All 3 compilers ✅ PASS — no issues

compliance_report.md

@@ -0,0 +1,64 @@
+## Spec compliance report
+
+| Compiler | Status | Issues |
+|----------|--------|--------|
+| l1a | ✅ PASS | None |
+| l1b | ✅ PASS | None |
+| l2 | ✅ PASS | None |
+
+---
+
+## Details
+
+### l1a Implementation
+- **Status:** ✅ PASS
+- **Files:** `core/src/core/org/jnode/vm/x86/compiler/l1a/X86BytecodeVisitor.java:2292-2341`
+- **Issues:** None
+
+**Analysis:**
+- **Stack effect**: Correctly pops two `IntItem` values from virtual stack (`vstack.popInt()` ×2) — matches JVM spec "..., value1, value2 → ..."
+- **Signed comparison**: Uses `X86Constants.JGE` (0x8D, "Jump if Greater or Equal" signed) — matches JVM spec "ge succeeds if and only if value1 ≥ value2 (signed)"
+- **Branch target**: Parser computes `address + gets2()` where `gets2()` reads signed 16-bit offset — matches JVM spec "(branchbyte1 << 8) | branchbyte2" as signed 16-bit
+- **Basic block barrier**: Flushes virtual stack via `vstack.push(eContext)` before emitting jump — correct for control-flow boundaries
+- **Operand handling**: `prepareForOperation` ensures `value1` (v1) is in a register; `value2` (v2) can be register, local, or constant — correct for x86 CMP dest, src
+- **Commutativity handling**: Only swaps for JE/JNE (equality); correctly does NOT swap for JGE/JG/JL/JLE — x86 CMP is dest−src, so operand order matters for signed relational ops
+- **Yield point**: Not emitted for forward branches; backward branches handled at caller level — acceptable
+
+### l1b Implementation
+- **Status:** ✅ PASS
+- **Files:** `core/src/core/org/jnode/vm/x86/compiler/l1b/X86BytecodeVisitor.java:2809-2884`
+- **Issues:** None
+
+**Analysis:**
+- Identical logic to l1a for `visit_if_icmp` helper and `visit_if_icmpge` dispatch
+- Adds bytecode counters (`countBytecode`, `countConstOps`) for metrics — no functional impact
+- Same constant-folding TODO comments as l1a — not a spec violation, optimization opportunity only
+
+### l2 Implementation
+- **Status:** ✅ PASS
+- **Files:** 
+  - `core/src/core/org/jnode/vm/x86/compiler/l2/GenericX86CodeGenerator.java:3787-3824` (codegen)
+  - `core/src/core/org/jnode/vm/compiler/ir/quad/BranchCondition.java:30-31` (IR enum)
+  - `core/src/core/org/jnode/vm/compiler/ir/IRGenerator.java:917` (IR generation)
+  - `core/src/core/org/jnode/vm/bytecode/BasicBlockFinder.java:218-223` (stack simulation)
+- **Issues:** None
+
+**Analysis:**
+- **IR representation**: `BranchCondition.IF_ICMPGE(">=", true)` — `binary=true` correctly indicates two stack operands
+- **IR generation**: `IRGenerator.visitBranchCondition` creates `ConditionalBranchQuad` with two stack slots (s1, s2) of type `Operand.INT` — matches stack effect
+- **Basic block finder**: Pops two `JvmType.INT` from type stack, calls `addBranch(address, true)` and `condYieldPoint` — correct control-flow construction
+- **Code generation**: `generateJumpForBinaryCondition` switches on `IF_ICMPGE` → emits `os.writeJCC(target, X86Constants.JGE)` — correct signed ≥
+- **Support checker**: `L2ByteCodeSupportChecker` does NOT override `visit_if_icmpge` — inherits empty impl from `BytecodeVisitorSupport`, meaning instruction is **supported** (unsupported ops explicitly throw)
+- **Branch offset**: Uses same parser (`BytecodeParser.gets2()`) for signed 16-bit target — correct
+
+---
+
+## JVM Spec References (Java SE 6)
+
+- **Opcode**: `if_icmpge` = 162 (0xA2)
+- **Operand Stack**: `..., value1, value2 → ...` (both `int`, both popped)
+- **Comparison**: Signed integer comparison; branch taken iff `value1 ≥ value2`
+- **Branch Offset**: Signed 16-bit constructed from `branchbyte1`, `branchbyte2`; target = `address + offset`
+- **Target Constraint**: Must be an opcode within the same method
+
+All three JNode compiler tiers implement the above faithfully. No spec violations found.