The AWK Programming Language (TAPL) seems at first glance to be a book on using AWK to fulfill your text processing needs. The first 100 or so pages do a great job of that, but then you reach Chapter 6, the first 5 pages of which describe a virtual machine and assembler whose AWK source fits just under a page, then you realize---that AWK is so much more.

This repository takes the assembler/interpreter on page 134 of TAPL and adds additional instructions and directives, without breaking backwards compatibility. It's still relatively small, though, under 200 lines.

The lines between Running... and Stopping. is the program output, in this case, it's 15.

$ awk -f vm int_sum.asm in.txt
Pass 1...done
Pass 2...done. 12 bytes total
Running program...
15

Stopping after 32 instructions.  Program counter at 10.

$ awk -f vm greet.asm
Pass 1...done
Pass 2...done. 85 bytes total
Running program...
What is your name? Brian Kernighan
Hello, Brian Kernighan.
Stopping after 206 instructions.  Program counter at 23.

start
        lda msg
        puts
        halt

msg string "hello, world!"

N.B. Instructions with opcode > 10 were written by me.

Boldface on opcodes <= 10 indicate changed behavior from the book

Not really operations, but make inline constants easy.

# Declare array "foo" of size 10
foo blk 10
# Declare variable "bar" with default value 5
bar const 5
# Declare string "msg" with contents "hello, world!"
msg string "hello, world!"

Mostly qualitative, not rigorous benchmarks. benchmark.asm and benchmark2.asm perform the same computation (decrement a number from 99999 to 0, 100 times), but benchmark.asm uses the accumulator to hold the value, whereas benchmark2.asm uses a memory cell. benchmark.asm runs in 60% of the instructions and takes 70% of the time compared to benchmark2.asm.

$ time awk -f vm benchmark.asm 
...
Stopping after 30000499 instructions.  Program counter at 12.
awk -f vm benchmark.asm  26.91s user 0.05s system 99% cpu 26.972 total

$ time awk -f vm benchmark2.asm
...
Stopping after 50000597 instructions.  Program counter at 13.
awk -f vm benchmark2.asm  36.26s user 0.00s system 99% cpu 36.258 total

• benchmark.asm (26.91/30000499) ≈ (0.897 µs/cycle)
• benchmark2.asm (36.26/50000597) ≈ (0.725 µs/cycle)

Using 0.725 µs/cycle, we find that the VM runs at a speed of 1.38 MHz. Not bad for 150 lines of Awk!

vm.awk doesn't depend on implementation specific features (such as the traversal order of associative arrays), so we can test out different Awk implementations, take, for instance, mawk:

$ time mawk -f vm.awk benchmark.asm
...
Stopping after 30000499 instructions.  Program counter at 12.
mawk -f vm.awk benchmark.asm  15.72s user 0.00s system 99% cpu 15.731 total

$ time mawk -f vm.awk benchmark2.asm
...
Stopping after 50000597 instructions.  Program counter at 13.
mawk -f vm.awk benchmark2.asm  21.00s user 0.01s system 99% cpu 21.007 total

• benchmark.asm (15.72/30000499) ≈ (0.524 µs/cycle)
• benchmark2.asm (21.00/50000597) ≈ (0.420 µs/cycle)

Now the VM runs at 2.38 MHz!

• Multiplication, division
• Memory indirection operations
• Multiple registers
• Reading/writing ASCII characters
• Stacks (push, pop)
  • Subroutine calls

• MIX computer (from The Art of Computer Programming by Donald Knuth)
• Z80 (without interrupts)
• R216