Bad performance of PARI when compiling with GCC for SPARC 4.2.0

Question

Hi,
I've compiled pari (2.3.3) with gcc 4.2.0 and with an old gcc 3.3.2 on a Sun Fire V240 and Solaris 10.
The performance of pari compiled with gcc 4.2.0 is terrible. I've tried it with '-O3 -fast' and without this flags with the same result.
First compiled with gcc 3.3.2 without any special CFLAGS and run make bench:
* Testing objets for gp-sta..TIME=4 for gp-dyn..TIME=4
* Testing analyz for gp-sta..TIME=82 for gp-dyn..TIME=80
* Testing number for gp-sta..TIME=63 for gp-dyn..TIME=62
* Testing polyser for gp-sta..TIME=19 for gp-dyn..TIME=19
* Testing linear for gp-sta..TIME=30 for gp-dyn..TIME=29
* Testing elliptic for gp-sta..TIME=50 for gp-dyn..TIME=51
* Testing sumiter for gp-sta..TIME=45 for gp-dyn..TIME=47
* Testing graph for gp-sta..TIME=25 for gp-dyn..TIME=25
* Testing program for gp-sta..TIME=94 for gp-dyn..TIME=94
* Testing trans for gp-sta..TIME=228 for gp-dyn..TIME=225
* Testing nfields for gp-sta..TIME=434 for gp-dyn..TIME=433
+++ Total bench for gp-sta is 726
+++ Total bench for gp-dyn is 722
Then compiled with gcc 4.2.0 again without any special flags and run make bench:
* Testing objets for gp-sta..TIME=4 for gp-dyn..TIME=4
* Testing analyz for gp-sta..TIME=82 for gp-dyn..TIME=83
* Testing number for gp-sta..TIME=62 for gp-dyn..TIME=65
* Testing polyser for gp-sta..TIME=19 for gp-dyn..TIME=19
* Testing linear for gp-sta..BUG [1686975] for gp-dyn..BUG [1608146]
* Testing elliptic for gp-sta..TIME=50 for gp-dyn..TIME=52
* Testing sumiter for gp-sta..TIME=51 for gp-dyn..TIME=50
* Testing graph for gp-sta..TIME=28 for gp-dyn..TIME=28
* Testing program for gp-sta..TIME=98 for gp-dyn..TIME=97
* Testing trans for gp-sta..BUG [613000] for gp-dyn..BUG [606558]
* Testing nfields for gp-sta..BUG [3648091] for gp-dyn..BUG [3716170]
+++ [BUG] Total bench for gp-sta is 3029987
+++ [BUG] Total bench for gp-dyn is 2958336
Do anybody know this problem or can tell me what the problem could be?
Thanks!!
Rainer W.

Answer 1

Here are the compiler options I used to compile it with gcc4.
C compiler is /usr/local/gcc4/gcc/bin/gcc -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer
Executable linker is /usr/local/gcc4/gcc/bin/gcc -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer
Dynamic Lib linker is /usr/local/gcc4/gcc/bin/gcc -shared -mimpure-text $(CFLAGS) $(DLCFLAGS) -Wl,-G,-h,$(LIBPARI_SONAME)
If I use the -fast option I get the same result may be a little bit worse.
Here are some compilation lines:
Making gp in Osolaris-sparcv9
make[2]: Entering directory `/no_backup/pari/pari-2.3.3/Osolaris-sparcv9'
File ../src/funclist not changed.
../config/genkernel ../src/kernel/sparcv8_micro/asm0-common.h ../src/kernel/sparcv8_micro/asm0.h > parilvl0.h
cat ../src/kernel/none/tune.h ../src/kernel/none/int.h ../src/kernel/none/level1.h > parilvl1.h
cat parilvl0.h parilvl1.h > pariinl.h
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -I../src/language -I/usr/local/include -o gp.o ../src/gp/gp.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -I../src/graph -o gp_init.o ../src/gp/gp_init.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -I../src/language -I/usr/local/include -o gp_rl.o ../src/gp/gp_rl.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -DDL_DFLT_NAME=NULL -o highlvl.o ../src/gp/highlvl.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o whatnow.o ../src/gp/whatnow.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -I/usr/openwin/include -o plotX.o ../src/graph/plotX.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o anal.o ../src/language/anal.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o compat.o ../src/language/compat.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o default.o ../src/language/default.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o errmsg.o ../src/language/errmsg.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o es.o ../src/language/es.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o init.o ../src/language/init.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o intnum.o ../src/language/intnum.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o members.o ../src/language/members.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o sumiter.o ../src/language/sumiter.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o aprcl.o ../src/modules/aprcl.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o elldata.o ../src/modules/elldata.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o elliptic.o ../src/modules/elliptic.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o galois.o ../src/modules/galois.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o groupid.o ../src/modules/groupid.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o kummer.o ../src/modules/kummer.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o mpqs.o ../src/modules/mpqs.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o nffactor.o ../src/modules/nffactor.c
/usr/local/gcc4/gcc/bin/gcc -c -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer -I. -I../src/headers -o part.o ../src/modules/part.c