diff --git a/Makefile b/Makefile
index d6e2d54..6832ada 100644
--- a/Makefile
+++ b/Makefile
@@ -1,6 +1,6 @@
 WASM2WAT=wasm2wat
 WAT2WASM=wat2wasm
-WAT2WASM_FLAGS=
+WAT2WASM_FLAGS=--enable-mutable-globals
 ifeq ($(DEBUG),1)
 WAT2WASM_FLAGS=--debug-names
 endif
diff --git a/src/shell/WAForth.js b/src/shell/WAForth.js
index f24a72c..9cbfdb0 100644
--- a/src/shell/WAForth.js
+++ b/src/shell/WAForth.js
@@ -25,6 +25,7 @@ class WAForth {
     const { skipPrelude } = options;
     let table;
     let memory;
+    let tos;
     const buffer = (this.buffer = []);
 
     return WebAssembly.instantiate(this.wasmModule, {
@@ -89,7 +90,7 @@ class WAForth {
           // console.log("Load", index, this.arrayToBase64(data));
           var module = new WebAssembly.Module(data);
           new WebAssembly.Instance(module, {
-            env: { table, memory, tos: -1 }
+            env: { table, memory, tos }
           });
         }
       }
@@ -97,6 +98,7 @@ class WAForth {
       this.core = instance.instance;
       table = this.core.exports.table;
       memory = this.core.exports.memory;
+      tos = this.core.exports.tos;
       if (!skipPrelude) {
         this.core.exports.loadPrelude();
       }
diff --git a/src/tools/quadruple.wat b/src/tools/quadruple.wat
index eff6fb9..3c0e236 100644
--- a/src/tools/quadruple.wat
+++ b/src/tools/quadruple.wat
@@ -3,7 +3,7 @@
 (module $quadruple
   (import "env" "table" (table 4 anyfunc))
   (import "env" "memory" (memory 1))
-  (import "env" "tos" (global $tos i32))
+  (import "env" "tos" (global $tos (mut i32)))
 
   (type $void (func))
   (type $push (func (param i32)))
@@ -16,13 +16,16 @@
     (local $incr1 i32)
 
     ;; Push
-    (call_indirect (type $push) (i32.const 43) (i32.const 1))
+    (i32.store (get_global $tos) (i32.const 43))
+    (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
 
     ;; Word call
     (call_indirect (type $push) (i32.const 10) (i32.const 9))
 
     ;; Conditional
-    (if (i32.ne (call_indirect (type $pop) (i32.const 2)) (i32.const 0))
+    (i32.load (get_global $tos))
+    (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+    (if (i32.ne (i32.const 0))
       (then
         (nop)
         (nop))
diff --git a/src/waforth.wat b/src/waforth.wat
index 9d537d8..0879b69 100644
--- a/src/waforth.wat
+++ b/src/waforth.wat
@@ -56,7 +56,7 @@
       "\u0003\u0065\u006E\u0076" "\u0006\u006d\u0065\u006d\u006f\u0072\u0079" ;; 'env' . 'memory'
         "\u0002" "\u0000" "\u0001" ;; memory
       "\u0003\u0065\u006E\u0076" "\u0003\u0074\u006f\u0073" ;; 'env' . 'tos'
-        "\u0003" "\u007F" "\u0000" ;; global, i32, immutable
+        "\u0003" "\u007F" "\u0001" ;; global, i32, mutable
 
     
     "\u0003" "\u0002" ;; Function section
@@ -109,6 +109,9 @@
 (define !compileCallIndex 7)
 (define !tableStartIndex 8)
 
+;; Predefined imported globals
+(define !tosIndex 0)
+
 (define !dictionaryLatest 0)
 (define !dictionaryTop !dictionaryBase)
 
@@ -179,7 +182,7 @@
   (type $word (func))
   (type $dataWord (func (param i32)))
 
-  (global $tos (mut i32) (i32.const !stackBase))
+  (global $tos (export "tos") (mut i32) (i32.const !stackBase))
   (global $tors (mut i32) (i32.const !returnStackBase))
   (global $inputBufferSize (mut i32) (i32.const 0))
   (global $inputBufferBase (mut i32) (i32.const !inputBufferBase))
@@ -1873,10 +1876,40 @@ EOF
     (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
     (set_global $cp (call $leb128 (get_global $cp) (get_local $n))))
 
+  (func $emitSetGlobal (param $n i32)
+    (i32.store8 (get_global $cp) (i32.const 0x24))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (set_global $cp (call $leb128 (get_global $cp) (get_local $n))))
+
+  (func $emitGetGlobal (param $n i32)
+    (i32.store8 (get_global $cp) (i32.const 0x23))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (set_global $cp (call $leb128 (get_global $cp) (get_local $n))))
+
+  (func $emitStore
+    (i32.store8 (get_global $cp) (i32.const 0x36))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (i32.store8 (get_global $cp) (i32.const 0x02))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (i32.store8 (get_global $cp) (i32.const 0x00))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1))))
+
+  (func $emitLoad
+    (i32.store8 (get_global $cp) (i32.const 0x28))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (i32.store8 (get_global $cp) (i32.const 0x02))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1)))
+    (i32.store8 (get_global $cp) (i32.const 0x00))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1))))
+
   (func $emitAdd
     (i32.store8 (get_global $cp) (i32.const 0x6a))
     (set_global $cp (i32.add (get_global $cp) (i32.const 1))))
 
+  (func $emitSub
+    (i32.store8 (get_global $cp) (i32.const 0x6b))
+    (set_global $cp (i32.add (get_global $cp) (i32.const 1))))
+
   (func $emitEqualsZero
     (i32.store8 (get_global $cp) (i32.const 0x45))
     (set_global $cp (i32.add (get_global $cp) (i32.const 1))))
@@ -2074,47 +2107,62 @@ EOF
     (call $here) 
     (call $over) 
     (call $erase)
-    (call $push (i32.const 2))
+    (i32.store (get_global $tos) (i32.const 2))
+    (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
     (block $endLoop1
       (loop $loop1
         (call $two-dupe) 
         (call $dupe) 
         (call $star) 
         (call $greater-than)
-        (br_if $endLoop1 (i32.eqz (call $pop)))
+        (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+        (br_if $endLoop1 (i32.eqz (i32.load (get_global $tos))))
         (call $dupe) 
         (call $sieve_prime)
-        (if (i32.ne (call $pop) (i32.const 0))
+        (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+        (if (i32.ne (i32.load (get_global $tos)) (i32.const 0))
           (block
             (call $two-dupe) 
             (call $dupe) 
             (call $star)
-            (set_local $i (call $pop))
-            (set_local $end (call $pop))
+            (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+            (set_local $i (i32.load (get_global $tos)))
+            (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+            (set_local $end (i32.load (get_global $tos)))
             (block $endLoop2
               (loop $loop2
-                (call $push (get_local $i))
+                (i32.store (get_global $tos) (get_local $i))
+                (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
                 (call $sieve_composite) 
                 (call $dupe)
-                (set_local $i (i32.add (call $pop) (get_local $i)))
+                (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+                (set_local $i (i32.add (i32.load (get_global $tos)) (get_local $i)))
                 (br_if $endLoop2 (i32.ge_s (get_local $i) (get_local $end)))
                 (br $loop2)))))
         (call $one-plus)
         (br $loop1)))
     (call $drop) 
-    (call $push (i32.const 1))
+    (i32.store (get_global $tos) (i32.const 1))
+    (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
     (call $swap) 
-    (call $push (i32.const 2))
-    (set_local $i (call $pop))
-    (set_local $end (call $pop))
+    (i32.store (get_global $tos) (i32.const 2))
+    (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
+
+    (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+    (set_local $i (i32.load (get_global $tos)))
+    (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+    (set_local $end (i32.load (get_global $tos)))
     (block $endLoop3
       (loop $loop3
-        (call $push (get_local $i))
+        (i32.store (get_global $tos) (get_local $i))
+        (set_global $tos (i32.add (get_global $tos) (i32.const 4)))
         (call $sieve_prime) 
-        (if (i32.ne (call $pop) (i32.const 0))
+        (set_global $tos (i32.sub (get_global $tos) (i32.const 4)))
+        (if (i32.ne (i32.load (get_global $tos)) (i32.const 0))
         (block
           (call $drop)
-          (call $push (get_local $i))))
+          (i32.store (get_global $tos) (get_local $i))
+          (set_global $tos (i32.add (get_global $tos) (i32.const 4)))))
         (set_local $i (i32.add (i32.const 1) (get_local $i)))
         (br_if $endLoop3 (i32.ge_s (get_local $i) (get_local $end)))
         (br $loop3))))
@@ -2141,9 +2189,6 @@ EOF
   (data (i32.const !moduleHeaderBase) !moduleHeader)
   (data (i32.const !preludeDataBase)  !preludeData)
 
-  (func (export "tos") (result i32)
-    (get_global $tos))
-
   (func (export "interpret") (result i32)
     (local $result i32)
     (call $refill)
diff --git a/tests/tests.js b/tests/tests.js
index e5b12c9..7cdfc26 100644
--- a/tests/tests.js
+++ b/tests/tests.js
@@ -100,14 +100,14 @@ function loadTests(wasmModule, arrayToBase64) {
 
     function here() {
       run("HERE");
-      const result = memory[core.tos() / 4 - 1];
+      const result = memory[core.tos.value / 4 - 1];
       run("DROP");
       return result;
     }
 
     function latest() {
       run("LATEST");
-      const result = memory[core.tos() / 4 - 1];
+      const result = memory[core.tos.value / 4 - 1];
       run("DROP");
       return result;
     }