From 83c9fb328170c42bf2056d33021c55b2dc668ada Mon Sep 17 00:00:00 2001
From: Uwe Schulzweida <uwe.schulzweida@mpimet.mpg.de>
Date: Fri, 27 Jun 2014 12:11:48 +0000
Subject: [PATCH] cgribexlib update
---
configure | 20 +++++++-------
configure.ac | 2 +-
src/cgribexlib.c | 69 +++++++++++++++++++++++++++++++++++++++++++-----
3 files changed, 74 insertions(+), 17 deletions(-)
diff --git a/configure b/configure
index 6c33d4053..ffea8d681 100755
--- a/configure
+++ b/configure
@@ -1,6 +1,6 @@
#! /bin/sh
# Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.68 for cdi 1.6.4rc4.
+# Generated by GNU Autoconf 2.68 for cdi 1.6.4.
#
# Report bugs to <http://code.zmaw.de/projects/cdi>.
#
@@ -570,8 +570,8 @@ MAKEFLAGS=
# Identity of this package.
PACKAGE_NAME='cdi'
PACKAGE_TARNAME='cdi'
-PACKAGE_VERSION='1.6.4rc4'
-PACKAGE_STRING='cdi 1.6.4rc4'
+PACKAGE_VERSION='1.6.4'
+PACKAGE_STRING='cdi 1.6.4'
PACKAGE_BUGREPORT='http://code.zmaw.de/projects/cdi'
PACKAGE_URL=''
@@ -1442,7 +1442,7 @@ if test "$ac_init_help" = "long"; then
# Omit some internal or obsolete options to make the list less imposing.
# This message is too long to be a string in the A/UX 3.1 sh.
cat <<_ACEOF
-\`configure' configures cdi 1.6.4rc4 to adapt to many kinds of systems.
+\`configure' configures cdi 1.6.4 to adapt to many kinds of systems.
Usage: $0 [OPTION]... [VAR=VALUE]...
@@ -1512,7 +1512,7 @@ fi
if test -n "$ac_init_help"; then
case $ac_init_help in
- short | recursive ) echo "Configuration of cdi 1.6.4rc4:";;
+ short | recursive ) echo "Configuration of cdi 1.6.4:";;
esac
cat <<\_ACEOF
@@ -1699,7 +1699,7 @@ fi
test -n "$ac_init_help" && exit $ac_status
if $ac_init_version; then
cat <<\_ACEOF
-cdi configure 1.6.4rc4
+cdi configure 1.6.4
generated by GNU Autoconf 2.68
Copyright (C) 2010 Free Software Foundation, Inc.
@@ -2460,7 +2460,7 @@ cat >config.log <<_ACEOF
This file contains any messages produced by compilers while
running configure, to aid debugging if configure makes a mistake.
-It was created by cdi $as_me 1.6.4rc4, which was
+It was created by cdi $as_me 1.6.4, which was
generated by GNU Autoconf 2.68. Invocation command line was
$ $0 $@
@@ -3405,7 +3405,7 @@ fi
# Define the identity of the package.
PACKAGE='cdi'
- VERSION='1.6.4rc4'
+ VERSION='1.6.4'
cat >>confdefs.h <<_ACEOF
@@ -28417,7 +28417,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
# report actual input values of CONFIG_FILES etc. instead of their
# values after options handling.
ac_log="
-This file was extended by cdi $as_me 1.6.4rc4, which was
+This file was extended by cdi $as_me 1.6.4, which was
generated by GNU Autoconf 2.68. Invocation command line was
CONFIG_FILES = $CONFIG_FILES
@@ -28483,7 +28483,7 @@ _ACEOF
cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
ac_cs_version="\\
-cdi config.status 1.6.4rc4
+cdi config.status 1.6.4
configured by $0, generated by GNU Autoconf 2.68,
with options \\"\$ac_cs_config\\"
diff --git a/configure.ac b/configure.ac
index fbd297c23..86b5c9ccf 100644
--- a/configure.ac
+++ b/configure.ac
@@ -4,7 +4,7 @@
# autoconf 2.68
# libtool 2.4.2
-AC_INIT([cdi], [1.6.4rc4], [http://code.zmaw.de/projects/cdi])
+AC_INIT([cdi], [1.6.4], [http://code.zmaw.de/projects/cdi])
echo "configuring ${PACKAGE_NAME} ${PACKAGE_VERSION}"
diff --git a/src/cgribexlib.c b/src/cgribexlib.c
index ae06fe140..6e968f88f 100644
--- a/src/cgribexlib.c
+++ b/src/cgribexlib.c
@@ -1,5 +1,5 @@
-/* Automatically generated by m214003 at 2014-05-28, do not edit */
+/* Automatically generated by m214003 at 2014-06-27, do not edit */
/* CGRIBEXLIB_VERSION="1.6.4" */
@@ -640,14 +640,16 @@ gcc -g -Wall -O3 -march=native -std=c99 -DTEST_MINMAXVAL minmax_val.c
orig : fmin: -500000 fmax: 499999 time: 4.84s
sse2 : fmin: -500000 fmax: 499999 time: 4.82s
-gcc -g -Wall -O3 -march=native -std=c99 -DTEST_MINMAXVAL minmax_val.c
+gcc -g -Wall -O3 -march=native -std=c99 -DTEST_MINMAXVAL -fopenmp -DOMP_SIMD minmax_val.c
result on thunder5 (gcc 4.7.2):
orig : fmin: -500000 fmax: 499999 time: 3.10s
+ simd : fmin: -500000 fmax: 499999 time: 3.23s # omp simd in gcc 4.9
avx : fmin: -500000 fmax: 499999 time: 2.84s
-icc -g -Wall -O3 -march=native -std=c99 -vec-report=1 -DTEST_MINMAXVAL minmax_val.c
- result on thunder5 (icc 13.1.2):
+icc -g -Wall -O3 -march=native -std=c99 -vec-report=1 -DTEST_MINMAXVAL -openmp -DOMP_SIMD minmax_val.c
+ result on thunder5 (icc 14.0.2):
orig : fmin: -500000 fmax: 499999 time: 2.83s
+ simd : fmin: -500000 fmax: 499999 time: 2.83s
avx : fmin: -500000 fmax: 499999 time: 2.92s
xlc_r -g -O3 -qhot -q64 -qarch=auto -qtune=auto -qreport -DTEST_MINMAXVAL minmax_val.c
@@ -978,6 +980,43 @@ void minmax_val_orig(const double *restrict data, long idatasize, double *fmin,
#pragma GCC pop_options
#endif
+// TEST
+#if defined(OMP_SIMD)
+
+//#pragma omp declare reduction(xmin : double : omp_out = omp:in > omp_out ? omp_out : omp_in)
+// initializer( omp_priv = { largenumber })
+//#pragma omp declare reduction(xmax : double : omp_out = omp:in < omp_out ? omp_out : omp_in)
+// initializer( omp_priv = { -largenumber })
+
+#if defined(GNUC_PUSH_POP)
+#pragma GCC push_options
+#pragma GCC optimize ("O3", "fast-math")
+#endif
+static
+void minmax_val_simd(const double *restrict data, long idatasize, double *fmin, double *fmax)
+{
+ size_t i;
+ size_t datasize = idatasize;
+ double dmin = *fmin, dmax = *fmax;
+
+#if defined(_OPENMP)
+ //#pragma omp simd reduction(xmin:dmin) reduction(xmax:dmax)
+#pragma omp simd
+#endif
+ for ( i = 0; i < datasize; ++i )
+ {
+ dmin = dmin < data[i] ? dmin : data[i];
+ dmax = dmax > data[i] ? dmax : data[i];
+ }
+
+ *fmin = dmin;
+ *fmax = dmax;
+}
+#if defined(GNUC_PUSH_POP)
+#pragma GCC pop_options
+#endif
+#endif
+
static
void minmax_val(const double *restrict data, long idatasize, double *fmin, double *fmax)
{
@@ -1083,6 +1122,10 @@ int main(void)
double fmin, fmax;
double t_begin, t_end;
+#if defined(_OPENMP)
+ printf("_OPENMP=%d\n", _OPENMP);
+#endif
+
#if defined(__ICC)
printf("icc\n");
#elif defined(__clang__)
@@ -1093,7 +1136,10 @@ int main(void)
data = (double*) malloc(datasize*sizeof(double));
- for ( long i = datasize-1; i >= 0; i-- ) data[i] = (double) (-datasize/2 + i);
+ // for ( long i = datasize-1; i >= 0; i-- ) data[i] = (double) (-datasize/2 + i);
+ for ( long i = 0; i < datasize/2; i++ ) data[i] = (double) (i);
+ for ( long i = datasize/2; i < datasize; i++ ) data[i] = (double) (-datasize + i);
+ // for ( long i = 0; i < datasize; i++ ) printf("%d %g\n", i, data[i]);
t_begin = dtime();
for ( int i = 0; i < NRUN; ++i )
@@ -1113,6 +1159,17 @@ int main(void)
t_end = dtime();
printf("orig : fmin: %ld fmax: %ld time: %6.2fs\n", (long)fmin, (long) fmax, t_end-t_begin);
+#if defined(OMP_SIMD)
+ t_begin = dtime();
+ for ( int i = 0; i < NRUN; ++i )
+ {
+ fmin = fmax = data[0];
+ minmax_val_simd(data, datasize, &fmin, &fmax);
+ }
+ t_end = dtime();
+ printf("simd : fmin: %ld fmax: %ld time: %6.2fs\n", (long)fmin, (long) fmax, t_end-t_begin);
+#endif
+
#if defined(_ENABLE_AVX)
t_begin = dtime();
for ( int i = 0; i < NRUN; ++i )
@@ -10842,7 +10899,7 @@ int gribUnzip(unsigned char *dbuf, long dbufsize, unsigned char *sbuf, long sbu
return (gribLen);
}
-static const char grb_libvers[] = "1.6.4" " of ""May 28 2014"" ""11:22:51";
+static const char grb_libvers[] = "1.6.4" " of ""Jun 27 2014"" ""14:00:04";
const char *
cgribexLibraryVersion(void)
{
--
GitLab