diff --git a/compiler/src/prog8/compiler/target/c64/codegen/assignment/AssignmentAsmGen.kt b/compiler/src/prog8/compiler/target/c64/codegen/assignment/AssignmentAsmGen.kt
index 522e65fea..764c554e0 100644
--- a/compiler/src/prog8/compiler/target/c64/codegen/assignment/AssignmentAsmGen.kt
+++ b/compiler/src/prog8/compiler/target/c64/codegen/assignment/AssignmentAsmGen.kt
@@ -211,6 +211,10 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                         // Everything else just evaluate via the stack.
                         // (we can't use the assignment helper functions to do it via registers here,
                         // because the code here is the implementation of exactly that...)
+                        if(value.parent is Return) {
+                            if (this.asmgen.options.slowCodegenWarnings)
+                                println("warning: slow stack evaluation used for return: $value  target=${assign.target.kind} at ${value.position}")
+                        }
                         asmgen.translateExpression(value)
                         if(assign.target.datatype in WordDatatypes && assign.source.datatype in ByteDatatypes)
                             asmgen.signExtendStackLsb(assign.source.datatype)
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index fedaccf6e..cd192b509 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -2,7 +2,8 @@
 TODO
 ====
 
-- check the various math routines that are marked TODO in prog8_lib.asm for correctness...
+- fix errors in arith examples
+
 - see if we can group some errors together for instance the (now single) errors about unidentified symbols
 - Cx16 target: support full-screen 640x480 and 320x240 graphics? That requires our own custom graphics routines though to draw lines.
 - make it possible to use cpu opcodes such as 'nop' as variable names by prefixing all asm vars with something such as '_'
diff --git a/examples/arithmetic/div.p8 b/examples/arithmetic/div.p8
index 28573758f..e8fafacf1 100644
--- a/examples/arithmetic/div.p8
+++ b/examples/arithmetic/div.p8
@@ -21,8 +21,8 @@ main {
         div_word(-20000,500,-40)
         div_word(-2222,2,-1111)
 
-        div_float(0,1,0)
-        div_float(999.9,111.0,9.008108108108107)
+        div_float(0,1,0)                            ; TODO FIX ERROR
+        div_float(999.9,111.0,9.008108108108107)    ; TODO FIX ERROR
     }
 
     sub div_ubyte(ubyte a1, ubyte a2, ubyte c) {
diff --git a/examples/arithmetic/minus.p8 b/examples/arithmetic/minus.p8
index 511fb3a11..bd3d9ff68 100644
--- a/examples/arithmetic/minus.p8
+++ b/examples/arithmetic/minus.p8
@@ -1,5 +1,6 @@
 %import floats
 %import textio
+%import test_stack
 %zeropage basicsafe
 
 main {
@@ -28,9 +29,11 @@ main {
         minus_word(0,-3333,3333)
         minus_word(-3333,0,-3333)
 
-        minus_float(0,0,0)
-        minus_float(2.5,1.5,1.0)
-        minus_float(-1.5,3.5,-5.0)
+        minus_float(0,0,0)              ; TODO FIX ERROR
+        minus_float(2.5,1.5,1.0)        ; TODO FIX ERROR
+        minus_float(-1.5,3.5,-5.0)      ; TODO FIX ERROR
+
+        test_stack.test()
     }
 
     sub minus_ubyte(ubyte a1, ubyte a2, ubyte c) {
@@ -95,12 +98,13 @@ main {
 
     sub minus_float(float  a1, float a2, float  c) {
         float r = a1-a2
-        if abs(r-c)<0.00001
+        if abs(r-c)<0.00001             ; TODO FIX COMPARISON  (it works when only comparing a var)
             txt.print(" ok  ")
-        else
+        else {
             txt.print("err! ")
+        }
 
-        txt.print("float ")
+        txt.print("  float ")
         floats.print_f(a1)
         txt.print(" - ")
         floats.print_f(a2)
diff --git a/examples/arithmetic/mult.p8 b/examples/arithmetic/mult.p8
index d12c07e1a..ab417f027 100644
--- a/examples/arithmetic/mult.p8
+++ b/examples/arithmetic/mult.p8
@@ -22,9 +22,9 @@ main {
         mul_word(-10,1000,-10000)
         mul_word(1,-3333,-3333)
 
-        mul_float(0,0,0)
-        mul_float(2.5,10,25)
-        mul_float(-1.5,10,-15)
+        mul_float(0,0,0)         ; TODO FIX ERROR
+        mul_float(2.5,10,25)     ; TODO FIX ERROR
+        mul_float(-1.5,10,-15)   ; TODO FIX ERROR
     }
 
     sub mul_ubyte(ubyte a1, ubyte a2, ubyte c) {
diff --git a/examples/arithmetic/plus.p8 b/examples/arithmetic/plus.p8
index d7a6788de..ec452602a 100644
--- a/examples/arithmetic/plus.p8
+++ b/examples/arithmetic/plus.p8
@@ -25,6 +25,7 @@ main {
         plus_word(0,-3333,-3333)
         plus_word(-3333,0,-3333)
 
+         ; TODO FIX ERRORs in float
         plus_float(0,0,0)
         plus_float(1.5,2.5,4.0)
         plus_float(-1.5,3.5,2.0)
diff --git a/examples/test.p8 b/examples/test.p8
index e5810e61b..9e3112745 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -6,5 +6,154 @@
 main {
 
     sub start() {
+        uword uw
+        uword xx
+        word ww
+        word yy
+
+        uw = shiftruw7()
+        xx+=uw
+        uw = shiftruw8()
+        xx+=uw
+        uw = shiftruw9()
+        xx+=uw
+        uw = shiftruw10()
+        xx+=uw
+        uw = shiftruw11()
+        xx+=uw
+        uw = shiftruw12()
+        xx+=uw
+        uw = shiftruw13()
+        xx+=uw
+        uw = shiftruw14()
+        xx+=uw
+        uw = shiftruw15()
+        xx+=uw
+
+        ww = shiftrsw7()
+        yy+=ww
+        ww = shiftrsw8()
+        yy+=ww
+        ww = shiftrsw9()
+        yy+=ww
+        ww = shiftrsw10()
+        yy+=ww
+        ww = shiftrsw11()
+        yy+=ww
+        ww = shiftrsw12()
+        yy+=ww
+        ww = shiftrsw13()
+        yy+=ww
+        ww = shiftrsw14()
+        yy+=ww
+        ww = shiftrsw15()
+        yy+=ww
+        ww = shiftrsw16()       ; TODO why sub not replaced by const?
+        yy+=ww
+        ww = shiftrsw17()       ; TODO why sub not replaced by const?
+        yy+=ww
+
     }
+
+
+        sub shiftruw7() -> uword {
+            uword q = $a49f
+            return q >> 7
+        }
+
+        sub shiftruw8() -> uword {
+            uword q = $a49f
+            return (q >> 8)         ; TODO fix slow?  (and for all below)
+        }
+
+        sub shiftruw9() -> uword {
+            uword q = $a49f
+            return (q >> 9)
+        }
+
+        sub shiftruw10() -> uword {
+            uword q = $a49f
+            return (q >> 10)
+        }
+
+        sub shiftruw11() -> uword {
+            uword q = $a49f
+            return (q >> 11)
+        }
+
+        sub shiftruw12() -> uword {
+            uword q = $a49f
+            return (q >> 12)
+        }
+
+        sub shiftruw13() -> uword {
+            uword q = $a49f
+            return (q >> 13)
+        }
+
+        sub shiftruw14() -> uword {
+            uword q = $a49f
+            return (q >> 14)
+        }
+
+        sub shiftruw15() -> uword {
+            uword q = $a49f
+            return (q >> 15)
+        }
+
+    sub shiftrsw7() -> word {
+        word q = -12345
+        return q >> 7
+    }
+
+    sub shiftrsw8() -> word {
+        word q = -12345
+        return (q >> 8)         ; TODO why not marked slow?  What code is generated?  Also for all below.
+    }
+
+    sub shiftrsw9() -> word {
+        word q = -12345
+        return (q >> 9)
+    }
+
+    sub shiftrsw10() -> word {
+        word q = -12345
+        return (q >> 10)
+    }
+
+    sub shiftrsw11() -> word {
+        word q = -12345
+        return (q >> 11)
+    }
+
+    sub shiftrsw12() -> word {
+        word q = -12345
+        return (q >> 12)
+    }
+
+    sub shiftrsw13() -> word {
+        word q = -12345
+        return (q >> 13)
+    }
+
+    sub shiftrsw14() -> word {
+        word q = -12345
+        return (q >> 14)
+    }
+
+    sub shiftrsw15() -> word {
+        word q = -12345
+        return (q >> 15)
+    }
+
+    sub shiftrsw16() -> word {
+        word q = -12345
+        return (q >> 16)
+    }
+
+    sub shiftrsw17() -> word {
+        word q = -12345
+        return (q >> 17)
+    }
+
 }