Script vg_perf produces too much output when running with several Valgrind tools and comparing performance for several Valgrind directories. Consider for example: -- Running tests in perf ---------------------------------------------- bigcode1 valgrind-reg_alloc3:0.09s no: 1.6s (18.0x, -----) me: 3.1s (35.0x, -----) ca:12.7s (140.8x, -----) he: 2.0s (22.8x, -----) ca: 3.7s (41.2x, -----) dr: 2.0s (22.2x, -----) ma: 1.9s (20.7x, -----) bigcode1 valgrind-linux:0.09s no: 1.6s (17.8x, 1.2%) me: 3.0s (33.3x, 4.8%) ca:12.7s (140.7x, 0.1%) he: 2.0s (22.2x, 2.4%) ca: 3.6s (40.2x, 2.4%) dr: 1.9s (21.7x, 2.5%) ma: 1.9s (20.7x, 0.0%) and you are lost. New command line option "--terse" produces, well, terse output: -- Running tests in vanilla/perf -------------------------------------- -- bigcode1 -- vanilla : no: ----- me: ----- ca: ----- he: ----- ca: ----- dr: ----- ma: ----- ra3-005 : no: 0.0% me: -4.7% ca: 0.2% he: -1.1% ca: -1.5% dr: -0.6% ma: 0.0% ra3-005-J : no: 0.0% me: 0.4% ca: 0.3% he: -0.5% ca: -0.9% dr: 0.0% ma: 0.6% ra3-006 : no: 0.0% me: 0.4% ca: 0.2% he: -0.0% ca: -1.2% dr: 0.0% ma: 0.0% ra3-007 : no: 0.0% me: 0.7% ca: 0.5% he: -0.5% ca: -0.6% dr: 0.0% ma: 1.2%
Created attachment 106668 [details] proposed patch
Committed as SVN r16461.