| Age | Commit message (Collapse) | Author | Files | Lines | |
|---|---|---|---|---|---|
| 2015-08-31 | core: update the header | Martin Peres | 1 | -3/+5 | |
| 2015-08-24 | core: use --format=fuller when generating the patch | Martin Peres | 1 | -1/+1 | |
| 2015-08-18 | core: Improve comparison with uninitialized earlier run data | Chris Wilson | 1 | -11/+6 | |
| Check against an undefined PrevTest result so that a failed read of the last run does not print out garbage to the console. | |||||
| 2015-08-15 | core: Add -B option (inverse of -b) to allow excluding arbitrary tests | Chris Wilson | 1 | -3/+16 | |
| As we have multiple different versions of a test, for example disabling rendering in order to purely measure the CPU overhead, we want an easy method to exclude those from a round of testing. | |||||
| 2015-08-14 | core: fix the feature allowing to pass a list of commits | Martin Peres | 1 | -11/+12 | |
| 2015-08-13 | core: Allow tests to invert the ratio | Chris Wilson | 1 | -1/+8 | |
| Not all tests are fps (or operations per second) and instead measure duration. For these, we need to invert the ratio between tests. | |||||
| 2015-08-13 | core: Make the tests folder an optional argument | Chris Wilson | 1 | -44/+49 | |
| By default keep using ezBenchDir/tests.d/, but allow the user to override. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> | |||||
| 2015-08-13 | core: Adjust geometric mean awk script for older versions | Chris Wilson | 1 | -1/+1 | |
| Older versions of gawk do not have the '**' operator so to perform the equivalent pow() we need to use the logarithm. Note that we pass in a single line and need to iterate over fields in that line as opposed to summing over multiple lines. | |||||
| 2015-08-13 | core: save the commits as we go | Martin Peres | 1 | -0/+2 | |
| 2015-07-30 | core: add a new variable containing the benchmark's name | Martin Peres | 1 | -0/+2 | |
| 2015-07-30 | core: add global pre/post hooks for all the benchmarks | Martin Peres | 1 | -0/+2 | |
| 2015-07-28 | Provide a conf.d for splitting up the callback definitions | Chris Wilson | 1 | -1/+4 | |
| 2015-07-28 | Use LC_LOCALE=C for numeric interchange | Chris Wilson | 1 | -1/+1 | |
| Otherwise those of us in non-US regions get warning messages about unknown locale! | |||||
| 2015-07-28 | core: fix the geometric mean computation | Martin Peres | 1 | -1/+1 | |
| 2015-07-28 | core: force the usage of the us locale for numeric operations | Martin Peres | 1 | -0/+3 | |
| 2015-07-27 | Run the ezbench_pre_hook before querying the tests | Chris Wilson | 1 | -3/+3 | |
| This way I can put my machine setup inside the pre-hook for the run. | |||||
| 2015-07-27 | Detect test run failure and highlight it | Chris Wilson | 1 | -1/+8 | |
| If the run fails to return any values or reported an error, keep the error logs and highlight the failure in the console. | |||||
| 2015-07-27 | Seed the output results with the final commit from the previous round | Chris Wilson | 1 | -3/+14 | |
| When we split the testing over multiple runs, it is nice for the pretty console output if we lookup the final result and use that for our diff. | |||||
| 2015-07-27 | Only source the .test files once | Chris Wilson | 1 | -45/+43 | |
| As the .test scripts grow more complex, we only want to do one pass through them. | |||||
| 2015-07-27 | Allow .test to define multiple tests at once | Chris Wilson | 1 | -13/+17 | |
| 2015-07-27 | Allow the user to provide an optional run name to concatenate results | Chris Wilson | 1 | -4/+7 | |
| v2: ()Martin Peres) - add the documentation in show-help | |||||
| 2015-07-27 | Allow multiple -b options | Chris Wilson | 1 | -1/+1 | |
| Append the next -b selection onto the list of benchmarks. | |||||
| 2015-07-27 | Provide a conf.d for splitting up the callback definitions | Chris Wilson | 1 | -1/+2 | |
| 2015-07-27 | Allow arbitrary git commit ids and ranges to be specified on the commandline | Chris Wilson | 1 | -5/+18 | |
| 2015-07-27 | Don't add a 0 result to the list for geometric mean processing | Chris Wilson | 1 | -1/+1 | |
| 2015-07-27 | Remove extra unguarded compile_pre_hook | Chris Wilson | 1 | -2/+0 | |
| Just a typo, nothing to see here. | |||||
| 2015-07-23 | core: use the long version for the year for the log dir (2015 instead of 15) | Martin Peres | 1 | -1/+1 | |
| 2015-07-22 | Allow individual tests to provide a hook to produce a global statistic | Chris Wilson | 1 | -10/+16 | |
| For the slightly more complex tests that do not just measure a constant workload and so need something different than a mean average, let them provide a hook to convert their output into a single statistic useful for tracking differences between commits. The full output is still available for detailed analysis. | |||||
| 2015-07-22 | Allow the user to disable compilation | Chris Wilson | 1 | -17/+33 | |
| v2: (Martin Peres) - reuse the -m option instead - document the behaviour - make sure the compilation time is not included in the estimation This is a small step towards making the benchmarking two phase in order to allow for say benchmarking the kernel over many commits (where we need to reboot between commits). | |||||
| 2015-07-22 | Save the per-run FPS values where we amalgamate | Chris Wilson | 1 | -3/+4 | |
| v2: (Martin Peres) - make benchmarks only output one fps reading per round - fix gen_report to ignore the per-run values | |||||
| 2015-07-22 | Use globstar to enable subdirectories in tests.d | Chris Wilson | 1 | -2/+7 | |
| 2015-07-21 | Be flexible in choosing where to benchmark upto | Chris Wilson | 1 | -4/+9 | |
| Allow the user to pass in the target commit. This is just a first step to allow the user to parse in arbitrary git ranges. | |||||
| 2015-07-21 | Make the FPS difference computations robust against divide-by-zero | Chris Wilson | 1 | -8/+8 | |
| 2015-07-21 | Reset the FPS summation between runs | Chris Wilson | 1 | -0/+1 | |
| If we don't remember to reset the FPS counter, the geometric mean gradually averages the result of the entire test run - rendering its comparison very confusing. | |||||
| 2015-07-20 | core: dump the commit list as we go to enable offline analysis | Martin Peres | 1 | -0/+5 | |
| 2015-07-20 | core: switch to the finish date instead of just the execution time | Martin Peres | 1 | -1/+2 | |
| 2015-07-20 | core: general style fixes | Martin Peres | 1 | -11/+12 | |
| 2015-07-20 | Finish each commit by showing the geometric mean | Chris Wilson | 1 | -0/+25 | |
| The geometric mean helps to summarise all of the individual tests into a single traffic. | |||||
| 2015-07-20 | Abort if we fail to stash the current dirty state | Chris Wilson | 1 | -0/+5 | |
| We never want to lose work-in-progress, so abort if we cannot stash it away. | |||||
| 2015-07-20 | Add some color to the fps diff output | Chris Wilson | 1 | -12/+30 | |
| I could say that the color provide a quick guide as to how the runs are fairing, but really it just makes the output pretty. Just as important is to just the text alignment so it becomes easier to scan through the numbers. | |||||
| 2015-07-20 | Place the repo check before setting up the termination handler | Chris Wilson | 1 | -12/+12 | |
| If we don't have a git repository we exit early and have nothing to restore. | |||||
| 2015-07-20 | Replace git log parsing with git plumbing | Chris Wilson | 1 | -7/+4 | |
| Instead of piping 'git log' output through a text pipeline, just use the git plumbing commands directly to generate the commit lists. | |||||
| 2015-07-20 | Try benchmarking any work-in-progress | Chris Wilson | 1 | -4/+13 | |
| 2015-07-20 | Preserve local modifications | Chris Wilson | 1 | -0/+5 | |
| Next challenge will be to benchmark them as well | |||||
| 2015-07-20 | Reverse computation of perf changes | Chris Wilson | 1 | -1/+1 | |
| An increase from 6000 to 15000 fps should be reported as an 150% improvement not, a mere, 60%. | |||||
| 2015-07-20 | core: redirect the tests' output to an error file | Martin Peres | 1 | -4/+10 | |
| 2015-07-20 | Use 'git show --format' to generate oneline commit identifier | Chris Wilson | 1 | -2/+1 | |
| Using the format for git show we can do the name generation in a single step, throw in a bit of color and avoid including an extra character into the commit id when copy-pasting. | |||||
| 2015-07-17 | core: save the logs in different folders | Martin Peres | 1 | -5/+5 | |
| 2015-07-16 | core: let the tests make sure that they do not print anything in stderr | Martin Peres | 1 | -1/+1 | |
| This makes it easier to debug what the heck is going wrong when it does :) | |||||
| 2015-07-16 | core: add support for hooks | Martin Peres | 1 | -1/+24 | |
 |
index : ~ickle/ezbench | |
| Unnamed repository; edit this file 'description' to name the repository. | ickle |
| summaryrefslogtreecommitdiff |
path: root/ezbench.sh