R5 encoding bug fixes

Remove extra branch-to-next

3 files changed, 21 insertions, 10 deletions

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 756735ad..f2a2449e 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/codegen/Encoding.java
@@ -183,13 +183,17 @@ public class Encoding {
         if( bb.nOuts()==0 ) return; // StopNode
         if( !(bb instanceof IfNode iff) ) {
             CFGNode next = bb instanceof ReturnNode ? (CFGNode)bb.out(bb.nOuts()-1) : bb.uctrl();
-            // If the *next* BB has already been visited, we require an
-            // unconditional backwards jump here
+            // If the *next* BB has already been visited, we may need an
+            // unconditional jump here
             if( visit.get(next._nid) && !(next instanceof StopNode) ) {
-                CFGNode jmp = _code._mach.jump();
-                jmp.setDefX(0,bb);
-                next.setDef(next._inputs.find(bb),jmp);
-                rpo.add(jmp);
+                // If all the blocks, in RPO order, to our target
+                // are empty, we will fall in and not need a jump.
+                if( next instanceof LoopNode || !isEmptyBackwardsScan(rpo,next) ) {
+                    CFGNode jmp = _code._mach.jump();
+                    jmp.setDefX(0,bb);
+                    next.setDef(next._inputs.find(bb),jmp);
+                    rpo.add(jmp);
+                }
             }
             _rpo_cfg(next,visit,rpo);
         } else {
@@ -232,6 +236,12 @@ public class Encoding {
         rpo.add(bb);
     }
 
+    private static boolean isEmptyBackwardsScan(Ary<CFGNode> rpo, CFGNode next) {
+        for( int i=rpo._len-1; rpo.at(i)!=next; i-- )
+            if( rpo.at(i).nOuts()!=1 )
+                return false;
+        return true;
+    }
 
     // Write encoding bits in order into a big byte array.
     // Record opcode start and length.
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/BranchRISC.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/BranchRISC.java
index a0a4b6aa..ef1792eb 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/BranchRISC.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/BranchRISC.java
@@ -43,7 +43,7 @@ public class BranchRISC extends IfNode implements MachNode, RIPRelSize {
         // Todo: relocs (for offset - immf)
         short src1 = enc.reg(in(1));
         short src2 = in(2)==null ? (short)riscv.ZERO : enc.reg(in(2));
-        enc.add4(riscv.b_type(0x63, riscv.jumpop(_bop), src1, src2, 0));
+        enc.add4(riscv.b_type(riscv.OP_BRANCH, riscv.jumpop(_bop), src1, src2, 0));
     }
 
     // Delta is from opcode start
@@ -58,7 +58,7 @@ public class BranchRISC extends IfNode implements MachNode, RIPRelSize {
         short src1 = enc.reg(in(1));
         short src2 = in(2)==null ? (short)riscv.ZERO : enc.reg(in(2));
         if( opLen==4 ) {
-            enc.patch4(opStart,riscv.b_type(0x63, riscv.jumpop(_bop), src1, src2, delta));
+            enc.patch4(opStart,riscv.b_type(riscv.OP_BRANCH, riscv.jumpop(_bop), src1, src2, delta));
         } else {
             throw Utils.TODO();
         }
diff --git a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/riscv.java b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/riscv.java
index 03569709..0d382c6d 100644
--- a/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/riscv.java
+++ b/chapter21/src/main/java/com/seaofnodes/simple/node/cpus/riscv/riscv.java
@@ -106,6 +106,7 @@ public class riscv extends Machine {
     public static int OP    = 0b01_100_11;
     public static int OP_FP = 0b10_100_11;
     public static int OP_CUSTOM0 = 0b00_010_11;
+    public static int OP_BRANCH  = 0b11_000_11;
 
     //I_type opcode: 0010 0011
     public static int I_TYPE = 0x13;
@@ -178,11 +179,11 @@ public class riscv extends Machine {
         // 31 30 29 28 27 26 25 24-20 19-15 14-12 11 10  9  8  7  6-0
         // 12 10  9  8  7  6  5 SRC2  SRC1  FUNC3  4  3  2  1 11  OP
         int imm4_1 = (delta>> 1) & 0xF;
-        int imm10_5= (delta>> 5) &0x2F;
+        int imm10_5= (delta>> 5) &0x3F;
         int imm11  = (delta>>11) &   1;
         int imm12  = (delta>>12) &   1;
         int imm5 = imm4_1<<1 | imm11;
-        int imm7 = imm12<<7 | imm10_5;
+        int imm7 = imm12<<6 | imm10_5;
         return (imm7 << 25 ) | (rs2 << 20) | (rs1 << 15) | (func3 << 12) | (imm5 << 7) | opcode;
     }