diff options
author | jasplin <qt-info@nokia.com> | 2009-07-02 11:15:36 +0200 |
---|---|---|
committer | jasplin <qt-info@nokia.com> | 2009-07-02 11:15:36 +0200 |
commit | ca7c2e2aeb17b0d3e1ffacccaf8dcbebf08717e7 (patch) | |
tree | fcea447aa1cca0b0f1856b7743907e4cc4c6f5a6 | |
parent | 26e344db23b0dcf55ed9b38c1727e175dc4ed62b (diff) |
Improved documentation.
-rw-r--r-- | tools/bmcompare/main.cpp | 15 |
1 files changed, 8 insertions, 7 deletions
diff --git a/tools/bmcompare/main.cpp b/tools/bmcompare/main.cpp index 66c2a2d..d9bee83 100644 --- a/tools/bmcompare/main.cpp +++ b/tools/bmcompare/main.cpp @@ -25,12 +25,13 @@ The bmcompare tool provides quick comparison of benchmark results against a reference. - Benchmark results are specified as XML-files of the same format as those produced by QTestLib. + Benchmark results are specified as XML files of the same format as those produced by QTestLib. A typical use case is to get a quick overview of the effect of a certain optimization: - First, run the benchmarks (using QTestLib) on the unoptimized code and dump results to - XML files (there is typically one file per metric (wall time, callgrind events ...)). + XML files (there is typically one file per measurement back-end + (wall time, callgrind events ...)). These files constitute the 'reference' against which other results will be compared. - Second, optimize the code one or more times (e.g. you may want to compare multiple optimization alternatives at once) each time running through the benchmarks and producing @@ -43,7 +44,7 @@ while each -cmp option specifies that the following file names belong to the next set of comparable results. - The output is a table like this: + The output is dumped to stdout as a table like this: <function> <tag> <metric 1> <value> <value> ... @@ -63,16 +64,16 @@ _data function. If several identical function/tag combinations are detected, only the first one is used. - The rows in a group indicate results for each unique metric found for this function/tag - combination (if several identical metrics are found for a combination, only the first one - is used). + The rows in a group indicate results for each unique metric found for this + function/tag combination. If several identical metrics are found for the + combination, only the first one is used. Horizontally, there is (to the right of the metric name) one column per set of comparable results, i.e. one column per occurrence of '-cmp' in the command-line arguments. By default, the value is the absolute percentage of the reference (e.g. 95% means a 5% - decrease for the metric in question). By passing -diff on the command-line, the value is + decrease for the metric in question). By passing -diff on the command line, the value is presented as a percentage difference instead (e.g. -5% and 5% mean a 5% decrease and increase respectively). |