Skip to content

Latest commit

 

History

History
105 lines (89 loc) · 3.41 KB

Sample_Binaries.md

File metadata and controls

105 lines (89 loc) · 3.41 KB
Raw

Sample Binaries

The sample binaries...

  • examples/bin/sum
  • examples/bin/sort
  • examples/bin/sieve
  • examples/bin/negate
  • examples/bin/puts
  • examples/bin/cat
  • examples/bin/3sat
  • examples/bin/gzip
  • examples/bin/gunzip
  • examples/bin/zcat
  • examples/bin/sizeof
  • examples/bin/dry2
  • examples/bin/dry2o
  • examples/bin/dry2nr

were created using the RISC-V cross compiler at https://github.com/riscv-collab/riscv-gnu-toolchain, following the directions under the "Installation (Newlib)" section; see: https://github.com/riscv-collab/riscv-gnu-toolchain#installation-newlib. Consider, for instance, the source for the sum program:

/* examples/src/sum.c */
#include <stdio.h>
#include <stdlib.h>

int
main(int argc, char ** argv)
{
    int sum = 0;
    int x = 1;
    for (; x < argc; x+= 1) sum += atoi(argv[x]);
    printf("%d\n", sum);
    return 0;
}

which is compiled accordingly:

riscv64-unknown-elf-gcc -o sum -static -march=rv64g sum.c

Please note that the binary is statically linked, since the simulator at this stage makes NO effort to accommodate dynamic linking.

Outside benchmarks

The gzip source was downloaded from https://ftp.gnu.org/gnu/gzip/gzip-1.2.4.tar.gz, and built with slight modifications according to a build recipe chronicled here.

Unlike the other sample programs, gzip is not a mere toy, but, rather, a real benchmark that does real work. I have used it successfully to compress my .bash_history file from about 375 KB down to about 15 KB with the Jabara sample model; to wit:

cd ${HOME}/src/nebula/pipelines/jabara ; \
LOGDIR=/tmp/log/$( basename $( pwd ) ) ; \
mkdir -p ${LOGDIR} ; \
rm -f ${LOGDIR}/* ; \
cp -f ${HOME}/.bash_history /tmp/.bash_history ; \
time ( \
    python3 ../../launcher.py \
        --log ${LOGDIR} \
        --service ../../toolbox/stats.py:localhost:22:-1:-1 \
        --config \
            stats:output_filename:${LOGDIR}/stats.json \
            mainmem:filename:${LOGDIR}/mainmem.raw \
            mainmem:capacity:$(( 2**32 )) \
        -- \
        $(( 1000 + ${RANDOM} )) \
        localhost.nebula \
        /home/john/src/nebula/examples/bin/gzip /tmp/.bash_history \
)

On my laptop the simulation, executed 71.2 million instructions at a rate of about 26,000 instructions/second, and completed in a bit less than 46 minutes. As expected, this emitted nothing to my terminal, but wrote /tmp/.bash_history.gz and deleted /tmp/.bash_history.

And uncompressing...

cd ${HOME}/src/nebula/pipelines/jabara ; \
LOGDIR=/tmp/log/$( basename $( pwd ) ) ; \
mkdir -p ${LOGDIR} ; \
rm -f ${LOGDIR}/* ; \
time ( \
    python3 ../../launcher.py \
        --log ${LOGDIR} \
        --service ../../toolbox/stats.py:localhost:22:-1:-1 \
        --config \
            stats:output_filename:${LOGDIR}/stats.json \
            mainmem:filename:${LOGDIR}/mainmem.raw \
            mainmem:capacity:$(( 2**32 )) \
        -- \
        $(( 1000 + ${RANDOM} )) \
        localhost.nebula \
        /home/john/src/nebula/examples/bin/gunzip /tmp/.bash_history.gz \
)

...executed 12 million instructions at about 25,000 instructions/second, and completed in a little more than 8 minutes, writing /tmp/.bash_history and deleting /tmp/.bash_history.gz.