R5 fcmp, flt load from pool

exportELF(null) now patches the cpool entries as well as globals.
Repeated cpool entries folded.
Fixed FltRISC, FCMP encoding.

10 files changed, 90 insertions, 96 deletions

diff --git a/chapter21/src/main/java/com/seaofnodes/simple/codegen/CodeGen.java b/chapter21/src/main/java/com/seaofnodes/simple/codegen/CodeGen.java
index 146c016a..2e1a4709 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/codegen/CodeGen.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/codegen/CodeGen.java
@@ -379,17 +379,16 @@ public class CodeGen {
     public int _tEncode;
     public Encoding _encoding;   // Encoding object
     public void preEncode() {  } // overridden by alternative ports
-    public CodeGen encode(boolean jit) {
+    public CodeGen encode() {
         assert _phase == Phase.RegAlloc;
         _phase = Phase.Encoding;
         long t0 = System.currentTimeMillis();
         _encoding = new Encoding(this);
         preEncode();
-        _encoding.encode(jit);
+        _encoding.encode();
         _tEncode = (int)(System.currentTimeMillis() - t0);
         return this;
     }
-    public CodeGen encode() { return encode(false); }
 
     // ---------------------------
     // Exporting to external formats
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java b/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java
index 118ca47c..395b3d46 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java
@@ -153,7 +153,7 @@ public class Encoding {
         _bigCons.put(relo,new Relo(relo,t,(byte)off,(byte)elf));
     }
 
-    void encode(boolean jit) {
+    void encode() {
         // Basic block layout: invert branches to keep blocks in-order; insert
         // unconditional jumps.  Attempt to keep backwards branches taken,
         // forwards not-taken (this is the default prediction on most
@@ -170,10 +170,6 @@ public class Encoding {
 
         // Patch RIP-relative and local encodings now.
         patchLocalRelocations();
-
-        // If JIT'ing (in memory execution), patch the global relocations now
-        if( jit )
-            patchGlobalRelocations();
     }
 
     // Basic block layout: invert branches to keep blocks in-order; insert
@@ -382,14 +378,10 @@ public class Encoding {
 
     // Write the constant pool into the BAOS and optionally patch locally
     void writeConstantPool( BAOS bits, boolean patch ) {
-        HashSet<Relo> ts = new HashSet<>();
-        for( Relo t : _bigCons.values() ) {
-            if( ts.contains(t) )
-                throw Utils.TODO(); // Dup!  Compress!
-            ts.add(t);
-        }
         padN(16,bits);
 
+        HashMap<Type,Integer> targets = new HashMap<>();
+
         // By log size
         for( int log = 3; log >= 0; log-- ) {
             // Write the 8-byte constants
@@ -397,17 +389,21 @@ public class Encoding {
                 Relo relo = _bigCons.get(op);
                 if( relo._t.log_size()==log ) {
                     // Map from relo to constant start and patch
-                    relo._target = bits.size();
+                    Integer target = targets.get(relo._t);
+                    if( target==null ) {
+                        targets.put(relo._t,target = bits.size());
+                        // Put constant into code space.
+                        if( relo._t instanceof TypeTuple tt ) // Constant tuples put all entries
+                            for( Type tx : tt._types )
+                                addN(log,tx,bits);
+                        else
+                            addN(log,relo._t,bits);
+                    }
+                    relo._target = target;
                     relo._opStart= _opStart[op._nid];
                     // Go ahead and locally patch in-memory
                     if( patch )
                         ((RIPRelSize)op).patch(this, relo._opStart, _opLen[op._nid], relo._target - relo._opStart);
-                    // Put constant into code space.
-                    if( relo._t instanceof TypeTuple tt ) // Constant tuples put all entries
-                        for( Type tx : tt._types )
-                            addN(log,tx,bits);
-                    else
-                        addN(log,relo._t,bits);
                 }
             }
         }
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/codegen/LinkMem.java b/chapter21/src/main/java/com/seaofnodes/simple/codegen/LinkMem.java
index 07a5dd12..e1e849e6 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/codegen/LinkMem.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/codegen/LinkMem.java
@@ -1,14 +1,22 @@
 package com.seaofnodes.simple.codegen;
 
+import com.seaofnodes.simple.Utils;
+import com.seaofnodes.simple.node.Node;
+
 // In-memory linking
 public class LinkMem {
     final CodeGen _code;
     LinkMem( CodeGen code ) { _code = code; }
 
     public CodeGen link() {
+        Encoding enc = _code._encoding;
+
+        // Patch external calls internally
+        enc.patchGlobalRelocations();
+
         // Write any large constants into a constant pool; they
         // are accessed by RIP-relative addressing.
-        _code._encoding.writeConstantPool(_code._encoding._bits,true);
+        enc.writeConstantPool(enc._bits,true);
 
         return _code;
     }
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/FltRISC.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/FltRISC.java
index 0a7b02e3..35de7e04 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/FltRISC.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/FltRISC.java
@@ -22,7 +22,7 @@ public class FltRISC extends ConstantNode implements MachNode, RIPRelSize {
         // AUIPC dst,#hi20_constant_pool
         enc.add4(riscv.u_type(riscv.OP_AUIPC, tmp, 0));
         // Load dst,[dst+#low12_constant_pool]
-        enc.add4(riscv.i_type(riscv.OP_LOAD, dst, 0b11, tmp, 0));
+        enc.add4(riscv.i_type(riscv.OP_LOADFP, dst, 0b11, tmp, 0));
     }
     @Override public RegMask killmap() { return new RegMask(riscv.T6); }
 
@@ -35,9 +35,9 @@ public class FltRISC extends ConstantNode implements MachNode, RIPRelSize {
         short dst = (short)(enc.reg(this) - riscv.F_OFFSET);
         short tmp = (short)riscv.T6;
         // AUIPC dst,#hi20_constant_pool
-        enc.patch4(opStart  , riscv.u_type(riscv.OP_AUIPC, tmp, delta));
+        enc.patch4(opStart  , riscv.u_type(riscv.OP_AUIPC, tmp, delta>>12));
         // Load dst,[dst+#low12_constant_pool]
-        enc.patch4(opStart+4, riscv.i_type(riscv.OP_LOAD, dst, 0b11, tmp, delta));
+        enc.patch4(opStart+4, riscv.i_type(riscv.OP_LOADFP, dst, 0b11, tmp, delta & 0xFFF));
     }
 
     @Override public void asm(CodeGen code, SB sb) {
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/SetFRISC.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/SetFRISC.java
index 64440536..9ee19e42 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/SetFRISC.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/SetFRISC.java
@@ -23,7 +23,7 @@ public class SetFRISC extends MachConcreteNode implements MachNode {
         short dst  =         enc.reg(this );
         short src1 = (short)(enc.reg(in(1))-riscv.F_OFFSET);
         short src2 = (short)(enc.reg(in(2))-riscv.F_OFFSET);
-        int body = riscv.r_type(riscv.OP_FP,dst,riscv.fsetop(_bop),src1,src2,0);
+        int body = riscv.r_type(riscv.OP_FP,dst,riscv.fsetop(_bop),src1,src2,0b1010001/*FCMP*/);
         enc.add4(body);
     }
 
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/print/ASMPrinter.java b/chapter21/src/main/java/com/seaofnodes/simple/print/ASMPrinter.java
index 902aa3c2..8f3c9bb2 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/print/ASMPrinter.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/print/ASMPrinter.java
@@ -6,6 +6,7 @@ import com.seaofnodes.simple.codegen.CodeGen;
 import com.seaofnodes.simple.codegen.Encoding;
 import com.seaofnodes.simple.node.*;
 import com.seaofnodes.simple.type.*;
+import java.util.HashSet;
 
 public abstract class ASMPrinter {
 
@@ -26,11 +27,15 @@ public abstract class ASMPrinter {
         // constant pool
         Encoding enc = code._encoding;
         if(  enc!=null && !enc._bigCons.isEmpty() ) {
+            iadr = (iadr+15)&-16; // pad to 16
+            HashSet<Type> targets = new HashSet<>();
             sb.p("--- Constant Pool ------").nl();
             // By log size
             for( int log = 3; log >= 0; log-- ) {
                 for( Node op : enc._bigCons.keySet() ) {
                     Encoding.Relo relo = enc._bigCons.get(op);
+                    if( targets.contains(relo._t) ) continue;
+                    targets.add(relo._t);
                     if( relo._t.log_size()==log ) {
                         sb.hex2(iadr).p("  ");
                         if( relo._t instanceof TypeTuple tt ) {
diff --git a/chapter21/src/test/java/com/seaofnodes/simple/Chapter21Test.java b/chapter21/src/test/java/com/seaofnodes/simple/Chapter21Test.java
index faf21b4f..cf088835 100644
--- a/chapter21/src/test/java/com/seaofnodes/simple/Chapter21Test.java
+++ b/chapter21/src/test/java/com/seaofnodes/simple/Chapter21Test.java
@@ -101,13 +101,14 @@ public class Chapter21Test {
 8  2.828427
 9  3.000000
 """;
-        TestC.run("newtonFloat",result);
+        //TestC.run("newtonFloat",result);
+
         EvalRisc5 R5 = TestRisc5.build("newtonFloat", 0);
-        R5.fregs[riscv.FA0 - riscv.F_OFFSET] = 4.0;
-        int trap = R5.step(100);
+        R5.fregs[riscv.FA0 - riscv.F_OFFSET] = 3.0;
+        int trap = R5.step(1000);
         assertEquals(0,trap);
         // Return register A0 holds fib(8)==55
-        assertEquals(2.0,R5.fregs[riscv.FA0 - riscv.F_OFFSET], 0.00001);
+        assertEquals(1.732051,R5.fregs[riscv.FA0 - riscv.F_OFFSET], 0.00001);
     }
 
     @Test public void testSieve() throws IOException {
diff --git a/chapter21/src/test/java/com/seaofnodes/simple/EvalRisc5.java b/chapter21/src/test/java/com/seaofnodes/simple/EvalRisc5.java
index 50e52095..e48dd34c 100644
--- a/chapter21/src/test/java/com/seaofnodes/simple/EvalRisc5.java
+++ b/chapter21/src/test/java/com/seaofnodes/simple/EvalRisc5.java
@@ -47,6 +47,10 @@ import java.util.Arrays;
     public int  ld4s(int x) { return ld2z(x) | ld2s(x+2)<<16; }
     public long ld8 (int x) { return (long)ld4s(x)&0xFFFFFFFFL | ((long)ld4s(x+4))<<32; }
 
+    public float  ld4f(int x) { return Float.intBitsToFloat(ld4s(x)); }
+    public double ld8f(int x) { return Double.longBitsToDouble(ld8(x)); }
+
+
     public void st1(int x, int  val) { _buf[x] = (byte)val; }
     public void st2(int x, int  val) { st1(x,val); st1(x+1,val>> 8); }
     public void st4(int x, int  val) { st2(x,val); st2(x+2,val>>16); }
@@ -85,17 +89,6 @@ import java.util.Arrays;
             int rdid = (ir >> 7) & 0x1f;
 
             switch( ir & 0x7f ) {
-            // floats
-            case 0x7: {
-                is_f = true;
-                // does not handle rsp case
-
-                long rs1 = (ir >> 15) & 0x1f;
-                // Todo: look for stack case
-                if (rs1 == 2) break;
-                frval = ir & 0x7FF000;
-                break;
-            }
             case 0x37: // LUI (0b0110111)
                 rval = ( ir & 0xfffff000 );
                 break;
@@ -174,6 +167,27 @@ import java.util.Arrays;
                 }
                 break;
             }
+            // floats
+            case 0x7: {
+                is_f = true;
+                int rs1 = (int)regs[(ir >> 15) & 0x1f]; // Address chop to 32b
+                int imm = ir >> 20;
+                int imm_se = imm | (( (imm & 0x800)!=0 ) ? 0xfffff000 : 0 );
+                int rsval = rs1 + imm_se;
+
+                if( rsval >= _buf.length-3 ) {
+                    trap = (5+1);
+                    rval = rsval;
+                } else {
+                    switch( ( ir >> 12 ) & 0x7 ) {
+                    case 2: frval = ld4f( rsval ); break;
+                    case 3: frval = ld8f( rsval ); break;
+                    default: trap = (2+1);
+                    }
+                }
+                break;
+            }
+
             case 0x23: { // Store 0b0100011
                 int  rs1 = (int)regs[(ir >> 15) & 0x1f]; // Address chop to 32b
                 long rs2 = regs[(ir >> 20) & 0x1f];
@@ -279,66 +293,37 @@ import java.util.Arrays;
             }
 
             case 0x53: {
-                // differentiate between  fcvt.d.w  and  fadd.d
                 is_f = true;
-                int rs1;
-                int rs1_int;
-                int rs2;
-                int func5 = (ir >> 27) & 0x1F;;
+                int rs1  = (ir >> 15) & 0x1F;
+                int rs2  = (ir >> 20) & 0x1F;
+                double lhs = fregs[rs1];
+                double rhs = fregs[rs2];
+                int func5 = (ir >> 27) & 0x1F;
                 int func2 = (ir >> 25) & 0x3;
-//                if(func5 == 0 && func2 == 1) {
-//                }
-                if(func5 == 0) {
-                    // fadd.d
-                    rs1  = (ir >> 15) & 0x1F;
-                    rs2  = (ir >> 20) & 0x1F;
-                    double lhs = fregs[rs1];
-                    double rhs = fregs[rs2];
-                    frval = lhs + rhs;
+                if( func2 != 1 ) throw Utils.TODO(); // 1==double
+                switch( func5 ) {
+                case  0:  frval = lhs + rhs;  break;  // fadd.d
+                case  1:  frval = lhs - rhs;  break;  // fsub.d
+                case  2:  frval = lhs * rhs;  break;  // fmul.d
+                case  3:  frval = lhs / rhs;  break;  // fdiv.d
+                case  5:  frval = Math.min(lhs,rhs); break; // fmin.d, also used for move
+                case 20:
+                    is_f = false; // Output into a GPR
+                    rval = switch( (ir >> 12) & 7 ) {
+                    case 0 -> lhs <= rhs ? 1 : 0;
+                    case 1 -> lhs == rhs ? 1 : 0;
+                    case 2 -> lhs <  rhs ? 1 : 0;
+                    default -> throw Utils.TODO();
+                    };
                     break;
-                } else if(func5 == 26) {
-                    //fcvt.d.w
-                    rs1_int  = (ir >> 15) & 0x1F;
-                    long value = regs[rs1_int];
-                    frval = (double)value;
-                    break;
-                } else if (func5 == 5) {
-                    // fmin.d
-                    rs1  = (ir >> 15) & 0x1F;
-                    rs2  = (ir >> 20) & 0x1F;
-                    // just pick one as they are the same
-                    frval = fregs[rs1];
-                    break;
-                }  else if(func5 == 3) {
-                    // fdiv.d
-                    rs1  = (ir >> 15) & 0x1F;
-                    rs2  = (ir >> 20) & 0x1F;
-                    double lhs = fregs[rs1];
-                    double rhs = fregs[rs2];
-                    frval = lhs / rhs;
-                    break;
-                } else if(func5 == 2) {
-                    // fmul.d
-                    rs1  = (ir >> 15) & 0x1F;
-                    rs2  = (ir >> 20) & 0x1F;
-                    double lhs = fregs[rs1];
-                    double rhs = fregs[rs2];
-                    frval = lhs * rhs;
-                    break;
-
-                } else if(func5 == 1) {
-                    // fsub.d
-                    rs1  = (ir >> 15) & 0x1F;
-                    rs2  = (ir >> 20) & 0x1F;
-                    double lhs = fregs[rs1];
-                    double rhs = fregs[rs2];
-                    frval = lhs - rhs;
-                    break;
-                } else {
+                case 26:  frval = (double)regs[rs1]; break; // fcvt.w.d
+                default:
                     trap  = (2+1); // Fault: Invalid opcode.
                     throw Utils.TODO();
                 }
+                break;
             }
+
             default: trap = (2+1); // Fault: Invalid opcode.
             }
 
diff --git a/chapter21/src/test/java/com/seaofnodes/simple/TestArm64.java b/chapter21/src/test/java/com/seaofnodes/simple/TestArm64.java
index 96aa5b74..1325f1e1 100644
--- a/chapter21/src/test/java/com/seaofnodes/simple/TestArm64.java
+++ b/chapter21/src/test/java/com/seaofnodes/simple/TestArm64.java
@@ -19,7 +19,7 @@ public class TestArm64 {
     public static EvalArm64 build(String dir, String file, int arg) throws IOException {
         // Compile and export Simple
         String src = Files.readString(Path.of(dir+"/"+file+".smp"));
-        CodeGen code = new CodeGen(src).driver(CodeGen.Phase.RegAlloc,"arm", "SystemV").encode(true);
+        CodeGen code = new CodeGen(src).driver("arm", "SystemV",null);
 
         // Image
         byte[] image = new byte[1<<20]; // A megabyte (1024*1024 bytes)
diff --git a/chapter21/src/test/java/com/seaofnodes/simple/TestRisc5.java b/chapter21/src/test/java/com/seaofnodes/simple/TestRisc5.java
index 8b9b8884..a72dbd1d 100644
--- a/chapter21/src/test/java/com/seaofnodes/simple/TestRisc5.java
+++ b/chapter21/src/test/java/com/seaofnodes/simple/TestRisc5.java
@@ -18,7 +18,7 @@ public abstract class TestRisc5 {
     public static EvalRisc5 build( String dir, String file, int arg ) throws IOException {
         // Compile and export Simple
         String src = Files.readString(Path.of(dir+"/"+file+".smp"));
-        CodeGen code = new CodeGen(src).driver(CodeGen.Phase.RegAlloc,"riscv", "SystemV").encode(true);
+        CodeGen code = new CodeGen(src).driver("riscv", "SystemV",null);
 
         // Image
         byte[] image = new byte[1<<20]; // A megabyte (1024*1024 bytes)