[Bug c/19133] New: march=athlon can produce slower code than march=i686

["ornati at fastwebnet dot it" <gcc-bugzilla at gcc dot gnu dot org>]

"gcc -v"
Reading specs from /usr/lib/gcc-lib/i686-pc-linux-gnu/3.3.4/specs
Configured with: /var/tmp/portage/gcc-3.3.4-r1/work/gcc-3.3.4/configure
--prefix=/usr --bindir=/usr/i686-pc-linux-gnu/gcc-bin/3.3
--includedir=/usr/lib/gcc-lib/i686-pc-linux-gnu/3.3.4/include
--datadir=/usr/share/gcc-data/i686-pc-linux-gnu/3.3
--mandir=/usr/share/gcc-data/i686-pc-linux-gnu/3.3/man
--infodir=/usr/share/gcc-data/i686-pc-linux-gnu/3.3/info --enable-shared
--host=i686-pc-linux-gnu --target=i686-pc-linux-gnu --with-system-zlib
--enable-languages=c,c++ --enable-threads=posix --enable-long-long
--disable-checking --disable-libunwind-exceptions --enable-cstdio=stdio
--enable-version-specific-runtime-libs
--with-gxx-include-dir=/usr/lib/gcc-lib/i686-pc-linux-gnu/3.3.4/include/g++-v3
--with-local-prefix=/usr/local --enable-shared --enable-nls
--without-included-gettext --disable-multilib --enable-__cxa_atexit
--enable-clocale=generic
Thread model: posix
gcc version 3.3.4 20040623 (Gentoo Linux 3.3.4-r1, ssp-3.3.2-2, pie-8.7.6)


"cat /proc/cpu"
processor	: 0
vendor_id	: AuthenticAMD
cpu family	: 6
model		: 3
model name	: AMD Duron(tm) Processor
stepping	: 1
cpu MHz		: 756.798
cache size	: 64 KB
fdiv_bug	: no
hlt_bug		: no
f00f_bug	: no
coma_bug	: no
fpu		: yes
fpu_exception	: yes
cpuid level	: 1
wp		: yes
flags		: fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx
fxsr syscall mmxext 3dnowext 3dnow
bogomips	: 1490.94


Playing with compiler optimizations I've found a "test-case" where
"-march=athlon" produces an executable that is ~18% slower than that produced
with "-march=i686".

"Commands used for compilation"
gcc -Wall -O2 -march=i686 -o test-i686 test.c
gcc -Wall -O2 -march=athlon -o test-athlon test.c

"test.c"
#define N 30000

static int A[N];
static int B[N];

int main(void)
{
	int i, j, x;

	for (i = 0; i < N; ++i) {
		x = 0;
		for (j=0; j < N; ++j)
			if (A[j] < A[i] && B[j] > x)
				x = A[j];
		A[i] = x + 1;
	}
	return 0;
}


Execution time:

paolo@tux gcc_test $ time ./test-i686 

real    0m6.149s
user    0m6.015s
sys     0m0.004s
paolo@tux gcc_test $ time ./test-athlon 

real    0m7.270s
user    0m7.120s
sys     0m0.004s


Anoter interesting thing is that if I only do this change:
--- test.c.orig	2004-12-22 21:52:46.208140096 +0100
+++ test.c	2004-12-22 21:55:01.069638032 +0100
@@ -10,7 +10,7 @@
 	for (i = 0; i < N; ++i) {
 		x = 0;
 		for (j=0; j < N; ++j)
-			if (A[j] < A[i] && B[j] > x)
+			if (A[j] < A[i] && A[j] > x)
 				x = A[j];
 		A[i] = x + 1;
 	}


The "test-athlon" version returns fast as "test-i686" one:

paolo@tux gcc_test $ time ./test-i686 

real    0m6.134s
user    0m6.015s
sys     0m0.003s
paolo@tux gcc_test $ time ./test-athlon 

real    0m6.128s
user    0m6.010s
sys     0m0.006s


I've tried also gcc 3.4.3 and it shows the same behaviour.

Shouldn't "-march=athlon" optimize for AMD K7 processors? ;-)

-- 
           Summary: march=athlon can produce slower code than march=i686
           Product: gcc
           Version: 3.3.4
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ornati at fastwebnet dot it
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19133