[llvm] ConstraintElim: add dry-run routine to fail early (PR #99670)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

Tweaking the patch was pretty straight-forward actually. There are no binary changes on SPEC2017 after tweaking the factor to 1.6. Compile time improvement is 0.6%. As I mentioned, there is no way to absolutely guarantee that there won't be binary changes in a theoretical degenerate case, but I think the trade-off I've chosen is practical, and will allow us to move forward.

```
Program                                       compile_time
                                              baseline     patch    diff
External/S...96.specrand_fs/996.specrand_fs       1.62         1.72   6.0%
External/S...99.specrand_ir/999.specrand_ir       1.67         1.76   5.6%
External/S...P2017speed/644.nab_s/644.nab_s     102.36       108.04   5.5%
External/S...NT2017rate/505.mcf_r/505.mcf_r      22.42        23.25   3.7%
External/S...T2017speed/605.mcf_s/605.mcf_s      22.29        22.98   3.1%
External/S...017speed/625.x264_s/625.x264_s     365.08       373.04   2.2%
External/S...te/520.omnetpp_r/520.omnetpp_r    3009.42      3027.51   0.6%
External/S...17rate/541.leela_r/541.leela_r     355.39       357.28   0.5%
External/S...te/526.blender_r/526.blender_r    7542.34      7568.07   0.3%
External/S...rate/510.parest_r/510.parest_r   10715.75     10749.72   0.3%
External/S...rate/511.povray_r/511.povray_r     802.93       805.45   0.3%
External/S...ed/638.imagick_s/638.imagick_s    1035.18      1038.16   0.3%
External/S...ed/620.omnetpp_s/620.omnetpp_s    3026.72      3032.78   0.2%
External/S...23.xalancbmk_r/523.xalancbmk_r    6407.85      6414.59   0.1%
Bitcode/Be...hmarks/Halide/blur/halide_blur       1.43         1.43   0.0%
External/S...T2017speed/602.gcc_s/602.gcc_s    6570.48      6569.85  -0.0%
External/S...23.xalancbmk_s/623.xalancbmk_s    6397.61      6396.53  -0.0%
External/S...INT2017speed/657.xz_s/657.xz_s     129.01       128.97  -0.0%
External/S...2017rate/525.x264_r/525.x264_r     369.01       368.63  -0.1%
External/S...00.perlbench_r/500.perlbench_r    1074.48      1071.83  -0.2%
External/S...NT2017rate/502.gcc_r/502.gcc_r    6543.43      6523.43  -0.3%
External/S...te/538.imagick_r/538.imagick_r    1136.65      1132.24  -0.4%
External/S...2017rate/508.namd_r/508.namd_r     539.89       537.10  -0.5%
External/S...7speed/641.leela_s/641.leela_s     353.56       350.23  -0.9%
External/S...98.specrand_is/998.specrand_is       1.77         1.75  -1.0%
External/S...CINT2017rate/557.xz_r/557.xz_r     131.92       129.86  -1.6%
External/S...00.perlbench_s/600.perlbench_s    1194.66      1174.82  -1.7%
External/S...FP2017rate/544.nab_r/544.nab_r     113.08       111.06  -1.8%
External/S...31.deepsjeng_s/631.deepsjeng_s      76.04        74.23  -2.4%
External/S...31.deepsjeng_r/531.deepsjeng_r      76.10        71.79  -5.7%
External/S...FP2017rate/519.lbm_r/519.lbm_r       8.87         8.23  -7.2%
External/S...97.specrand_fr/997.specrand_fr       1.65         1.53  -7.3%
External/S...P2017speed/619.lbm_s/619.lbm_s       9.61         8.05 -16.3%
                           Geomean difference                        -0.6%
```

https://github.com/llvm/llvm-project/pull/99670