Small peep to drop simple store-updates out of loops

5 files changed, 53 insertions, 5 deletions

diff --git a/chapter25/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java b/chapter25/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java
index 3e618472..dfe8794d 100644
--- a/chapter25/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java
+++ b/chapter25/src/main/java/com/seaofnodes/simple/node/MemMergeNode.java
@@ -104,6 +104,9 @@ public class MemMergeNode extends Node {
 
         // Collapse stacked merged-mem
         if( in(1) instanceof MemMergeNode mem ) {
+            // A dead memory cycle can lose its last use when we bypass mem.
+            // Keep the peephole's result alive until the worklist sees it.
+            keep();
             // Goal is to swap my default mem with mem's default mem
             for( int i=2; i<mem.nIns(); i++ ) {
                 if( mem.in(i) != null ) {
@@ -114,7 +117,7 @@ public class MemMergeNode extends Node {
             }
             CodeGen.CODE.add(mem.in(1));
             setDef(1,mem.in(1));
-            return this;
+            return unkeep();
         }
 
         return progress ? this : null;
diff --git a/chapter25/src/main/java/com/seaofnodes/simple/node/MemPhiNode.java b/chapter25/src/main/java/com/seaofnodes/simple/node/MemPhiNode.java
index e8e2592c..cdfaaf97 100644
--- a/chapter25/src/main/java/com/seaofnodes/simple/node/MemPhiNode.java
+++ b/chapter25/src/main/java/com/seaofnodes/simple/node/MemPhiNode.java
@@ -61,6 +61,11 @@ public class MemPhiNode extends PhiNode {
                 setDef(i,mem.alias(_alias));
                 return this;
             }
+        if( nIns()==3 && region() instanceof LoopNode loop && !loop.inProgress() &&
+            in(2) instanceof StoreNode st && st.mem()==this ) {
+            Node sink = st.sink(this);
+            if( sink!=null ) return sink;
+        }
         return super.idealize();
     }
 
diff --git a/chapter25/src/main/java/com/seaofnodes/simple/node/StoreNode.java b/chapter25/src/main/java/com/seaofnodes/simple/node/StoreNode.java
index 415ade34..4917e890 100644
--- a/chapter25/src/main/java/com/seaofnodes/simple/node/StoreNode.java
+++ b/chapter25/src/main/java/com/seaofnodes/simple/node/StoreNode.java
@@ -249,6 +249,46 @@ public class StoreNode extends MemOpNode {
         };
     }
 
+    // Phi(mem, Store(phi,ptr,off,val)) -> Store(mem,ptr,off,Phi(load(mem),val)).
+    // With no observers of the backedge store, only the last value matters.
+    // The entry load preserves the field even when the loop takes zero trips.
+    Node sink(MemPhiNode phi) {
+        phi.addDepForwards(this);
+        if( nOuts()!=1 || _alias==1 || _size==0 || _init ) return null;
+        phi.addDep(ptr());
+        // Start with a fixed field of an allocation dominating the loop.
+        // In particular, do not sink stores through a loop-varying address.
+        if( !(ptr() instanceof ProjNode p) || !(p.in(0) instanceof NewNode obj) ||
+            !(ptr()._type instanceof TypeMemPtr tmp) || tmp._obj.isAry() ||
+            !(off() instanceof ConstantNode) || err()!=null ) return null;
+        LoopNode loop = (LoopNode)phi.region();
+        // Bulk splitting identifies precise slices by Region/alias. Finish it
+        // before replacing this memory point with a Store.
+        for( Node use : loop._outputs )
+            if( use instanceof BulkMemPhiNode ) {
+                phi.addDepForwards(use);
+                return null;
+            }
+        phi.addDep(obj);
+        CFGNode ctrl = loop.entry();
+        while( ctrl!=null && ctrl!=obj.cfg0() ) {
+            phi.addDep(ctrl);
+            ctrl=ctrl.idom(phi);
+        }
+        if( ctrl==null ) return null;
+
+        Node mem = phi.in(1);
+        Node init = new LoadNode(_loc,_name,_alias,null,mem,ptr(),off()).peephole();
+        Node value = CodeGen.CODE.add(new PhiNode(_name,loop,init,val()).peephole());
+        Node sink = new StoreNode(_loc,_name,_alias,_con,null,mem,ptr(),off(),value,_init,_size);
+        // Break the memory cycle, killing the old store. Keep the Phi alive
+        // even if that store was its last use, until our caller replaces it.
+        phi.keep();
+        phi.setDef(2,mem);
+        phi.unkeep();
+        return sink;
+    }
+
     // Check that "mem" has no uses except "this"
     private boolean checkOnlyUse(Node mem) {
         if( mem.nOuts()==1 ) return true;
diff --git a/chapter25/src/test/java/com/seaofnodes/simple/Chapter21Test.java b/chapter25/src/test/java/com/seaofnodes/simple/Chapter21Test.java
index bfce93b9..4d610a42 100644
--- a/chapter25/src/test/java/com/seaofnodes/simple/Chapter21Test.java
+++ b/chapter25/src/test/java/com/seaofnodes/simple/Chapter21Test.java
@@ -283,8 +283,8 @@ return new _s0.v1;
     @Test public void testInfinite() {
         String src = "struct _S { int i; }; !_S !s = new _S; while(1) s.i++; return s.i;";
         testCPU(src,"x86_64_v2", "SystemV",0,"return Top;");
-        testCPU(src,"riscv"    , "SystemV",2,"return Top;");
-        testCPU(src,"arm"      , "SystemV",2,"return Top;");
+        testCPU(src,"riscv"    , "SystemV",0,"return Top;");
+        testCPU(src,"arm"      , "SystemV",0,"return Top;");
     }
 
     @Test
diff --git a/chapter25/src/test/java/com/seaofnodes/simple/Chapter23Test.java b/chapter25/src/test/java/com/seaofnodes/simple/Chapter23Test.java
index 7c46567a..053621bd 100644
--- a/chapter25/src/test/java/com/seaofnodes/simple/Chapter23Test.java
+++ b/chapter25/src/test/java/com/seaofnodes/simple/Chapter23Test.java
@@ -279,7 +279,7 @@ return _s.peek('q');
         CodeGen code = new CodeGen(src).parse().opto().typeCheck();
         assertEquals("return Phi(Region,0,1);", code.print());
         assertEquals("1", Eval2.eval(code, 0));
-        testCPU(src,"x86_64_v2", "win64",1,null);
+        testCPU(src,"x86_64_v2", "win64",23,null);
     };
 
     @Test
@@ -334,7 +334,7 @@ return _s.require('[');
         CodeGen code = new CodeGen(src).parse().opto().typeCheck();
         assertEquals("return (.[]==91);", code.print());
         assertEquals("1", Eval2.eval(code, 0));
-        testCPU(src,"x86_64_v2", "win64", 0, null);
+        testCPU(src,"x86_64_v2", "win64",4, null);
     };