SPECINT 2006: Trunk generic -O2 - standard vs LTO vs LTO+PGO
-O2
AMD EPYC 8635P (zen5 based Sorano)
Run-times
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
86.20%
401.bzip2
100.00%
98.75%
403.gcc
100.00%
92.46%
429.mcf
100.00%
99.33%
445.gobmk
100.00%
101.70%
456.hmmer
100.00%
57.14%
458.sjeng
100.00%
101.91%
462.libquantum
100.00%
110.36%
464.h264ref
100.00%
80.37%
471.omnetpp
100.00%
106.94%
473.astar
100.00%
101.75%
483.xalancbmk
100.00%
108.47%
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 86.20% |
| 401.bzip2 | 100.00% | 98.75% |
| 403.gcc | 100.00% | 92.46% |
| 429.mcf | 100.00% | 99.33% |
| 445.gobmk | 100.00% | 101.70% |
| 456.hmmer | 100.00% | 57.14% |
| 458.sjeng | 100.00% | 101.91% |
| 462.libquantum | 100.00% | 110.36% |
| 464.h264ref | 100.00% | 80.37% |
| 471.omnetpp | 100.00% | 106.94% |
| 473.astar | 100.00% | 101.75% |
| 483.xalancbmk | 100.00% | 108.47% |
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.
Non-debug executable sizes
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
93.02%
401.bzip2
100.00%
109.64%
403.gcc
100.00%
93.08%
429.mcf
100.00%
100.79%
445.gobmk
100.00%
99.18%
456.hmmer
100.00%
87.40%
458.sjeng
100.00%
100.13%
462.libquantum
100.00%
85.16%
464.h264ref
100.00%
96.86%
471.omnetpp
100.00%
75.60%
473.astar
100.00%
92.30%
483.xalancbmk
100.00%
65.63%
Smaller binary sizes are always better.
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 93.02% |
| 401.bzip2 | 100.00% | 109.64% |
| 403.gcc | 100.00% | 93.08% |
| 429.mcf | 100.00% | 100.79% |
| 445.gobmk | 100.00% | 99.18% |
| 456.hmmer | 100.00% | 87.40% |
| 458.sjeng | 100.00% | 100.13% |
| 462.libquantum | 100.00% | 85.16% |
| 464.h264ref | 100.00% | 96.86% |
| 471.omnetpp | 100.00% | 75.60% |
| 473.astar | 100.00% | 92.30% |
| 483.xalancbmk | 100.00% | 65.63% |
AMD EPYC 75F3 (zen3 based high-frequency Milan)
Run-times
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
82.49%
401.bzip2
100.00%
96.84%
403.gcc
100.00%
89.14%
429.mcf
100.00%
99.34%
445.gobmk
100.00%
97.75%
456.hmmer
100.00%
55.69%
458.sjeng
100.00%
98.19%
462.libquantum
100.00%
97.94%
464.h264ref
100.00%
85.54%
471.omnetpp
100.00%
96.65%
473.astar
100.00%
97.73%
483.xalancbmk
100.00%
98.78%
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 82.49% |
| 401.bzip2 | 100.00% | 96.84% |
| 403.gcc | 100.00% | 89.14% |
| 429.mcf | 100.00% | 99.34% |
| 445.gobmk | 100.00% | 97.75% |
| 456.hmmer | 100.00% | 55.69% |
| 458.sjeng | 100.00% | 98.19% |
| 462.libquantum | 100.00% | 97.94% |
| 464.h264ref | 100.00% | 85.54% |
| 471.omnetpp | 100.00% | 96.65% |
| 473.astar | 100.00% | 97.73% |
| 483.xalancbmk | 100.00% | 98.78% |
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.
Non-debug executable sizes
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
93.02%
401.bzip2
100.00%
109.62%
403.gcc
100.00%
93.08%
429.mcf
100.00%
101.06%
445.gobmk
100.00%
99.18%
456.hmmer
100.00%
87.40%
458.sjeng
100.00%
100.13%
462.libquantum
100.00%
85.18%
464.h264ref
100.00%
96.86%
471.omnetpp
100.00%
75.61%
473.astar
100.00%
92.14%
483.xalancbmk
100.00%
65.63%
Smaller binary sizes are always better.
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 93.02% |
| 401.bzip2 | 100.00% | 109.62% |
| 403.gcc | 100.00% | 93.08% |
| 429.mcf | 100.00% | 101.06% |
| 445.gobmk | 100.00% | 99.18% |
| 456.hmmer | 100.00% | 87.40% |
| 458.sjeng | 100.00% | 100.13% |
| 462.libquantum | 100.00% | 85.18% |
| 464.h264ref | 100.00% | 96.86% |
| 471.omnetpp | 100.00% | 75.61% |
| 473.astar | 100.00% | 92.14% |
| 483.xalancbmk | 100.00% | 65.63% |