[llvm] [SLP] Improve block traversal in getSpillCost() (PR #128620)

Mikhail R. Gadelha via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 24 19:02:11 PST 2025


mikhailramalho wrote:

Current compile regression of SPEC:
```
Program                                       compile_time              
                                              lhs          rhs     diff 
INT2017rate/502.gcc_r/502.gcc_r               1213.14      1296.42  6.9%
INT2017speed/602.gcc_s/602.gcc_s              1226.93      1296.10  5.6%
FP2017spee...96.specrand_fs/996.specrand_fs      1.61         1.69  5.2%
FP2017rate/508.namd_r/508.namd_r                83.55        87.05  4.2%
FP2017rate/538.imagick_r/538.imagick_r         240.84       244.25  1.4%
INT2017rat...00.perlbench_r/500.perlbench_r    215.02       217.01  0.9%
FP2017speed/638.imagick_s/638.imagick_s        226.80       228.73  0.9%
FP2017rate/526.blender_r/526.blender_r        1910.20      1923.96  0.7%
INT2017spe...00.perlbench_s/600.perlbench_s    217.00       218.35  0.6%
INT2017spe...31.deepsjeng_s/631.deepsjeng_s     26.76        26.89  0.5%
INT2017speed/657.xz_s/657.xz_s                  87.78        88.22  0.5%
FP2017rate/510.parest_r/510.parest_r          1838.97      1847.39  0.5%
FP2017rate/511.povray_r/511.povray_r           200.60       201.43  0.4%
INT2017spe...23.xalancbmk_s/623.xalancbmk_s   1536.31      1542.30  0.4%
INT2017rate/525.x264_r/525.x264_r               81.10        81.36  0.3%
INT2017speed/625.x264_s/625.x264_s              81.27        81.51  0.3%
FP2017rate/544.nab_r/544.nab_r                  30.93        31.00  0.2%
INT2017rate/505.mcf_r/505.mcf_r                 12.43        12.46  0.2%
INT2017speed/641.leela_s/641.leela_s            68.88        69.01  0.2%
INT2017speed/605.mcf_s/605.mcf_s                12.48        12.49  0.1%
INT2017rat...99.specrand_ir/999.specrand_ir      1.78         1.78  0.0%
INT2017rat...23.xalancbmk_r/523.xalancbmk_r   1539.76      1539.29 -0.0%
INT2017rate/520.omnetpp_r/520.omnetpp_r        631.83       631.48 -0.1%
INT2017spe...ed/620.omnetpp_s/620.omnetpp_s    630.56       629.81 -0.1%
INT2017rate/557.xz_r/557.xz_r                   88.09        87.89 -0.2%
INT2017rat...31.deepsjeng_r/531.deepsjeng_r     26.95        26.80 -0.6%
FP2017rate...97.specrand_fr/997.specrand_fr      1.78         1.77 -0.6%
INT2017rate/541.leela_r/541.leela_r             69.31        68.84 -0.7%
FP2017speed/644.nab_s/644.nab_s                 29.05        28.74 -1.1%
FP2017speed/619.lbm_s/619.lbm_s                  2.78         2.75 -1.3%
INT2017spe...98.specrand_is/998.specrand_is      1.80         1.78 -1.4%
FP2017rate/519.lbm_r/519.lbm_r                   2.85         2.81 -1.4%
                           Geomean difference                       0.7%


https://github.com/llvm/llvm-project/pull/128620


More information about the llvm-commits mailing list