Yet another (fast) IVM emulator

This ivm64 emulator takes an ivm64 binary and emulates its execution. It benefits from two kind of optimizations:

• Optimized execution of common instruction patterns generated by the C/C++ compiler
• Parallel execution of output instructions based on multiple processes

Requirements

For versions with I/O, library png is needed.

In a deb based linux distribution, it can be installed as:

sudo apt update
sudo apt install libpng-dev

How to compile?

In order to have the emulator ivm64-emu, build the project by calling make:

  $ make # generates ivm64-emu and some other executables
  gcc ivm_emu.c -o ivm64-emu               -DSTEPCOUNT -DWITH_IO -lpng
  gcc ivm_emu.c -o ivm64-emu-trace_compact -DSTEPCOUNT -DWITH_IO -DNOOPT -DVERBOSE=3 -lpng

  gcc ivm_emu.c -o ivm64-emu-no-io    -DSTEPCOUNT
  gcc ivm_emu.c -o ivm64-emu-parallel -DSTEPCOUNT -DWITH_IO -DPARALLEL_OUTPUT -lpng

  gcc ivm_emu.c -o ivm64-emu-histo    -DSTEPCOUNT -DWITH_IO -DNOOPT -DVERBOSE=1 -DHISTOGRAM -lpng
  gcc ivm_emu.c -o ivm64-emu-trace2   -DSTEPCOUNT -DWITH_IO -DNOOPT -DVERBOSE=2 -lpng
  gcc ivm_emu.c -o ivm64-emu-trace4   -DSTEPCOUNT -DWITH_IO -DNOOPT -DVERBOSE=4 -lpng

Also you can build your customized emulator combining different options. The following ones are for debugging and they can make the execution slower:

  gcc  -DFPE_ENABLED ivm_emu.c # Generate exception when dividing by zero
  gcc  -DNOOPT       ivm_emu.c # Disable optimizations
  gcc  -DSTEPCOUNT   ivm_emu.c # Enable instruction count, show probes if used
  gcc  -DVERBOSE=1   ivm_emu.c # Some extra info
  gcc  -DVERBOSE=2   ivm_emu.c # Trace mode (show instruction sequence)
  gcc  -DVERBOSE=3   ivm_emu.c # Trace mode w/ compact output (similar to original ivm app)
  gcc  -DVERBOSE=4   ivm_emu.c # Detailed trace mode (instr. and stack content sequence)
  gcc  -DHISTOGRAM   ivm_emu.c # Enable insn. pattern histogram

The number of processes for the version with parallel output is 8 by default. If compiled with -DNUM_THREADS=N1, N1 is used instead of the default value. If set the environment variable NUM_THREADS=N2, N2 is used instead of N1 or default. In any case, the parallel version uses at least 2 threads, in general: 1 thread for emulation and (N-1) thread for io.

How to execute?

Options keep compatibility with the original ivm implementation.

ivm64-emu [-m <size in bytes>] [-a <arg file> [-a <env file>]] [-i <input dir>] [-o <output dir>] <ivm_binary_file>

This is the meaning of the options:

• -m <size in bytes>: sets the size in bytes of the emulated virtual machine memory
• -a <arg file>: specifies an argument file (in case of a ivm code generated by the ivm64-gcc compiler, the c run time crt0 parses this argument file as common linux process arguments found in file /proc/<pid>/comdline; additionally a second -a option allows specifying an environment file that is processed by crt0 as the same format of linux /proc/<pid>/environment)
• -i <input dir>: in this directory, input instructions will find the data
• -o <output dir>: in this directory, output instructions will write data

Debugging

In addition to ivm64-emu, version 2.0 can be compiled to generate executables that are useful for debugging.

Examples of this capabilities are: ivm64-emu-histo, and ivm64-emu-trace_compact. These last two allow the execution showing the histogram of instructions or a traced execution with a compact output, respectively.

Execution times

These experiments were conducted on an Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz, where the emulator was compiled with gcc 7.4.0.

Code static_unboxing was compiled with gcc 10.2.0-ivm64-2.0. This code has no I/O instructions but put_char.

Code ex4_short_video.s is one of the example codes provided with the ivm application.

Binaries generated with ivm v0.37.

benchmark	no optimization	optimized	optimized + parallel io
static_unboxing	1m30s	35s
ex4_short_video.s (512 frames)	7s	3.6s	1.8s