This is an old revision of the document!

Star Forth Test Suite

Current version: 1.0

Version	Pass/Fail	Test	Expected Results
1.0 260313	PASS	signed > `-3 0 > .`	`0`
1.0 260313	PASS	: ; `: TEST 42 . ;`	`42`
1.0 260313	PASS	IF/THEN `: SIGN DUP 0 > IF 43 EMIT THEN . ;` `5 SIGN` `-3 SIGN` 0 SIGN	`+5` `-3` `0`
1.0 260313	PASS	DO/LOOP/I `: TENS 10 0 DO I . LOOP ;` `TENS`	`0 1 2 3 4 5 6 7 8 9`
1.0 260313	PASS	BEGIN/UNTIL `: COUNT3 3 BEGIN DUP . 1 - DUP 0 = UNTIL DROP ;` `COUNT3`	`3 2 1`
1.0 260313	PASS	BEGIN/WHILE `: COUNTDOWN BEGIN DUP WHILE DUP . 1 - REPEAT DROP ;` `5 COUNTDOWN` `0 COUNTDOWN`	`5 4 3 2 1` `0`
1.0 260313	PASS	`HERE .`	`$030800`
1.0 260313	PASS	`HEX $030800 @ .`	`0`
1.0 260313	PASS	`$030800 C@ .`	`0`
1.0 260313	PASS	`42 HERE ! HERE @ .`	`42`
1.0 260313	PASS	`: GREET .“ Hello, World!” CR ;` `GREET`	`Hello, world!`
1.0 260313	PASS	`: FIZZBUZZ 21 1 DO I 15 MOD 0 = IF .“ FizzBuzz” ELSE I 3 MOD 0 = IF .“ Fizz” ELSE I 5 MOD 0 = IF .“ Buzz” ELSE I . THEN THEN THEN CR LOOP ;`	not shown (but it works)
1.0 260313	PASS	`: SIMPLE 500000 0 DO LOOP ;` `SIMPLE`	~6 seconds (170,000 w/s) at 4MHz
1.0 260313	PASS	`: BENCH 500000 0 DO I DUP * DROP LOOP ;` `BENCH`	22 seconds (140,000 w/s) at 4Mhz

Benchmarking

The SIMPLE and BENCH tests indicate SD/FORTH is:

75x faster than Stellar BASIC (1,800 lines/sec) × ~5 ops/line ≈ 9,000 ops/sec)
5x to 7x faster than a real C64 running DurexForth (~25,000 words/sec)
Comparable to a 4 MHz Z80 running Forth850 or an 8MHz Z80B running CamelForth.

However, this is an upper end. The following MUL and DIV heavy benchmarks operate a bit slower. Here's a summary of the followinf programs:

Benchmark	Words/Sec.	Notes
Bare Loop	176,000	Only dispatch overhead
Stackbench	160,000	Only stack operations
DUP * DROP	136,000	Multiply, Stack Ops
ADD only	120,000	Same as Mixed but no DIV or MUL
Mixed no DIV	100,000	Same as Mixed but no DIV
FizzBench	97,000	3x MOD, branches
Mixed +-*/	90,000	Division, Heavy Stack

The estimate is that 70% overhead is dispatch. So with a touch of opt I think the system wants to run at 180,000 words/sec, but gets tinted down by various factors. Stack operations are cheap, then come other types of operation, then MUL, DIV and MOD are considered 'heavy'. This is not unusual, and is in fact a great sign; this is a very fast FORTH.

STACKBENCH

    : STACKBENCH ( -- )
      1 2 3
      300000 0 DO
        DUP DROP SWAP OVER DROP
      LOOP
      DROP DROP DROP ;

    STACKBENCH

FIZZBENCH

    : FIZZBENCH 0 10000 0
        DO I 15 MOD 0 = IF 1 + ELSE
            I 3 MOD 0 = IF 1 + ELSE
                I 5 MOD 0 = IF 1 + THEN
                THEN
            THEN 
        LOOP
    ;
    : FB10 10 0 DO FIZZBENCH LOOP . ." hits" CR ;
    FB10

This executes in 18.5 seconds, indicating 97,000 words/sec.

BENCH-ARITH

    : ARITH ( n -- n )
      DUP 42 +
      DUP 7 -
      DUP 3 *
      DUP 2 /
      DROP DROP DROP DROP ;

    : BENCH-ARITH ( -- )
      ." Arithmetic benchmark" CR
      100000 0 DO
        I ARITH DROP
      LOOP
      ." Done" CR ;

    BENCH-ARITH

Same without DIV:

    : ARITH3 ( n -- n )
      DUP 42 +
      DUP 7 -
      DUP 3 *
      DROP DROP DROP ;

    : BENCH3 ( -- )
      ." Add/Sub/Mul benchmark" CR
      100000 0 DO
        I ARITH3 DROP
      LOOP
      ." Done" CR ;

    BENCH3

Appledog

Table of Contents

Star Forth Test Suite

Benchmarking

STACKBENCH

FIZZBENCH

BENCH-ARITH