diff --git a/compiler/res/prog8lib/math.p8 b/compiler/res/prog8lib/math.p8
index 086faa33c..15469f79b 100644
--- a/compiler/res/prog8lib/math.p8
+++ b/compiler/res/prog8lib/math.p8
@@ -484,4 +484,49 @@ log2_tab
         }}
     }
 
+    sub crc16(uword data, uword length) -> uword {
+        ; calculates the CRC16 (XMODEM) checksum of the buffer.
+        cx16.r1 = data  ; make sure pointer is in zp (on cx16)
+        cx16.r0 = 0  ; the crc value
+        repeat length {
+            cx16.r0H ^= @(cx16.r1)
+            repeat 8 {
+                if cx16.r0H & $80 {
+                    cx16.r0 <<= 1
+                    cx16.r0 ^= $1021
+                }
+                else
+                    cx16.r0<<=1
+            }
+            cx16.r1++
+        }
+        return cx16.r0
+    }
+
+    sub crc32(uword data, uword length) {
+        ; Calculates the CRC-32 (POSIX) checksum of the buffer.
+        ; because prog8 doesn't have 32 bits integers, we have to split up the calculation over 2 words.
+        ; result stored in cx16.r0 (low word) and cx16.r1 (high word)
+        cx16.r2 = data  ; make sure pointer is in zp (on cx16)
+        cx16.r1 = 0
+        cx16.r0 = 0
+        repeat length {
+            cx16.r1H ^= @(cx16.r2)
+            repeat 8 {
+                if cx16.r1H & $80 {
+                    cx16.r0 <<= 1
+                    rol(cx16.r1)
+                    cx16.r1 ^= $04c1
+                    cx16.r0 ^= $1db7
+                }
+                else {
+                    cx16.r0 <<= 1
+                    rol(cx16.r1)
+                }
+            }
+            cx16.r2++
+        }
+        cx16.r1 ^= $ffff
+        cx16.r0 ^= $ffff
+    }
 }
diff --git a/compiler/res/prog8lib/virtual/math.p8 b/compiler/res/prog8lib/virtual/math.p8
index a9d0aacc9..194c30187 100644
--- a/compiler/res/prog8lib/virtual/math.p8
+++ b/compiler/res/prog8lib/virtual/math.p8
@@ -293,4 +293,45 @@ math {
         return w2-w1
     }
 
+    sub crc16(uword data, uword length) -> uword {
+        ; calculates the CRC16 (XMODEM) checksum of the buffer.
+        cx16.r0 = 0  ; the crc value
+        repeat length {
+            cx16.r0H ^= @(data)
+            repeat 8 {
+                if cx16.r0H & $80
+                    cx16.r0 = (cx16.r0<<1)^$1021
+                else
+                    cx16.r0<<=1
+            }
+            data++
+        }
+        return cx16.r0
+    }
+
+    sub crc32(uword data, uword length) {
+        ; Calculates the CRC-32 (POSIX) checksum of the buffer.
+        ; because prog8 doesn't have 32 bits integers, we have to split up the calculation over 2 words.
+        ; result stored in cx16.r0 (low word) and cx16.r1 (high word)
+        cx16.r1 = 0
+        cx16.r0 = 0
+        repeat length {
+            cx16.r1H ^= @(data)
+            repeat 8 {
+                if cx16.r1H & $80 {
+                    cx16.r0 <<= 1
+                    rol(cx16.r1)
+                    cx16.r1 ^= $04c1
+                    cx16.r0 ^= $1db7
+                }
+                else {
+                    cx16.r0 <<= 1
+                    rol(cx16.r1)
+                }
+            }
+            data++
+        }
+        cx16.r1 ^= $ffff
+        cx16.r0 ^= $ffff
+    }
 }
diff --git a/docs/source/libraries.rst b/docs/source/libraries.rst
index ff439e2cc..cb9a03227 100644
--- a/docs/source/libraries.rst
+++ b/docs/source/libraries.rst
@@ -491,6 +491,15 @@ but perhaps the provided ones can be of service too.
     It does not work for the verafx multiplication routines on the Commander X16!
     These have a different way to obtain the upper 16 bits of the result: just read cx16.r0.
 
+``crc16 (uword data, uword length) -> uword``
+    Returns a CRC-16 (XMODEM) checksum over the given data buffer.
+    Note: on the Commander X16, there is a CRC-16 routine in the kernal: cx16.memory_crc().
+    That one is faster, but yields different results. It is unclear what flavour of crc it is calculating.
+
+``crc32 (uword data, uword length)``
+    Calculates a CRC-32 (POSIX) checksum over the given data buffer.
+    The 32 bits result is stored in cx16.r0 (low word) and cx16.r1 (high word).
+
 
 cx16logo
 --------
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index 44332bfdc..b03a69203 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -2,8 +2,6 @@
 TODO
 ====
 
-- add crc8 and crc16 and crc32 to math
-- fix crc* bench routines to no longer depend on the kernal rom version (use a bin file)
 - make internalCast() not complain anymore about signed <-> unsigned conversions
 - fix bitshift.p8
 
diff --git a/examples/bench8/crc16.p8 b/examples/bench8/crc16.p8
deleted file mode 100644
index ab741e818..000000000
--- a/examples/bench8/crc16.p8
+++ /dev/null
@@ -1,32 +0,0 @@
-%import textio
-%import floats
-
-main {
-
-    sub crc16(uword data, uword length) -> uword {
-        uword crc = 0
-        repeat length {
-            crc ^= mkword(@(data), 0)
-            repeat 8 {
-                if crc & $8000
-                    crc = (crc<<1)^$1021
-                else
-                    crc<<=1
-            }
-            data++
-        }
-        return crc
-    }
-
-    sub start() {
-        txt.print("calculating (expecting $ffd0)...\n")
-        txt.print("  if mismatch: first check if kernal is maybe updated?\n")
-        cbm.SETTIM(0,0,0)
-        uword crc = crc16($e000, $2000)
-        txt.print_uwhex(crc, true)
-        txt.nl()
-        floats.print(cbm.RDTIM16() / 60.0)
-        txt.print(" seconds")
-        sys.wait(9999)
-    }
-}
diff --git a/examples/bench8/crc32.p8 b/examples/bench8/crc32.p8
deleted file mode 100644
index 22ec8b854..000000000
--- a/examples/bench8/crc32.p8
+++ /dev/null
@@ -1,44 +0,0 @@
-%import textio
-%import floats
-
-main {
-    sub crc32(uword data, uword length) {
-        ; because prog8 doesn't have 32 bits integers, we have to split up the calucation over 2 words.
-        ; result in cx16.r0 (high word) and cx1.r1 (low word).
-        cx16.r0 = 0
-        cx16.r1 = 0
-        repeat length {
-            cx16.r0 ^= mkword(@(data), 0)
-            repeat 8 {
-                if cx16.r0 & $8000 {
-                    sys.clear_carry()
-                    rol(cx16.r1)
-                    rol(cx16.r0)
-                    cx16.r0 ^= $04c1
-                    cx16.r1 ^= $1db7
-                }
-                else {
-                    sys.clear_carry()
-                    rol(cx16.r1)
-                    rol(cx16.r0)
-                }
-            }
-            data++
-        }
-        cx16.r0 ^= $ffff
-        cx16.r1 ^= $ffff
-    }
-
-    sub start() {
-        txt.print("calculating (expecting $e1fa84c6)...\n")
-        txt.print("  if mismatch: first check if kernal is maybe updated?\n")
-        cbm.SETTIM(0,0,0)
-        crc32($e000, $2000)
-        txt.print_uwhex(cx16.r0, true)
-        txt.print_uwhex(cx16.r1, false)
-        txt.nl()
-        floats.print(cbm.RDTIM16() / 60.0)
-        txt.print(" seconds")
-        sys.wait(9999)
-    }
-}
diff --git a/examples/bench8/crc8.p8 b/examples/bench8/crc8.p8
deleted file mode 100644
index 73613d461..000000000
--- a/examples/bench8/crc8.p8
+++ /dev/null
@@ -1,32 +0,0 @@
-%import textio
-%import floats
-
-main {
-
-    sub crc8(uword data, uword length) -> ubyte {
-        ubyte crc = 0
-        repeat length {
-            crc ^= @(data)
-            repeat 8 {
-                if crc & $80
-                    crc = (crc<<1)^$1d
-                else
-                    crc<<=1
-            }
-            data++
-        }
-        return crc
-   }
-
-    sub start() {
-        txt.print("calculating (expecting $a2)...\n")
-        txt.print("  if mismatch: first check if kernal is maybe updated?\n")
-        cbm.SETTIM(0,0,0)
-        ubyte crc = crc8($e000, $2000)
-        txt.print_ubhex(crc, true)
-        txt.nl()
-        floats.print(cbm.RDTIM16() / 60.0)
-        txt.print(" seconds")
-        sys.wait(9999)
-    }
-}
diff --git a/examples/bench8/pow.p8 b/examples/bench8/pow.p8
deleted file mode 100644
index a992ce866..000000000
--- a/examples/bench8/pow.p8
+++ /dev/null
@@ -1,42 +0,0 @@
-%import textio
-%import floats
-
-main {
-    const ubyte N_ITER = 10
-    const ubyte SIZE = 32
-
-    float[SIZE] array = 0.0
-
-    sub testpow(float x, uword y) -> float {
-        float tmp = x
-        if y==0
-            return 1
-        repeat y-1 {
-            tmp *= x
-        }
-        return tmp
-    }
-
-    sub start() {
-        txt.print("calculating (expecting 3.614007e+12)...\n")
-        cbm.SETTIM(0,0,0)
-
-        float res=0.0
-        uword i
-        ubyte j
-        for i in 0 to N_ITER-1 {
-            for j in 0 to SIZE-1 {
-                array[j] += testpow(2.5/(i+1.0), j)
-            }
-        }
-        for j in 0 to SIZE-1 {
-            res += array[j]
-        }
-
-        floats.print(res)
-        txt.nl()
-        floats.print(cbm.RDTIM16() / 60.0)
-        txt.print(" seconds")
-        sys.wait(9999)
-    }
-}
diff --git a/examples/bench8/readme.txt b/examples/bench8/readme.txt
deleted file mode 100644
index a0d5f746f..000000000
--- a/examples/bench8/readme.txt
+++ /dev/null
@@ -1 +0,0 @@
-Contains several benchmarks from https://gglabs.us/node/2293 ported to prog8.
diff --git a/examples/bench8/sieve-bit.p8 b/examples/bench8/sieve-bit.p8
deleted file mode 100644
index e45f32a87..000000000
--- a/examples/bench8/sieve-bit.p8
+++ /dev/null
@@ -1,64 +0,0 @@
-%import textio
-%import floats
-
-main {
-    const ubyte N_ITER = 4
-    const uword SIZE = 16000
-
-    uword @zp flags_ptr = memory("flags", SIZE/8+1, $100)
-    ubyte[] bitv = [ $01, $02, $04, $08, $10, $20, $40, $80 ]
-
-    sub start() {
-        txt.print_ub(N_ITER)
-        txt.print(" iterations, calculating... (expecting 3431)\n")
-
-        cbm.SETTIM(0, 0, 0)
-        uword prime_count
-        repeat N_ITER {
-            prime_count = sieve()
-        }
-
-        txt.print_uw(prime_count)
-        txt.print(" primes\n")
-
-        float time = cbm.RDTIM16() as float / 60.0
-        floats.print(time)
-        txt.print(" sec total = ")
-        floats.print(time/N_ITER)
-        txt.print(" sec per iteration\n")
-        sys.wait(9999)
-    }
-
-    sub check_flag(uword idx) -> ubyte
-    {
-        return flags_ptr[idx/8] & bitv[lsb(idx)&7]
-    }
-
-    sub clear_flag(uword idx)
-    {
-        flags_ptr[idx/8] &= ~bitv[lsb(idx)&7]
-    }
-
-    sub sieve() -> uword {
-        uword prime
-        uword k
-        uword count=0
-        uword i
-        sys.memset(flags_ptr, SIZE/8+1, $ff)
-
-        for i in 0 to SIZE-1 {
-            if check_flag(i) {
-                prime = i*2 + 3
-                k = i + prime
-                while k < SIZE {
-                    clear_flag(k)
-                    k += prime
-                }
-;                    txt.print_uw(prime)
-;                    txt.spc()
-                count++
-            }
-        }
-        return count
-    }
-}
diff --git a/examples/bench8/sieve.p8 b/examples/bench8/sieve.p8
deleted file mode 100644
index 0a5ba03e1..000000000
--- a/examples/bench8/sieve.p8
+++ /dev/null
@@ -1,52 +0,0 @@
-%import textio
-%import floats
-
-; The "Byte Sieve" test.  https://en.wikipedia.org/wiki/Byte_Sieve
-; Note: this program can be compiled for multiple target systems.
-
-main {
-    sub start() {
-
-
-        const ubyte ITERS = 10
-        uword count
-        uword i
-        uword prime
-        uword k
-        const uword SIZEPL = 8191
-        uword @zp flags_ptr = memory("flags", SIZEPL, $100)
-
-        txt.print_ub(ITERS)
-        txt.print(" iterations, calculating...\n")
-
-        cbm.SETTIM(0, 0, 0)
-
-        repeat ITERS {
-            sys.memset(flags_ptr, SIZEPL, 1)
-            count = 0
-            for i in 0 to SIZEPL-1 {
-                if flags_ptr[i] {
-                    prime = i*2 + 3
-                    k = i + prime
-                    while k < SIZEPL {
-                        flags_ptr[k] = false
-                        k += prime
-                    }
-;                    txt.print_uw(prime)
-;                    txt.spc()
-                    count++
-                }
-            }
-        }
-
-        txt.print_uw(count)
-        txt.print(" primes\n")
-
-        float time = cbm.RDTIM16() as float / 60.0
-        floats.print(time)
-        txt.print(" sec total = ")
-        floats.print(time/ITERS)
-        txt.print(" sec per iteration\n")
-        sys.wait(9999)
-    }
-}
diff --git a/examples/test.p8 b/examples/test.p8
index 3be700990..72259fc75 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -1,12 +1,41 @@
-%import floats
 %import math
-%import string
-%zeropage basicsafe
+%import textio
+%zeropage dontuse
 
 main {
     sub start() {
-        float fl = 1.2  ; no other assignments
-        cx16.r0L = string.isdigit(math.diff(119, floats.floor(floats.deg(fl)) as ubyte))
-        cx16.r1L = string.isletter(math.diff(119, floats.floor(floats.deg(1.2)) as ubyte))
+        str poem_data = iso:"Once upon a midnight dreary, while I pondered, weak and weary,"+
+                        iso:"Over many a quaint and curious volume of forgotten lore-"+
+                        iso:"While I nodded, nearly napping, suddenly there came a tapping,"+
+                        iso:"As of some one gently rapping, rapping at my chamber door. ..."
+        uword size = len(poem_data)
+
+        cbm.SETTIM(0,0,0)
+        repeat 20 {
+            cx16.r9 = math.crc16(poem_data, size)
+        }
+        txt.print_uwhex(cx16.r9, true)
+        txt.spc()
+        txt.print_uw(cbm.RDTIM16())
+        txt.nl()
+
+        cbm.SETTIM(0,0,0)
+        repeat 20 {
+            cx16.r9 = cx16.memory_crc(poem_data, size)      ; faster but I can't figure out the flavour of crc algorithm it uses, it's not any on https://crccalc.com/
+        }
+        txt.print_uwhex(cx16.r9, true)
+        txt.spc()
+        txt.print_uw(cbm.RDTIM16())
+        txt.nl()
+
+        cbm.SETTIM(0,0,0)
+        repeat 20 {
+            math.crc32(poem_data, size)
+        }
+        txt.print_uwhex(cx16.r1, true)
+        txt.print_uwhex(cx16.r0, false)
+        txt.spc()
+        txt.print_uw(cbm.RDTIM16())
+        txt.nl()
     }
 }