FlashForth is a Forth stamp system implemented on the Microchip 8-bit PIC18F and 16-bit PIC24, 30, 33 and the Atmel Atmega microcontroller families.

FlashForth also works on the popular Arduino UNO and MEGA boards.

FF (FlashForth) allows you to write and debug complex real-time applications. The complete system including the compiler is executing on the microcontroller.

A Forth interpreter, compiler, assembler, multitasker and user definable interrupts are provided.

A computer with a terminal emulator is used for communicating with FF via a serial or USB link.

The Forth source files are edited and saved on the computer and uploaded to the microcontroller as Forth source code.

A special ff-shell, written in python, provides command line editing, history and file transfer capabilities.

All microcontroller registers and memories can be read and written from the command line.

When the application is ready, the application word address can be stored in the turnkey vector, and your autonomous embedded application has been set up.

FlashForth is mostly compatible with the ANS'94 standard. This guide describes the differences, and features not covered by the ANS'94 standard.

FlashForth is licensed according to the Gnu Public License v3.

If you feel that FlashForth has been useful for you, please contribute with a donation.

Design philosophy

The main idea with FlashForth is to enable a robust Forth stamp system that is easy to use.

The requirements were:


FF can be downloaded from http://www.sourceforge.net/projects/flashforth.

Download FlashForth 5.0. Expand the downloaded archive into a folder.
The git repository contains the latest developments.
The git repository can be cloned from SourceForge and GitHub.

git clone git://flashforth.git.sourceforge.net/gitroot/flashforth/flashforth
git clone https://github.com/oh2aun/flashforth.git

Check the device specific pages for further installation details.

Case sensitivity

FF is case sensitive. All FF core words are written in lowercase letters and hex numbers must be entered in lower case also.

Although the FF core words are in lower case, in documentation the FF core words may be written in UPPERCASE to make it clear that a Forth word is referred to.

The Interpreter

The Forth interpreter is a normal Forth interpreter. It parses words delimited by space and tries to find the word in the dictionary. TAB is ignored. If the word is found it is executed, if not, FF tries to convert it to a number according to the current base (or base prefix), and put the number on the stack. If that fails ABORT is called.

The interpreter can only be used by the OPERATOR task, not by the background tasks.

After interpreting a line, QUIT prints Ok and executes the deferred word PROMPT.
By default PROMPT executes .ST which prints a number base prefix symbol, the current data area memory type and the parameter stack contents.
If you don't want to see the info from .ST, you can re vector PROMPT to for example CHARS which does nothing.

' chars is prompt
' .st is prompt 


FF is a 16-bit Forth and the single precision math operations are consequently 16-bit.

FF also supports double precision 32-bit math.

UM/MOD, M+ and UM* are used as base for the extra 32-bit double precision math words that can be loaded from math.txt.

For working with 48 and 64 bit numbers there are words in qmath.txt for PIC18.
The words UT* UT/ UT/* have 48 bits precision.
The words UQ* UQ/MOD QM+ D>Q have 64 bits precision.
The 64 bit words are implemented for PIC18, 24, 30 and 33.

Number conversion

FF supports single precision 16-bit and double precision 32-bit number conversion.

Double precision numbers are identified by a trailing dot.

Input numbers can be prefixed by % # $ to achieve binary decimal and hexadecimal number conversion without changing BASE.

Output numbers are always converted according to BASE.

The Compiler

FF is a subroutine threaded Forth with native code generation.

Code is always compiled to flash memory. PIC and ATMEGA can only execute code from flash.

Constants, variable addresses and literals are compiled as native code.

DUP and 0= before IF WHILE UNTIL are optimized away.

All the structured conditional words generate native code.

: and ] puts FF in compilation state. ;  ;I and [ enters the interpreter state.

The maximum word name length is 15 characters.

Words that should not be interpreted have a 'compile only' bit in the header. Interpreting these words will result in an ABORT and restarting the interpreter, that is jumping to QUIT.

The compiler performs tail call -> goto optimisation at the end of a word.

It is possible to call the word beeing currently defined.
If the word is the last word before ';' it will result in a branch back to beginning of the word.
If the current word is called earlier, it will result in recursion.

The data space

The harvard architecture microcontrollers have separate memory spaces for ram, flash and eeprom.

In order to make it possible for words like CMOVE to be used on all kinds of memories, these are mapped into one 64 Kbyte memory space. Like this it is avoided to make separate words for copying beteen different kinds of memories.

The memory spaces are mapped by FF in software:


FLASH is mapped to  $0000 - $ebff
EEPROM is mapped to $ec00 - $efff
RAM is mapped to    $f000 - $ffff 

PIC24, dsPIC30, dsPIC33:

RAM is mapped to    $0000   - RAMSIZE-1
FLASH is mapped to  RAMSIZE - $fbff
EEPROM is mapped to $fc00   - $ffff 


RAM is mapped to    $0000   - (RAMSIZE-1)
FLASH is mapped to  (0xffff - FLASHSIZE + 1) - 0xffff

In FlashForth data space can be allocated by CREATE ALLOT VARIABLE 2VARIABLE , C, VALUE DEFER .

Each memory type has its own allocation pointer. The words RAM EEPROM FLASH sets the data area from which the succeeding allocations will be made.

@ ! C@ C! and derived memory access words can be used transparently with all types of memory.

The exception to this rule are the words MSET MCLR MTST BSET BCLR BTST that can only address RAM.
These words are typically used for setting and clearing bits in registers.

For PIC18 the mapping overhead for ram @ and ! is 4 instruction cycles.
For PIC24 the mapping overhead for ram @ and ! is 3 instruction cycles.
For Atmega the mapping overhead for ram @ and ! is 2 instruction cycles.

\ A variable in eeprom
eeprom variable var2

\ A variable in ram
ram variable var1

It is not recommended to create variables in eeprom unless these are updated fairly seldom.

Data areas in flash are normally used for constant data and constant execution vectors.

Typically flash cells can be written over 10000 times until it may fail. Eeprom cells can typically be written over 100000 times until it may fail.

As an example here is a word which creates character arrays in the current data space.

: carray: ( n "name" -- ) 
  create allot 
  does> + 

Create a 20 character array in eeprom called CALIBRATE.

decimal 20 carray: calibrate

It is good to always set the data space context back to ram after flash or eeprom has been used.

Compile a word which creates indexed cell arrays.

: array: ( n "name" -- )
  create cells allot 
  does> swap 2* + 
ram 20 array: cnt               \ Creates the array cnt ( size 20 cells )
1233 10 cnt !                   \ Store 1233 in table index 10
10 cnt @                        \ Fetch from index 10 

Create a table in flash with constant data.

flash create flash-table $1234 , $3456 , #12345 , %1010101010 ,

Create a table in eeprom with some data.

eeprom create eeprom-table $e123 , $e456 , #8888 , %1110111011101110 ,

In order to prevent accidental writes to flash or eeprom the word FL- can be used to prevent writing. FL+ can be used for enabling the writes again.

Typically FL- is used in the TURNKEY word when the embedded application starts up. FL+/FL- is then used specifically in those words that store something to eeprom or flash.

Inlining of words

FF can compile location independent assembler primitives as inline code. The shortest of these words have the inline bit set in the word header for automatic inlining.

Individual words can be inlined by prefixing the word with INLINE.

: newswap inline swap ; 

When compiling a new word that should be inlined automatically, the inline flag can be set with the word INLINED.

: 1+
  [ Sminus w, a, swapf, ]  \ Decrement stack pointer with one
  [ Splus  f, a, infsnz, ] \ Add lower byte, skip next instruction if the result was nonzero
  [ Srw    f, a, incf, ]   \ Add high byte
; inlined                  \ Set the inline header flag

Also words defined by CONSTANT, VARIABLE, 2CONSTANT and 2VARIABLE can be inlined. They compile the constant and the variable address as inline literal code.

If you append the definition with INLINED, the compiler will later compile the constant as an inline literal.

34 constant thirtyfour inlined
: literal-34 thirtyfour ;

See the device specific pages for further details.

Dictionary search order

The core dictionary is always searched before the user dictionary. It is not possible to redefine existing words. These measures have been taken to make the system more robust and to make it possible to recover to the basic state, without the need to flash the chip again. It also makes the code clearer, since there will not be two words with the same name.

Dictionary management

In order to recover to an earlier dictionary and memory allocation state, use MARKER. Always before defining new words define a marker. Otherwise you may need to return to an earlier marker or to say EMPTY which will empty the dictionary and reset all memory allocations to default values. A marker will restore TURNKEY, DP and LATEST. IRQ is not affected.

FORGET can be used to forget a user word, but FORGET can only adjust the FLASH DP. This means that allotted EEPROM or RAM will not be reclaimed if you use FORGET.

Note that the eeprom variables TURNKEY, DP, LATEST are cached in ram during interpretation of a input line and also during compilation state. This makes compilations run faster, and there will be less wear of the eeprom.

Since FF refuses to redefine words, certain words, typically one line definitions, can be compiled from several source files. The first compilation is accepted, and the others rejected. This is quite practical for having some short definition in many files, so that you can compile exactly the words one or more applications needs.

For example i2c_base.txt and task-test.txt both have defined PORTC, but PORTC will be compiled only once, even if both files are loaded to FF.

The word FL- can be used for disabling writes to flash and eeprom. It is useful for making sure that no writes to flash or eeprom occur.

Boot sequence

After processor reset a check is made to see if a turnkey word should be executed. If the eeprom value TURNKEY contains a nonzero value, it must be an address of a valid user word. Unless the user presses ESC within the turnkey timeout, the user word is executed. If ESC is pressed, the user word will not execute. Instead the forth interpreter is entered.

' my_application is turnkey 

If your TURNKEY word is crashing, press ESC and as a first command give:

false is turnkey 

This will disable the TURNKEY and allow you to make corrections.

When FlashForth start it prints out a variable amount of characters indicating the restart reason:

P = Power on  reset (ALL)
B = Brown out reset (ALL)
W = Watchdog timeout reset (ALL)
S = Software Reset Instruction (PIC18, PIC24)
O = Return stack overflow (PIC18, PIC24)
U = Return stack underflow (PIC18)
E = External reset (Atmega, PIC24)
M = Math error (Divide by zero) (ALL)
A = Address Error (PIC24) 

Interrupt handling

See the device specific pages

Background tasks

FF can execute background tasks concurrently with the operator task.

The task switching is made cooperatively by executing PAUSE. PAUSE is executed in I/O words KEY and EMIT so that background tasks can run while the console is waiting for input or queuing for output. MS executes PAUSE while it waits for the specified delay to pass.

If IDLE_MODE in the configuration file is enabled, the processor will enter the idle powersaving mode in PAUSE in the OPERATOR task.
If you want to allow the processor to go into idle mode use the word IDLE. If not, use the word BUSY.
An interrupt will exit the idle mode and the processor will run until the next time idle mode is entered.

The percentage of time that the processor is busy can be read by the LOAD word. The integration interval is 256 milliseconds.

The words for tasking can be loaded from task.txt.

The user area lives in ram. The user area is initialized from the task definition in flash.

TASK: creates a new task and defines the stack sizes and the additional user area size and the tibsize.
Tibsize can be set to zero for background tasks, except if numeric output is used.
The end of the TIB is shared with the HOLD buffer. A task that uses “. U. <# # #s #>” etc., will
need a small TIB for number formatting.

Each task has its own PAD which starts at end of TIB. When allocating ram you must allot space for the PAD if it is being used.

The FF kernel does not use PAD. So if you want to use PAD in the OPERATOR task it is up to you to allot space for PAD. This non-standard behavior exists to save ram.

TINIT initializes a task with the XT of the task loop. It also initializes the task user area.

RUN makes the task run. It inserts the task in the round-robin linked list.

END ends a task. It removes the task from the round robin linked list.

SINGLE ends all tasks except the operator task.

TASKS lists all running tasks.

The tasking commands may only be executed from the operator task.

KEY, KEY?, EMIT can be deferred and used in a background task to interact for example with a keyboard and a LCD display.

\ Task loop for displaying data on the LCD display
: lcd_display ( -- )
  ['] lcd_emit 'emit !     \ Use LCD emit
    #00 lcd_at             \ Position cursor at beginning of first line
    ." Ticks: " ticks u.4  \ Display the current number of ticks

$10 $20 $20 $0 task: lcd_task  \ tibSize stackSize retrunStackSize addSize --
' lcd_display lcd_task tinit
lcd_task run


I have always found DO..LOOP cumbersome to use. I wanted to separate the loop count and the index handling. The FF core implements FOR..NEXT and a re-entrant P register.

DO ?DO LOOP +LOOP LEAVE I J UNLOOP can be added from Forth source code.

FOR..NEXT loops exactly the amount of times specified ( also 0 ) .

: star [char] * emit ; ok <$,ram>
star * ok <@,ram>
: stars for star next ; ok <$,ram>
10 stars ****************ok <$,ram>
0 stars ok<$,ram>

The loop count is held on top of the return stack and it can be fetched by R@. ENDIT sets the loop count to 0, so that NEXT will terminate the loop.

: test
    r@ . r@ 4 =
; ok
test 9 8 7 6 5 4 ok 

If you EXIT a FOR..NEXT loop you must drop the loop count with RDROP.

: test
    r@ 4 =
      rdrop exit
    r@ .
; ok
test 9 8 7 6 5 ok 

The P register

The P register can be used as a variable or as a pointer. It can be used in conjunction with FOR..NEXT or at any other time.

!P>R pushes the current P value on the return stack and sets a new value to P.

In a definition !P>R and R>P should always be used to allow proper nesting of words.

R>P pops a value into P from the return stack.

!P sets a new value into P. Use !P only from the command line, or between !P>R and R>P in a definition.

@P lets you fetch the value of P.

P+ increments P by one.

P2+ increments P by two.

P++ ( n -- ) adds n to P.

P@ P! PC@ PC! are used to access memory via the pointer.

Always remember to balance the return stack in all branches of your code.

\ CMOVE src dst u -- copy u bytes from src to dst
\ The source address is kept on the parameter stack.
\ The destination address is kept in the P register.
: cmove
  swap !p>r
     c@+ pc! p+
  r>p drop

Interacting with FlashForth

Forth source code can be interpreted and compiled by loading it via a terminal emulator.

FF supports terminal communication via UART or USB serial emulation.
The USB serial emulation requires that you use a PIC18F chip with inbuilt USB tranceiver.

The default UART setting is 38400, 1, N, XON/XOFF. It is mandatory to enable inband flow control (XON/XOFF). When FF stores data in flash, the chip will stop responding for up to 20 milliseconds. XOFF will prevent the terminal emulator from sending characters to FF while data is being stored into flash.

CTS/RTS (HW flow control) is available as a compilation option.

USB serial emulation flow control is handled by the USB protocol. Use HW flow control in the terminal emulator.

Minicom with linux works OK without any extra TX delays. Forth source files can be sent to the PIC using the 'send ascii file' (CTRL-A S) function.

With Windows TeraTerm works just fine.

The FlashForth python shell (ff-shell.py) can be used for a more convenient interaction with FlashForth. The shell has a command line editor and command line history. It can send files to over the serial link with the #send directive. It also has a directive, #warm, for sending control-o over the serial link to force warm start of FlashForth. The python shell waits for a line to be completely interpreted by FlashForth until the next line is sent. This reduces potential flow control problems. For example the Arduino boards flawed USB-serial converter communicates quite reliably with the ff-shell.

Note that the Arduino boards USB-serial converter prevents the XON/XOFF flow control from working properly.
The ff-shell can be used to remedy the situation or a 1 millisecond delay can be inserted between each character sent to the Arduino board.
At least TeraTerm can do that.

FlashForth recognises CRLF or only CR as end of line in ACCEPT. LF and CR are not echoed by ACCEPT

U1- and U2- can be used for disabling flow control if the end application can not support flow control on the serial interface.

Problems ?

Normally communication with the PC and writing to flash works very reliably, but...

If to you see a vertical bar '|' output from FlashForth, it means that the UART RX interrupt buffer has overflowed.

It is usually caused by the PC reacting slowly on XOFF.
'setserial /dev/ttyS0 low_latency' improves the situation on Linux.
On Windows, disabling the UART buffers improves the situation.
Another alternative is to use TeraTerm with an intercharacter delay of a few milliseconds.

Increasing the UART RX interrupt buffer size and sending XOFF for a small buffer fill level can also improve the situation.

If you see an extra '~' output it means that a serial framing or overrun error has occurred.

If you see a '^' output from FlashForth, it means that the verification of a program memory write has failed. FlashForth will try to write the same buffer only once, then an warm start will be made.


Many thanks to Joe Ennis, W7NET, and Pete Zawasky, AG7C, for pushing FlashForth hard and bugging me with trouble reports.

Thanks to Igor, OM1ZZ for contributing a proto board and two PICs for the 24 and 33 PIC series.

Thanks to Brian Howell of WCU for contributing a PicKit 2.

Thanks to Microchip for contributing a PicKit 3, a Microstick II and a MicrostickPlus.

FlashForth Users

Western Carolina University, Kimmel School, Electrical and Computer Engineering Technology
is using FlashForth with PIC18F and dsPIC30F for teaching microcontroller and DSP concepts.

University of Queensland.

PZEF Company.


And many more.