rurban

From af38a34ef19b7ebc912d4d8e688feadb2a82e983 Mon Sep 17 00:00:00 2001
From: Reini Urban <[email protected]>
Date: Thu, 19 Jun 2014 08:40:29 -0600
Subject: [PATCH 1/2] optimize return in pmc methods

---
 lib/Parrot/Pmc2c/PCCMETHOD.pm | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git lib/Parrot/Pmc2c/PCCMETHOD.pm lib/Parrot/Pmc2c/PCCMETHOD.pm

rurban

static  void
Parrot_Float_nci_acos(PARROT_INTERP, ARGMOD(PMC *_self))
{
    PMC * const _ctx         = CURRENT_CONTEXT(interp);
    PMC * const _call_object = Parrot_pcc_get_signature(interp, _ctx);

    { /* BEGIN PARMS SCOPE */
        
        /* Parrot_pcc_fill_params_from_c_args(interp, _call_object, "Pi",
           &_self); */

rurban

diff --git src/pmc/integer.pmc src/pmc/integer.pmc
index e43e90b..3ca3679 100644
--- src/pmc/integer.pmc
+++ src/pmc/integer.pmc
@@ -417,17 +417,16 @@ Adds C<value> to C<SELF> inplace.

     MULTI void i_add(Complex value) {
         const INTVAL a = SELF.get_integer();
-        UNUSED(a);
-

rurban

# -*- perl -*- for parrot pmc2c WB debugging
# perl -d -Ilib tools/build/pmc2c.pl --dump  src/pmc/fixedintegerarray.pmc
sub afterinit {
  push @DB::typeahead, (
    "c 51", "s",
    #"b postpone Parrot::Pmc2c::PMC::Null::pre_method_gen",
    "b Parrot::Pmc2c::Dumper::dump_pmc",
    "b Parrot::Pmc2c::Emitter::add_write_barrier",
    "b Parrot::Pmc2c::PCCMETHOD::rewrite_RETURNs"
    );

rurban

#!/bin/sh
make-4.0 -j4 -s clean archclean
# clang is the fastest
perl Configure.pl --optimize --cc="clang" --link="clang" --ld="clang" --cxx="clang" \
    --maintainer

set -x
sed -i 's/-shared -O2/-shared -Wl,-O1/' Makefile config_lib.pir lib/Parrot/Config/Generated.pm
# highly specific to my system
sed -i 's|-fstack-protector -L/usr/local/lib|-fstack-protector -L/usr/local/lib -Wl,--as-needed -Wl,-z,relro -Wl,-z,now|' Makefile config_lib.pir lib/Parrot/Config/Generated.pm

rurban

Just trying performance and quality of the Intel HW CRC32
In real code this should be checked in perl_init for CPU features.

I guess the hash function is not quite kosher yet, but gives a
rough estimate for the data I'm looking for: performance and
number of collisions.

diff --git hv_func.h hv_func.h
index 191912a..38134bc 100644
--- hv_func.h

rurban

Note that OpenBSD also disabled the openssl freelist "optimization".

CFLAGS+= -DOPENSSL_NO_BUF_FREELISTS -DOPENSSL_NO_HEARTBEAT
http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libssl/ssl/Makefile?rev=1.29;content-type=text%2Fx-cvsweb-markup

I used:
CC= clang
CFLAG+= -O3 -fsanitize=address -fsanitize=undefined -DOPENSSL_NO_BUF_FREELISTS -g

I also created a more readable version at https://github.com/rurban/openssl

rurban

linux-x86_64 shared:

$ ./config
add -DOPENSSL_NO_BUF_FREELISTS to CFLAG in Makefile (because of http://article.gmane.org/gmane.os.openbsd.misc/211963)
$ make
$ export LD_LIBRARY_PATH=`pwd`
$ cd test

find out how to test openssl with valgrind, oops. nope. but make files creates a nice list of EXE targets which contain the tests.

rurban

cat ../bench.sh

#!/bin/sh
if [ -n $1 ]; then
  git clean -dxf 2>&1 >/dev/null
  git checkout $1 2>&1 >/dev/null
else
  make -s clean archclean 2>&1 >/dev/null
fi
perl Configure.pl --optimize --without-crypto --without-opengl 2>&1 >/dev/null \

rurban

# parrot-bench
# release secs
2.0	8.72
2.1	8.30
2.1.1	8.29
2.2	8.23
2.3	10.38
2.4	8.74
2.5	8.52
2.6	8.59