https://bugs.kde.org/show_bug.cgi?id=384987
--- Comment #6 from Ivo Raisr <iv...@ivosh.net> --- I've tested on amd64, ppc8le and arm64 architectures. In all cases, the produced code was more compact and overall performance better, when running inner Memcheck on perf/bz2. Numbers are given as instruction count; ratio as reported by Memcheck with '--stats=yes'. amd64: vanilla: 45,112,349,784 total; 165,978,807 reg alloc; ratio 15.5 v3-reoder: 44,943,765,809 total; 167,403,237 reg alloc; ratio 15.3 power8le: vanilla: 61,928,020,284 total; 351,285,156 reg alloc; ratio 17.0 v3-reorder: 61,919,130,481 total; 343,001,581 reg alloc; ratio 17.0 arm64 [callgrind does not work on this arch]: vanilla: ratio 14.7 v3-reorder: ratio 14.7 Manual inspection of top 200 SB profiled blocks showed VexExpansionRatio always few instructions better than in vanilla. -- You are receiving this mail because: You are watching all bug changes.