d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/I2f8X86.java @@ -0,0 +1,28 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class I2f8X86 extends MachConcreteNode implements MachNode { + I2f8X86(Node i2f8 ) { super(i2f8); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "i2f8 (flt)int_value" + + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p("(flt)").p(code.reg(in(1))); + } + + @Override public String op() { return "i2f8"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/IntX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/IntX86.java new file mode 100644 index 00000000..2cd27c14 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/IntX86.java @@ -0,0 +1,45 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.ConstantNode; +import com.seaofnodes.simple.node.MachNode; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; + +import java.io.ByteArrayOutputStream; + +// Integer constants +public class IntX86 extends ConstantNode implements MachNode { + + IntX86( ConstantNode con ) { + super(con); + } + + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // General int registers + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // Human-readable form appended to the SB. Things like the encoding, + // indentation, leading address or block labels not printed here. + // Just something like "ld4\tR17=[R18+12] // Load array base". + // General form: "op\tdst=src+src" + @Override public void asm(CodeGen code, SB sb) { + String reg = code.reg(this); + if( _con == Type.NIL || _con == TypeInteger.ZERO ) + sb.p(reg).p(",").p(reg); + else + _con.print(sb.p(reg).p(" #")); + } + + @Override public String op() { + if( _con == Type.NIL || _con == TypeInteger.ZERO ) + return "xor"; + return "ldi"; // Some fancier encoding + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/JmpX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/JmpX86.java new file mode 100644 index 00000000..7cf4444c --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/JmpX86.java @@ -0,0 +1,46 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +// Jump on flags, uses flags +public class JmpX86 extends IfNode implements MachNode { + final String _bop; + JmpX86( IfNode iff, String bop ) { + super(iff); + _bop = bop; + } + + @Override public String label() { return op(); } + + @Override public void postSelect() { + Node set = in(1); + Node cmp = set.in(1); + // Bypass an expected Set and just reference the cmp directly + if( set instanceof SetX86 && (cmp instanceof CmpX86 || cmp instanceof CmpIX86 || cmp instanceof CmpMemX86) ) + _inputs.set(1,cmp); + else + throw Utils.TODO(); + } + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.FLAGS_MASK; } + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + @Override public void asm(CodeGen code, SB sb) { + String src = code.reg(in(1)); + if( src!="FLAGS" ) sb.p(src); + } + + @Override public String op() { return "j"+_bop; } + + @Override public String comment() { + return "L"+cproj(1)._nid+", L"+cproj(0)._nid; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LeaX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LeaX86.java new file mode 100644 index 00000000..a49b8d7f --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LeaX86.java @@ -0,0 +1,42 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class LeaX86 extends MachConcreteNode implements MachNode { + final int _scale; + final long _offset; + LeaX86( Node add, Node base, Node idx, int scale, long offset ) { + super(add); + assert scale==1 || scale==2 || scale==4 || scale==8; + _inputs.pop(); + _inputs.pop(); + _inputs.push(base); + _inputs.push(idx); + _scale = scale; + _offset = offset; + } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "lea dst = base + 4*idx + 12" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = "); + if( in(1) != null ) + sb.p(code.reg(in(1))).p(" + "); + sb.p(code.reg(in(2))).p("*").p(_scale); + if( _offset!=0 ) sb.p(" + #").p(_offset); + } + + @Override public String op() { return "lea"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LoadX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LoadX86.java new file mode 100644 index 00000000..02a9a4fe --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/LoadX86.java @@ -0,0 +1,25 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.LoadNode; +import com.seaofnodes.simple.node.Node; + +public class LoadX86 extends MemOpX86 { + LoadX86( LoadNode ld, Node base, Node idx, int off, int scale ) { + super(ld,ld, base, idx, off, scale, 0); + } + + @Override public String op() { return "ld"+_sz; } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // General form: "ldN dst,[base + idx<<2 + 12]" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(","); + asm_address(code,sb); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemAddX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemAddX86.java new file mode 100644 index 00000000..418f63f3 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemAddX86.java @@ -0,0 +1,28 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.*; + +public class MemAddX86 extends MemOpX86 { + MemAddX86( StoreNode st, Node base, Node idx, int off, int scale, int imm, Node val ) { + super(st, st, base, idx, off, scale, imm, val ); + } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + + // General form: "add [base + idx<<2 + 12] += src" + @Override public void asm(CodeGen code, SB sb) { + asm_address(code,sb); + if( val()==null ) { + if( _imm != 1 && _imm != -1 ) sb.p(" += #").p(_imm); + } else sb.p(" += ").p(code.reg(val())); + } + + @Override public String op() { + return (_imm == 1 ? "inc" : (_imm == -1 ? "dec" : "add")) + _sz; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemOpX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemOpX86.java new file mode 100644 index 00000000..1e678a0b --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MemOpX86.java @@ -0,0 +1,87 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; +import java.io.ByteArrayOutputStream; +import java.lang.StringBuilder; +import java.util.BitSet; + +// Generic X86 memory operand base. +// inputs: +// ctrl - will appear for possibly RCE (i.e. array ops) +// mem - memory dependence edge +// base - Basic object pointer +// idx/null - Scaled index offset, or null if none +// val/null - Value to store, as part of a op-to-mem or op-from-mem. Null for loads, or if an immediate is being used +// Constants: +// offset - offset added to base. Can be zero. +// scale - scale on index; only 0,1,2,4,8 allowed, 0 is only when index is null +// imm - immediate value to store or op-to-mem, and only when val is null +public abstract class MemOpX86 extends MemOpNode implements MachNode { + final int _off; // Limit 32 bits + final int _scale; // Limit 0,1,2,3 + final int _imm; // Limit 32 bits + final char _sz = (char)('0'+(1<<_declaredType.log_size())); + MemOpX86( Node op, MemOpNode mop, Node base, Node idx, int off, int scale, int imm ) { + super(op,mop); + assert base._type instanceof TypeMemPtr && !(base instanceof AddNode); + assert (idx==null && scale==0) || (idx!=null && 0<= scale && scale<=3); + + // Copy memory parts from eg the LoadNode over the opcode, e.g. an Add + if( op != mop ) { + _inputs.set(0,mop.in(0)); // Control from mem op + _inputs.set(1,mop.in(1)); // Memory from mem op + _inputs.set(2,base); // Base handed in + } + + assert ptr() == base; + _inputs.setX(3,idx); + _off = off; + _scale = scale; + _imm = imm; + } + + // Store-based flavors have a value edge + MemOpX86( Node op, MemOpNode mop, Node base, Node idx, int off, int scale, int imm, Node val ) { + this(op,mop,base,idx,off,scale,imm); + _inputs.setX(4,val); + } + + Node idx() { return in(3); } + Node val() { return in(4); } + + @Override public StringBuilder _printMach(StringBuilder sb, BitSet visited) { return sb.append(".").append(_name); } + + @Override public String label() { return op(); } + @Override public Type compute() { throw Utils.TODO(); } + @Override public Node idealize() { throw Utils.TODO(); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { + if( i==1 ) return RegMask.EMPTY; // Memory + if( i==2 ) return x86_64_v2.RMASK; // base + if( i==3 ) return x86_64_v2.RMASK; // index + if( i==4 ) return x86_64_v2.RMASK; // value + throw Utils.TODO(); + } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { throw Utils.TODO(); } + + @Override public int encoding(ByteArrayOutputStream bytes) { throw Utils.TODO(); } + + + // "[base + idx<<2 + 12]" + SB asm_address(CodeGen code, SB sb) { + sb.p("[").p(code.reg(ptr())); + if( idx() != null ) { + sb.p("+").p(code.reg(idx())); + if( _scale != 0 ) + sb.p("<<").p(_scale); + } + if( _off != 0 ) + sb.p("+").p(_off); + return sb.p("]"); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulFX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulFX86.java new file mode 100644 index 00000000..aef813e1 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulFX86.java @@ -0,0 +1,30 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeFloat; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class MulFX86 extends MachConcreteNode implements MachNode { + MulFX86( Node mulf) { super(mulf); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.XMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "mulf dst *= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" * ").p(code.reg(in(2))); + } + + @Override public String op() { return "mulf"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulIX86.java new file mode 100644 index 00000000..7508cf4c --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulIX86.java @@ -0,0 +1,36 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class MulIX86 extends MachConcreteNode implements MachNode { + final TypeInteger _ti; + MulIX86(Node mul, TypeInteger ti) {super(mul); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + // assert i==1; + return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + // General form + // General form: "muli dst * #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" * #"); + _ti.print(sb); + } + + @Override public String op() { return "muli"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulX86.java new file mode 100644 index 00000000..15fd9496 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/MulX86.java @@ -0,0 +1,30 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class MulX86 extends MachConcreteNode implements MachNode { + + MulX86( Node mul ) { super(mul); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "mul dst *= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" * ").p(code.reg(in(2))); + } + + @Override public String op() { return "mul"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/NewX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/NewX86.java new file mode 100644 index 00000000..4396a917 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/NewX86.java @@ -0,0 +1,40 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import java.io.ByteArrayOutputStream; + +public class NewX86 extends NewNode implements MachNode { + + // A pre-zeroed chunk of memory. + NewX86( NewNode nnn ) { super(nnn); } + + // Register mask allowed on input i, the size + @Override public RegMask regmap(int i) { + // Size + if( i==1 ) return x86_64_v2.RDI_MASK; + // All the memory alias edges + return RegMask.EMPTY; + } + @Override public RegMask outregmap() { throw Utils.TODO(); } + + // Register mask allowed as a result. Pointer result in standard calling + // convention. + @Override public RegMask outregmap(int i) { + if( i == 1 ) return x86_64_v2.RET_MASK; + // All the memory aliases edges + return RegMask.EMPTY; + } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "alloc #bytes" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(size())); + } + + @Override public String op() { return "alloc"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/OrIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/OrIX86.java new file mode 100644 index 00000000..513962c2 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/OrIX86.java @@ -0,0 +1,44 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.MachConcreteNode; +import com.seaofnodes.simple.node.MachNode; +import com.seaofnodes.simple.node.Node; +import com.seaofnodes.simple.type.TypeInteger; + +import java.io.ByteArrayOutputStream; + +public class OrIX86 extends MachConcreteNode implements MachNode { + + final TypeInteger _ti; + OrIX86(Node or, TypeInteger ti) {super(or); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + //assert i==1; + return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "ori ori | #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" | #"); + _ti.print(sb); + } + + @Override public String op() { return "ori"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ParmX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ParmX86.java new file mode 100644 index 00000000..8a439644 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ParmX86.java @@ -0,0 +1,24 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.ParmNode; +import com.seaofnodes.simple.node.MachNode; +import java.io.ByteArrayOutputStream; + +public class ParmX86 extends ParmNode implements MachNode { + final RegMask _rmask; + ParmX86( ParmNode parm ) { + super(parm); + _rmask = x86_64_v2.callInMask(_idx); + } + + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // Register mask allowed as a result. Calling convention register + @Override public RegMask outregmap() { return _rmask; } + + // Encoding is appended into the byte array. Returns size + @Override public int encoding(ByteArrayOutputStream bytes) { return 0; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ProjX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ProjX86.java new file mode 100644 index 00000000..2849b232 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ProjX86.java @@ -0,0 +1,23 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.ProjNode; +import com.seaofnodes.simple.node.MachNode; +import java.io.ByteArrayOutputStream; + +public class ProjX86 extends ProjNode implements MachNode { + + ProjX86( ProjNode p ) { super(p); } + + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java new file mode 100644 index 00000000..e6219ff4 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java @@ -0,0 +1,48 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.FunNode; +import com.seaofnodes.simple.node.ReturnNode; +import com.seaofnodes.simple.node.MachNode; +import java.io.ByteArrayOutputStream; + +// Return +public class RetX86 extends ReturnNode implements MachNode { + RetX86( ReturnNode ret, FunNode fun ) { super(ret, fun); fun.setRet(this); } + + // Correct Nodes outside the normal edges + @Override public void postSelect() { + FunNode fun = (FunNode)rpc().in(0); + _fun = fun; + fun.setRet(this); + } + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + return switch( i ) { + case 0 -> RegMask.EMPTY; + case 1 -> RegMask.EMPTY; + case 2 -> x86_64_v2.RET_MASK; + case 3 -> RegMask.EMPTY; // RPC is always on stack + default -> throw Utils.TODO(); + }; + } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // Human-readable form appended to the SB. Things like the encoding, + // indentation, leading address or block labels not printed here. + // Just something like "ld4\tR17=[R18+12] // Load array base". + // General form: "op\tdst=src+src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(in(2))); + } + + @Override public String op() { return "ret"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SarX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SarX86.java new file mode 100644 index 00000000..3d907eef --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SarX86.java @@ -0,0 +1,40 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; +import java.lang.StringBuilder; + + +public class SarX86 extends MachConcreteNode implements MachNode { + final TypeInteger _ti; + SarX86(Node sar, TypeInteger ti) {super(sar); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + // assert i==1; + return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "sari dst << #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" >> #"); + _ti.print(sb); + } + + @Override public String op() { return "sari"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SetX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SetX86.java new file mode 100644 index 00000000..25367e03 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SetX86.java @@ -0,0 +1,39 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +// Compare immediate. Sets flags. +// Corresponds to the x86 instruction "sete && setne". +// Use result of comparison without jump. +public class SetX86 extends MachConcreteNode implements MachNode { + final String _bop; // One of <,<=,== + // Constructor expects input is an X86 and not an Ideal node. + SetX86( Node cmp, String bop ) { + super(cmp); + _inputs.setLen(1); // Pop the cmp inputs + // Replace with the matched cmp + _inputs.push(cmp); + _bop = bop; + } + + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.FLAGS_MASK; } + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)); + String src = code.reg(in(1)); + if( src!="FLAGS" ) sb.p(" = ").p(src); + } + + @Override public String op() { return "set"+_bop; } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShlIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShlIX86.java new file mode 100644 index 00000000..61e0d262 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShlIX86.java @@ -0,0 +1,38 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; + +public class ShlIX86 extends MachConcreteNode implements MachNode{ + final TypeInteger _ti; + ShlIX86(Node shl, TypeInteger ti) {super(shl); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + //assert i==1; + return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "shli dst << #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" << #"); + _ti.print(sb); + } + + @Override public String op() { return "shli"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShrIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShrIX86.java new file mode 100644 index 00000000..08cae93e --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/ShrIX86.java @@ -0,0 +1,39 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; +import java.util.BitSet; +import java.lang.StringBuilder; + +public class ShrIX86 extends MachConcreteNode implements MachNode{ + final TypeInteger _ti; + ShrIX86(Node shr, TypeInteger ti) {super(shr); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { + //assert i==1; + return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "shri dst >>> #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" >>> #"); + _ti.print(sb); + } + + @Override public String op() { return "shri"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/StoreX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/StoreX86.java new file mode 100644 index 00000000..c9191b87 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/StoreX86.java @@ -0,0 +1,30 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.node.Node; +import com.seaofnodes.simple.node.StoreNode; +import java.util.BitSet; + +public class StoreX86 extends MemOpX86 { + StoreX86( StoreNode st, Node base, Node idx, int off, int scale, int imm, Node val ) { + super(st,st, base, idx, off, scale, imm, val); + } + + @Override public String op() { return "st"+_sz; } + + @Override public StringBuilder _printMach(StringBuilder sb, BitSet visited) { + return sb.append(".").append(_name).append("=").append(val()==null ? _imm : val()).append(";"); + } + + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return RegMask.EMPTY; } + + // General form: "stN [base + idx<<2 + 12],val" + @Override public void asm(CodeGen code, SB sb) { + asm_address(code,sb).p(","); + if( val()==null ) sb.p("#").p(_imm); + else sb.p(code.reg(val())); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubFX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubFX86.java new file mode 100644 index 00000000..ed19ad79 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubFX86.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import java.io.ByteArrayOutputStream; + +public class SubFX86 extends MachConcreteNode implements MachNode { + SubFX86( Node subf) { super(subf); } + + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.XMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.XMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "subf dst -= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" - ").p(code.reg(in(2))); + } + + @Override public String op() { return "subf"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubX86.java new file mode 100644 index 00000000..cac59d3b --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/SubX86.java @@ -0,0 +1,29 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.TypeInteger; +import java.io.ByteArrayOutputStream; + +public class SubX86 extends MachConcreteNode implements MachNode { + SubX86( Node sub ) { super(sub); } + + // Register mask allowed on input i. + @Override public RegMask regmap(int i) { assert i==1 || i==2; return x86_64_v2.RMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form: "sub dst -= src" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" - ").p(code.reg(in(2))); + } + + @Override public String op() { return "sub"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/TFPX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/TFPX86.java new file mode 100644 index 00000000..bd88011f --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/TFPX86.java @@ -0,0 +1,41 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.*; +import com.seaofnodes.simple.node.ConstantNode; +import com.seaofnodes.simple.node.MachNode; +import com.seaofnodes.simple.type.Type; +import com.seaofnodes.simple.type.TypeFunPtr; + +import java.io.ByteArrayOutputStream; + +// Function constants +public class TFPX86 extends ConstantNode implements MachNode { + TFPX86( ConstantNode con ) { super(con); } + // Register mask allowed on input i. 0 for no register. + @Override public RegMask regmap(int i) { return RegMask.EMPTY; } + // General int registers + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // Human-readable form appended to the SB. Things like the encoding, + // indentation, leading address or block labels not printed here. + // Just something like "ld4\tR17=[R18+12] // Load array base". + // General form: "op\tdst=src+src" + @Override public void asm(CodeGen code, SB sb) { + String reg = code.reg(this); + if( _con == Type.NIL ) + sb.p(reg).p(",").p(reg); + else + _con.print(sb.p(reg).p(" #")); + } + + @Override public String op() { + if( _con == Type.NIL ) + return "xor"; + return "ldx"; // Some fancier encoding + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/XorIX86.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/XorIX86.java new file mode 100644 index 00000000..795fa814 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/XorIX86.java @@ -0,0 +1,42 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.MachConcreteNode; +import com.seaofnodes.simple.node.MachNode; +import com.seaofnodes.simple.node.Node; +import com.seaofnodes.simple.type.TypeInteger; + +import java.io.ByteArrayOutputStream; + +public class XorIX86 extends MachConcreteNode implements MachNode { + + final TypeInteger _ti; + XorIX86(Node xor, TypeInteger ti) {super(xor); _inputs.pop(); _ti = ti;} + + // Register mask allowed on input i. + // This is the normal calling convention + @Override public RegMask regmap(int i) { assert i==1; return x86_64_v2.WMASK; } + // Register mask allowed as a result. 0 for no register. + @Override public RegMask outregmap() { return x86_64_v2.WMASK; } + + + // Output is same register as input#1 + @Override public int twoAddress() { return 1; } + + // Encoding is appended into the byte array; size is returned + @Override public int encoding(ByteArrayOutputStream bytes) { + throw Utils.TODO(); + } + + // General form + // General form: "ori xori ^ #imm" + @Override public void asm(CodeGen code, SB sb) { + sb.p(code.reg(this)).p(" = ").p(code.reg(in(1))).p(" ^ #"); + _ti.print(sb); + } + + @Override public String op() { return "xori"; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java new file mode 100644 index 00000000..45ae19dc --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java @@ -0,0 +1,418 @@ +package com.seaofnodes.simple.node.cpus.x86_64_v2; + +import com.seaofnodes.simple.CodeGen; +import com.seaofnodes.simple.Machine; +import com.seaofnodes.simple.RegMask; +import com.seaofnodes.simple.Utils; +import com.seaofnodes.simple.node.*; +import com.seaofnodes.simple.type.*; + +public class x86_64_v2 extends Machine { + // X86-64 V2. Includes e.g. SSE4.2 and POPCNT. + @Override public String name() { return "x86_64_v2"; } + + public static int RAX = 0, RCX = 1, RDX = 2, RBX = 3, RSP = 4, RBP = 5, RSI = 6, RDI = 7; + public static int R08 = 8, R09 = 9, R10 = 10, R11 = 11, R12 = 12, R13 = 13, R14 = 14, R15 = 15; + public static int FLAGS = 32; + + public static int XMM0 = 16, XMM1 = 17, XMM2 = 18, XMM3 = 19, XMM4 = 20, XMM5 = 21, XMM6 = 22, XMM7 = 23; + public static int XMM8 = 24, XMM9 = 25, XMM10 = 26, XMM11 = 27, XMM12 = 28, XMM13 = 29, XMM14 = 30, XMM15 = 31; + + // General purpose register mask: pointers and ints, not floats + public static RegMask RMASK = new RegMask(0b1111111111111111); + // No RSP in the *write* general set. + public static RegMask WMASK = new RegMask(0b1111111111101111); + // Xmm register mask + public static RegMask XMASK = new RegMask( 0b1111111111111111L << XMM0); + + public static RegMask FLAGS_MASK = new RegMask(1L<<FLAGS); + + // Return single int/ptr register + public static RegMask RET_MASK = new RegMask(RAX); + + public static RegMask RDI_MASK = new RegMask(1L<<RDI); + public static RegMask RCX_MASK = new RegMask(1L<<RCX); + public static RegMask RDX_MASK = new RegMask(1L<<RDX); + public static RegMask R08_MASK = new RegMask(1L<<R08); + public static RegMask R09_MASK = new RegMask(1L<<R09); + public static RegMask RSI_MASK = new RegMask(1L<<RSI); + + // Calling conv metadata + public int GPR_COUNT_CONV_WIN64 = 4; // RCX, RDX, R9, R9 + public int XMM_COUNT_CONV_WIN64 = 4; // XMM0L, XMM1L, XMM2L, XMM3L + + public int GPR_COUNT_CONV_SYSTEM_V = 6; // RDI, RSI, RDX, RCX, R8, R9 + public int XMM_COUNT_CONV_SYSTEM_V = 4; // XMM0, XMM1, XMM2, XMM3 .... + // Human-readable name for a register number, e.g. "RAX". + // Hard crash for bad register number, fix yer bugs! + public static final String[] REGS = new String[] { + "rax" , "rcx" , "rdx" , "rbx" , "rsp" , "rbp" , "rsi" , "rdi" , + "r8" , "r9" , "r10" , "r11" , "r12" , "r13" , "r14" , "r15" , + "xmm0", "xmm1", "xmm2" , "xmm3" , "xmm4" , "xmm5" , "xmm6" , "xmm7" , + "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15", + }; + @Override public String reg( int reg ) { return REGS[reg]; } + + // Calling convention; returns a machine-specific register + // for incoming argument idx. + // index 0 for control, 1 for memory, real args start at index 2 + static int callInArg( int idx ) { + return switch( CodeGen.CODE._callingConv ) { + case CodeGen.CallingConv.SystemV -> callInArgSystemV(idx); + case CodeGen.CallingConv.Win64 -> callInArgWin64 (idx); + }; + } + + // WIN64(param passing) + static RegMask[] CALLINMASK_WIN64 = new RegMask[] { + RegMask.EMPTY, + RegMask.EMPTY, + RCX_MASK, + RDX_MASK, + R08_MASK, + R09_MASK, + RegMask.EMPTY, + RegMask.EMPTY + }; + static int[] CALLINARG_WIN64 = new int[] { + 0, // Control, no register + 0, // Memory, no register + RCX, // + RDX, + R08, + R09, + 0, + 0, + }; + static int callInArgWin64( int idx ) { return CALLINARG_WIN64[idx]; } + + // caller saved(win64) + public static final long WIN64_ABI_CALLER_SAVED = + (1L << RAX) | (1L << RCX) | (1L << RDX) | (1L << R08) | (1L << R09) | (1L << R10) | (1L << R11); + + // callee saved(win64) + public static final long WIN64_ABI_CALLEE_SAVED = ~WIN64_ABI_CALLER_SAVED; + + // SystemV(param passing) + static RegMask[] CALLINMASK_SYSTEMV = new RegMask[] { + RegMask.EMPTY, + RegMask.EMPTY, + RDI_MASK, + RSI_MASK, + RDX_MASK, + RCX_MASK, + R08_MASK, + R09_MASK + }; + static int[] CALLINARG_SYSTEMV = new int[] { + 0, // Control, no register + 0, // Memory, no register + RDI, + RSI, + RDX, + RCX, + R08, + R09, + }; + static int callInArgSystemV( int idx ) { return CALLINARG_SYSTEMV[idx]; } + + // caller saved(systemv) + // caller saved(win64) + public static final long SYSTEMV_ABI_CALLER_SAVED = + (1L << RAX) | (1L << RDI) | (1L << RSI) | (1L << RCX) | (1L << RDX) | (1L << R08) | (1L << R09) << (1L << R10) << (1L << R11); + // callee saved(systemv) + public static final long SYSTEMV_ABI_CALLE_SAVED = ~SYSTEMV_ABI_CALLER_SAVED; + + + static RegMask callInMask( int idx ) { + return switch( CodeGen.CODE._callingConv ) { + case CodeGen.CallingConv.SystemV -> CALLINMASK_SYSTEMV[idx]; + case CodeGen.CallingConv.Win64 -> CALLINMASK_WIN64 [idx]; + }; + } + + // Create a split op; any register to any register, including stack slots + @Override public Node split() { + throw Utils.TODO(); + } + + // Return a MachNode unconditional branch + @Override public CFGNode jump() { + throw Utils.TODO(); + } + + // Break an infinite loop + @Override public IfNode never( CFGNode ctrl ) { + throw Utils.TODO(); + } + + // Instruction selection + @Override public Node instSelect( Node n ) { + return switch( n ) { + case AddFNode addf -> addf(addf); + case AddNode add -> add(add); + case AndNode and -> and(and); + case BoolNode bool -> cmp(bool); + case CallEndNode cend -> new CallEndNode((CallNode)cend.in(0)); + case CallNode call -> call(call); + case CProjNode c -> new CProjNode(c); + case ConstantNode con -> con(con); + case DivFNode divf -> new DivFX86(divf); + case DivNode div -> new DivX86(div); + case FunNode fun -> new FunX86(fun); + case IfNode iff -> jmp(iff); + case LoadNode ld -> ld(ld); + case MemMergeNode mem -> new MemMergeNode(mem); + case MulFNode mulf -> new MulFX86(mulf); + case MulNode mul -> mul(mul); + case NewNode nnn -> new NewX86(nnn); + case OrNode or -> or(or); + case ParmNode parm -> new ParmX86(parm); + case PhiNode phi -> new PhiNode(phi); + case ProjNode prj -> prj(prj); + case ReadOnlyNode read -> new ReadOnlyNode(read); + case ReturnNode ret -> new RetX86(ret,ret.fun()); + case SarNode sar -> sar(sar); + case ShlNode shl -> shl(shl); + case ShrNode shr -> shr(shr); + case StartNode start -> new StartNode(start); + case StopNode stop -> new StopNode(stop); + case StoreNode st -> st(st); + case SubFNode subf -> new SubFX86(subf); + case SubNode sub -> sub(sub); + case ToFloatNode tfn -> i2f8(tfn); + case XorNode xor -> xor(xor); + + case LoopNode loop -> new LoopNode(loop); + case RegionNode region-> new RegionNode(region); + default -> throw Utils.TODO(); + }; + } + + + // Attempt a full LEA-style break down. + private Node add( AddNode add ) { + Node lhs = add.in(1); + Node rhs = add.in(2); + if( lhs instanceof LoadNode ld && ld.nOuts()==1 ) + return new AddMemX86(add,address(ld),ld.ptr(),idx,off,scale, imm(rhs),val); + + if( rhs instanceof LoadNode ld && ld.nOuts()==1 ) + throw Utils.TODO(); // Swap load sides + + // Attempt a full LEA-style break down. + // Returns one of AddX86, AddIX86, LeaX86, or LHS + if( rhs instanceof ConstantNode off && off._con instanceof TypeInteger toff ) { + if( lhs instanceof AddNode ladd ) + // ((base + (idx << scale)) + off) + return _lea(add,ladd.in(1),ladd.in(2),toff.value()); + if( lhs instanceof ShlNode shift ) + // (idx << scale) + off; no base + return _lea(add,null,shift,toff.value()); + + // lhs + rhs1 + if( toff.value()==0 ) return add; + return new AddIX86(add, toff); + } + return _lea(add,lhs,rhs,0); + } + + + private Node addf( AddFNode addf ) { + if( addf.in(1) instanceof LoadNode ld && ld.nOuts()==1 ) + return new AddFMemX86(addf,address(ld),ld.ptr(),idx,off,scale, addf.in(2)); + + if( addf.in(2) instanceof LoadNode ld && ld.nOuts()==1 ) + throw Utils.TODO(); // Swap load sides + + return new AddFX86(addf); + } + + + private Node _lea( Node add, Node base, Node idx, long off ) { + int scale = 1; + if( base instanceof ShlNode && !(idx instanceof ShlNode) ) throw Utils.TODO(); // Bug in canonicalization, should on RHS + if( idx instanceof ShlNode shift && shift.in(2) instanceof ConstantNode shfcon && + shfcon._con instanceof TypeInteger tscale && 0 <= tscale.value() && tscale.value() <= 3 ) { + idx = shift.in(1); + scale = 1 << ((int)tscale.value()); + } + // (base + idx) + off + return off==0 && scale==1 + ? new AddX86(add) + : new LeaX86(add,base,idx,scale,off); + } + + + private Node and(AndNode and) { + if( and.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti ) + return new AndIX86(and, ti); + throw Utils.TODO(); + } + + private Node call(CallNode call) { + if( call.fptr() instanceof ConstantNode con && con._con instanceof TypeFunPtr tfp ) + return new CallX86(call, tfp); + return new CallRX86( call ); + } + + // Because X86 flags, a normal ideal Bool is 2 X86 ops: a "cmp" and at "setz". + // Ideal If reading from a setz will skip it and use the "cmp" instead. + private Node cmp( BoolNode bool ) { + Node cmp = _cmp(bool); + return new SetX86(cmp,bool.op()); + } + private Node _cmp( BoolNode bool ) { + Node lhs = bool.in(1); + Node rhs = bool.in(2); + if( lhs instanceof LoadNode ld && ld.nOuts()==1 ) + return new CmpMemX86(bool,address(ld),ld.ptr(),idx,off,scale, imm(rhs),val,false); + + if( rhs instanceof LoadNode ld && ld.nOuts()==1 ) + return new CmpMemX86(bool,address(ld),ld.ptr(),idx,off,scale, imm(lhs),val,true); + + return rhs instanceof ConstantNode con && con._con instanceof TypeInteger ti + ? new CmpIX86(bool, ti) + : new CmpX86(bool); + } + + private Node con( ConstantNode con ) { + return switch( con._con ) { + case TypeInteger ti -> new IntX86(con); + case TypeFloat tf -> new FltX86(con); + case TypeFunPtr tfp -> new TFPX86(con); + case TypeMemPtr tmp -> new ConstantNode(con); + case TypeNil tn -> throw Utils.TODO(); + // TOP, BOTTOM, XCtrl, Ctrl, etc. Never any executable code. + case Type t -> new ConstantNode(con); + }; + } + + private Node i2f8(ToFloatNode tfn) { + assert tfn.in(1)._type instanceof TypeInteger ti; + return new I2f8X86(tfn); + } + + private Node jmp( IfNode iff ) { + // If/Bool combos will match to a Cmp/Set which sets flags. + // Most general arith ops will also set flags, which the Jmp needs directly. + // Loads do not set the flags, and will need an explicit TEST + if( !(iff.in(1) instanceof BoolNode) ) + iff.setDef(1,new BoolNode.EQ(iff.in(1),new ConstantNode(TypeInteger.ZERO))); + return new JmpX86(iff, ((BoolNode)iff.in(1)).op()); + } + + private Node ld( LoadNode ld ) { + return new LoadX86(address(ld),ld.ptr(),idx,off,scale); + } + + private Node mul(MulNode mul) { + return mul.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti + ? new MulIX86(mul, ti) + : new MulX86(mul); + } + + private Node or(OrNode or) { + if( or.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti) + return new OrIX86(or, ti); + throw Utils.TODO(); + } + + private Node prj( ProjNode prj ) { + return new ProjX86(prj); + } + + private Node sar( SarNode sar ) { + if( sar.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti) + return new SarX86(sar, ti); + throw Utils.TODO(); + } + + private Node shl( ShlNode shl ) { + if( shl.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti ) + return new ShlIX86(shl, ti); + throw Utils.TODO(); + } + + private Node shr(ShrNode shr) { + if( shr.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti ) + return new ShrIX86(shr, ti); + throw Utils.TODO(); + } + + private Node st( StoreNode st ) { + // Look for "*ptr op= val" + Node op = st.val(); + if( op instanceof AddNode ) { + if( op.in(1) instanceof LoadNode ld && + ld.in(0)==st.in(0) && + ld.mem()==st.mem() && + ld.ptr()==st.ptr() && + ld.off()==st.off() ) { + if( op instanceof AddNode ) { + return new MemAddX86(address(st),st.ptr(),idx,off,scale,imm(op.in(2)),val); + } + throw Utils.TODO(); + } + } + + return new StoreX86(address(st),st.ptr(),idx,off,scale,imm(st.val()),val); + } + + private Node sub( SubNode sub ) { + return sub.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti + ? new AddIX86(sub, TypeInteger.constant(-ti.value())) + : new SubX86(sub); + } + + private Node xor(XorNode xor) { + if(xor.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti) + return new XorIX86(xor, ti); + throw Utils.TODO(); + } + + + // Gather X86 addressing mode bits prior to constructing. This is a + // builder pattern, but saving the bits in a *local* *global* here to keep + // mess contained. + private static int off, scale, imm; + private static Node idx, val; + private <N extends MemOpNode> N address( N mop ) { + off = scale = imm = 0; // Reset + idx = val = null; + Node base = mop.ptr(); + // Skip/throw-away a ReadOnly, only used to typecheck + if( base instanceof ReadOnlyNode read ) base = read.in(1); + assert !(base instanceof AddNode) && base._type instanceof TypeMemPtr; // Base ptr always, not some derived + if( mop.off() instanceof AddNode add && add.in(2) instanceof ConstantNode con && con._con instanceof TypeInteger ti ) { + off = (int)ti.value(); + assert off == ti.value(); // In 32-bit range + idx = add.in(1); + if( idx instanceof ShlNode shift && shift.in(2) instanceof ConstantNode shfcon && + shfcon._con instanceof TypeInteger tscale && 0 <= tscale.value() && tscale.value() <= 3 ) { + idx = shift.in(1); + scale = (int)tscale.value(); + } + } else { + if( mop.off() instanceof ConstantNode con && con._con instanceof TypeInteger ti ) { + off = (int)ti.value(); + assert off == ti.value(); // In 32-bit range + } else { + idx = mop.off(); + } + } + return mop; + } + private int imm( Node xval ) { + assert val==null && imm==0; + if( xval instanceof ConstantNode con && con._con instanceof TypeInteger ti ) { + val = null; + imm = (int)ti.value(); + assert imm == ti.value(); // In 32-bit range + } else { + val = xval; + } + return imm; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/Field.java b/chapter20/src/main/java/com/seaofnodes/simple/type/Field.java new file mode 100644 index 00000000..ffa9d2a6 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/Field.java @@ -0,0 +1,74 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Represents a field in a struct. This is not a Type in the type system. + */ +public class Field extends Type { + + // The pair {fieldName,type} uniquely identifies a field. + + // Field name + public final String _fname; + // Type of the field + public final Type _type; + // Unique memory alias, not sensibly part of a "type" but very convenient here. + public final int _alias; + // Field must be written to exactly once, no more, no less + public final boolean _final; + + private Field(String fname, Type type, int alias, boolean xfinal ) { + super(TFLD); + _fname = fname; + _type = type; + _alias = alias; + _final = xfinal; + } + // Make with existing alias + public static Field make( String fname, Type type, int alias, boolean xfinal ) { + return new Field(fname,type,alias,xfinal).intern(); + } + public Field makeFrom( Type type ) { + return type == _type ? this : new Field(_fname,type,_alias,_final).intern(); + } + @Override public Field makeRO() { return _final ? this : make(_fname,_type.makeRO(),_alias,true); } + @Override public boolean isFinal() { return _final && _type.isFinal(); } + + public static final Field TEST = make("test",Type.NIL,-2,false); + public static final Field TEST2= make("test",Type.NIL,-2,true); + public static void gather(ArrayList<Type> ts) { ts.add(TEST); ts.add(TEST2); } + + @Override Field xmeet( Type that ) { + Field fld = (Field)that; // Invariant + assert _fname.equals(fld._fname); + assert _alias==fld._alias; + return make(_fname,_type.meet(fld._type),_alias,_final | fld._final); + } + + @Override + public Field dual() { return make(_fname,_type.dual(),_alias,!_final); } + + @Override public Field glb() { + Type glb = _type.glb(); + return (glb==_type && _final) ? this : make(_fname,glb,_alias,true); + } + + // Override in subclasses + int hash() { return _fname.hashCode() ^ _type.hashCode() ^ _alias ^ (_final ? 1024 : 0); } + + boolean eq(Type t) { + Field f = (Field)t; + return _fname.equals(f._fname) && _type==f._type && _alias==f._alias && _final==f._final; + } + + + @Override + public SB print( SB sb ) { + return _type.print(sb.p(_final?"":"!").p(_fname).p(":").p(_alias).p(" : ")); + } + + @Override public String str() { return _fname; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/Type.java b/chapter20/src/main/java/com/seaofnodes/simple/type/Type.java new file mode 100644 index 00000000..847235dd --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/Type.java @@ -0,0 +1,235 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; +import java.util.HashMap; + +/** + * These types are part of a Monotone Analysis Framework, + * @see <a href="https://www.cse.psu.edu/~gxt29/teaching/cse597s21/slides/08monotoneFramework.pdf">see for example this set of slides</a>. + * <p> + * The types form a lattice; @see <a href="https://en.wikipedia.org/wiki/Lattice_(order)">a symmetric complete bounded (ranked) lattice.</a> + * <p> + * This wild lattice theory will be needed later to allow us to easily beef up + * the analysis and optimization of the Simple compiler... but we don't need it + * now, just know that it is coming along in a later Chapter. + * <p>g + * One of the fun things here is that while the theory is deep and subtle, the + * actual implementation is darn near trivial and is generally really obvious + * what we're doing with it. Right now, it's just simple integer math to do + * simple constant folding e.g. 1+2 == 3 stuff. + */ + +public class Type { + static final HashMap<Type,Type> INTERN = new HashMap<>(); + + // ---------------------------------------------------------- + // Simple types are implemented fully here. "Simple" means: the code and + // type hierarchy are simple, not that the Type is conceptually simple. + static final byte TBOT = 0; // Bottom (ALL) + static final byte TTOP = 1; // Top (ANY) + static final byte TCTRL = 2; // Ctrl flow bottom + static final byte TXCTRL = 3; // Ctrl flow top (mini-lattice: any-xctrl-ctrl-all) + static final byte TNIL = 4; // low null of all flavors + static final byte TXNIL = 5; // high or choice null + static final byte TSIMPLE = 6; // End of the Simple Types + + static final byte TPTR = 7; // All nil-able scalar values + static final byte TINT = 8; // All Integers; see TypeInteger + static final byte TFLT = 9; // All Floats ; see TypeFloat + static final byte TMEMPTR =10; // Memory pointer to a struct type + static final byte TFUNPTR =11; // Function pointer; unique signature and code address (just a bit) + static final byte TTUPLE =12; // Tuples; finite collections of unrelated Types, kept in parallel + static final byte TMEM =13; // All memory (alias 0) or A slice of memory - with specific alias + static final byte TSTRUCT =14; // Structs; tuples with named fields + static final byte TFLD =15; // Named fields in structs + static final byte TRPC =16; // Return Program Control (Return PC or RPC) + + public final byte _type; + + public boolean is_simple() { return _type < TSIMPLE; } + private static final String[] STRS = new String[]{"Bot","Top","Ctrl","~Ctrl","null","~nil"}; + protected Type(byte type) { _type = type; } + + public static final Type BOTTOM = new Type( TBOT ).intern(); // ALL + public static final Type TOP = new Type( TTOP ).intern(); // ANY + public static final Type CONTROL = new Type( TCTRL ).intern(); // Ctrl + public static final Type XCONTROL = new Type( TXCTRL ).intern(); // ~Ctrl + public static final Type NIL = new Type( TNIL ).intern(); // low null of all flavors + public static final Type XNIL = new Type( TXNIL ).intern(); // high or choice null + public static Type[] gather() { + ArrayList<Type> ts = new ArrayList<>(); + ts.add(BOTTOM); + ts.add(CONTROL); + ts.add(NIL); + ts.add(XNIL); + TypeNil.gather(ts); + TypePtr.gather(ts); + TypeInteger.gather(ts); + TypeFloat.gather(ts); + TypeMemPtr.gather(ts); + TypeFunPtr.gather(ts); + TypeMem.gather(ts); + Field.gather(ts); + TypeStruct.gather(ts); + TypeTuple.gather(ts); + TypeRPC.gather(ts); + int sz = ts.size(); + for( int i = 0; i < sz; i++ ) + ts.add(ts.get(i).dual()); + return ts.toArray(new Type[ts.size()]); + } + + // Is high or on the lattice centerline. + public boolean isHigh () { return _type==TTOP || _type==TXCTRL || _type==TXNIL; } + public boolean isHighOrConst() { return isHigh() || isConstant(); } + + // Strict constant values, things on the lattice centerline. + // Excludes both high and low values + public boolean isConstant() { return _type==TNIL; } + + // ---------------------------------------------------------- + + // Notes on Type interning: At the moment it is not easy to reset the + // interned types because we hold static references to several types and + // these are scattered around. This means the INTERN cache will retain all + // types from every run of the Parser. For this to work correctly types + // must be rigorous about defining when they are the same. Also types need + // to be immutable once defined. The rationale for interning is + // *correctness* with cyclic type definitions. Simple structural recursive + // checks go exponential with merely sharing, but with cycles they will + // stack overflow and crash. Interning means we do not need to have sharing + // checks with every type compare, only during interning. + + // Factory method which interns "this" + public <T extends Type> T intern() { + T nnn = (T)INTERN.get(this); + if( nnn==null ) + INTERN.put(nnn=(T)this,this); + return nnn; + } + + private int _hash; // Hash cache; not-zero when set. + @Override + public final int hashCode() { + if( _hash!=0 ) return _hash; + _hash = hash(); + if( _hash==0 ) _hash = 0xDEADBEEF; // Bad hash from subclass; use some junk thing + return _hash; + } + // Override in subclasses + int hash() { return _type; } + + @Override + public final boolean equals( Object o ) { + if( o==this ) return true; + if( !(o instanceof Type t)) return false; + if( _type != t._type ) return false; + return eq(t); + } + // Overridden in subclasses; subclass can assume "this!=t" and java classes are same + boolean eq(Type t) { return true; } + + + // ---------------------------------------------------------- + public final Type meet(Type t) { + // Shortcut for the self case + if( t == this ) return this; + // Same-type is always safe in the subclasses + if( _type==t._type ) return xmeet(t); + // TypeNil vs TypeNil meet + if( this instanceof TypeNil ptr0 && t instanceof TypeNil ptr1 ) + return ptr0.nmeet(ptr1); + // Reverse; xmeet 2nd arg is never "is_simple" and never equal to "this". + if( is_simple() ) return this.xmeet(t ); + if( t.is_simple() ) return t .xmeet(this); + return Type.BOTTOM; // Mixing 2 unrelated types + } + + // Compute meet right now. Overridden in subclasses. + // Handle cases where 'this.is_simple()' and unequal to 't'. + // Subclassed xmeet calls can assert that '!t.is_simple()'. + Type xmeet(Type t) { + assert is_simple(); // Should be overridden in subclass + // ANY meet anything is thing; thing meet ALL is ALL + if( _type==TBOT || t._type==TTOP ) return this; + if( _type==TTOP || t._type==TBOT ) return t; + + // RHS TypeNil vs NIL/XNIL + if( _type== TNIL ) return t instanceof TypeNil ptr ? ptr.meet0() : (t._type==TXNIL ? TypePtr.PTR : BOTTOM); + if( _type== TXNIL ) return t instanceof TypeNil ptr ? ptr.meetX() : (t._type== TNIL ? TypePtr.PTR : BOTTOM); + // 'this' is only {TCTRL,TXCTRL} + // Other non-simple RHS things bottom out + if( !t.is_simple() ) return BOTTOM; + // If RHS is NIL/XNIL + if( t._type==TNIL || t._type==TXNIL ) return BOTTOM; + // Both are {TCTRL,TXCTRL} and unequal to each other + return _type==TCTRL || t._type==TCTRL ? CONTROL : XCONTROL; + } + + public Type dual() { + return switch( _type ) { + case TBOT -> TOP; + case TTOP -> BOTTOM; + case TCTRL ->XCONTROL; + case TXCTRL-> CONTROL; + case TNIL ->XNIL; + case TXNIL -> NIL; + default -> throw Utils.TODO(); // Should not reach here + }; + } + + // ---------------------------------------------------------- + // Our lattice is defined with a MEET and a DUAL. + // JOIN is dual of meet of both duals. + public final Type join(Type t) { + if( this==t ) return this; + return dual().meet(t.dual()).dual(); + } + + // True if this "isa" t; e.g. 17 isa TypeInteger.BOT + public boolean isa( Type t ) { return meet(t)==t; } + + // True if this "isa" t up to named structures + public boolean shallowISA( Type t ) { return isa(t); } + + public Type nonZero() { return TypePtr.NPTR; } + + // Make a zero version of this type, 0 for integers and null for pointers. + public Type makeZero() { return Type.NIL; } + + /** Compute greatest lower bound in the lattice */ + public Type glb() { return Type.BOTTOM; } + + // Is forward-reference + public boolean isFRef() { return false; } + + // All reachable struct Fields are final + public boolean isFinal() { return true; } + + // Make all reachable struct Fields final + public Type makeRO() { return this; } + + // ---------------------------------------------------------- + + // Size in bits to hold an instance of this type. + // Sizes are expected to be between 1 and 64 bits. + // Size 0 means this either takes no space (such as a known-zero field) + // or isn't a scalar to be stored in memory. + public int log_size() { throw Utils.TODO(); } + + // ---------------------------------------------------------- + // Useful in the debugger, which calls toString everywhere. + // This is a more verbose dev-friendly print. + @Override + public final String toString() { + return print(new SB()).toString(); + } + + public SB print(SB sb) { return sb.p(str()); } + public SB gprint(SB sb) { return print(sb); } + + // This is used by error messages, and is a shorted print. + public String str() { return STRS[_type]; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFloat.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFloat.java new file mode 100644 index 00000000..aa205149 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFloat.java @@ -0,0 +1,95 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Float Type + */ +public class TypeFloat extends Type { + + public final static TypeFloat ONE = constant(1.0); + public final static TypeFloat FZERO = constant(0.0); + public final static TypeFloat F32 = make(32, 0); + public final static TypeFloat F64 = make(64, 0); + + // - high -64, high -32, con 0, low +32, low +64 + public final byte _sz; + + /** + * The constant value or 0 + */ + public final double _con; + + private TypeFloat(byte sz, double con) { + super(TFLT); + _sz = sz; + _con = con; + } + private static TypeFloat make(int sz, double con) { + return new TypeFloat((byte)sz,con).intern(); + } + + public static TypeFloat constant(double con) { return make(0, con); } + + public static void gather(ArrayList<Type> ts) { ts.add(F64); ts.add(F32); ts.add(constant(3.141592653589793)); } + + @Override public String str() { + return switch( _sz ) { + case -64 -> "~flt"; + case -32 -> "~f32"; + case 0 -> ""+_con+((float)_con==_con ? "f" : ""); + case 32 -> "f32"; + case 64 -> "flt"; + default -> throw Utils.TODO(); + }; + } + private boolean isF32() { return ((float)_con)==_con; } + + @Override public boolean isHigh () { return _sz< 0; } + @Override public boolean isConstant() { return _sz==0; } + @Override public int log_size() { return _sz==32 || _sz==-32 ? 2 : 3; } + + public double value() { return _con; } + + @Override + public TypeFloat xmeet(Type other) { + TypeFloat f = (TypeFloat)other; + // Invariant from caller: 'this' != 'other' and same class (TypeFloat). + TypeFloat i = (TypeFloat)other; // Contract + // Larger size in i1, smaller in i0 + TypeFloat i0 = _sz < i._sz ? this : i; + TypeFloat i1 = _sz < i._sz ? i : this; + + if( i1._sz== 64 ) return F64; + if( i0._sz==-64 ) return i1; + if( i1._sz== 32 ) + return i0._sz==0 && !i0.isF32() ? F64 : F32; + if( i1._sz!= 0 ) return i1; + // i1 is a constant + if( i0._sz==-32 ) + return i1.isF32() ? i1 : F64; + // Since both are constants, and are never equals (contract) unequals + // constants fall to bottom + return i0.isF32() && i1.isF32() ? F32 : F64; + } + + @Override + public Type dual() { + return isConstant() ? this : make(-_sz,0); // Constants are a self-dual + } + + @Override public Type makeZero() { return FZERO; } + @Override public Type glb() { return F64; } + + @Override + int hash() { return (int)(Double.hashCode(_con) ^ _sz ^ (1<<17)); } + @Override + public boolean eq( Type t ) { + TypeFloat i = (TypeFloat)t; // Contract + // Allow NaN to check for equality + return _sz==i._sz && Double.doubleToLongBits(_con)==Double.doubleToLongBits(i._con); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFunPtr.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFunPtr.java new file mode 100644 index 00000000..be801c85 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeFunPtr.java @@ -0,0 +1,104 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; +import java.util.HashMap; + +/** + * Represents a Pointer to a function. + * <p></p> + * Functions have argument types and a return type, making a function + * signature. Within a signature there can be many instances of functions, and + * each function is labeled with a small integer constant. A TFP can represent + * a single signature and a set of functions; the set might contain only 1 + * function or the zeroth function (the null ptr). + * <p></p> + * Functions with different signatures cannot be mixed, and will result in a + * bottom function type, which can only be null-checked. + */ +public class TypeFunPtr extends TypeNil { + // A TypeFunPtr is Signature and a set of functions. + public final TypeTuple _sig; + public final Type _ret; + // Cheesy easy implementation for a small set; 1 bit per unique function + // within the TypeTuple. Can be upgraded to a BitSet for larger classes of + // functions. Negative means "these 63 concrete bits plus infinite unknown more" + public final long _fidxs; // 63 unique functions per signature + + private TypeFunPtr(byte nil, TypeTuple sig, Type ret, long fidxs) { + super(TFUNPTR,nil); + assert sig != null; + _sig = sig; + _ret = ret; + _fidxs = fidxs; + } + + public static TypeFunPtr make( byte nil, TypeTuple sig, Type ret, long fidxs ) { return new TypeFunPtr(nil,sig,ret,fidxs).intern(); } + public static TypeFunPtr make( boolean nil, TypeTuple sig, Type ret ) { return make((byte)(nil ? 3 : 2),sig,ret,-1); } + @Override TypeFunPtr makeFrom( byte nil ) { return nil ==_nil ? this : make( nil,_sig,_ret, _fidxs); } + public TypeFunPtr makeFrom( Type ret ) { return ret ==_ret ? this : make( _nil,_sig, ret, _fidxs); } + public TypeFunPtr makeFrom( int fidx ) { return make((byte)2, _sig,_ret,1L<<fidx ); } + + public static TypeFunPtr BOT = make((byte)3,TypeTuple.BOT,Type.BOTTOM,-1); + public static TypeFunPtr TEST = make((byte)2,TypeTuple.TEST,TypeInteger.BOT,1); + public static TypeFunPtr TEST0 = make((byte)3,TypeTuple.TEST,TypeInteger.BOT,3); + public static void gather(ArrayList<Type> ts) { ts.add(TEST); ts.add(TEST0); ts.add(BOT); } + + @Override + Type xmeet(Type t) { + TypeFunPtr that = (TypeFunPtr) t; + return TypeFunPtr.make(xmeet0(that),(TypeTuple)_sig.meet(that._sig), _ret.meet(that._ret), _fidxs | that._fidxs); + } + + @Override + public TypeFunPtr dual() { return TypeFunPtr.make(dual0(), _sig.dual(), _ret.dual(), ~_fidxs); } + + // RHS is NIL; do not deep-dual when crossing the centerline + @Override public Type meet0() { return _nil==3 ? this : make((byte)3,_sig,_ret,_fidxs); } + + @Override public TypeFunPtr glb() { return make((byte)3,_sig,_ret,-1L); } + + @Override public boolean isHigh () { return _nil <= 1 || (_nil==2 && _fidxs==0); } + @Override public boolean isConstant() { return (_nil==2 && Long.bitCount(_fidxs)==1) || (_nil==3 && _fidxs==0); } + + @Override public int log_size() { return 2; } // (1<<2)==4-byte pointers + + public Type arg(int i) { return _sig._types[i]; } + public long fidxs() { return _fidxs; } + public Type ret() { return _ret; } + public int nargs() { return _sig._types.length; } + + @Override + int hash() { return Utils.fold(_sig.hashCode() ^ _ret.hashCode() ^ _fidxs ^ super.hash()); } + + @Override + boolean eq(Type t) { + TypeFunPtr ptr = (TypeFunPtr)t; // Invariant + return _sig == ptr._sig && _ret == ptr._ret && _fidxs == ptr._fidxs && super.eq(ptr); + } + + @Override public String str() { return print(new SB()).toString(); } + @Override public SB print(SB sb) { return _print(sb,false,true); } + public SB print(SB sb, boolean n) { return _print(sb,false,n); } + @Override public SB gprint(SB sb) { return _print(sb,true ,true); } + private static SB _print(SB sb, boolean g, Type t) { return g ? t.gprint(sb) : t.print(sb); } + private SB _print(SB sb, boolean g, boolean n) { + sb.p(x()).p("{ "); + if( _sig._types!=null ) + for( Type t : _sig._types ) + _print(sb,g,t).p(" "); + _print(sb.p(g ? "→ " : "-> "),g,_ret).p(" #"); + if( isHigh() ) sb.p("~"); + long fidxs = isHigh() ? ~_fidxs : _fidxs; + String fidx = fidxs==0 ? "" + : Long.bitCount(fidxs) == 1 ? ""+Long.numberOfTrailingZeros(fidxs) + : fidxs == -1 ? "ALL" + : "b"+Long.toBinaryString(fidxs); // Just some function bits + return sb.p(fidx).p("}").p(q()); + } + + // Usage: for( long fidxs=fidxs(); fidxs!=0; fidxs=nextFIDX(fidxs) { int fidxs = Long.numberOfTrailingZeros(fidxs); ... } + public static long nextFIDX(long fidxs) { return fidxs & (fidxs-1); } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeInteger.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeInteger.java new file mode 100644 index 00000000..cccb85a5 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeInteger.java @@ -0,0 +1,128 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Integer Type + */ +public class TypeInteger extends Type { + + public final static TypeInteger ZERO= make(0,0); + public final static TypeInteger FALSE=ZERO; + public final static TypeInteger TRUE= make(1,1); + + public final static TypeInteger I1 = make(-1,0); + public final static TypeInteger I8 = make(-128,127); + public final static TypeInteger I16 = make(-32768,32767); + public final static TypeInteger I32 = make(-1L<<31,(1L<<31)-1); + public final static TypeInteger BOT = make(Long.MIN_VALUE,Long.MAX_VALUE); + public final static TypeInteger TOP = BOT.dual(); + + public final static TypeInteger U1 = make(0,1); + public final static TypeInteger BOOL= U1; + public final static TypeInteger U8 = make(0,255); + public final static TypeInteger U16 = make(0,65535); + public final static TypeInteger U32 = make(0,(1L<<32)-1); + + /** + * Describes an integer *range* - everything from min to max; both min and + * max are inclusive. If min==max, this is a constant. + * <br> + * If min <= max, this is a below center (towards bottom). + * If min > max, this is an above center (towards top). + */ + public final long _min, _max; + + private TypeInteger(long min, long max) { + super(TINT); + _min = min; + _max = max; + } + + // Strict non-zero contract + public static TypeInteger make(long lo, long hi) { return new TypeInteger(lo,hi).intern(); } + + public static TypeInteger constant(long con) { return make(con, con); } + + public static void gather(ArrayList<Type> ts) { ts.add(I32); ts.add(BOT); ts.add(U1); ts.add(I1); ts.add(U8); } + + @Override public String str() { + if( isConstant() ) return ""+_min; + long lo = _min, hi = _max; + String x = ""; + if( hi < lo ) { + lo = _max; hi = _min; + x = "~"; + } + return x+_str(lo,hi); + } + private static String _str(long lo, long hi) { + if( lo==Long.MIN_VALUE && hi==Long.MAX_VALUE ) return "int"; + if( lo== 0 && hi== 1 ) return "bool"; + if( lo== -1 && hi== 0 ) return "i1"; + if( lo== -128 && hi== 127 ) return "i8"; + if( lo== -32768 && hi== 32767 ) return "i16"; + if( lo== -1L<<31 && hi==(1L<<31)-1 ) return "i32"; + if( lo== 0 && hi== 255 ) return "u8"; + if( lo== 0 && hi== 65535 ) return "u16"; + if( lo== 0 && hi==(1L<<32)-1 ) return "u32"; + return "["+lo+"-"+hi+"]"; + } + + @Override public boolean isHigh () { return _min > _max; } + @Override public boolean isConstant() { return _min == _max; } + + @Override public int log_size() { + if( this==I8 || this==U8 || this==BOOL ) return 0; // 1<<0 == 1 bytes + if( this==I16 || this==U16 ) return 1; // 1<<1 == 2 bytes + if( this==I32 || this==U32 ) return 2; // 1<<2 == 4 bytes + if( this==BOT ) return 3; // 1<<3 == 8 bytes + if( isHighOrConst() ) return 0; + throw Utils.TODO(); + } + + public long value() { assert isConstant(); return _min; } + + // AND-mask of forced zeros. e.g. unsigned types will return their mask; + // u8 will return 0xFF. But also a range of 16-18 (0x10-0x12) will return + // 0x13 - no value in the range {16,17,18} will allow bit 0x04 to be set. + public long mask() { + if( isHigh() ) return 0; + if( isConstant() ) return _min; + // Those bit positions which differ min to max + long x = _min ^ _max; + // Highest '1' bit in the differ set. Since the range is from min to + // max, all values below here are possible. + long ff1 = Long.highestOneBit(x); + // Make a all-1's mask from ff1, and set over the same bits (either min + // or max is ok). + long mask = _min | (ff1-1) | ff1; + return mask; + } + + @Override + public Type xmeet(Type other) { + // Invariant from caller: 'this' != 'other' and same class (TypeInteger) + TypeInteger i = (TypeInteger)other; // Contract + return make(Math.min(_min,i._min), Math.max(_max,i._max)); + } + + @Override public TypeInteger dual() { return make(_max,_min); } + + @Override public TypeInteger nonZero() { + if( isHigh() ) return this; + if( this==ZERO ) return null; // No sane answer + if( _min==0 ) return make(1,Math.max(_max,1)); // specifically good on BOOL + if( _max==0 ) return make(_min,-1); + return this; + } + @Override public Type makeZero() { return ZERO; } + @Override public TypeInteger glb() { return BOT; } + @Override int hash() { return Utils.fold(_min) * Utils.fold(_max); } + @Override public boolean eq( Type t ) { + TypeInteger i = (TypeInteger)t; // Contract + return _min==i._min && _max==i._max; + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMem.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMem.java new file mode 100644 index 00000000..9ed4c7a8 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMem.java @@ -0,0 +1,74 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.lang.Long; +import java.util.BitSet; +import java.util.ArrayList; + + +/** + * Represents a slice of memory corresponding to a set of aliases + */ +public class TypeMem extends Type { + + // Which slice of memory? + // 0 means TOP, no slice. + // 0 means BOT, all memory. + // N means slice#N. + public final int _alias; + public final Type _t; // Memory contents, some scalar type + + private TypeMem(int alias, Type t) { + super(TMEM); + assert alias!=0 || (t==Type.TOP || t==Type.BOTTOM); + _alias = alias; + _t = t; + } + + public static TypeMem make(int alias, Type t) { return new TypeMem(alias,t).intern(); } + public static final TypeMem TOP = make(0, Type.TOP ); + public static final TypeMem BOT = make(0, Type.BOTTOM); + + public static void gather(ArrayList<Type> ts) { ts.add(make(1,Type.NIL)); ts.add(make(1,TypeInteger.ZERO)); ts.add(BOT); } + + @Override + TypeMem xmeet(Type t) { + TypeMem that = (TypeMem) t; // Invariant: TypeMem and unequal + if( this==TOP ) return that; + if( that==TOP ) return this; + if( this==BOT ) return BOT; + if( that==BOT ) return BOT; + int alias = _alias==that._alias ? _alias : 0; + Type mt = _t.meet(that._t); + return make(alias,mt); + } + + @Override + public Type dual() { + return make(_alias,_t.dual()); + } + + @Override public boolean isHigh() { return _t.isHigh(); } + @Override public Type glb() { return make(_alias,_t.glb()); } + + @Override int hash() { return 9876543 + _alias + _t.hashCode(); } + + @Override boolean eq(Type t) { + TypeMem that = (TypeMem) t; // Invariant + return _alias == that._alias && _t == that._t; + } + + @Override public SB print(SB sb) { + sb.p("#"); + if( _alias==0 ) return sb.p(_t._type==TTOP ? "TOP" : "BOT"); + return _t.print(sb.p(_alias).p(":")); + } + @Override public SB gprint(SB sb) { + sb.p("#"); + if( _alias==0 ) return sb.p(_t._type==TTOP ? "TOP" : "BOT"); + return _t.gprint(sb.p(_alias).p(":")); + } + + @Override public String str() { return toString(); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMemPtr.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMemPtr.java new file mode 100644 index 00000000..d7a8b17d --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeMemPtr.java @@ -0,0 +1,105 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Represents a Pointer to memory. + * + * Null is generic pointer to non-existent memory. + * *void is a non-Null pointer to all possible refs, both structs and arrays. + * Pointers can be to specific struct and array types, or a union with Null. + * The distinguished *$BOT ptr represents union of *void and Null. + * The distinguished *$TOP ptr represents the dual of *$BOT. + */ +public class TypeMemPtr extends TypeNil { + // A TypeMemPtr is pair (obj,nil) + // where obj is a TypeStruct, possibly TypeStruct.BOT/TOP + // where nil is an explicit null is allowed or not + + // Examples: + // (Person,false) - a not-nil Person + // (Person,true ) - a Person or a nil + // (BOT ,false) - a not-nil void* (unspecified struct) + // (TOP ,true ) - a nil + + public final TypeStruct _obj; + + private TypeMemPtr(byte nil, TypeStruct obj) { + super(TMEMPTR,nil); + assert obj!=null; + _obj = obj; + } + static TypeMemPtr make(byte nil, TypeStruct obj) { return new TypeMemPtr(nil, obj).intern(); } + public static TypeMemPtr makeNullable(TypeStruct obj) { return make((byte)3, obj); } + public static TypeMemPtr make(TypeStruct obj) { return new TypeMemPtr((byte)2, obj).intern(); } + public TypeMemPtr makeFrom(TypeStruct obj) { return obj==_obj ? this : make(_nil, obj); } + public TypeMemPtr makeNullable() { return makeFrom((byte)3); } + @Override TypeMemPtr makeFrom(byte nil) { return nil==_nil ? this : make(nil,_obj); } + @Override public TypeMemPtr makeRO() { return makeFrom(_obj.makeRO()); } + @Override public boolean isFinal() { return _obj.isFinal(); } + + // An abstract pointer, pointing to either a Struct or an Array. + // Can also be null or not, so 4 choices {TOP,BOT} x {nil,not} + public static TypeMemPtr BOT = make((byte)3, TypeStruct.BOT); + public static TypeMemPtr TOP = BOT.dual(); + public static TypeMemPtr NOTBOT = make((byte)2,TypeStruct.BOT); + + public static TypeMemPtr TEST= make((byte)2, TypeStruct.TEST); + public static void gather(ArrayList<Type> ts) { ts.add(NOTBOT); ts.add(BOT); ts.add(TEST); } + + @Override + public TypeNil xmeet(Type t) { + TypeMemPtr that = (TypeMemPtr) t; + return TypeMemPtr.make(xmeet0(that), (TypeStruct)_obj.meet(that._obj)); + } + + @Override + public TypeMemPtr dual() { return TypeMemPtr.make( dual0(), _obj.dual()); } + + // RHS is NIL; do not deep-dual when crossing the centerline + @Override Type meet0() { return _nil==3 ? this : make((byte)3,_obj); } + + + // True if this "isa" t up to named structures + @Override public boolean shallowISA( Type t ) { + if( !(t instanceof TypeMemPtr that) ) return false; + if( this==that ) return true; + if( xmeet0(that)!=that._nil ) return false; + if( _obj==that._obj ) return true; + if( _obj._name.equals(that._obj._name) ) + return true; // Shallow, do not follow matching names, just assume ok + throw Utils.TODO(); // return _obj.shallowISA(that._obj); + } + + @Override public TypeMemPtr glb() { return make((byte)3,_obj.glb()); } + // Is forward-reference + @Override public boolean isFRef() { return _obj.isFRef(); } + + @Override public int log_size() { return 2; } // (1<<2)==4-byte pointers + + @Override int hash() { return _obj.hashCode() ^ super.hash(); } + + @Override boolean eq(Type t) { + TypeMemPtr ptr = (TypeMemPtr)t; // Invariant + return _obj == ptr._obj && super.eq(ptr); + } + + @Override public String str() { + if( this== NOTBOT) return "*void"; + if( this== BOT) return "*void?"; + return x()+"*"+_obj.str()+q(); + } + + @Override public SB print(SB sb) { + if( this== NOTBOT) return sb.p("*void"); + if( this== BOT) return sb.p("*void?"); + return _obj.print(sb.p(x()).p("*")).p(q()); + } + @Override public SB gprint(SB sb) { + if( this== NOTBOT) return sb.p("*void"); + if( this== BOT) return sb.p("*void?"); + return _obj.gprint(sb.p(x()).p("*")).p(q()); + } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeNil.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeNil.java new file mode 100644 index 00000000..aee623ba --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeNil.java @@ -0,0 +1,62 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Nil-able Scalar types + */ +public abstract class TypeNil extends Type { + // 0 = high-subclass choice nil + // 1 = high-subclass no nil + // 2 = low -subclass no nil + // 3 = low -subclass also nil + final byte _nil; + + TypeNil(byte t, byte nil ) { super(t); _nil = nil; } + + public static void gather(ArrayList<Type> ts) { } + + abstract TypeNil makeFrom(byte nil); + + byte xmeet0(TypeNil that) { return (byte)Math.max(_nil,that._nil); } + + byte dual0() { return (byte)(3-_nil); } + + // RHS is NIL + abstract Type meet0(); + + // RHS is XNIL + Type meetX() { + return _nil==0 ? XNIL : (_nil<=2 ? TypePtr.NPTR : TypePtr.PTR); + } + + Type nmeet(TypeNil tn) { + // Invariants: both are TypeNil subclasses and unequal classes. + // If this is TypePtr, we went to TypePtr.nmeet and not here. + // If that is TypePtr, this is not (invariant); reverse and go again. + if( tn instanceof TypePtr ts ) return ts.nmeet(this); + + // Two mismatched TypeNil, no Scalar. + if( _nil==0 && tn._nil==0 ) return XNIL; + if( _nil<=2 && tn._nil<=2 ) return TypePtr.NPTR; + return TypePtr.PTR; + } + + @Override public boolean isHigh () { return _nil <= 1; } + @Override public boolean isConstant () { return false; } + @Override public boolean isHighOrConst() { return isHigh() || isConstant(); } + + @Override public Type glb() { return Type.NIL; } + + public boolean notNull() { return _nil==1 || _nil==2; } + public boolean nullable() { return _nil==3; } + + final String q() { return _nil == 1 || _nil == 2 ? "" : "?"; } + final String x() { return isHigh() ? "~" : ""; } + + int hash() { return _nil<<17; } + + boolean eq(TypeNil ptr) { return _nil == ptr._nil; } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypePtr.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypePtr.java new file mode 100644 index 00000000..e9091c25 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypePtr.java @@ -0,0 +1,53 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Represents a Scalar; a single register-sized value. + */ +public class TypePtr extends TypeNil { + + private TypePtr(byte nil) { super(TPTR,nil); } + + // An abstract pointer, pointing to either a Struct or an Array. + // Can also be null or not, so 4 choices {TOP,BOT} x {nil,not} + public static TypePtr XPTR = new TypePtr((byte)0).intern(); + public static TypePtr XNPTR= new TypePtr((byte)1).intern(); + public static TypePtr NPTR = new TypePtr((byte)2).intern(); + public static TypePtr PTR = new TypePtr((byte)3).intern(); + private static final TypePtr[] PTRS = new TypePtr[]{XPTR,XNPTR,NPTR,PTR}; + + public static void gather(ArrayList<Type> ts) { ts.add(PTR); ts.add(NPTR); } + + TypeNil makeFrom(byte nil) { throw Utils.TODO(); } + + @Override public TypeNil xmeet(Type t) { + TypePtr that = (TypePtr) t; + return PTRS[xmeet0(that)]; + } + + @Override public TypePtr dual() { return PTRS[dual0()]; } + + // High scalar loses, low scalar wins + @Override TypeNil nmeet(TypeNil tn) { + if( _nil==0 ) return tn; // High scalar loses + if( _nil==1 ) return tn.makeFrom(xmeet0(tn)); // High scalar loses + if( _nil==2 ) return tn._nil==3 ? PTR : NPTR; // Low scalar wins + return PTR; // this + } + + + // RHS is NIL + @Override Type meet0() { return isHigh() ? NIL : PTR; } + // RHS is XNIL + // 0->xscalar, 1->nscalar, 2->nscalar, 3->scalar + @Override Type meetX() { return _nil==0 ? XNIL : (_nil==3 ? PTR : NPTR); } + + @Override public TypePtr glb() { return PTR; } + + private static final String[] STRS = new String[]{"~ptr","~nptr","nptr","ptr"}; + @Override public String str() { return STRS[_nil]; } + @Override public SB print(SB sb) { return sb.p(str()); } +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeRPC.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeRPC.java new file mode 100644 index 00000000..ec2f5f0a --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeRPC.java @@ -0,0 +1,100 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; +import java.util.BitSet; +import java.util.HashSet; + +/** + * Return Program Control or Return PC or RPC + */ +public class TypeRPC extends Type { + + // A set of CallEndNode IDs (or StopNode); commonly just one. + // Basically a sparse bit set + final HashSet<Integer> _rpcs; + + // If true, invert the meaning of the bits + final boolean _any; + + private TypeRPC(boolean any, HashSet<Integer> rpcs) { + super(TRPC); + _any = any; + _rpcs = rpcs; + } + private static TypeRPC make(boolean any, HashSet<Integer> rpcs) { + return new TypeRPC(any,rpcs).intern(); + } + + public static TypeRPC constant(int cend) { + HashSet<Integer> rpcs = new HashSet<>(); + rpcs.add(cend); + return make(false,rpcs); + } + + public static final TypeRPC BOT = make(true,new HashSet<>()); + private static final TypeRPC TEST2 = constant(2); + private static final TypeRPC TEST3 = constant(2); + + public static void gather(ArrayList<Type> ts) { ts.add(BOT); ts.add(TEST2); ts.add(TEST3); } + + @Override public String str() { + if( _rpcs.isEmpty() ) + return _any ? "$[ALL]" : "$[]"; + if( _rpcs.size()==1 ) + for( Integer rpc : _rpcs ) + return _any ? "$[-"+rpc+"]" : "$["+rpc+"]"; + return "$["+(_any ? "-" : "")+_rpcs+"]"; + } + + @Override public boolean isConstant() { + return !_any && _rpcs.size()==1; + } + + @Override + public TypeRPC xmeet(Type other) { + TypeRPC rpc = (TypeRPC)other; + // If the two sets are equal, the _any must be unequal (invariant), + // so they cancel and all bits are set. + if( _rpcs.equals(rpc._rpcs) ) + return BOT; + // Classic union of bit sets (which might be infinite). + HashSet<Integer> lhs = _rpcs, rhs = rpc._rpcs; + // Smaller on left + if( lhs.size() > rhs.size() ) { lhs = rpc._rpcs; rhs = _rpcs; } + + HashSet<Integer> rpcs = new HashSet<>(); + boolean any = true; + // If both sets are infinite, intersect. + if( _any && rpc._any ) { + for( Integer i : lhs ) if( rhs.contains(i) ) rpcs.add(i); + + } else if( !_any && !rpc._any ) { + // if neither set is infinite, union. + rpcs.addAll( lhs ); + rpcs.addAll( rhs ); + any = false; + } else { + // If one is infinite, subtract the other from it. + HashSet<Integer> sub = _any ? rpc._rpcs : _rpcs; + HashSet<Integer> inf = _any ? _rpcs : rpc._rpcs; + for( Integer i : inf ) + if( inf.contains(i) && !sub.contains(i) ) + rpcs.add(i); + } + return make(any,rpcs); + } + + @Override + public Type dual() { return make(!_any,_rpcs); } + + @Override + int hash() { return _rpcs.hashCode() ^ (_any ? -1 : 0) ; } + @Override + public boolean eq( Type t ) { + TypeRPC rpc = (TypeRPC)t; // Contract + return _any==rpc._any && _rpcs.equals(rpc._rpcs); + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeStruct.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeStruct.java new file mode 100644 index 00000000..6c5cb8d4 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeStruct.java @@ -0,0 +1,248 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import com.seaofnodes.simple.Utils; +import java.util.ArrayList; + +/** + * Represents a struct type. + */ +public class TypeStruct extends Type { + + // A Struct has a name and a set of fields; the fields themselves have + // names, types and aliases. The name has no semantic meaning, but is + // useful for debugging. Briefly during parsing a Struct can be a + // forward-ref; in this case the _fields array is null. + + // Its illegal to attempt to load a field from a forward-ref struct. + + // Example: "int rez = new S.x; struct S { int x; } return rez;" // Error, S not defined + // Rewrite: "struct S { int x; } int rez = new S.x; return rez;" // Ok, no forward ref + // + // During the normal optimization run, struct types "bottom out" at further + // struct references, so we don't have to handle e.g. cyclic types. The + // "bottom out" is again the forward-ref struct. + public final String _name; + public final Field[] _fields; + + TypeStruct(String name, Field[] fields) { + super(TSTRUCT); + _name = name; + _fields = fields; + } + + // All fields directly listed + public static TypeStruct make(String name, Field... fields) { return new TypeStruct(name, fields).intern(); } + public static final TypeStruct TOP = make("$TOP",new Field[0]); + public static final TypeStruct BOT = make("$BOT",new Field[0]); + public static final TypeStruct TEST = make("test",new Field[]{Field.TEST}); + // Forward-ref version + public static TypeStruct makeFRef(String name) { return make(name, (Field[])null); } + // Make a read-only version + @Override public TypeStruct makeRO() { + if( isFinal() ) return this; + Field[] flds = new Field[_fields.length]; + for( int i=0; i<flds.length; i++ ) + flds[i] = _fields[i].makeRO(); + return make(_name,flds); + } + + // Array + public static TypeStruct makeAry(TypeInteger len, int lenAlias, Type body, int bodyAlias) { + assert body instanceof TypeInteger || body instanceof TypeFloat || (body instanceof TypeNil tn && tn.nullable()); + assert len.isa(TypeInteger.U32); + return make("[" + body.str() + "]", + Field.make("#" ,len , lenAlias,true ), + Field.make("[]",body,bodyAlias,false)); + } + + // A pair of self-cyclic types + private static final TypeStruct S1F = makeFRef("S1"); + private static final TypeStruct S2F = makeFRef("S2"); + public static final TypeStruct S1 = make("S1", Field.make("a", TypeInteger.BOT, -1, false), Field.make("s2",TypeMemPtr.make((byte)2,S2F),-2, false) ); + private static final TypeStruct S2 = make("S2", Field.make("b", TypeFloat .F64, -3, false), Field.make("s1",TypeMemPtr.make((byte)2,S1F),-4, false) ); + + private static final TypeStruct ARY = makeAry(TypeInteger.U32,-1,TypeInteger.BOT,-2); + + public static void gather(ArrayList<Type> ts) { ts.add(TEST); ts.add(BOT); ts.add(S1); ts.add(S2); ts.add(ARY); } + + // Find field index by name + public int find(String fname) { + for( int i=0; i<_fields.length; i++ ) + if( _fields[i]._fname.equals(fname) ) + return i; + return -1; + } + public Field field(String fname) { + int idx = find(fname); + return idx== -1 ? null : _fields[idx]; + } + + // Find field index by alias + public int findAlias( int alias ) { + for( int i=0; i<_fields.length; i++ ) + if( _fields[i]._alias==alias ) + return i; + return -1; + } + + + @Override + Type xmeet(Type t) { + TypeStruct that = (TypeStruct) t; + if( this==TOP ) return that; + if( that==TOP ) return this; + if( this==BOT ) return BOT; + if( that==BOT ) return BOT; + + // Within the same compilation unit, struct names are unique. If the + // names differ, its different structs. Across many compilation units, + // structs with the same name but different field layouts can be + // interned... which begs the question: + // "What is the meet of structs from two different compilation units?" + // And the answer is: "don't ask". + if( !_name.equals(that._name) ) + return BOT; // It's a struct; that's about all we know + if( this._fields==null ) return that; + if( that._fields==null ) return this; + if( _fields.length != that._fields.length ) return BOT; + // Just do field meets + Field[] flds = new Field[_fields.length]; + for( int i=0; i<_fields.length; i++ ) { + Field f0 = _fields[i], f1 = that._fields[i]; + if( !f0._fname.equals(f1._fname) || f0._alias != f1._alias ) + return BOT; + flds[i] = (Field)f0.meet(f1); + } + return make(_name,flds); + } + + @Override + public TypeStruct dual() { + if( this==TOP ) return BOT; + if( this==BOT ) return TOP; + if( _fields == null ) return this; + Field[] flds = new Field[_fields.length]; + for( int i=0; i<_fields.length; i++ ) + flds[i] = _fields[i].dual(); + return make(_name,flds); + } + + // Keeps the same struct, but lower-bounds all fields. + @Override public TypeStruct glb() { + if( _glb() ) return this; + // Need to glb each field + Field[] flds = new Field[_fields.length]; + for( int i=0; i<_fields.length; i++ ) + flds[i] = _fields[i].glb(); + return make(_name,flds); + } + private boolean _glb() { + if( _fields!=null ) + for( Field f : _fields ) + if( f.glb() != f ) + return false; + return true; + } + + // Is forward-reference + @Override public boolean isFRef() { return _fields==null; } + // All fields are final + @Override public boolean isFinal() { + if( _fields==null ) return true; + for( Field fld : _fields ) + if( !fld.isFinal() ) + return false; + return true; + } + + @Override + boolean eq(Type t) { + TypeStruct ts = (TypeStruct)t; // Invariant + if( !_name.equals(ts._name) ) + return false; + if( _fields == ts._fields ) return true; + if( _fields==null || ts._fields==null ) return false; + if( _fields.length!=ts._fields.length ) + return false; + for( int i = 0; i < _fields.length; i++ ) + if( _fields[i] != ts._fields[i] ) + return false; + return true; + } + + @Override + int hash() { + long hash = _name.hashCode(); + if( _fields != null ) + for( Field f : _fields ) + hash = Utils.rot(hash,13) ^ f.hashCode(); + return Utils.fold(hash); + } + + @Override + public SB print(SB sb) { + sb.p(_name); + if( _fields == null ) return sb; // Forward reference struct, just print the name + sb.p(" {"); + for( Field f : _fields ) + f._type.print(sb).p(f._final ? " " : " !").p(f._fname).p("; "); + return sb.p("}"); + } + @Override public SB gprint( SB sb ) { return sb.p(_name); } + + @Override public String str() { return _name; } + + + public boolean isAry() { return _fields.length==2 && _fields[1]._fname.equals("[]"); } + + public int aryBase() { + assert isAry(); + if( _offs==null ) _offs = offsets(); + return _offs[1]; + } + public int aryScale() { + assert isAry(); + return _fields[1]._type.log_size(); + } + + + // Field offsets as packed byte offsets + private int[] _offs; // Lazily computed and not part of the type semantics + public int offset(int idx) { + if( _offs==null ) _offs = offsets(); + return _offs[idx]; + } + private int[] offsets() { // Field byte offsets + // Compute a layout for a collection of fields + assert _fields != null; // No forward refs + + // Array layout is different: len,[pad],body... + if( isAry() ) + return _offs = new int[]{ 0, _fields[1]._type.log_size() < 3 ? 4 : 8 }; + + // Compute a layout + int[] cnts = new int[4]; // Count of fields at log field size + for( Field f : _fields ) + cnts[f._type.log_size()]++; // Log size is 0(byte), 1(i16/u16), 2(i32/f32), 3(i64/dbl) + int off = 0, idx = 0; // Base common struct fields go here, e.g. Mark/Klass + // Compute offsets to the start of each power-of-2 aligned fields. + int[] offs = new int[4]; + for( int i=3; i>=0; i-- ) { + offs[i] = off; + off += cnts[i]<<i; + } + // Assign offsets to all fields. + // Really a hidden radix sort. + _offs = new int[_fields.length+1]; + for( Field f : _fields ) { + int log = f._type.log_size(); + _offs[idx++] = offs[log]; // Field offset + offs[log] += 1<<log; // Next field offset at same alignment + cnts[log]--; // Count down, should be all zero at end + } + _offs[_fields.length] = (off+7)& ~7; // Round out to max alignment + return _offs; + } + +} diff --git a/chapter20/src/main/java/com/seaofnodes/simple/type/TypeTuple.java b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeTuple.java new file mode 100644 index 00000000..6d3a2b69 --- /dev/null +++ b/chapter20/src/main/java/com/seaofnodes/simple/type/TypeTuple.java @@ -0,0 +1,97 @@ +package com.seaofnodes.simple.type; + +import com.seaofnodes.simple.SB; +import java.util.ArrayList; + +public class TypeTuple extends Type { + + public final Type[] _types; + + private TypeTuple(Type[] types) { super(TTUPLE); _types = types; } + public static TypeTuple make(Type... types) { return new TypeTuple(types).intern(); } + + public static final TypeTuple BOT = new TypeTuple(new Type[0]).intern(); + public static final TypeTuple TOP = new TypeTuple(null).intern(); + + public static final TypeTuple TEST = make(TypeInteger.BOT,TypeMemPtr.TEST); + public static final TypeTuple START= make(Type.CONTROL,TypeMem.TOP,TypeInteger.BOT); + public static final TypeTuple MAIN = make(TypeInteger.BOT); + public static final TypeTuple RET = make(Type.CONTROL,TypeMem.BOT,Type.BOTTOM); + + public static final TypeTuple IF_BOTH = make(new Type[]{Type. CONTROL,Type. CONTROL}); + public static final TypeTuple IF_NEITHER = make(new Type[]{Type.XCONTROL,Type.XCONTROL}); + public static final TypeTuple IF_TRUE = make(new Type[]{Type. CONTROL,Type.XCONTROL}); + public static final TypeTuple IF_FALSE = make(new Type[]{Type.XCONTROL,Type. CONTROL}); + + public static void gather(ArrayList<Type> ts) { ts.add(BOT); ts.add(TEST); ts.add(START); ts.add(MAIN); ts.add(IF_TRUE); } + + @Override + Type xmeet(Type other) { + TypeTuple tt = (TypeTuple)other; // contract from xmeet + if( this==TOP ) return other; + if( tt ==TOP ) return this ; + if( _types.length != tt._types.length ) + return BOT; + Type[] ts = new Type[_types.length]; + for( int i=0; i<_types.length; i++ ) + ts[i] = _types[i].meet(tt._types[i]); + return make(ts); + } + + @Override public TypeTuple dual() { + if( this==TOP ) return BOT; + if( this==BOT ) return TOP; + Type[] ts = new Type[_types.length]; + for( int i=0; i<_types.length; i++ ) + ts[i] = _types[i].dual(); + return make(ts); + } + + @Override + public Type glb() { + Type[] ts = new Type[_types.length]; + for( int i=0; i<_types.length; i++ ) + ts[i] = _types[i].glb(); + return make(ts); + } + + @Override public String str() { return print(new SB()).toString(); } + + @Override public SB print(SB sb) { + if( this==TOP ) return sb.p("[TOP]"); + if( this==BOT ) return sb.p("[BOT]"); + sb.p("[ "); + for( Type t : _types ) + t.print(sb).p(", "); + return sb.unchar(2).p("]"); + } + @Override public SB gprint(SB sb) { + if( this==TOP ) return sb.p("[TOP]"); + if( this==BOT ) return sb.p("[BOT]"); + sb.p("[ "); + for( Type t : _types ) + t.gprint(sb).p(", "); + return sb.unchar(2).p("]"); + } + + @Override + int hash() { + int sum = 0; + if( _types!=null ) for( Type type : _types ) sum ^= type.hashCode(); + return sum; + } + + @Override + boolean eq( Type t ) { + TypeTuple tt = (TypeTuple)t; // Contract + if( _types==null && tt._types==null ) return true; + if( _types==null || tt._types==null ) return false; + if( _types.length != tt._types.length ) return false; + for( int i=0; i<_types.length; i++ ) + if( _types[i]!=tt._types[i] ) + return false; + return true; + } + + +} |