|
We used SimpleScalar to gather all of our
results. Here is the SimpleScalar
configuration file used to gather these runs. The branch predictor used in
these simulations is a combination of a bimodal predictor and a 2 level local
predictor, each with 8K tables as well as the chooser table. The cache
configuration consists of a level-1 data cache and instruction cache, and a
level-2 unified cache. The data cache is 16K 4-way associative, and instruction
cache is 8K 2-way associative with 32 byte lines. The unified cache is 1 Meg
with 4-way associativity with 64 byte lines.
The results for in this web page for different
configurations are provided in two tables, one for SPEC INT and the other for
SPEC FP.
The following table shows the true IPC for
the above configuration for the full simulation of the corresponding reference
SPEC 2000 inputs.
| name |
IPC |
| bzip2-graphic-ref |
1.642 |
| bzip2-program-ref |
1.767 |
| bzip2-source-ref |
1.541 |
| crafty-ref |
0.569 |
| eon-cook-ref |
0.862 |
| eon-rushmeier-ref |
0.876 |
| gap-ref |
0.750 |
| gcc-00-166-ref |
1.397 |
| gcc-00-200-ref |
1.048 |
| gcc-00-expr-ref |
1.066 |
| gcc-00-integrate-ref |
1.431 |
| gcc-00-scilab-ref |
0.987 |
| gzip-graphic-ref |
1.961 |
| gzip-log-ref |
1.477 |
| gzip-program-ref |
1.344 |
| gzip-random-ref |
1.961 |
| gzip-source-ref |
1.351 |
| mcf-ref |
0.094 |
| parser-ref |
0.958 |
| perlbmk-diffmail-ref |
0.881 |
| perlbmk-makerand-ref |
0.967 |
| perlbmk-perfect-ref |
1.325 |
| perlbmk-splitmail-ref |
1.353 |
| twolf-ref |
0.636 |
| vortex-one-ref |
1.271 |
| vortex-three-ref |
1.266 |
| vortex-two-ref |
1.139 |
| vpr-route-ref |
0.687 |
| name |
IPC |
| ammp-ref |
0.938 |
| applu-ref |
0.831 |
| apsi-ref |
1.006 |
| art-110 |
0.464 |
| art-470 |
0.466 |
| equake-ref |
0.311 |
| facerec-ref |
1.445 |
| fma3d-ref |
0.537 |
| galgel-ref |
1.333 |
| lucas-ref |
0.775 |
| mesa-ref |
1.279 |
| mgrid-ref |
0.986 |
| sixtrack-ref |
2.463 |
| swim-ref |
0.694 |
| wupwise-ref |
1.821 |
100M Interval Size Multiple SimPoint Error Rates
These results show the average percent
error for the Standard and Early Multiple Simulation Points using the
new SimPoint 3.0 algorithm with an interval size of 100 million. The
Table shows the number of points, percent of the full program
execution that we need to fast forward through to reach the last
simulation point (Perc Exec), and the error in CPI between the CPI
estimated using SimPoint and the baseline CPI for the full execution of the
program as shown above.
|
name
|
error
|
numpoints
|
percent_exec
|
| ammp-ref |
1.31% |
8 |
85% |
| applu-ref |
0.49% |
9 |
83% |
| apsi-ref |
4.52% |
9 |
80% |
| art-110 |
0.46% |
6 |
69% |
| art-470 |
0.40% |
6 |
71% |
| bzip2-graphic |
1.15% |
9 |
99% |
| bzip2-program |
1.21% |
9 |
69% |
| bzip2-source |
0.23% |
8 |
62% |
| crafty-ref |
0.27% |
8 |
99% |
| eon-cook |
1.03% |
8 |
93% |
| eon-kajiya |
0.03% |
8 |
73% |
| eon-rushmeier |
0.56% |
7 |
92% |
| equake-ref |
0.09% |
5 |
85% |
| facerec-ref |
0.60% |
8 |
78% |
| fma3d-ref |
0.53% |
6 |
65% |
| galgel-ref |
2.90% |
9 |
85% |
| gap-ref |
2.75% |
7 |
96% |
| gcc-166 |
2.62% |
7 |
95% |
| gcc-200 |
5.19% |
9 |
86% |
| gcc-expr |
4.87% |
9 |
82% |
| gcc-integrate |
1.27% |
8 |
90% |
| gcc-scilab |
0.05% |
10 |
72% |
| gzip-graphic |
0.06% |
8 |
63% |
| gzip-log |
0.24% |
8 |
83% |
| gzip-program |
4.51% |
7 |
68% |
| gzip-random |
0.03% |
5 |
88% |
| gzip-source |
0.07% |
7 |
79% |
| lucas-ref |
2.60% |
9 |
89% |
| mcf-ref |
3.71% |
7 |
80% |
| mesa-ref |
0.10% |
6 |
99% |
| mgrid-ref |
0.11% |
8 |
98% |
| parser-ref |
4.36% |
7 |
26% |
| perlbmk-diffmail |
1.40% |
7 |
99% |
| perlbmk-makerand |
0.01% |
5 |
100% |
| perlbmk-perfect |
0.04% |
8 |
8% |
| perlbmk-splitmail |
1.98% |
4 |
84% |
| sixtrack-ref |
0.21% |
9 |
69% |
| swim-ref |
0.45% |
9 |
86% |
| twolf-ref |
0.62% |
8 |
39% |
| vortex-one |
1.46% |
9 |
99% |
| vortex-three |
0.57% |
9 |
98% |
| vortex-two |
0.09% |
9 |
89% |
| vpr-route |
5.47% |
6 |
97% |
| wupwise-ref |
2.38% |
7 |
77% |
| Max |
5.47% |
10 |
100% |
| Avg |
1.43% |
7 |
80% |
10M Interval Size Multiple SimPoint Error Rates
Error rates when an interval size of 10 million was used.
|
name
|
error
|
numpoints
|
percent_exec
|
| ammp-ref |
0.81% |
30 |
95% |
| applu-ref |
0.15% |
28 |
98% |
| apsi-ref |
1.19% |
30 |
83% |
| art-110 |
0.29% |
30 |
99% |
| art-470 |
0.79% |
30 |
96% |
| bzip2-graphic |
0.46% |
30 |
99% |
| bzip2-program |
1.44% |
30 |
91% |
| bzip2-source |
4.65% |
30 |
97% |
| crafty-ref |
0.00% |
30 |
99% |
| eon-cook |
0.72% |
30 |
95% |
| eon-kajiya |
0.13% |
30 |
99% |
| eon-rushmeier |
0.73% |
30 |
99% |
| equake-ref |
0.07% |
30 |
98% |
| facerec-ref |
0.08% |
30 |
98% |
| fma3d-ref |
0.64% |
30 |
98% |
| galgel-ref |
2.14% |
30 |
90% |
| gap-ref |
0.92% |
30 |
99% |
| gcc-166 |
5.25% |
30 |
100% |
| gcc-200 |
7.91% |
30 |
89% |
| gcc-expr |
0.95% |
30 |
97% |
| gcc-integrate |
1.08% |
30 |
99% |
| gcc-scilab |
1.11% |
30 |
86% |
| gzip-graphic |
0.17% |
30 |
98% |
| gzip-log |
0.04% |
30 |
91% |
| gzip-program |
3.52% |
30 |
94% |
| gzip-random |
0.04% |
30 |
94% |
| gzip-source |
0.11% |
30 |
96% |
| lucas-ref |
0.72% |
30 |
99% |
| mcf-ref |
5.19% |
30 |
99% |
| mesa-ref |
0.01% |
30 |
99% |
| mgrid-ref |
0.06% |
30 |
96% |
| parser-ref |
3.21% |
30 |
40% |
| perlbmk-diffmail |
0.96% |
30 |
99% |
| perlbmk-makerand |
0.03% |
28 |
100% |
| perlbmk-perfect |
0.04% |
30 |
9% |
| perlbmk-splitmail |
5.24% |
30 |
98% |
| sixtrack-ref |
0.09% |
30 |
95% |
| swim-ref |
0.34% |
24 |
84% |
| twolf-ref |
0.70% |
30 |
41% |
| vortex-one |
0.25% |
30 |
99% |
| vortex-three |
0.22% |
30 |
99% |
| vortex-two |
0.15% |
30 |
99% |
| vpr-route |
3.85% |
30 |
99% |
| wupwise-ref |
0.06% |
30 |
99% |
| Max |
7.91% |
30 |
100% |
| Avg |
1.28% |
29 |
92% |
100M Interval Size Single SimPoint Error Rates
The results below are the error rates for 100 Million Single Simulation
Points using the Early SimPoint Algorithm.
|
name
|
Original SimPoint
|
Early SimPoint
|
| ammp-ref |
5.06% |
6.34% |
| applu-ref |
3.1% |
0.44% |
| apsi-ref |
2.25% |
3.27% |
| art-110 |
0.05% |
0.06% |
| art-470 |
0% |
0.14% |
| equake-ref |
0.03% |
0.03% |
| facerec-ref |
2.66% |
5.51% |
| fma3d-ref |
2.56% |
8.82% |
| lucas-ref |
2.73% |
2.54% |
| mesa-ref |
0.02% |
1.51% |
| mgrid-ref |
0.18% |
0.24% |
| sixtrack-ref |
0.07% |
0.07% |
| swim-ref |
1.5% |
0.93% |
| wupwise-ref |
3.01% |
5.56% |
| bzip2-graphic-ref |
6.46% |
8.76% |
| bzip2-program-ref |
1.65% |
5.69% |
| crafty-ref |
1.54% |
10.33% |
| eon-rushmeier-ref |
2.04% |
7.46% |
| gap-ref |
0.28% |
0.28% |
| gcc-00-166-ref |
8.36% |
0.43% |
| gcc-00-integrate-ref |
0.48% |
3.9% |
| gzip-graphic-ref |
1.24% |
1.16% |
| gzip-program-ref |
4.99% |
7.72% |
| parser-ref |
0.16% |
8.84% |
| perlbmk-makerand-ref |
0% |
0.51% |
| vortex-two-ref |
3.26% |
5.23% |
| vpr-route-ref |
3.45% |
3.45% |
| Avg |
2.12% |
3.67% |
| Max |
8.36% |
10.33% |
| 
