diff --git a/include/cyclone/runtime.h b/include/cyclone/runtime.h
index 0d322c22..2bbf3195 100644
--- a/include/cyclone/runtime.h
+++ b/include/cyclone/runtime.h
@@ -263,7 +263,7 @@ object Cyc_sub(void *data, object cont, int argc, object n, ...);
 object Cyc_mul(void *data, object cont, int argc, object n, ...);
 object Cyc_div(void *data, object cont, int argc, object n, ...);
 object Cyc_fast_sum(void *data, object ptr, object x, object y);
-//object Cyc_fast_sub(void *data, object ptr, object x, object y);
+object Cyc_fast_sub(void *data, object ptr, object x, object y);
 object Cyc_bit_unset(void *data, object n1, object n2); 
 object Cyc_bit_set(void *data, object n1, object n2);
 object Cyc_num_op_va_list(void *data, int argc,
diff --git a/runtime.c b/runtime.c
index 4dad5e34..fc2b5ddf 100644
--- a/runtime.c
+++ b/runtime.c
@@ -2294,17 +2294,6 @@ void FUNC_APPLY(void *data, int argc, object clo, object cont, object n, ...) {
     return_closcall1(data, cont, result); \
 }
 
-/*
-TODO: need compiler to allocate a common type on the stack, and pass
-a pointer to it as ptr. This would allow us to have a function that can
-be inlined.
-
-TODO: with above, will want compiler to replace suitable instances of +
-with something else to be compiled to this, such as Cyc:+.
-I think it is good enough to replace all calls to + with only 2 params.
-might be able to condense calls with more than 2 params down to multiple
-calls to Cyc:+, but lets do that afterwards
-*/
 object Cyc_fast_sum(void *data, object ptr, object x, object y) {
   // x is int (assume value types for integers)
   if (obj_is_int(x)){
@@ -2335,25 +2324,25 @@ object Cyc_fast_sum(void *data, object ptr, object x, object y) {
   return NULL;
 }
 
-/*object Cyc_fast_sub(void *data, object ptr, object x, object y) {
+object Cyc_fast_sub(void *data, object ptr, object x, object y) {
   // x is int (assume value types for integers)
   if (obj_is_int(x)){
     if (obj_is_int(y)){
       int z = obj_obj2int(x) - obj_obj2int(y);
-      _return_closcall1(data, cont, obj_int2obj(z));
+      return obj_int2obj(z);
     } else if (is_object_type(y) && type_of(y) == double_tag) {
-      make_double(d, (double)(obj_obj2int(x)) - double_value(y));
-      _return_closcall1(data, cont, &d);
+      assign_double(ptr, (double)(obj_obj2int(x)) - double_value(y));
+      return ptr;
     }
   }
   // x is double
   if (is_object_type(x) && type_of(x) == double_tag) {
     if (obj_is_int(y)){
-      make_double(d, (double)(obj_obj2int(y)) - double_value(x));
-      _return_closcall1(data, cont, &d);
+      assign_double(ptr, (double)(obj_obj2int(y)) - double_value(x));
+      return ptr;
     } else if (is_object_type(y) && type_of(y) == double_tag) {
-      make_double(d, double_value(x) - double_value(y));
-      _return_closcall1(data, cont, &d);
+      assign_double(ptr, double_value(x) - double_value(y));
+      return ptr;
     }
   }
   // still here, raise an error 
@@ -2363,7 +2352,7 @@ object Cyc_fast_sum(void *data, object ptr, object x, object y) {
   make_pair(c0, &s, &c1);
   Cyc_rt_raise(data, &c0);
   return NULL;
-}*/
+}
 
 object Cyc_div_op(void *data, common_type * x, object y)
 {
diff --git a/scheme/cyclone/primitives.sld b/scheme/cyclone/primitives.sld
index f3763082..2728a5f7 100644
--- a/scheme/cyclone/primitives.sld
+++ b/scheme/cyclone/primitives.sld
@@ -62,6 +62,7 @@
          Cyc-stdin
          Cyc-stderr
          Cyc-fast-plus
+         Cyc-fast-sub
          +
          -
          *
@@ -182,6 +183,7 @@
          (Cyc-stdin 0 0)
          (Cyc-stderr 0 0)
          (Cyc-fast-plus 2 2)
+         (Cyc-fast-sub 2 2)
          (- 1 #f)
          (/ 1 #f)
          (= 2 #f)
@@ -409,6 +411,7 @@
          ((eq? p 'Cyc-stdin)             "Cyc_stdin")
          ((eq? p 'Cyc-stderr)            "Cyc_stderr")
          ((eq? p 'Cyc-fast-plus)         "Cyc_fast_sum")
+         ((eq? p 'Cyc-fast-sub)          "Cyc_fast_sub")
          ((eq? p '+)                     "Cyc_sum")
          ((eq? p '-)                     "Cyc_sub")
          ((eq? p '*)                     "Cyc_mul")
@@ -531,6 +534,7 @@
     (define (prim/data-arg? p)
       (member p '(
         Cyc-fast-plus
+        Cyc-fast-sub
         +
         -
         *
@@ -607,6 +611,7 @@
     (define (prim/c-var-pointer p)
       (cond
         ((eq? p 'Cyc-fast-plus) "common_type")
+        ((eq? p 'Cyc-fast-sub) "common_type")
         (else #f)))
 
     ;; Determine if primitive assigns (allocates) a C variable
@@ -619,6 +624,7 @@
         ((eq? p 'open-input-file) "port_type")
         ((eq? p 'open-output-file) "port_type")
         ((eq? p 'Cyc-fast-plus) "object")
+        ((eq? p 'Cyc-fast-sub) "object")
         ((eq? p '+) "object")
         ((eq? p '-) "object")
         ((eq? p '*) "object")
@@ -674,6 +680,7 @@
                  symbol->string number->string 
                  substring
                  Cyc-fast-plus
+                 Cyc-fast-sub
                  + - * / apply 
                  = > < >= <=
                  command-line-arguments
@@ -728,12 +735,11 @@
               (member exp '())))
 
     (define (prim:inline-convert-prim-call prim-call)
-      ;(write `(prim:inline-convert-prim-call ,prim-call))
-      ;(newline)
       (cond
-        ((and (equal? (car prim-call) '+)
-              (= (length prim-call) 3))
-          (cons 'Cyc-fast-plus (cdr prim-call)))
+        ((and (equal? (car prim-call) '+) (= (length prim-call) 3))
+         (cons 'Cyc-fast-plus (cdr prim-call)))
+        ((and (equal? (car prim-call) '-) (= (length prim-call) 3))
+         (cons 'Cyc-fast-sub (cdr prim-call)))
         (else
-          prim-call)))
+         prim-call)))
 ))