perf(bop): train on l1 prefetch request

[Maxpicca-Li]

(cherry picked from OpenXiangShan/CoupledL2@c61837e )

performance influence: also including l1 prefetcher train source changed.

comment	kmhv3-base	fix-base	rate1/0
400.perlbench	19.620	19.669	0.250%
401.bzip2	10.160	10.150	-0.098%
403.gcc	19.400	19.395	-0.026%
429.mcf	16.632	16.663	0.186%
445.gobmk	13.216	13.240	0.182%
456.hmmer	18.142	18.137	-0.028%
458.sjeng	13.114	13.089	-0.191%
462.libquantum	46.489	46.360	-0.277%
464.h264ref	22.780	22.784	0.018%
471.omnetpp	13.651	13.854	1.487%
473.astar	16.614	16.607	-0.042%
483.xalancbmk	23.969	23.865	-0.434%
SPECint2006/GHz	18.010	18.025	0.083%
410.bwaves	28.890	29.257	1.270%
416.gamess	19.035	19.010	-0.131%
433.milc	33.613	33.856	0.723%
434.zeusmp	23.912	24.016	0.435%
435.gromacs	12.640	12.643	0.024%
436.cactusADM	25.752	25.793	0.159%
437.leslie3d	20.955	21.112	0.749%
444.namd	14.653	14.674	0.143%
447.dealII	15.643	15.579	-0.409%
450.soplex	17.334	17.558	1.292%
453.povray	23.861	23.848	-0.054%
454.Calculix	18.721	18.831	0.588%
459.GemsFDTD	28.256	29.394	4.027%
465.tonto	17.151	17.116	-0.204%
470.lbm	42.926	43.371	1.037%
481.wrf	15.859	15.898	0.246%
482.sphinx3	19.680	19.674	-0.030%
SPECfp2006/GHz	21.180	21.302	0.576%
SPEC2006/GHz	19.806	19.879	0.369%