m5 x86_se spec2k status

32-bit results

as of 4 November 2009

The following results assume the following syscall related patches have been made:

access = ignoreFunc, times = timesFunc, setrlimit = ignoreFunc, _llseek() is hooked up, rt_sigprocmask = ignoreFunc, ugetrlimit = ignoreFunc, dup = dupFunc, mprotect = ignoreFunc
ioctl() is implemented
getrlimit() is implemented, getrusage() is implemented

Times given are running i686 compiled spec2k benchmarks (with compiler option -O2 -static) under m5.opt using config se.py on a cluster of 3.46GHz Pentium D machines.

Runs Correctly to Completion

benchmark	native	simple nocache	slowdown
gcc.expr	0:03.2	5:42:46	6427x
gcc.integrate	0:03.5	5:50:11	6003x
gcc.166	0:10.4	18:48:12	6509x
gzip.log	0:13.1	20:23:11	5602x
gcc.scilab	0:17.7	29:57:27	6093x
gzip.source	0:27.8	40:01:59	5184x
gzip.random	0:24.9	44:32:49	6440x
gcc.200	0:31.3	53:54:01	6199x
gzip.graphic	0:31.2	54:59:35	6345x
bzip2.source	0:45.3	63:18:22	5031x
bzip2.program	0:45.4	73:45:46
gzip.program	0:49.5	75:46:29
bzip2.graphic	0:56.4	84:43:40
crafty	1:37	141:50:20
gap	1:24	154:04:29
parser	3:10	271:24:21

Still Running

mesa

apsi

art.470

art.110

Benchmark Fails

Benchmark	Error message	Reason
mcf	n/a	wrong output

Benchmark is broken with modern compilers

vortex.*

System Call Missing

benchmark	Missing syscall
fma3d	dup
mgrid	gettid
perl.*	getdents64

Missing Floating Point

instruction	benchmarks
f2xm1	eon
fabs	art,galgel,sixtrack
fadd	crafty,eon,swim,twolf
faddp	apsi,art,galgel,mcf,wupwise
fchs	eon,mgrid,wupwise
fcmove	galgel
fcomp	galgel
fcos	swim
fdiv	crafty,eon,galgel,mgrid,swim,twolf
fdivp	gcc,mgrid,swim
fdivr	apsi,art,galgel,swim
fdivrp	apsi,art,eon,galgel,twolf,wupwise
fild	apsi,art,crafty,eon,mcf,gcc,swim,twolf,wupwise
fistp	applu,crafty,eon,mgrid,sixtrack,twolf,wupwise
flcdw_Mw	all
fld	art,crafty,eon,fma3d,galgel,mgrid,sixtrack,swim,twolf,vpr
fld1	apsi,art,fma3d,galgel,mgrid,twolf,vpr
fldlg2	galgel
fldz	art,eon,galgel,perlbmk,sixtrack,twolf,vpr,wupwise
fmul	apsi,art,eon,galgel,gcc,mgrid,sixtrack,swim,twolf
fmulp	art,eon,gcc,mcf,twolf
fnstcw_Mw	all
fnstsw	eon,mgrid,twolf,vpr
fpatan	apsi,sixtrack,swim
fprem1	apsi
frndint	eon
fscale	eon
fsin	apsi,swim
fsqrt	art,mgrid,sixtrack,wupwise
fst	eon
fstp	eon,galgel,vpr
fsub	art,sixtrack,swim,twolf
fsubp	eon
fsubr	eon,twolf,wupwise
fsubrp	wupwise
fucom	apsi,art,crafty,eon,galgel,mgrid,twolf,wupwise
fucomip	art
fucomp	art,twolf
fucompp	art,mgrid,sixtrack,vpr,wupwise
fwait	fma3d,galgel
fxam	galgel
fxch	apsi,art,crafty,eon,galgel,gcc,mgrid,sixtrack,swim,twolf,wupwise
fyl2x	galgel
fylx2p1	eon

Errors that can occur

Pagetable Error (this has been fixed with recent m5 changes and shoud no longer occur)

fatal: PageTable::allocate: address 0xffffd000 already mapped @ cycle 659293000 [allocate:build/X86_SE/mem/page_table.cc, line 81].

This is caused by mmap_end staring very close to the stack.

Back to main m5 status page