SPECINT 2006: Trunk native -Ofast - standard vs LTO vs LTO+PGO
-Ofast -march=native -mtune=native
AMD EPYC 8635P (zen5 based Sorano)
Run-times
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
81.19%
401.bzip2
100.00%
97.74%
403.gcc
100.00%
90.85%
429.mcf
100.00%
102.47%
445.gobmk
100.00%
103.40%
456.hmmer
100.00%
103.55%
458.sjeng
100.00%
103.38%
462.libquantum
100.00%
108.88%
464.h264ref
100.00%
90.11%
471.omnetpp
100.00%
90.32%
473.astar
100.00%
FE
483.xalancbmk
100.00%
119.21%
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 81.19% |
| 401.bzip2 | 100.00% | 97.74% |
| 403.gcc | 100.00% | 90.85% |
| 429.mcf | 100.00% | 102.47% |
| 445.gobmk | 100.00% | 103.40% |
| 456.hmmer | 100.00% | 103.55% |
| 458.sjeng | 100.00% | 103.38% |
| 462.libquantum | 100.00% | 108.88% |
| 464.h264ref | 100.00% | 90.11% |
| 471.omnetpp | 100.00% | 90.32% |
| 473.astar | 100.00% | FE |
| 483.xalancbmk | 100.00% | 119.21% |
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%
75.68%
401.bzip2
100.00%
73.21%
403.gcc
100.00%
85.62%
429.mcf
100.00%
106.24%
445.gobmk
100.00%
93.69%
456.hmmer
100.00%
70.01%
458.sjeng
100.00%
97.90%
462.libquantum
100.00%
78.15%
464.h264ref
100.00%
67.77%
471.omnetpp
100.00%
74.78%
473.astar
100.00%
FE
483.xalancbmk
100.00%
62.50%
Smaller binary sizes are always better.
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 75.68% |
| 401.bzip2 | 100.00% | 73.21% |
| 403.gcc | 100.00% | 85.62% |
| 429.mcf | 100.00% | 106.24% |
| 445.gobmk | 100.00% | 93.69% |
| 456.hmmer | 100.00% | 70.01% |
| 458.sjeng | 100.00% | 97.90% |
| 462.libquantum | 100.00% | 78.15% |
| 464.h264ref | 100.00% | 67.77% |
| 471.omnetpp | 100.00% | 74.78% |
| 473.astar | 100.00% | FE |
| 483.xalancbmk | 100.00% | 62.50% |
AMD EPYC 75F3 (zen3 based high-frequency Milan)
Run-times
Benchmark
LTO
LTO+PGO
400.perlbench
100.00%
84.43%
401.bzip2
100.00%
96.66%
403.gcc
100.00%
83.64%
429.mcf
100.00%
100.87%
445.gobmk
100.00%
97.75%
456.hmmer
100.00%
102.89%
458.sjeng
100.00%
100.82%
462.libquantum
100.00%
96.83%
464.h264ref
100.00%
91.01%
471.omnetpp
100.00%
94.95%
473.astar
100.00%
FE
483.xalancbmk
100.00%
123.08%
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 84.43% |
| 401.bzip2 | 100.00% | 96.66% |
| 403.gcc | 100.00% | 83.64% |
| 429.mcf | 100.00% | 100.87% |
| 445.gobmk | 100.00% | 97.75% |
| 456.hmmer | 100.00% | 102.89% |
| 458.sjeng | 100.00% | 100.82% |
| 462.libquantum | 100.00% | 96.83% |
| 464.h264ref | 100.00% | 91.01% |
| 471.omnetpp | 100.00% | 94.95% |
| 473.astar | 100.00% | FE |
| 483.xalancbmk | 100.00% | 123.08% |
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%
75.78%
401.bzip2
100.00%
70.73%
403.gcc
100.00%
85.91%
429.mcf
100.00%
107.64%
445.gobmk
100.00%
93.81%
456.hmmer
100.00%
71.84%
458.sjeng
100.00%
98.04%
462.libquantum
100.00%
74.15%
464.h264ref
100.00%
66.35%
471.omnetpp
100.00%
74.87%
473.astar
100.00%
FE
483.xalancbmk
100.00%
61.32%
Smaller binary sizes are always better.
| Benchmark | LTO | LTO+PGO |
|---|---|---|
| 400.perlbench | 100.00% | 75.78% |
| 401.bzip2 | 100.00% | 70.73% |
| 403.gcc | 100.00% | 85.91% |
| 429.mcf | 100.00% | 107.64% |
| 445.gobmk | 100.00% | 93.81% |
| 456.hmmer | 100.00% | 71.84% |
| 458.sjeng | 100.00% | 98.04% |
| 462.libquantum | 100.00% | 74.15% |
| 464.h264ref | 100.00% | 66.35% |
| 471.omnetpp | 100.00% | 74.87% |
| 473.astar | 100.00% | FE |
| 483.xalancbmk | 100.00% | 61.32% |