Fix sieve

RA returns maxSlot+1
FunX86 prints as `subi`, uses Sub; Ret uses Add
X86 knows that always RPC is passed on stack "as if" a stack arg.
sieve uses unsigned ints.
Driver expects an array pointer

5 files changed, 17 insertions, 21 deletions

diff --git a/chapter21/src/main/java/com/seaofnodes/simple/codegen/RegAlloc.java b/chapter21/src/main/java/com/seaofnodes/simple/codegen/RegAlloc.java
index 9042de3b..c84e6011 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/codegen/RegAlloc.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/codegen/RegAlloc.java
@@ -550,7 +550,7 @@ public class RegAlloc {
                 Node n = bb.out(j);
                 if( lrg(n)!=null ) {
                     int slot = _code._mach.stackSlot(lrg(n)._reg);
-                    maxSlot = Math.max(maxSlot,slot);
+                    maxSlot = Math.max(maxSlot,slot+1);
                 }
 
                 if( !(n instanceof SplitNode lo) ) continue;
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/FunNode.java b/chapter21/src/main/java/com/seaofnodes/simple/node/FunNode.java
index 8db30e7d..2e32b75e 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/FunNode.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/FunNode.java
@@ -184,7 +184,7 @@ public class FunNode extends RegionNode {
     public short _maxSlot = -1;
     // Max argument slot, set during early Encoding.
     public short _maxArgSlot = -1;
-    // Frame adjust, including padding, st during early Encoding
+    // Frame adjust, including padding, set during early Encoding
     public short _frameAdjust;
 
 
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java
index 923fa362..15805fee 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/FunX86.java
@@ -7,7 +7,7 @@ import com.seaofnodes.simple.node.MachNode;
 
 public class FunX86 extends FunNode implements MachNode {
     FunX86( FunNode fun ) { super(fun); }
-    @Override public String op() { return "addi"; }
+    @Override public String op() { return "subi"; }
     @Override public RegMask regmap(int i) { return null; }
     @Override public RegMask outregmap() { return null; }
     @Override public void postSelect(CodeGen code) { code.link(this);  }
@@ -16,15 +16,16 @@ public class FunX86 extends FunNode implements MachNode {
     // X86, with too many args & spills:
     // | CALLER |
     // |  argN  | // slot 1, required by callER
-    // +--------+
+    // +--------+ // Old frame
     // |  RPC   | // slot 0, required by callER
+    // +--------+ // New frame top
+    // |  PAD16 |
     // | callee | // slot 3, callEE
     // | callee | // slot 2, callEE
-    // |  PAD16 |
     // +--------+
     @Override public void encoding( Encoding enc ) {
         // Stack frame size: _maxSlot - max(arg), padded to 16.
-        _maxArgSlot = x86_64_v2.maxSlot(enc._fun.sig());
+        _maxArgSlot = x86_64_v2.maxArgSlot(enc._fun.sig());
         int sz = _maxSlot - _maxArgSlot;
         // If non-leaf function, pad to 16b
         if( _hasCalls ) sz = ((sz+1) & -2)+1;
@@ -34,11 +35,11 @@ public class FunX86 extends FunNode implements MachNode {
 
         // opcode: 0x83, addi rsp with immediate 8
         enc.add1( x86_64_v2.REX_W ).add1( 0x83 );
-        enc.add1( x86_64_v2.modrm(x86_64_v2.MOD.DIRECT, 0, x86_64_v2.RSP) );
+        enc.add1( x86_64_v2.modrm(x86_64_v2.MOD.DIRECT, 0b101, x86_64_v2.RSP) );
         enc.add1(sz*8);
     }
 
     @Override public void asm(CodeGen code, SB sb) {
-        sb.p("rsp += #").p(_frameAdjust*8);
+        sb.p("rsp -= #").p(_frameAdjust*8);
     }
 }
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java
index a9da15c6..4faf7cb4 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/RetX86.java
@@ -22,7 +22,7 @@ public class RetX86 extends ReturnNode implements MachNode {
         if( frameAdjust > 0 ) {
             enc.add1( x86_64_v2.REX_W ).add1( 0x83 );
             enc.add1( x86_64_v2.modrm(x86_64_v2.MOD.DIRECT, 0, x86_64_v2.RSP) );
-            enc.add1(frameAdjust*-8);
+            enc.add1(frameAdjust*8);
         }
         enc.add1(0xC3);
     }
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java
index d9bca291..4770bfbb 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/x86_64_v2/x86_64_v2.java
@@ -59,10 +59,10 @@ public class x86_64_v2 extends Machine {
     public static int REX_WB = 0x49;
 
     public enum MOD {
-        INDIRECT, //  [mem]
-        INDIRECT_disp8, // [mem + 0x12]
-        INDIRECT_disp32,// [mem + 0x12345678]
-        DIRECT,          // mem
+        INDIRECT,               //  [mem]
+        INDIRECT_disp8,         // [mem + 0x12]
+        INDIRECT_disp32,        // [mem + 0x12345678]
+        DIRECT,                 // mem
     };
 
     // opcode included here
@@ -249,8 +249,8 @@ public class x86_64_v2 extends Machine {
         throw Utils.TODO(); // Pass on stack slot
     }
 
-    // Return the max stack slot used by this signature, or 0
-    static short maxSlot( TypeFunPtr tfp ) {
+    // Return the max stack slot used by this signature, or 1 (for RPC)
+    static short maxArgSlot( TypeFunPtr tfp ) {
         // Count floats in signature up to index
         int fcnt=0;
         for( int i=0; i<tfp.nargs(); i++ )
@@ -265,12 +265,7 @@ public class x86_64_v2 extends Machine {
         };
         if( tfp.nargs()-fcnt >= cargs.length )
             throw Utils.TODO();
-        return 0;               // No stack args
-    }
-
-    // Return single int/ptr register.  Used by CallEnd output and Return input.
-    static RegMask retMask( TypeFunPtr tfp ) {
-        return tfp.ret() instanceof TypeFloat ? XMM0_MASK : RAX_MASK;
+        return 1;               // No stack args, plus RPC
     }