blob: f08e2828022a08226c8dbe6e029413fe2a140b03 (plain
Naming Convention of i-g-t Tests and Subtests
To facilitate easy test selection with piglit we need a somewhat consistent
naming scheme for tests and subtests.
core_: Test for core drm ioctls and behaviour.
kms_: Used for modesetting tests.
drm_: Tests for libdrm behaviour, currently just testing the buffer cache
gem_: Used for all kinds of GEM tests.
prime_: Used for buffer sharing tests, both for self-importing (used by
dri3/wayland) and actual multi-gpu tests.
drv_: Tests for overall driver behaviour like module reload, s/r, debugfs files.
pm_: Tests for power management features like runtime PM, tuning knobs in sysfs
and also performance tuning features.
gen3_: Used by Chris' gen3 specific tiling/fencing tests. Generally tests that
only run on some platforms don't have a specific prefix but just skip on
platforms where the test doesn't apply.
debugfs_/sysfs_: Mostly for tests that use sysfs/debugfs but tend to tests all
sorts of things. Please consider using a more appropriate prefix from above if
the main point isn't to test sysfs/debugfs, but a driver subsystem/feature.
igt_: Testcase which test the i-g-t infrastructure itself and which are all run
through "make check" while building i-g-t.
Much more powerful for filtering sets of tests are patterns anywhere in either
the test or subtest name.
hang: Tests that provoke gpu hangs
swap: Tests that force their full working sets through swap. Dreadfully slow on
machines with spinning rust and tons of memory.
thrash: Tests that tend to have really slow forward progress due to
gtt/memory/.. thrashing. Mostly used to stress-test error-handling corner-cases.
crc: Tests that use the display CRC infrastructure to check the results.
tiled/tiling: Tests that exercise behaviour on tiled buffers.
normal/uncached/snooped: Usual 3 variants for tests that use different coherency
modes for the buffer objects they're using.
rte: _R_un_t_ime _e_nviroment checks. For testcases which will fail if the
machine isn't configured properly there should be a first subtest to just check
ctx: Tests that exercise the hw context support.