added new performance flags, extended PIC handling to size 4, introduced fast paths for ObjFastIntRangeIterator, small-arity arguments, index access, and unary operations

docs/perf_guide.md

@@ -10,15 +10,35 @@ Optimizations are controlled by runtime‑mutable flags in `net.sergeych.lyng.Pe
 
 All flags are `var` and can be flipped at runtime (e.g., from tests or host apps) for A/B comparisons.
 
+### Workload presets (JVM‑first)
+
+To simplify switching between recommended flag sets for different workloads, use `net.sergeych.lyng.PerfProfiles`:
+
+```
+val snap = PerfProfiles.apply(PerfProfiles.Preset.BENCH)  // or BASELINE / BOOKS
+// ... run workload ...
+PerfProfiles.restore(snap)  // restore previous flags
+```
+
+- BASELINE: restores platform defaults from `PerfDefaults` (good rollback point).
+- BENCH: expression‑heavy micro‑bench focus (aggressive R‑value and PIC optimizations on JVM).
+- BOOKS: documentation workloads (prefers simpler paths; disables some PIC/arg builder features shown neutral/negative for this load in A/B).
+
 ## Key flags
 
 - `LOCAL_SLOT_PIC` — Runtime cache in `LocalVarRef` to avoid repeated name→slot lookups per frame (ON JVM default).
 - `EMIT_FAST_LOCAL_REFS` — Compiler emits `FastLocalVarRef` for identifiers known to be locals/params (ON JVM default).
 - `ARG_BUILDER` — Efficient argument building: small‑arity no‑alloc and pooled builder on JVM (ON JVM default).
+- `ARG_SMALL_ARITY_12` — Extends small‑arity no‑alloc call paths from 0–8 to 0–12 arguments (JVM‑first exploration; OFF by default). Use for codebases with many 9–12 arg calls; A/B before enabling.
 - `SKIP_ARGS_ON_NULL_RECEIVER` — Early return on optional‑null receivers before building args (semantics‑compatible). A/B only.
 - `SCOPE_POOL` — Scope frame pooling for calls (JVM, per‑thread ThreadLocal pool). ON by default on JVM; togglable at runtime.
 - `FIELD_PIC` — 2‑entry polymorphic inline cache for field reads/writes keyed by `(classId, layoutVersion)` (ON JVM default).
 - `METHOD_PIC` — 2‑entry PIC for instance method calls keyed by `(classId, layoutVersion)` (ON JVM default).
+- `FIELD_PIC_SIZE_4` — Increases Field PIC size from 2 to 4 entries (JVM-first tuning; OFF by default). Use for sites with >2 receiver shapes.
+- `METHOD_PIC_SIZE_4` — Increases Method PIC size from 2 to 4 entries (JVM-first tuning; OFF by default).
+- `PIC_ADAPTIVE_2_TO_4` — Adaptive growth of Field/Method PICs from 2→4 entries per-site when miss rate >20% over ≥256 accesses (JVM-first; OFF by default).
+- `INDEX_PIC` — Enables polymorphic inline cache for indexing (e.g., `a[i]`) and related fast paths. Defaults to follow `FIELD_PIC` on init; can be toggled independently.
+- `INDEX_PIC_SIZE_4` — Increases Index PIC size from 2 to 4 entries (JVM-first tuning). Default: ON for JVM; OFF elsewhere by default.
 - `PIC_DEBUG_COUNTERS` — Enable lightweight hit/miss counters via `PerfStats` (OFF by default).
 - `PRIMITIVE_FASTOPS` — Fast paths for `(ObjInt, ObjInt)` arithmetic/comparisons and `(ObjBool, ObjBool)` logic (ON JVM default).
 - `RVAL_FASTPATH` — Bypass `ObjRecord` in pure expression evaluation via `ObjRef.evalValue` (ON JVM default, OFF elsewhere).
@@ -28,11 +48,21 @@ See `src/commonMain/kotlin/net/sergeych/lyng/PerfFlags.kt` and `PerfDefaults.*.k
 ## Where optimizations apply
 
 - Locals: `FastLocalVarRef`, `LocalVarRef` per‑frame cache (PIC).
-- Calls: small‑arity zero‑alloc paths (0–8 args), pooled builder (JVM), and child frame pooling (optional).
+- Calls: small‑arity zero‑alloc paths (0–8 args; optionally 0–12 with `ARG_SMALL_ARITY_12`), pooled builder (JVM), and child frame pooling (optional).
-- Properties/methods: Field/Method PICs with receiver shape `(classId, layoutVersion)` and handle‑aware caches.
+- Properties/methods: Field/Method PICs with receiver shape `(classId, layoutVersion)` and handle‑aware caches; configurable 2→4 entries under flags.
 - Expressions: R‑value fast paths in hot nodes (`UnaryOpRef`, `BinaryOpRef`, `ElvisRef`, logical ops, `RangeRef`, `IndexRef` read, `FieldRef` receiver eval, `ListLiteralRef` elements, `CallRef` callee, `MethodCallRef` receiver, assignment RHS).
 - Primitives: Direct boolean/int ops where safe.
 
+### Compiler constant folding (conservative)
+- The compiler folds a safe subset of literal‑only expressions at compile time to reduce runtime work:
+  - Integer arithmetic: `+ - * / %` (division/modulo only when divisor ≠ 0).
+  - Bitwise integer ops: `& ^ | << >>`.
+  - Comparisons and equality for ints/strings/chars: `== != < <= > >=`.
+  - Boolean logic for literal booleans: `|| &&` and unary `!`.
+  - String concatenation of literal strings: `"a" + "b"`.
+- Non‑literal operands or side‑effecting constructs are not folded.
+- Semantics remain unchanged; tests verify parity.
+
 ## Running JVM micro‑benchmarks
 
 Each benchmark prints timings with `[DEBUG_LOG]` and includes correctness assertions to prevent dead‑code elimination.
@@ -46,8 +76,11 @@ Run individual tests to avoid multiplatform matrices:
 ./gradlew :lynglib:jvmTest --tests CallSplatBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests PicBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests PicInvalidationJvmTest
+./gradlew :lynglib:jvmTest --tests PicAdaptiveABTest
 ./gradlew :lynglib:jvmTest --tests ArithmeticBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests ExpressionBenchmarkTest
+./gradlew :lynglib:jvmTest --tests IndexPicABTest
+./gradlew :lynglib:jvmTest --tests IndexWritePathABTest
 ./gradlew :lynglib:jvmTest --tests CallPoolingBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests MethodPoolingBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests MixedBenchmarkTest
@@ -62,6 +95,18 @@ Typical output (example):
 
 Lower time is better. Run the same bench with a flag OFF vs ON to compare.
 
+### Optional JFR allocation profiling (JVM)
+
+When running end‑to‑end “book” workloads or heavier benches, you can enable JFR to capture allocation and GC details:
+
+```
+./gradlew :lynglib:jvmTest --tests BookAllocationProfileTest -Dlyng.jfr=true \
+  -Dlyng.profile.warmup=1 -Dlyng.profile.repeats=3 -Dlyng.profile.shuffle=true
+```
+
+- Dumps are saved to `lynglib/build/jfr_*.jfr` if the JVM supports Flight Recorder.
+- The test also records GC counts/time and median time/heap deltas to `lynglib/build/book_alloc_profile.txt`.
+
 ## Toggling flags in tests
 
 Flags are mutable at runtime, e.g.:
@@ -88,17 +133,46 @@ Available counters in `PerfStats`:
 
 - Field PIC: `fieldPicHit`, `fieldPicMiss`, `fieldPicSetHit`, `fieldPicSetMiss`
 - Method PIC: `methodPicHit`, `methodPicMiss`
+- Index PIC: `indexPicHit`, `indexPicMiss`
 - Locals: `localVarPicHit`, `localVarPicMiss`, `fastLocalHit`, `fastLocalMiss`
 - Primitive ops: `primitiveFastOpsHit`
 
 Print a summary at the end of a bench/test as needed. Remember to turn counters OFF after the test.
 
+## A/B scenarios and guidance (JVM)
+
+### Adaptive PIC (fields/methods)
+- Flags: `FIELD_PIC=true`, `METHOD_PIC=true`, `FIELD_PIC_SIZE_4=false`, `METHOD_PIC_SIZE_4=false`, toggle `PIC_ADAPTIVE_2_TO_4` OFF vs ON.
+- Benchmarks: `PicBenchmarkTest`, `MixedBenchmarkTest`, `PicAdaptiveABTest`.
+- Expect wins at sites with >2 receiver shapes; counters should show fewer misses with adaptivity ON.
+
+### Index PIC and size
+- Flags: toggle `INDEX_PIC` OFF vs ON; then `INDEX_PIC_SIZE_4` OFF vs ON.
+- Benchmarks: `ExpressionBenchmarkTest` (list indexing) and `IndexPicABTest` (string/map indexing).
+- Expect wins when the same index shape recurs; counters should show higher `indexPicHit`.
+
+### Index WRITE paths (Map and List)
+- Flags: toggle `INDEX_PIC` OFF vs ON; then `INDEX_PIC_SIZE_4` OFF vs ON.
+- Benchmark: `IndexWritePathABTest` (Map[String] put, List[Int] set) — writes results to `lynglib/build/index_write_ab_results.txt`.
+- Direct fast‑paths are used on R‑value paths where safe and semantics‑preserving (e.g., optional‑chaining no‑ops on null receivers; bounds exceptions unchanged).
+
 ## Guidance per flag (JVM)
 
 - Keep `RVAL_FASTPATH = true` unless debugging a suspected expression‑semantics issue.
 - Use `SCOPE_POOL = true` only for benchmarks or once pooling passes the deep stress tests and broader validation; currently OFF by default.
 - `FIELD_PIC` and `METHOD_PIC` should remain ON; they are validated with invalidation tests.
+- Consider enabling `FIELD_PIC_SIZE_4`/`METHOD_PIC_SIZE_4` for sites with 3–4 receiver shapes; measure first.
+- `PIC_ADAPTIVE_2_TO_4` is useful on polymorphic sites and may outperform fixed size 2 on mixed-shape workloads. Validate with `PicAdaptiveABTest`.
+- `INDEX_PIC` is generally beneficial on JVM; leave ON when measuring index‑heavy workloads.
+- `INDEX_PIC_SIZE_4` is ON by default on JVM as A/B showed consistent wins on String[Int] and Map[String] workloads. You can disable it by setting `PerfFlags.INDEX_PIC_SIZE_4 = false` if needed.
 - `ARG_BUILDER` should remain ON; switch OFF only to get a baseline.
+- `ARG_SMALL_ARITY_12` is experimental and OFF by default. Enable it only if your workload frequently calls functions with 9–12 arguments and A/B shows consistent wins.
+
+### Workload‑specific recommendations (JVM)
+
+- “Books”/documentation loads (BookTest): prefer simpler paths; in A/B these often benefit from the BOOKS preset (e.g., `ARG_BUILDER=false`, `SCOPE_POOL=false`, `INDEX_PIC=false`). Use `PerfProfiles.apply(PerfProfiles.Preset.BOOKS)` before the run and `restore(...)` after.
+- Expression‑heavy benches: use the BENCH preset (PICs and R‑value fast‑paths enabled, `INDEX_PIC_SIZE_4=true`).
+- Always verify with local A/B on your environment; rollback is a flag flip or applying BASELINE.
 
 ## Notes on correctness & safety
 
@@ -111,6 +185,12 @@ Print a summary at the end of a bench/test as needed. Remember to turn counters
 - Non‑JVM defaults keep `RVAL_FASTPATH=false` for now; other low‑risk flags may be ON.
 - Once JVM path is fully validated and measured, add lightweight benches for JS/Wasm/Native and enable flags incrementally.
 
+## Range fast iteration (experimental)
+
+- Flag: `RANGE_FAST_ITER` (default OFF).
+- When enabled and applicable, simple ascending integer ranges (`0..n`, `0..<n`) use a specialized non‑allocating iterator (`ObjFastIntRangeIterator`).
+- Benchmark: `RangeIterationBenchmarkTest` records OFF/ON timings for inclusive, exclusive, reversed, negative, and empty ranges. Semantics are preserved; non‑int or complex ranges fall back to the generic iterator.
+
 ## Troubleshooting
 
 - If a benchmark shows regressions, flip related flags OFF to isolate the source (e.g., `ARG_BUILDER`, `RVAL_FASTPATH`, `FIELD_PIC`, `METHOD_PIC`).

lynglib/src/androidMain/kotlin/net/sergeych/lyng/PerfDefaults.android.kt

@@ -27,6 +27,11 @@ actual object PerfDefaults {
 
     actual val FIELD_PIC: Boolean = true
     actual val METHOD_PIC: Boolean = true
+    actual val FIELD_PIC_SIZE_4: Boolean = false
+    actual val METHOD_PIC_SIZE_4: Boolean = false
+    actual val PIC_ADAPTIVE_2_TO_4: Boolean = false
+    actual val PIC_ADAPTIVE_METHODS_ONLY: Boolean = false
+    actual val PIC_ADAPTIVE_HEURISTIC: Boolean = false
 
     actual val PIC_DEBUG_COUNTERS: Boolean = false
 
@@ -34,4 +39,7 @@ actual object PerfDefaults {
     actual val RVAL_FASTPATH: Boolean = true
     // Regex caching aligns with JVM behavior on Android (Dalvik/ART)
     actual val REGEX_CACHE: Boolean = true
+    actual val ARG_SMALL_ARITY_12: Boolean = false
+    actual val INDEX_PIC_SIZE_4: Boolean = false
+    actual val RANGE_FAST_ITER: Boolean = false
 }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/Arguments.kt

@@ -26,12 +26,13 @@ import net.sergeych.lyng.obj.ObjList
  suspend fun Collection<ParsedArgument>.toArguments(scope: Scope, tailBlockMode: Boolean): Arguments {
      // Small-arity fast path (no splats) to reduce allocations
      if (PerfFlags.ARG_BUILDER) {
+        val limit = if (PerfFlags.ARG_SMALL_ARITY_12) 12 else 8
          var hasSplat = false
          var count = 0
          for (pa in this) {
              if (pa.isSplat) { hasSplat = true; break }
              count++
-             if (count > 8) break
+            if (count > limit) break
          }
          if (!hasSplat && count == this.size) {
              val quick = when (count) {
@@ -93,6 +94,60 @@ import net.sergeych.lyng.obj.ObjList
                      val a7 = this.elementAt(7).value.execute(scope)
                      Arguments(listOf(a0, a1, a2, a3, a4, a5, a6, a7), tailBlockMode)
                  }
+                9 -> if (PerfFlags.ARG_SMALL_ARITY_12) {
+                    val a0 = this.elementAt(0).value.execute(scope)
+                    val a1 = this.elementAt(1).value.execute(scope)
+                    val a2 = this.elementAt(2).value.execute(scope)
+                    val a3 = this.elementAt(3).value.execute(scope)
+                    val a4 = this.elementAt(4).value.execute(scope)
+                    val a5 = this.elementAt(5).value.execute(scope)
+                    val a6 = this.elementAt(6).value.execute(scope)
+                    val a7 = this.elementAt(7).value.execute(scope)
+                    val a8 = this.elementAt(8).value.execute(scope)
+                    Arguments(listOf(a0, a1, a2, a3, a4, a5, a6, a7, a8), tailBlockMode)
+                } else null
+                10 -> if (PerfFlags.ARG_SMALL_ARITY_12) {
+                    val a0 = this.elementAt(0).value.execute(scope)
+                    val a1 = this.elementAt(1).value.execute(scope)
+                    val a2 = this.elementAt(2).value.execute(scope)
+                    val a3 = this.elementAt(3).value.execute(scope)
+                    val a4 = this.elementAt(4).value.execute(scope)
+                    val a5 = this.elementAt(5).value.execute(scope)
+                    val a6 = this.elementAt(6).value.execute(scope)
+                    val a7 = this.elementAt(7).value.execute(scope)
+                    val a8 = this.elementAt(8).value.execute(scope)
+                    val a9 = this.elementAt(9).value.execute(scope)
+                    Arguments(listOf(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9), tailBlockMode)
+                } else null
+                11 -> if (PerfFlags.ARG_SMALL_ARITY_12) {
+                    val a0 = this.elementAt(0).value.execute(scope)
+                    val a1 = this.elementAt(1).value.execute(scope)
+                    val a2 = this.elementAt(2).value.execute(scope)
+                    val a3 = this.elementAt(3).value.execute(scope)
+                    val a4 = this.elementAt(4).value.execute(scope)
+                    val a5 = this.elementAt(5).value.execute(scope)
+                    val a6 = this.elementAt(6).value.execute(scope)
+                    val a7 = this.elementAt(7).value.execute(scope)
+                    val a8 = this.elementAt(8).value.execute(scope)
+                    val a9 = this.elementAt(9).value.execute(scope)
+                    val a10 = this.elementAt(10).value.execute(scope)
+                    Arguments(listOf(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10), tailBlockMode)
+                } else null
+                12 -> if (PerfFlags.ARG_SMALL_ARITY_12) {
+                    val a0 = this.elementAt(0).value.execute(scope)
+                    val a1 = this.elementAt(1).value.execute(scope)
+                    val a2 = this.elementAt(2).value.execute(scope)
+                    val a3 = this.elementAt(3).value.execute(scope)
+                    val a4 = this.elementAt(4).value.execute(scope)
+                    val a5 = this.elementAt(5).value.execute(scope)
+                    val a6 = this.elementAt(6).value.execute(scope)
+                    val a7 = this.elementAt(7).value.execute(scope)
+                    val a8 = this.elementAt(8).value.execute(scope)
+                    val a9 = this.elementAt(9).value.execute(scope)
+                    val a10 = this.elementAt(10).value.execute(scope)
+                    val a11 = this.elementAt(11).value.execute(scope)
+                    Arguments(listOf(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11), tailBlockMode)
+                } else null
                  else -> null
              }
              if (quick != null) return quick

lynglib/src/commonMain/kotlin/net/sergeych/lyng/Compiler.kt

@@ -1845,6 +1845,92 @@ class Compiler(
         }
 
         private var lastPriority = 0
+        // Helpers for conservative constant folding (literal-only). Only pure, side-effect-free ops.
+        private fun constOf(r: ObjRef): Obj? = (r as? ConstRef)?.constValue
+
+        private fun foldBinary(op: BinOp, aRef: ObjRef, bRef: ObjRef): Obj? {
+            val a = constOf(aRef) ?: return null
+            val b = constOf(bRef) ?: return null
+            return when (op) {
+                // Boolean logic
+                BinOp.OR -> if (a is ObjBool && b is ObjBool) if (a.value || b.value) ObjTrue else ObjFalse else null
+                BinOp.AND -> if (a is ObjBool && b is ObjBool) if (a.value && b.value) ObjTrue else ObjFalse else null
+
+                // Equality and comparisons for ints/strings/chars
+                BinOp.EQ -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value == b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value == b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value == b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+                BinOp.NEQ -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value != b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value != b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value != b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+                BinOp.LT -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value < b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value < b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value < b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+                BinOp.LTE -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value <= b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value <= b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value <= b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+                BinOp.GT -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value > b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value > b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value > b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+                BinOp.GTE -> when {
+                    a is ObjInt && b is ObjInt -> if (a.value >= b.value) ObjTrue else ObjFalse
+                    a is ObjString && b is ObjString -> if (a.value >= b.value) ObjTrue else ObjFalse
+                    a is ObjChar && b is ObjChar -> if (a.value >= b.value) ObjTrue else ObjFalse
+                    else -> null
+                }
+
+                // Arithmetic for ints only (keep semantics simple at compile time)
+                BinOp.PLUS -> when {
+                    a is ObjInt && b is ObjInt -> ObjInt(a.value + b.value)
+                    a is ObjString && b is ObjString -> ObjString(a.value + b.value)
+                    else -> null
+                }
+                BinOp.MINUS -> if (a is ObjInt && b is ObjInt) ObjInt(a.value - b.value) else null
+                BinOp.STAR -> if (a is ObjInt && b is ObjInt) ObjInt(a.value * b.value) else null
+                BinOp.SLASH -> if (a is ObjInt && b is ObjInt && b.value != 0L) ObjInt(a.value / b.value) else null
+                BinOp.PERCENT -> if (a is ObjInt && b is ObjInt && b.value != 0L) ObjInt(a.value % b.value) else null
+
+                // Bitwise for ints
+                BinOp.BAND -> if (a is ObjInt && b is ObjInt) ObjInt(a.value and b.value) else null
+                BinOp.BXOR -> if (a is ObjInt && b is ObjInt) ObjInt(a.value xor b.value) else null
+                BinOp.BOR -> if (a is ObjInt && b is ObjInt) ObjInt(a.value or b.value) else null
+                BinOp.SHL -> if (a is ObjInt && b is ObjInt) ObjInt(a.value shl (b.value.toInt() and 63)) else null
+                BinOp.SHR -> if (a is ObjInt && b is ObjInt) ObjInt(a.value shr (b.value.toInt() and 63)) else null
+
+                // Non-folded / side-effecting or type-dependent ops
+                BinOp.EQARROW, BinOp.REF_EQ, BinOp.REF_NEQ, BinOp.MATCH, BinOp.NOTMATCH,
+                BinOp.IN, BinOp.NOTIN, BinOp.IS, BinOp.NOTIS, BinOp.SHUTTLE -> null
+            }
+        }
+
+        private fun foldUnary(op: UnaryOp, aRef: ObjRef): Obj? {
+            val a = constOf(aRef) ?: return null
+            return when (op) {
+                UnaryOp.NOT -> if (a is ObjBool) if (!a.value) ObjTrue else ObjFalse else null
+                UnaryOp.NEGATE -> when (a) {
+                    is ObjInt -> ObjInt(-a.value)
+                    is ObjReal -> ObjReal(-a.value)
+                    else -> null
+                }
+                UnaryOp.BITNOT -> if (a is ObjInt) ObjInt(a.value.inv()) else null
+            }
+        }
+
         val allOps = listOf(
             // assignments, lowest priority
             Operator(Token.Type.ASSIGN, lastPriority) { pos, a, b ->
@@ -1867,6 +1953,7 @@ class Compiler(
             },
             // logical 1
             Operator(Token.Type.OR, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.OR, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 LogicalOrRef(a, b)
             },
             // logical 2
@@ -1875,12 +1962,15 @@ class Compiler(
             },
             // bitwise or/xor/and (tighter than &&, looser than equality)
             Operator(Token.Type.BITOR, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.BOR, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.BOR, a, b)
             },
             Operator(Token.Type.BITXOR, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.BXOR, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.BXOR, a, b)
             },
             Operator(Token.Type.BITAND, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.BAND, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.BAND, a, b)
             },
             // equality/not equality and related
@@ -1888,9 +1978,11 @@ class Compiler(
                 BinaryOpRef(BinOp.EQARROW, a, b)
             },
             Operator(Token.Type.EQ, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.EQ, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.EQ, a, b)
             },
             Operator(Token.Type.NEQ, lastPriority) { _, a, b ->
+                foldBinary(BinOp.NEQ, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.NEQ, a, b)
             },
             Operator(Token.Type.REF_EQ, lastPriority) { _, a, b ->
@@ -1907,15 +1999,19 @@ class Compiler(
             },
             // relational <=,...
             Operator(Token.Type.LTE, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.LTE, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.LTE, a, b)
             },
             Operator(Token.Type.LT, lastPriority) { _, a, b ->
+                foldBinary(BinOp.LT, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.LT, a, b)
             },
             Operator(Token.Type.GTE, lastPriority) { _, a, b ->
+                foldBinary(BinOp.GTE, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.GTE, a, b)
             },
             Operator(Token.Type.GT, lastPriority) { _, a, b ->
+                foldBinary(BinOp.GT, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.GT, a, b)
             },
             // in, is:
@@ -1942,25 +2038,32 @@ class Compiler(
             },
             // shifts (tighter than shuttle, looser than +/-)
             Operator(Token.Type.SHL, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.SHL, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.SHL, a, b)
             },
             Operator(Token.Type.SHR, lastPriority) { _, a, b ->
+                foldBinary(BinOp.SHR, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.SHR, a, b)
             },
             // arithmetic
             Operator(Token.Type.PLUS, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.PLUS, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.PLUS, a, b)
             },
             Operator(Token.Type.MINUS, lastPriority) { _, a, b ->
+                foldBinary(BinOp.MINUS, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.MINUS, a, b)
             },
             Operator(Token.Type.STAR, ++lastPriority) { _, a, b ->
+                foldBinary(BinOp.STAR, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.STAR, a, b)
             },
             Operator(Token.Type.SLASH, lastPriority) { _, a, b ->
+                foldBinary(BinOp.SLASH, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.SLASH, a, b)
             },
             Operator(Token.Type.PERCENT, lastPriority) { _, a, b ->
+                foldBinary(BinOp.PERCENT, a, b)?.let { return@Operator ConstRef(it.asReadonly) }
                 BinaryOpRef(BinOp.PERCENT, a, b)
             },
         )

lynglib/src/commonMain/kotlin/net/sergeych/lyng/PerfDefaults.kt

@@ -32,6 +32,14 @@ expect object PerfDefaults {
 
     val FIELD_PIC: Boolean
     val METHOD_PIC: Boolean
+    // Optional larger PICs and adaptivity (JVM-first)
+    val FIELD_PIC_SIZE_4: Boolean
+    val METHOD_PIC_SIZE_4: Boolean
+    val PIC_ADAPTIVE_2_TO_4: Boolean
+    // Enable adaptivity only for methods (independent of fields)
+    val PIC_ADAPTIVE_METHODS_ONLY: Boolean
+    // Enable per-site heuristic to revert/avoid promotion when it shows no benefit
+    val PIC_ADAPTIVE_HEURISTIC: Boolean
 
     val PIC_DEBUG_COUNTERS: Boolean
 
@@ -40,4 +48,13 @@ expect object PerfDefaults {
 
     // Regex caching (JVM-first): small LRU for compiled patterns
     val REGEX_CACHE: Boolean
+
+    // Argument builder extended arity (JVM-first exploration)
+    val ARG_SMALL_ARITY_12: Boolean
+
+    // Index PIC sizing (JVM-first)
+    val INDEX_PIC_SIZE_4: Boolean
+
+    // Specialized non-allocating integer range iteration in hot loops
+    val RANGE_FAST_ITER: Boolean
 }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/PerfFlags.kt

@@ -30,6 +30,8 @@ object PerfFlags {
 
     // Enable more efficient argument building and bulk-copy for splats
     var ARG_BUILDER: Boolean = PerfDefaults.ARG_BUILDER
+    // Extend small-arity no-alloc argument paths to 9..12 (JVM-first; default via PerfDefaults)
+    var ARG_SMALL_ARITY_12: Boolean = PerfDefaults.ARG_SMALL_ARITY_12
     // Allow early-return in optional calls before building args (semantics-compatible). Present for A/B only.
     var SKIP_ARGS_ON_NULL_RECEIVER: Boolean = PerfDefaults.SKIP_ARGS_ON_NULL_RECEIVER
     // Enable pooling of Scope frames for calls (may be JVM-only optimization)
@@ -39,6 +41,23 @@ object PerfFlags {
     var FIELD_PIC: Boolean = PerfDefaults.FIELD_PIC
     var METHOD_PIC: Boolean = PerfDefaults.METHOD_PIC
 
+    // Optional: expand PIC size for fields/methods from 2 to 4 entries (JVM-first tuning)
+    // Initialized from platform defaults; still runtime-togglable.
+    var FIELD_PIC_SIZE_4: Boolean = PerfDefaults.FIELD_PIC_SIZE_4
+    var METHOD_PIC_SIZE_4: Boolean = PerfDefaults.METHOD_PIC_SIZE_4
+    // Adaptive growth from 2→4 entries per-site for field/method PICs when polymorphism is high (JVM-first)
+    var PIC_ADAPTIVE_2_TO_4: Boolean = PerfDefaults.PIC_ADAPTIVE_2_TO_4
+    // Adaptive growth for methods only (independent of fields)
+    var PIC_ADAPTIVE_METHODS_ONLY: Boolean = PerfDefaults.PIC_ADAPTIVE_METHODS_ONLY
+    // Heuristic to avoid or revert promotion when it shows no benefit (experimental)
+    var PIC_ADAPTIVE_HEURISTIC: Boolean = PerfDefaults.PIC_ADAPTIVE_HEURISTIC
+
+    // Index PIC/fast paths (JVM-first). By default follow FIELD_PIC enablement to avoid extra flags churn.
+    // Host apps/tests can flip independently if needed.
+    var INDEX_PIC: Boolean = FIELD_PIC
+    // Optional 4-entry PIC for IndexRef (JVM-first tuning); initialized from platform defaults
+    var INDEX_PIC_SIZE_4: Boolean = PerfDefaults.INDEX_PIC_SIZE_4
+
     // Debug/observability for PICs and fast paths (JVM-first)
     var PIC_DEBUG_COUNTERS: Boolean = PerfDefaults.PIC_DEBUG_COUNTERS
 
@@ -50,4 +69,7 @@ object PerfFlags {
 
     // Regex: enable small LRU cache for compiled patterns (JVM-first usage)
     var REGEX_CACHE: Boolean = PerfDefaults.REGEX_CACHE
+
+    // Specialized non-allocating integer range iteration in hot loops
+    var RANGE_FAST_ITER: Boolean = PerfDefaults.RANGE_FAST_ITER
 }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/PerfStats.kt

@@ -41,6 +41,10 @@ object PerfStats {
     // Primitive fast ops
     var primitiveFastOpsHit: Long = 0
 
+    // Index PIC
+    var indexPicHit: Long = 0
+    var indexPicMiss: Long = 0
+
     fun resetAll() {
         fieldPicHit = 0
         fieldPicMiss = 0
@@ -53,5 +57,7 @@ object PerfStats {
         fastLocalHit = 0
         fastLocalMiss = 0
         primitiveFastOpsHit = 0
+        indexPicHit = 0
+        indexPicMiss = 0
     }
 }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/obj/ObjRange.kt

@@ -161,6 +161,18 @@ class ObjRange(val start: Obj?, val end: Obj?, val isEndInclusive: Boolean) : Ob
             }
             addFn("iterator") {
                 val self = thisAs<ObjRange>()
+                if (net.sergeych.lyng.PerfFlags.RANGE_FAST_ITER) {
+                    val s = self.start
+                    val e = self.end
+                    if (s is ObjInt && e is ObjInt) {
+                        val start = s.value.toInt()
+                        val endExclusive = (if (self.isEndInclusive) e.value.toInt() + 1 else e.value.toInt())
+                        // Only for ascending simple ranges; fall back otherwise
+                        if (start <= endExclusive) {
+                            return@addFn ObjFastIntRangeIterator(start, endExclusive)
+                        }
+                    }
+                }
                 ObjRangeIterator(self).apply { init() }
             }
         }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/obj/ObjRangeIterator.kt

@@ -17,6 +17,7 @@
 
 package net.sergeych.lyng.obj
 
+import net.sergeych.lyng.PerfFlags
 import net.sergeych.lyng.Scope
 
 class ObjRangeIterator(val self: ObjRange) : Obj() {
@@ -65,4 +66,28 @@ class ObjRangeIterator(val self: ObjRange) : Obj() {
             }
         }
     }
+}
+
+/**
+ * Fast iterator for simple integer ranges (step +1). Returned only when
+ * [PerfFlags.RANGE_FAST_ITER] is enabled and the range is an ascending int range.
+ */
+class ObjFastIntRangeIterator(private val start: Int, private val endExclusive: Int) : Obj() {
+
+    private var cur: Int = start
+
+    override val objClass: ObjClass = type
+
+    fun hasNext(): Boolean = cur < endExclusive
+
+    fun next(scope: Scope): Obj =
+        if (cur < endExclusive) ObjInt(cur++.toLong())
+        else scope.raiseError(ObjIterationFinishedException(scope))
+
+    companion object {
+        val type = ObjClass("FastIntRangeIterator", ObjIterator).apply {
+            addFn("hasNext") { thisAs<ObjFastIntRangeIterator>().hasNext().toObj() }
+            addFn("next") { thisAs<ObjFastIntRangeIterator>().next(this) }
+        }
+    }
 }

lynglib/src/commonMain/kotlin/net/sergeych/lyng/obj/ObjRef.kt

@@ -62,7 +62,23 @@ enum class BinOp {
 /** R-value reference for unary operations. */
 class UnaryOpRef(private val op: UnaryOp, private val a: ObjRef) : ObjRef {
     override suspend fun get(scope: Scope): ObjRecord {
-        val v = if (net.sergeych.lyng.PerfFlags.RVAL_FASTPATH) a.evalValue(scope) else a.get(scope).value
+        val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
+        val v = if (fastRval) a.evalValue(scope) else a.get(scope).value
+        if (net.sergeych.lyng.PerfFlags.PRIMITIVE_FASTOPS) {
+            val rFast: Obj? = when (op) {
+                UnaryOp.NOT -> if (v is ObjBool) if (!v.value) ObjTrue else ObjFalse else null
+                UnaryOp.NEGATE -> when (v) {
+                    is ObjInt -> ObjInt(-v.value)
+                    is ObjReal -> ObjReal(-v.value)
+                    else -> null
+                }
+                UnaryOp.BITNOT -> if (v is ObjInt) ObjInt(v.value.inv()) else null
+            }
+            if (rFast != null) {
+                if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.primitiveFastOpsHit++
+                return rFast.asReadonly
+            }
+        }
         val r = when (op) {
             UnaryOp.NOT -> v.logicalNot(scope)
             UnaryOp.NEGATE -> v.negate(scope)
@@ -303,8 +319,9 @@ class IncDecRef(
 /** Elvis operator reference: a ?: b */
 class ElvisRef(private val left: ObjRef, private val right: ObjRef) : ObjRef {
     override suspend fun get(scope: Scope): ObjRecord {
-        val a = if (net.sergeych.lyng.PerfFlags.RVAL_FASTPATH) left.evalValue(scope) else left.get(scope).value
+        val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
-        val r = if (a != ObjNull) a else if (net.sergeych.lyng.PerfFlags.RVAL_FASTPATH) right.evalValue(scope) else right.get(scope).value
+        val a = if (fastRval) left.evalValue(scope) else left.get(scope).value
+        val r = if (a != ObjNull) a else if (fastRval) right.evalValue(scope) else right.get(scope).value
         return r.asReadonly
     }
 }
@@ -312,10 +329,12 @@ class ElvisRef(private val left: ObjRef, private val right: ObjRef) : ObjRef {
 /** Logical OR with short-circuit: a || b */
 class LogicalOrRef(private val left: ObjRef, private val right: ObjRef) : ObjRef {
     override suspend fun get(scope: Scope): ObjRecord {
-        val a = if (net.sergeych.lyng.PerfFlags.RVAL_FASTPATH) left.evalValue(scope) else left.get(scope).value
+        val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
+        val fastPrim = net.sergeych.lyng.PerfFlags.PRIMITIVE_FASTOPS
+        val a = if (fastRval) left.evalValue(scope) else left.get(scope).value
         if ((a as? ObjBool)?.value == true) return ObjTrue.asReadonly
-        val b = if (net.sergeych.lyng.PerfFlags.RVAL_FASTPATH) right.evalValue(scope) else right.get(scope).value
+        val b = if (fastRval) right.evalValue(scope) else right.get(scope).value
-        if (net.sergeych.lyng.PerfFlags.PRIMITIVE_FASTOPS) {
+        if (fastPrim) {
             if (a is ObjBool && b is ObjBool) {
                 return if (a.value || b.value) ObjTrue.asReadonly else ObjFalse.asReadonly
             }
@@ -347,6 +366,8 @@ class LogicalAndRef(private val left: ObjRef, private val right: ObjRef) : ObjRe
  */
 class ConstRef(private val record: ObjRecord) : ObjRef {
     override suspend fun get(scope: Scope): ObjRecord = record
+    // Expose constant value for compiler constant folding (pure, read-only)
+    val constValue: Obj get() = record.value
 }
 
 /**
@@ -373,6 +394,48 @@ class FieldRef(
     // Transient per-step cache to optimize read-then-write sequences within the same frame
     private var tKey: Long = 0L; private var tVer: Int = -1; private var tFrameId: Long = -1L; private var tRecord: ObjRecord? = null
 
+    // Adaptive PIC (2→4) for reads/writes
+    private var rAccesses: Int = 0; private var rMisses: Int = 0; private var rPromotedTo4: Boolean = false
+    private var wAccesses: Int = 0; private var wMisses: Int = 0; private var wPromotedTo4: Boolean = false
+    private inline fun size4ReadsEnabled(): Boolean =
+        net.sergeych.lyng.PerfFlags.FIELD_PIC_SIZE_4 ||
+            (net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4 && rPromotedTo4)
+    private inline fun size4WritesEnabled(): Boolean =
+        net.sergeych.lyng.PerfFlags.FIELD_PIC_SIZE_4 ||
+            (net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4 && wPromotedTo4)
+    private fun noteReadHit() {
+        if (!net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4) return
+        val a = (rAccesses + 1).coerceAtMost(1_000_000)
+        rAccesses = a
+    }
+    private fun noteReadMiss() {
+        if (!net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4) return
+        val a = (rAccesses + 1).coerceAtMost(1_000_000)
+        rAccesses = a
+        rMisses = (rMisses + 1).coerceAtMost(1_000_000)
+        if (!rPromotedTo4 && a >= 256) {
+            // promote if miss rate > 20%
+            if (rMisses * 100 / a > 20) rPromotedTo4 = true
+            // reset counters after decision
+            rAccesses = 0; rMisses = 0
+        }
+    }
+    private fun noteWriteHit() {
+        if (!net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4) return
+        val a = (wAccesses + 1).coerceAtMost(1_000_000)
+        wAccesses = a
+    }
+    private fun noteWriteMiss() {
+        if (!net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4) return
+        val a = (wAccesses + 1).coerceAtMost(1_000_000)
+        wAccesses = a
+        wMisses = (wMisses + 1).coerceAtMost(1_000_000)
+        if (!wPromotedTo4 && a >= 256) {
+            if (wMisses * 100 / a > 20) wPromotedTo4 = true
+            wAccesses = 0; wMisses = 0
+        }
+    }
+
     override suspend fun get(scope: Scope): ObjRecord {
         val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
         val fieldPic = net.sergeych.lyng.PerfFlags.FIELD_PIC
@@ -383,6 +446,7 @@ class FieldRef(
             val (key, ver) = receiverKeyAndVersion(base)
             rGetter1?.let { g -> if (key == rKey1 && ver == rVer1) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                noteReadHit()
                 val rec0 = g(base, scope)
                 if (base is ObjClass) {
                     val idx0 = base.classScope?.getSlotIndexOf(name)
@@ -392,6 +456,7 @@ class FieldRef(
             } }
             rGetter2?.let { g -> if (key == rKey2 && ver == rVer2) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                noteReadHit()
                 // move-to-front: promote 2→1
                 val tK = rKey2; val tV = rVer2; val tG = rGetter2
                 rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
@@ -403,8 +468,9 @@ class FieldRef(
                 } else { tRecord = null }
                 return rec0
             } }
-            rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
+            if (size4ReadsEnabled()) rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                noteReadHit()
                 // move-to-front: promote 3→1
                 val tK = rKey3; val tV = rVer3; val tG = rGetter3
                 rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
@@ -417,8 +483,9 @@ class FieldRef(
                 } else { tRecord = null }
                 return rec0
             } }
-            rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
+            if (size4ReadsEnabled()) rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                noteReadHit()
                 // move-to-front: promote 4→1
                 val tK = rKey4; val tV = rVer4; val tG = rGetter4
                 rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
@@ -434,6 +501,7 @@ class FieldRef(
             } }
             // Slow path
             if (picCounters) net.sergeych.lyng.PerfStats.fieldPicMiss++
+            noteReadMiss()
             val rec = try {
                 base.readField(scope, name)
             } catch (e: ExecutionError) {
@@ -444,9 +512,11 @@ class FieldRef(
                 rKey1 = key; rVer1 = ver; rGetter1 = { _, sc -> sc.raiseError(e.message ?: "no such field: $name") }
                 throw e
             }
-            // Install move-to-front with a handle-aware getter (shift 1→2→3→4; put new at 1)
+            // Install move-to-front with a handle-aware getter; honor PIC size flag
-            rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+            if (size4ReadsEnabled()) {
-            rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+            }
             rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
             when (base) {
                 is ObjClass -> {
@@ -499,18 +569,21 @@ class FieldRef(
             val (key, ver) = receiverKeyAndVersion(base)
             wSetter1?.let { s -> if (key == wKey1 && ver == wVer1) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicSetHit++
+                noteWriteHit()
                 return s(base, scope, newValue)
             } }
             wSetter2?.let { s -> if (key == wKey2 && ver == wVer2) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicSetHit++
+                noteWriteHit()
                 // move-to-front: promote 2→1
                 val tK = wKey2; val tV = wVer2; val tS = wSetter2
                 wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
                 wKey1 = tK; wVer1 = tV; wSetter1 = tS
                 return s(base, scope, newValue)
             } }
-            wSetter3?.let { s -> if (key == wKey3 && ver == wVer3) {
+            if (size4WritesEnabled()) wSetter3?.let { s -> if (key == wKey3 && ver == wVer3) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicSetHit++
+                noteWriteHit()
                 // move-to-front: promote 3→1
                 val tK = wKey3; val tV = wVer3; val tS = wSetter3
                 wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
@@ -518,8 +591,9 @@ class FieldRef(
                 wKey1 = tK; wVer1 = tV; wSetter1 = tS
                 return s(base, scope, newValue)
             } }
-            wSetter4?.let { s -> if (key == wKey4 && ver == wVer4) {
+            if (size4WritesEnabled()) wSetter4?.let { s -> if (key == wKey4 && ver == wVer4) {
                 if (picCounters) net.sergeych.lyng.PerfStats.fieldPicSetHit++
+                noteWriteHit()
                 // move-to-front: promote 4→1
                 val tK = wKey4; val tV = wVer4; val tS = wSetter4
                 wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
@@ -530,10 +604,13 @@ class FieldRef(
             } }
             // Slow path
             if (picCounters) net.sergeych.lyng.PerfStats.fieldPicSetMiss++
+            noteWriteMiss()
             base.writeField(scope, name, newValue)
-            // Install move-to-front with a handle-aware setter (shift 1→2→3→4; put new at 1)
+            // Install move-to-front with a handle-aware setter; honor PIC size flag
-            wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
+            if (size4WritesEnabled()) {
-            wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
+                wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
+                wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
+            }
             wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
             when (base) {
                 is ObjClass -> {
@@ -566,6 +643,63 @@ class FieldRef(
         is ObjClass -> obj.classId to obj.layoutVersion
         else -> 0L to -1 // no caching for primitives/dynamics without stable shape
     }
+
+    override suspend fun evalValue(scope: Scope): Obj {
+        // Mirror get(), but return raw Obj to avoid transient ObjRecord on R-value paths
+        val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
+        val fieldPic = net.sergeych.lyng.PerfFlags.FIELD_PIC
+        val picCounters = net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS
+        val base = if (fastRval) target.evalValue(scope) else target.get(scope).value
+        if (base == ObjNull && isOptional) return ObjNull
+        if (fieldPic) {
+            val (key, ver) = receiverKeyAndVersion(base)
+            rGetter1?.let { g -> if (key == rKey1 && ver == rVer1) {
+                if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                return g(base, scope).value
+            } }
+            rGetter2?.let { g -> if (key == rKey2 && ver == rVer2) {
+                if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                val tK = rKey2; val tV = rVer2; val tG = rGetter2
+                rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                rKey1 = tK; rVer1 = tV; rGetter1 = tG
+                return g(base, scope).value
+            } }
+            if (size4ReadsEnabled()) rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
+                if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                val tK = rKey3; val tV = rVer3; val tG = rGetter3
+                rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                rKey1 = tK; rVer1 = tV; rGetter1 = tG
+                return g(base, scope).value
+            } }
+            if (size4ReadsEnabled()) rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
+                if (picCounters) net.sergeych.lyng.PerfStats.fieldPicHit++
+                val tK = rKey4; val tV = rVer4; val tG = rGetter4
+                rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                rKey1 = tK; rVer1 = tV; rGetter1 = tG
+                return g(base, scope).value
+            } }
+            if (picCounters) net.sergeych.lyng.PerfStats.fieldPicMiss++
+            val rec = base.readField(scope, name)
+            // install primary generic getter for this shape
+            when (base) {
+                is ObjClass -> {
+                    rKey1 = base.classId; rVer1 = base.layoutVersion; rGetter1 = { obj, sc -> obj.readField(sc, name) }
+                }
+                is ObjInstance -> {
+                    val cls = base.objClass
+                    rKey1 = cls.classId; rVer1 = cls.layoutVersion; rGetter1 = { obj, sc -> obj.readField(sc, name) }
+                }
+                else -> {
+                    rKey1 = 0L; rVer1 = -1; rGetter1 = null
+                }
+            }
+            return rec.value
+        }
+        return base.readField(scope, name).value
+    }
 }
 
 /**
@@ -605,47 +739,139 @@ class IndexRef(
                 // Bounds checks are enforced by the underlying list access; exceptions propagate as before
                 return base.list[i].asMutable
             }
-            // Polymorphic inline cache for other common shapes
+            // String[Int] fast path
-            val (key, ver) = when (base) {
+            if (base is ObjString && idx is ObjInt) {
-                is ObjInstance -> base.objClass.classId to base.objClass.layoutVersion
+                val i = idx.toInt()
-                is ObjClass -> base.classId to base.layoutVersion
+                return ObjChar(base.value[i]).asMutable
-                else -> 0L to -1
             }
-            if (key != 0L) {
+            // Map[String] fast path (common case); return ObjNull if absent
-                rGetter1?.let { g -> if (key == rKey1 && ver == rVer1) return g(base, scope, idx).asMutable }
+            if (base is ObjMap && idx is ObjString) {
-                rGetter2?.let { g -> if (key == rKey2 && ver == rVer2) {
+                val v = base.map[idx] ?: ObjNull
-                    val tk = rKey2; val tv = rVer2; val tg = rGetter2
-                    rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
-                    rKey1 = tk; rVer1 = tv; rGetter1 = tg
-                    return g(base, scope, idx).asMutable
-                } }
-                rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
-                    val tk = rKey3; val tv = rVer3; val tg = rGetter3
-                    rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
-                    rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
-                    rKey1 = tk; rVer1 = tv; rGetter1 = tg
-                    return g(base, scope, idx).asMutable
-                } }
-                rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
-                    val tk = rKey4; val tv = rVer4; val tg = rGetter4
-                    rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
-                    rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
-                    rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
-                    rKey1 = tk; rVer1 = tv; rGetter1 = tg
-                    return g(base, scope, idx).asMutable
-                } }
-                // Miss: resolve and install generic handler
-                val v = base.getAt(scope, idx)
-                rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
-                rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
-                rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
-                rKey1 = key; rVer1 = ver; rGetter1 = { obj, sc, ix -> obj.getAt(sc, ix) }
                 return v.asMutable
             }
+            if (net.sergeych.lyng.PerfFlags.INDEX_PIC) {
+                // Polymorphic inline cache for other common shapes
+                val (key, ver) = when (base) {
+                    is ObjInstance -> base.objClass.classId to base.objClass.layoutVersion
+                    is ObjClass -> base.classId to base.layoutVersion
+                    else -> 0L to -1
+                }
+                if (key != 0L) {
+                    rGetter1?.let { g -> if (key == rKey1 && ver == rVer1) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        return g(base, scope, idx).asMutable
+                    } }
+                    rGetter2?.let { g -> if (key == rKey2 && ver == rVer2) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey2; val tv = rVer2; val tg = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx).asMutable
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey3; val tv = rVer3; val tg = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx).asMutable
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey4; val tv = rVer4; val tg = rGetter4
+                        rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx).asMutable
+                    } }
+                    // Miss: resolve and install generic handler
+                    if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicMiss++
+                    val v = base.getAt(scope, idx)
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) {
+                        rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                    }
+                    rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                    rKey1 = key; rVer1 = ver; rGetter1 = { obj, sc, ix -> obj.getAt(sc, ix) }
+                    return v.asMutable
+                }
+            }
         }
         return base.getAt(scope, idx).asMutable
     }
 
+    override suspend fun evalValue(scope: Scope): Obj {
+        val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
+        val base = if (fastRval) target.evalValue(scope) else target.get(scope).value
+        if (base == ObjNull && isOptional) return ObjNull
+        val idx = if (fastRval) index.evalValue(scope) else index.get(scope).value
+        if (fastRval) {
+            // Fast list[int] path
+            if (base is ObjList && idx is ObjInt) {
+                val i = idx.toInt()
+                return base.list[i]
+            }
+            // String[Int] fast path
+            if (base is ObjString && idx is ObjInt) {
+                val i = idx.toInt()
+                return ObjChar(base.value[i])
+            }
+            // Map[String] fast path
+            if (base is ObjMap && idx is ObjString) {
+                return base.map[idx] ?: ObjNull
+            }
+            if (net.sergeych.lyng.PerfFlags.INDEX_PIC) {
+                // PIC path analogous to get(), but returning raw Obj
+                val (key, ver) = when (base) {
+                    is ObjInstance -> base.objClass.classId to base.objClass.layoutVersion
+                    is ObjClass -> base.classId to base.layoutVersion
+                    else -> 0L to -1
+                }
+                if (key != 0L) {
+                    rGetter1?.let { g -> if (key == rKey1 && ver == rVer1) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        return g(base, scope, idx)
+                    } }
+                    rGetter2?.let { g -> if (key == rKey2 && ver == rVer2) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey2; val tv = rVer2; val tg = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx)
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) rGetter3?.let { g -> if (key == rKey3 && ver == rVer3) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey3; val tv = rVer3; val tg = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx)
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) rGetter4?.let { g -> if (key == rKey4 && ver == rVer4) {
+                        if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicHit++
+                        val tk = rKey4; val tv = rVer4; val tg = rGetter4
+                        rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                        rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                        rKey1 = tk; rVer1 = tv; rGetter1 = tg
+                        return g(base, scope, idx)
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.PIC_DEBUG_COUNTERS) net.sergeych.lyng.PerfStats.indexPicMiss++
+                    val v = base.getAt(scope, idx)
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) {
+                        rKey4 = rKey3; rVer4 = rVer3; rGetter4 = rGetter3
+                        rKey3 = rKey2; rVer3 = rVer2; rGetter3 = rGetter2
+                    }
+                    rKey2 = rKey1; rVer2 = rVer1; rGetter2 = rGetter1
+                    rKey1 = key; rVer1 = ver; rGetter1 = { obj, sc, ix -> obj.getAt(sc, ix) }
+                    return v
+                }
+            }
+        }
+        return base.getAt(scope, idx)
+    }
+
     override suspend fun setAt(pos: Pos, scope: Scope, newValue: Obj) {
         val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
         val base = if (fastRval) target.evalValue(scope) else target.get(scope).value
@@ -661,43 +887,52 @@ class IndexRef(
                 base.list[i] = newValue
                 return
             }
-            // Polymorphic inline cache for index write
+            // Direct write fast path for ObjMap + ObjString
-            val (key, ver) = when (base) {
+            if (base is ObjMap && idx is ObjString) {
-                is ObjInstance -> base.objClass.classId to base.objClass.layoutVersion
+                base.map[idx] = newValue
-                is ObjClass -> base.classId to base.layoutVersion
-                else -> 0L to -1
-            }
-            if (key != 0L) {
-                wSetter1?.let { s -> if (key == wKey1 && ver == wVer1) { s(base, scope, idx, newValue); return } }
-                wSetter2?.let { s -> if (key == wKey2 && ver == wVer2) {
-                    val tk = wKey2; val tv = wVer2; val ts = wSetter2
-                    wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
-                    wKey1 = tk; wVer1 = tv; wSetter1 = ts
-                    s(base, scope, idx, newValue); return
-                } }
-                wSetter3?.let { s -> if (key == wKey3 && ver == wVer3) {
-                    val tk = wKey3; val tv = wVer3; val ts = wSetter3
-                    wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
-                    wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
-                    wKey1 = tk; wVer1 = tv; wSetter1 = ts
-                    s(base, scope, idx, newValue); return
-                } }
-                wSetter4?.let { s -> if (key == wKey4 && ver == wVer4) {
-                    val tk = wKey4; val tv = wVer4; val ts = wSetter4
-                    wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
-                    wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
-                    wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
-                    wKey1 = tk; wVer1 = tv; wSetter1 = ts
-                    s(base, scope, idx, newValue); return
-                } }
-                // Miss: perform write and install generic handler
-                base.putAt(scope, idx, newValue)
-                wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
-                wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
-                wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
-                wKey1 = key; wVer1 = ver; wSetter1 = { obj, sc, ix, v -> obj.putAt(sc, ix, v) }
                 return
             }
+            if (net.sergeych.lyng.PerfFlags.INDEX_PIC) {
+                // Polymorphic inline cache for index write
+                val (key, ver) = when (base) {
+                    is ObjInstance -> base.objClass.classId to base.objClass.layoutVersion
+                    is ObjClass -> base.classId to base.layoutVersion
+                    else -> 0L to -1
+                }
+                if (key != 0L) {
+                    wSetter1?.let { s -> if (key == wKey1 && ver == wVer1) { s(base, scope, idx, newValue); return } }
+                    wSetter2?.let { s -> if (key == wKey2 && ver == wVer2) {
+                        val tk = wKey2; val tv = wVer2; val ts = wSetter2
+                        wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
+                        wKey1 = tk; wVer1 = tv; wSetter1 = ts
+                        s(base, scope, idx, newValue); return
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) wSetter3?.let { s -> if (key == wKey3 && ver == wVer3) {
+                        val tk = wKey3; val tv = wVer3; val ts = wSetter3
+                        wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
+                        wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
+                        wKey1 = tk; wVer1 = tv; wSetter1 = ts
+                        s(base, scope, idx, newValue); return
+                    } }
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) wSetter4?.let { s -> if (key == wKey4 && ver == wVer4) {
+                        val tk = wKey4; val tv = wVer4; val ts = wSetter4
+                        wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
+                        wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
+                        wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
+                        wKey1 = tk; wVer1 = tv; wSetter1 = ts
+                        s(base, scope, idx, newValue); return
+                    } }
+                    // Miss: perform write and install generic handler
+                    base.putAt(scope, idx, newValue)
+                    if (net.sergeych.lyng.PerfFlags.INDEX_PIC_SIZE_4) {
+                        wKey4 = wKey3; wVer4 = wVer3; wSetter4 = wSetter3
+                        wKey3 = wKey2; wVer3 = wVer2; wSetter3 = wSetter2
+                    }
+                    wKey2 = wKey1; wVer2 = wVer1; wSetter2 = wSetter1
+                    wKey1 = key; wVer1 = ver; wSetter1 = { obj, sc, ix, v -> obj.putAt(sc, ix, v) }
+                    return
+                }
+            }
         }
         base.putAt(scope, idx, newValue)
     }
@@ -752,6 +987,63 @@ class MethodCallRef(
     private var mKey3: Long = 0L; private var mVer3: Int = -1; private var mInvoker3: (suspend (Obj, Scope, Arguments) -> Obj)? = null
     private var mKey4: Long = 0L; private var mVer4: Int = -1; private var mInvoker4: (suspend (Obj, Scope, Arguments) -> Obj)? = null
 
+    // Adaptive PIC (2→4) for methods
+    private var mAccesses: Int = 0; private var mMisses: Int = 0; private var mPromotedTo4: Boolean = false
+    // Heuristic: windowed miss-rate tracking and temporary freeze back to size=2
+    private var mFreezeWindowsLeft: Int = 0
+    private var mWindowAccesses: Int = 0
+    private var mWindowMisses: Int = 0
+    private inline fun size4MethodsEnabled(): Boolean =
+        net.sergeych.lyng.PerfFlags.METHOD_PIC_SIZE_4 ||
+            ((net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4 || net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_METHODS_ONLY) && mPromotedTo4 && mFreezeWindowsLeft == 0)
+    private fun noteMethodHit() {
+        if (!(net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4 || net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_METHODS_ONLY)) return
+        val a = (mAccesses + 1).coerceAtMost(1_000_000)
+        mAccesses = a
+        if (net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_HEURISTIC) {
+            // Windowed tracking
+            mWindowAccesses = (mWindowAccesses + 1).coerceAtMost(1_000_000)
+            if (mWindowAccesses >= 256) endHeuristicWindow()
+        }
+    }
+    private fun noteMethodMiss() {
+        if (!(net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_2_TO_4 || net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_METHODS_ONLY)) return
+        val a = (mAccesses + 1).coerceAtMost(1_000_000)
+        mAccesses = a
+        mMisses = (mMisses + 1).coerceAtMost(1_000_000)
+        if (!mPromotedTo4 && mFreezeWindowsLeft == 0 && a >= 256) {
+            if (mMisses * 100 / a > 20) mPromotedTo4 = true
+            mAccesses = 0; mMisses = 0
+        }
+        if (net.sergeych.lyng.PerfFlags.PIC_ADAPTIVE_HEURISTIC) {
+            mWindowAccesses = (mWindowAccesses + 1).coerceAtMost(1_000_000)
+            mWindowMisses = (mWindowMisses + 1).coerceAtMost(1_000_000)
+            if (mWindowAccesses >= 256) endHeuristicWindow()
+        }
+    }
+
+    private fun endHeuristicWindow() {
+        // Called only when PIC_ADAPTIVE_HEURISTIC is true
+        val accesses = mWindowAccesses
+        val misses = mWindowMisses
+        // Reset window
+        mWindowAccesses = 0
+        mWindowMisses = 0
+        // Count down freeze if active
+        if (mFreezeWindowsLeft > 0) {
+            mFreezeWindowsLeft = (mFreezeWindowsLeft - 1).coerceAtLeast(0)
+            return
+        }
+        // If promoted, but still high miss rate, freeze back to 2 for a few windows
+        if (mPromotedTo4 && accesses >= 256) {
+            val rate = misses * 100 / accesses
+            if (rate >= 25) {
+                mPromotedTo4 = false
+                mFreezeWindowsLeft = 4 // freeze next 4 windows
+            }
+        }
+    }
+
     override suspend fun get(scope: Scope): ObjRecord {
         val fastRval = net.sergeych.lyng.PerfFlags.RVAL_FASTPATH
         val methodPic = net.sergeych.lyng.PerfFlags.METHOD_PIC
@@ -763,14 +1055,17 @@ class MethodCallRef(
             val (key, ver) = receiverKeyAndVersion(base)
             mInvoker1?.let { inv -> if (key == mKey1 && ver == mVer1) {
                 if (picCounters) net.sergeych.lyng.PerfStats.methodPicHit++
+                noteMethodHit()
                 return inv(base, scope, callArgs).asReadonly
             } }
             mInvoker2?.let { inv -> if (key == mKey2 && ver == mVer2) {
                 if (picCounters) net.sergeych.lyng.PerfStats.methodPicHit++
+                noteMethodHit()
                 return inv(base, scope, callArgs).asReadonly
             } }
-            mInvoker3?.let { inv -> if (key == mKey3 && ver == mVer3) {
+            if (size4MethodsEnabled()) mInvoker3?.let { inv -> if (key == mKey3 && ver == mVer3) {
                 if (picCounters) net.sergeych.lyng.PerfStats.methodPicHit++
+                noteMethodHit()
                 // move-to-front: promote 3→1
                 val tK = mKey3; val tV = mVer3; val tI = mInvoker3
                 mKey3 = mKey2; mVer3 = mVer2; mInvoker3 = mInvoker2
@@ -778,8 +1073,9 @@ class MethodCallRef(
                 mKey1 = tK; mVer1 = tV; mInvoker1 = tI
                 return inv(base, scope, callArgs).asReadonly
             } }
-            mInvoker4?.let { inv -> if (key == mKey4 && ver == mVer4) {
+            if (size4MethodsEnabled()) mInvoker4?.let { inv -> if (key == mKey4 && ver == mVer4) {
                 if (picCounters) net.sergeych.lyng.PerfStats.methodPicHit++
+                noteMethodHit()
                 // move-to-front: promote 4→1
                 val tK = mKey4; val tV = mVer4; val tI = mInvoker4
                 mKey4 = mKey3; mVer4 = mVer3; mInvoker4 = mInvoker3
@@ -790,6 +1086,7 @@ class MethodCallRef(
             } }
             // Slow path
             if (picCounters) net.sergeych.lyng.PerfStats.methodPicMiss++
+            noteMethodMiss()
             val result = try {
                 base.invokeInstanceMethod(scope, name, callArgs)
             } catch (e: ExecutionError) {
@@ -800,9 +1097,11 @@ class MethodCallRef(
                 mKey1 = key; mVer1 = ver; mInvoker1 = { _, sc, _ -> sc.raiseError(e.message ?: "method not found: $name") }
                 throw e
             }
-            // Install move-to-front with a handle-aware invoker: shift 1→2→3→4, put new at 1
+            // Install move-to-front with a handle-aware invoker; honor PIC size flag
-            mKey4 = mKey3; mVer4 = mVer3; mInvoker4 = mInvoker3
+            if (size4MethodsEnabled()) {
-            mKey3 = mKey2; mVer3 = mVer2; mInvoker3 = mInvoker2
+                mKey4 = mKey3; mVer4 = mVer3; mInvoker4 = mInvoker3
+                mKey3 = mKey2; mVer3 = mVer2; mInvoker3 = mInvoker2
+            }
             mKey2 = mKey1; mVer2 = mVer1; mInvoker2 = mInvoker1
             when (base) {
                 is ObjInstance -> {
@@ -889,6 +1188,18 @@ class LocalVarRef(private val name: String, private val atPos: Pos) : ObjRef {
         return scope[name] ?: scope.raiseError("symbol not defined: '$name'")
     }
 
+    override suspend fun evalValue(scope: Scope): Obj {
+        scope.pos = atPos
+        if (!PerfFlags.LOCAL_SLOT_PIC) {
+            scope.getSlotIndexOf(name)?.let { return scope.getSlotRecord(it).value }
+            return (scope[name] ?: scope.raiseError("symbol not defined: '$name'")).value
+        }
+        val hit = (cachedFrameId == scope.frameId && cachedSlot >= 0 && cachedSlot < scope.slotCount())
+        val slot = if (hit) cachedSlot else resolveSlot(scope)
+        if (slot >= 0) return scope.getSlotRecord(slot).value
+        return (scope[name] ?: scope.raiseError("symbol not defined: '$name'")).value
+    }
+
     override suspend fun setAt(pos: Pos, scope: Scope, newValue: Obj) {
         scope.pos = atPos
         if (!PerfFlags.LOCAL_SLOT_PIC) {
@@ -929,6 +1240,11 @@ class BoundLocalVarRef(
         return scope.getSlotRecord(slot)
     }
 
+    override suspend fun evalValue(scope: Scope): Obj {
+        scope.pos = atPos
+        return scope.getSlotRecord(slot).value
+    }
+
     override suspend fun setAt(pos: Pos, scope: Scope, newValue: Obj) {
         scope.pos = atPos
         val rec = scope.getSlotRecord(slot)
@@ -989,6 +1305,15 @@ class FastLocalVarRef(
         return actualOwner.getSlotRecord(slot)
     }
 
+    override suspend fun evalValue(scope: Scope): Obj {
+        scope.pos = atPos
+        val ownerValid = isOwnerValidFor(scope)
+        val slot = if (ownerValid && cachedSlot >= 0) cachedSlot else resolveSlotInAncestry(scope)
+        val actualOwner = cachedOwnerScope
+        if (slot < 0 || actualOwner == null) scope.raiseError("local '$name' is not available in this scope")
+        return actualOwner.getSlotRecord(slot).value
+    }
+
     override suspend fun setAt(pos: Pos, scope: Scope, newValue: Obj) {
         scope.pos = atPos
         val owner = if (isOwnerValidFor(scope)) cachedOwnerScope else null

lynglib/src/jsMain/kotlin/net/sergeych/lyng/PerfDefaults.js.kt

@@ -27,6 +27,11 @@ actual object PerfDefaults {
 
     actual val FIELD_PIC: Boolean = true
     actual val METHOD_PIC: Boolean = true
+    actual val FIELD_PIC_SIZE_4: Boolean = false
+    actual val METHOD_PIC_SIZE_4: Boolean = false
+    actual val PIC_ADAPTIVE_2_TO_4: Boolean = false
+    actual val PIC_ADAPTIVE_METHODS_ONLY: Boolean = false
+    actual val PIC_ADAPTIVE_HEURISTIC: Boolean = false
 
     actual val PIC_DEBUG_COUNTERS: Boolean = false
 
@@ -35,4 +40,7 @@ actual object PerfDefaults {
     actual val RVAL_FASTPATH: Boolean = false
     // Regex caching: disabled by default on JS until validated
     actual val REGEX_CACHE: Boolean = false
+    actual val ARG_SMALL_ARITY_12: Boolean = false
+    actual val INDEX_PIC_SIZE_4: Boolean = false
+    actual val RANGE_FAST_ITER: Boolean = false
 }

lynglib/src/jvmMain/kotlin/net/sergeych/lyng/PerfDefaults.jvm.kt

@@ -27,6 +27,11 @@ actual object PerfDefaults {
 
     actual val FIELD_PIC: Boolean = true
     actual val METHOD_PIC: Boolean = true
+    actual val FIELD_PIC_SIZE_4: Boolean = false
+    actual val METHOD_PIC_SIZE_4: Boolean = false
+    actual val PIC_ADAPTIVE_2_TO_4: Boolean = false
+    actual val PIC_ADAPTIVE_METHODS_ONLY: Boolean = false
+    actual val PIC_ADAPTIVE_HEURISTIC: Boolean = false
 
     actual val PIC_DEBUG_COUNTERS: Boolean = false
 
@@ -35,4 +40,13 @@ actual object PerfDefaults {
 
     // Regex caching (JVM-first): enabled by default on JVM
     actual val REGEX_CACHE: Boolean = true
+
+    // Extended small-arity calls 9..12 (experimental; keep OFF by default)
+    actual val ARG_SMALL_ARITY_12: Boolean = false
+
+    // Index PIC size (beneficial on JVM in A/B): enable size=4 by default
+    actual val INDEX_PIC_SIZE_4: Boolean = true
+
+    // Range fast-iteration (experimental; OFF by default)
+    actual val RANGE_FAST_ITER: Boolean = false
 }

lynglib/src/nativeMain/kotlin/net/sergeych/lyng/PerfDefaults.native.kt

@@ -27,6 +27,11 @@ actual object PerfDefaults {
 
     actual val FIELD_PIC: Boolean = true
     actual val METHOD_PIC: Boolean = true
+    actual val FIELD_PIC_SIZE_4: Boolean = false
+    actual val METHOD_PIC_SIZE_4: Boolean = false
+    actual val PIC_ADAPTIVE_2_TO_4: Boolean = false
+    actual val PIC_ADAPTIVE_METHODS_ONLY: Boolean = false
+    actual val PIC_ADAPTIVE_HEURISTIC: Boolean = false
 
     actual val PIC_DEBUG_COUNTERS: Boolean = false
 
@@ -35,4 +40,7 @@ actual object PerfDefaults {
     actual val RVAL_FASTPATH: Boolean = false
     // Regex caching: keep OFF by default on Native until benchmarks validate it
     actual val REGEX_CACHE: Boolean = false
+    actual val ARG_SMALL_ARITY_12: Boolean = false
+    actual val INDEX_PIC_SIZE_4: Boolean = false
+    actual val RANGE_FAST_ITER: Boolean = false
 }

lynglib/src/wasmJsMain/kotlin/net/sergeych/lyng/PerfDefaults.wasmJs.kt

@@ -27,6 +27,11 @@ actual object PerfDefaults {
 
     actual val FIELD_PIC: Boolean = true
     actual val METHOD_PIC: Boolean = true
+    actual val FIELD_PIC_SIZE_4: Boolean = false
+    actual val METHOD_PIC_SIZE_4: Boolean = false
+    actual val PIC_ADAPTIVE_2_TO_4: Boolean = false
+    actual val PIC_ADAPTIVE_METHODS_ONLY: Boolean = false
+    actual val PIC_ADAPTIVE_HEURISTIC: Boolean = false
 
     actual val PIC_DEBUG_COUNTERS: Boolean = false
 
@@ -35,4 +40,7 @@ actual object PerfDefaults {
     actual val RVAL_FASTPATH: Boolean = false
     // Regex caching: disabled by default on WasmJS until validated
     actual val REGEX_CACHE: Boolean = false
+    actual val ARG_SMALL_ARITY_12: Boolean = false
+    actual val INDEX_PIC_SIZE_4: Boolean = false
+    actual val RANGE_FAST_ITER: Boolean = false
 }