diff -r 31a9f89776a3 -r d47041b23498 progs/while-arrays/compile_bfc.sc --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/progs/while-arrays/compile_bfc.sc Mon Oct 26 10:27:01 2020 +0000 @@ -0,0 +1,310 @@ +// Some more interesting testcases for the +// WHILE compiler with arrays, it includes: +// +// - a small parser for WHILE programs +// +// - transpiles BF programs into WHILE programs +// and then compiles and runs them +// +// - the power-example is the Mandelbrot set: +// +// * 65k for the transpiled WHILE program +// * parsing uses around 30 secs using fastparse +// * the jasmin assembly file is 236k +// * the resulting Java program takes about 20 secs +// +// Call with (X being 0,1,..,4) +// +// amm compile_bfc.sc all +// amm compile_bfc.sc bfcX + + +// load the compiler +import $file.compile_arrays +import compile_arrays._ + +def time_needed[T](i: Int, code: => T) = { + val start = System.nanoTime() + for (j <- 2 to i) code + val result = code + val end = System.nanoTime() + ((end - start) / (i * 1.0e9), result) +} + +// for BF we have to change the write library-function, +// because in BF integers are written out as characters +val beginning = """ +.class public XXX.XXX +.super java/lang/Object + +.method public static write(I)V + .limit locals 1 + .limit stack 2 + getstatic java/lang/System/out Ljava/io/PrintStream; + iload 0 + i2c ; Int => Char + invokevirtual java/io/PrintStream/print(C)V ; println(I)V => print(C)V + return +.end method + +.method public static main([Ljava/lang/String;)V + .limit locals 200 + .limit stack 200 + +; COMPILED CODE STARTS + +""" + + +// modified main compilation function for blocks +def compile(bl: Block, class_name: String) : String = { + val instructions = compile_block(bl, Map())._1 + (beginning ++ instructions ++ ending).replace("XXX", class_name) +} + +// automating the above +import ammonite.ops._ + +def compile_to_file(bl: Block, class_name: String) : Unit = + write.over(pwd / s"$class_name.j", compile(bl, class_name)) + +def compile_and_run(bl: Block, class_name: String) : Unit = { + println(s"Start of compilation") + compile_to_file(bl, class_name) + println(s"generated $class_name.j file") + val (jasmin_time, _) = + time_needed(1, os.proc("java", "-jar", "jasmin.jar", s"$class_name.j").call()) + println(s"generated $class_name.class file (in $jasmin_time secs).") + val (running_time, output) = + time_needed(1, os.proc("java", s"${class_name}/${class_name}").call().out.text()) + println(output) + println(s"done (in $running_time secs).") +} + + +//===================================== +// Grammar Rules for WHILE with arrays +//===================================== + +import fastparse._ +import MultiLineWhitespace._ + +def lowercase [_ : P] = P( CharIn("a-z") ) +def uppercase[_ : P] = P( CharIn("A-Z") ) +def letter[_ : P] = P( lowercase | uppercase ) +def digit [_ : P] = P( CharIn("0-9") ) + +def Number[_ : P]: P[Int] = P( digit.rep(1) ).!.map(_.toInt) +def Ident[_ : P]: P[String] = P( letter ~ (letter | digit | "_").rep ).! + +// arithmetic expressions +def AExp[_ : P]: P[AExp] = + P( P(Te ~ "+" ~ AExp).map{ case (l, r) => Aop("+", l, r)} + | P(Te ~ "-" ~ AExp).map{ case (l, r) => Aop("-", l, r)} + | Te ) +def Te[_ : P]: P[AExp] = + P( P(Fa ~ "*" ~ Te).map{ case (l, r) => Aop("*", l, r)} + | Fa ) +def Fa[_ : P]: P[AExp] = + P( "(" ~ AExp ~ ")" + | P (Ident ~ "[" ~ AExp ~ "]").map{Ref.tupled} + | P(Number).map{Num} + | P(Ident).map{Var} ) + +// boolean expressions +def BExp[_ : P]: P[BExp] = + P( P(AExp ~ "=" ~ AExp).map{ case (x, z) => Bop("=", x, z)} + | P(AExp ~ "!=" ~ AExp).map{ case (x, z) => Bop("!=", x, z)} + | P(AExp ~ "<" ~ AExp).map{ case (x, z) => Bop("<", x, z)} + | P(AExp ~ ">" ~ AExp).map{ case (x, z) => Bop("<", z, x)} + | P("true").map{ _ => True} + | P("false").map{ _ => False} + | "(" ~ BExp ~ ")" ) + +// statements and blocks +def Stmt[_ : P]: P[Stmt] = + P( P("skip").map( _ => Skip) + | P(Ident ~ ":=" ~ AExp).map{Assign.tupled} + | P(Ident ~ "[" ~ AExp ~ "]" ~ ":=" ~ AExp).map{AssignA.tupled} + | P("if" ~ BExp ~ "then" ~ Block ~ "else" ~ Block).map{If.tupled} + | P("while" ~ BExp ~ "do" ~ Block).map{While.tupled} + | P("new(" ~ Ident ~ "[" ~ Number ~ "])").map{ArrayDef.tupled} + | P("write(" ~ Ident ~ ")").map{Write} ) + +def Stmts[_ : P]: P[Block] = + P( P(Stmt ~ ";" ~ Stmts).map{ case (x, z) => x :: z } + | P(Stmt).map{s => List(s)} ) + +def Block[_ : P]: P[Block] = + P( "{" ~ Stmts ~ "}" + | P(Stmt).map(s => List(s)) ) + +// some test cases for the parser + +//println(fastparse.parse("(1 + (2 + 5)) + 4", AExp(_)).get) +//println(fastparse.parse("1 + 2 + 3 + 45", AExp(_))) +//println(fastparse.parse("1 + 2 * 3", AExp(_))) +//println(fastparse.parse("x + 2 * 3", AExp(_))) +//println(fastparse.parse("x2 := 5 + a", Stmts(_))) +//println(fastparse.parse("x2 := 5 + a[3+a]", Stmts(_))) +//println(fastparse.parse("a[x2+3] := 5 + a[3+a]", Stmts(_))) +//println(fastparse.parse("{x := 5; y := 8}", Block(_))) +//println(fastparse.parse("if (false) then {x := 5} else {x := 10}", Block(_))) + +val fib = + """n := 10; + minus1 := 0; + minus2 := 1; + temp:=0; + while (n > 0) do { + temp := minus2; + minus2 := minus1 + minus2; + minus1 := temp; + n := n - 1}; + result := minus2; + write(result) + """ + +//println(fastparse.parse(fib, Stmts(_)).get.value) + + + +//====================================== +// BF transpiler into WHILE with arrays +//====================================== + +// simple BF instructions translation +def instr(c: Char) : String = c match { + case '>' => "ptr := ptr + 1;" + case '<' => "ptr := ptr - 1;" + case '+' => "mem[ptr] := mem[ptr] + 1;" + case '-' => "mem[ptr] := mem[ptr] - 1;" + case '.' => "x := mem[ptr]; write x;" + //case ',' => "XXX" // "ptr = getchar();\n" + case '[' => "while (mem[ptr] != 0) do {" + case ']' => "skip};" + case _ => "" +} + +def instrs(prog: String) : String = + prog.toList.map(instr).mkString + + +// Note: Unfortunately, the transpiled mandelbrot.bf program +// is so large that it does not fit inside the limitations of +// what the JVM imposes on methods (only 64K of instructions). +// Therefore some optimisations are first applied to +// BF programs before WHILE programs are created. The +// optimisations are +// +// - replacing BF-loops of the form [-] with a new 0-instruction +// - combining single increment/decrement instructions +// +// The size of the resulting .j-file is 270K. + + +def splice(cs: List[Char], acc: List[(Char, Int)]) : List[(Char, Int)] = (cs, acc) match { + case (Nil, acc) => acc + case (c :: cs, Nil) => splice(cs, List((c, 1))) + case (c :: cs, (d, n) :: acc) => + if (c == d) splice(cs, (c, n + 1) :: acc) + else splice(cs, (c, 1) :: (d, n) :: acc) +} + +def spl(s: String) = splice(s.toList, Nil).reverse + +def instr2(c: Char, n: Int) : String = c match { + case '>' => s"ptr := ptr + $n;" + case '<' => s"ptr := ptr - $n;" + case '0' => s"mem[ptr] := 0;" + case '+' => s"mem[ptr] := mem[ptr] + $n;" + case '-' => s"mem[ptr] := mem[ptr] - $n;" + case '.' => s"x := mem[ptr]; write(x);" + case '[' => "while (mem[ptr] != 0) do {" * n + case ']' => "skip};" * n + case _ => "" +} + +def instrs2(prog: String) : String = + spl(prog.replaceAll("""\[-\]""", "0")).map{ case (c, n) => instr2(c, n) }.mkString + +// adding the "header" to the BF program +def bf_str(prog: String) : String = { + "new(mem[30000]);" ++ + "ptr := 15000;" ++ + instrs2(prog) ++ + "skip" +} + + +def bf_run(prog: String, name: String) = { + println(s"BF pre-processing of $name") + val bf_string = bf_str(prog) + println(s"BF parsing (program length ${bf_string.length} characters)") + val (time, bf_prog) = + time_needed(1, fastparse.parse(bf_string, Stmts(_)).get.value) + println(s"BF generated WHILE program (needed $time secs for parsing)") + compile_and_run(bf_prog, name) +} + +// a benchmark program (counts down from 'Z' to 'A') +@doc(" Benchmark 'Z' to 'A'.") +@main +def bfc0() = bf_run(read(pwd / "benchmark.bf"), "bench") + + +@doc(" Sierpinski triangle.") +@main +def bfc1() = bf_run(read(pwd / "sierpinski.bf"), "sier") + +// Hello World +val bf2 = """++++++++[>++++[>++>+++>+++>+<<<<-]>+>+>->>+[<]<-] + >>.>---.+++++++..+++.>>.<-.<.+++.------.--------.>>+.>++.""" + +@doc(" Hello world.") +@main +def bfc2() = bf_run(bf2, "hello") + +// Fibonacci +val bf3 = """+++++++++++ + >+>>>>++++++++++++++++++++++++++++++++++++++++++++ + >++++++++++++++++++++++++++++++++<<<<<<[>[>>>>>>+> + +<<<<<<<-]>>>>>>>[<<<<<<<+>>>>>>>-]<[>++++++++++[- + <-[>>+>+<<<-]>>>[<<<+>>>-]+<[>[-]<[-]]>[<<[>>>+<<< + -]>>[-]]<<]>>>[>>+>+<<<-]>>>[<<<+>>>-]+<[>[-]<[-]] + >[<<+>>[-]]<<<<<<<]>>>>>[+++++++++++++++++++++++++ + +++++++++++++++++++++++.[-]]++++++++++<[->-<]>++++ + ++++++++++++++++++++++++++++++++++++++++++++.[-]<< + <<<<<<<<<<[>>>+>+<<<<-]>>>>[<<<<+>>>>-]<-[>>.>.<<< + [-]]<<[>>+>+<<<-]>>>[<<<+>>>-]<<[<+>-]>[<+>-]<<<-] + [-]++++++++++.""" + +@doc(" Fibonacci numbers.") +@main +def bfc3() = bf_run(bf3, "fibs") + +// Mandelbrot Set +//---------------- +// +// Note: Parsing of the generated WHILE program (around 60K in size) +// takes approximately 10 minutes to parse with our parser combinators, +// and approximately 30 seconds with Ammonite's fastparse. + +@doc(" Mandelbrot set.") +@main +def bfc4() = bf_run(read(pwd / "mandelbrot.bf"), "mandelbrot") + + +// this unfortunately hits the capacity of the JVM, even with optimisations +//@doc(" Coolatz serries up to 30.") +//@main +//def bfc5() = bf_run(read(pwd / "collatz.bf"), "coll") + + +// +@doc(" All benchmarks.") +@main +def all() = { bfc0(); bfc1(); bfc2(); bfc3(); bfc4() } + + +