use_lbr was enabled but range_count_map was empty!

[michaelrun]

I would like to see if there is any performance improvement on my workload using autoFDO, here are steps:

1. download and compile my workload
2. run and collect perf data: ../pmu-tools/ocperf.py record -C 0-13,112-125 -b -e br_inst_retired.near_taken:pp ( my workload run on those cores)
3. create gov file: /autofdo/build/create_gcov --binary=/myworkload --profile=./perf.data --gcov=./autofdo.gcov -gcov_version=1
4. recompile with profile: flags="-fauto-profile=/{HOME}/shell/autofdo.gcov -fuse-ld=gold"

in step 3, although it is successful, but there are many warnings like:

[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 5
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 4
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 2
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 2
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 5
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 5
[WARNING:/autofdo/third_party/perf_data_converter/src/quipper/perf_serializer.cc:604] Ignoring branch stack entry reserved bits: 4
[INFO:/autofdo/third_party/perf_data_converter/src/quipper/perf_reader.cc:1058] Number of events stored: 354553
[INFO:/autofdo/third_party/perf_data_converter/src/quipper/perf_parser.cc:274] Parser processed: 367 MMAP/MMAP2 events, 346 COMM events, 344 FORK events, 30 EXIT events, 351931 SAMPLE events, 351931 of these were mapped, 0 SAMPLE events with a data address, 0 of these were mapped
WARNING: Logging before InitGoogleLogging() is written to STDERR
W20240524 03:25:37.918485 104570 addr2line_inlinestack.cc:679] 2089 overlapping ranges
W20240524 03:25:38.148685 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.150476 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.166795 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.188889 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.317080 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.326786 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.346027 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.348367 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.383397 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.391921 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.504626 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.505682 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!
W20240524 03:25:38.506157 104570 profile.cc:102] use_lbr was enabled but range_count_map was empty!

also there are very limited performance improvement, like about 3-5%. I think it is possibly has something to do with the warning: use_lbr was enabled but range_count_map was empty! Am I correct? If it is, how to fix this issue?

[erozenfeld]

@michaelrun What version of GCC are you using?

[michaelrun]

I tried both gcc version 11.2.1 20210728 (Red Hat 11.2.1-1) (GCC) and gcc version 11.4.1 20231218 (Red Hat 11.4.1-3) (GCC)