diff --git a/compiler/res/prog8lib/cx16/syslib.p8 b/compiler/res/prog8lib/cx16/syslib.p8
index f30da824b..96719b84e 100644
--- a/compiler/res/prog8lib/cx16/syslib.p8
+++ b/compiler/res/prog8lib/cx16/syslib.p8
@@ -823,16 +823,61 @@ sys {
 
     asmsub memcopy(uword source @R0, uword target @R1, uword count @AY) clobbers(A,X,Y) {
         ; note: can't be inlined because is called from asm as well
+        ;       also: doesn't use cx16 ROM routine so this always works even when ROM is not banked in.
         %asm {{
-            sta  cx16.r2
-            sty  cx16.r2+1
-            jmp  cx16.memory_copy
+            cpy  #0
+            bne  _longcopy
+
+            ; copy <= 255 bytes
+            tay
+            bne  _copyshort
+            rts     ; nothing to copy
+
+_copyshort
+            ; decrease source and target pointers so we can simply index by Y
+            lda  cx16.r0
+            bne  +
+            dec  cx16.r0+1
++           dec  cx16.r0
+            lda  cx16.r1
+            bne  +
+            dec  cx16.r1+1
++           dec  cx16.r1
+
+-           lda  (cx16.r0),y
+            sta  (cx16.r1),y
+            dey
+            bne  -
+            rts
+
+_longcopy
+            pha                         ; lsb(count) = remainder in last page
+            tya
+            tax                         ; x = num pages (1+)
+            ldy  #0
+-           lda  (cx16.r0),y
+            sta  (cx16.r1),y
+            iny
+            bne  -
+            inc  cx16.r0+1
+            inc  cx16.r1+1
+            dex
+            bne  -
+            ply
+            bne  _copyshort
+            rts
         }}
     }
 
-    inline asmsub memset(uword mem @R0, uword numbytes @R1, ubyte value @A) clobbers(A,X,Y) {
+    asmsub memset(uword mem @R0, uword numbytes @R1, ubyte value @A) clobbers(A,X,Y) {
         %asm {{
-            jsr  cx16.memory_fill
+            ldy  cx16.r0
+            sty  P8ZP_SCRATCH_W1
+            ldy  cx16.r0+1
+            sty  P8ZP_SCRATCH_W1+1
+            ldx  cx16.r1
+            ldy  cx16.r1+1
+            jmp  prog8_lib.memset
         }}
     }
 
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index 86f439448..3a7b1ac67 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -3,8 +3,6 @@ TODO
 
 For next release
 ^^^^^^^^^^^^^^^^
-- move memcopy() logic to prog8_lib and call this from sys.memcopy() the same for all targets (so cx16 now also works when kernal is disabled!)
-- same for memset()
 - get rid of RAW_LOAD_ADDRESS  and make specifying the load address for RAW launcher mode required
 ...
 
@@ -12,7 +10,7 @@ For next release
 Need help with
 ^^^^^^^^^^^^^^
 - c128 target: various machine specific things (free zp locations, how banking works, getting the floating point routines working, ...)
-- other targets such as Atari 800XL: all required details about the machine, I have no clue whatsoever
+- atari target: more details details about the machine, fixing library routines. I have no clue whatsoever.
 - see the :ref:`portingguide` for details on what information is needed.
 
 
diff --git a/examples/test.p8 b/examples/test.p8
index 0aa8d4b90..421e1eedd 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -1,11 +1,48 @@
-%launcher none
+%import textio
+%zeropage kernalsafe
 
 main {
     sub start() {
-        uword @shared names_buffer = memory("filenames", 200, 0)
+        ubyte[20] array1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+        ubyte[20] array2 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
 
-        do {
-            ubyte @shared char = '*'
-        } until true
+        sys.set_irqd()
+
+        cx16.rombank(9)
+        sys.memset(array1, 20, 255)
+
+        cx16.rombank(0)
+        ubyte ii
+        for ii in array1 {
+            txt.print_ubhex(ii, false)
+            txt.spc()
+        }
+        txt.nl()
+        for ii in array2 {
+            txt.print_ubhex(ii, false)
+            txt.spc()
+        }
+        txt.nl()
+
+        cx16.rombank(9)
+        sys.memset(array1, 20, 255)
+        array2=array1
+
+        cx16.rombank(0)
+        for ii in array1 {
+            txt.print_ubhex(ii, false)
+            txt.spc()
+        }
+        txt.nl()
+        for ii in array2 {
+            txt.print_ubhex(ii, false)
+            txt.spc()
+        }
+        txt.nl()
+
+        cx16.rombank(4)
+
+        repeat {
+        }
     }
 }