1 Correctness Testing 2 =================== 3 4 Skia correctness testing is primarily served by a tool named DM. 5 This is a quickstart to building and running DM. 6 7 ~~~ 8 $ python bin/sync-and-gyp 9 $ ninja -C out/Debug dm 10 $ out/Debug/dm -v -w dm_output 11 ~~~ 12 13 When you run this, you may notice your CPU peg to 100% for a while, then taper 14 off to 1 or 2 active cores as the run finishes. This is intentional. DM is 15 very multithreaded, but some of the work, particularly GPU-backed work, is 16 still forced to run on a single thread. You can use `--threads N` to limit DM to 17 N threads if you like. This can sometimes be helpful on machines that have 18 relatively more CPU available than RAM. 19 20 As DM runs, you ought to see a giant spew of output that looks something like this. 21 ~~~ 22 Skipping nonrendering: Don't understand 'nonrendering'. 23 Skipping angle: Don't understand 'angle'. 24 Skipping nvprmsaa4: Could not create a surface. 25 492 srcs * 3 sinks + 382 tests == 1858 tasks 26 27 ( 25MB 1857) 1.36ms 8888 image mandrill_132x132_12x12.astc-5-subsets 28 ( 25MB 1856) 1.41ms 8888 image mandrill_132x132_6x6.astc-5-subsets 29 ( 25MB 1855) 1.35ms 8888 image mandrill_132x130_6x5.astc-5-subsets 30 ( 25MB 1854) 1.41ms 8888 image mandrill_132x130_12x10.astc-5-subsets 31 ( 25MB 1853) 151s 8888 image mandrill_130x132_10x6.astc-5-subsets 32 ( 25MB 1852) 154s 8888 image mandrill_130x130_5x5.astc-5-subsets 33 ... 34 ( 748MB 5) 9.43ms unit test GLInterfaceValidation 35 ( 748MB 4) 30.3ms unit test HalfFloatTextureTest 36 ( 748MB 3) 31.2ms unit test FloatingPointTextureTest 37 ( 748MB 2) 32.9ms unit test DeferredCanvas_GPU 38 ( 748MB 1) 49.4ms unit test ClipCache 39 ( 748MB 0) 37.2ms unit test Blur 40 ~~~ 41 Do not panic. 42 43 As you become more familiar with DM, this spew may be a bit annoying. If you 44 remove -v from the command line, DM will spin its progress on a single line 45 rather than print a new line for each status update. 46 47 Don't worry about the "Skipping something: Here's why." lines at startup. DM 48 supports many test configurations, which are not all appropriate for all 49 machines. These lines are a sort of FYI, mostly in case DM can't run some 50 configuration you might be expecting it to run. 51 52 The next line is an overview of the work DM is about to do. 53 ~~~ 54 492 srcs * 3 sinks + 382 tests == 1858 tasks 55 ~~~ 56 57 DM has found 382 unit tests (code linked in from tests/), and 492 other drawing 58 sources. These drawing sources may be GM integration tests (code linked in 59 from gm/), image files (from `--images`, which defaults to "resources") or .skp 60 files (from `--skps`, which defaults to "skps"). You can control the types of 61 sources DM will use with `--src` (default, "tests gm image skp"). 62 63 DM has found 3 usable ways to draw those 492 sources. This is controlled by 64 `--config`, which today defaults to "565 8888 gpu nonrendering angle nvprmsaa4". 65 DM has skipped nonrendering, angle, and nvprmssa4, leaving three usable configs: 66 565, 8888, and gpu. These three name different ways to draw using Skia: 67 68 - 565: draw using the software backend into a 16-bit RGB bitmap 69 - 8888: draw using the software backend into a 32-bit RGBA bitmap 70 - gpu: draw using the GPU backend (Ganesh) into a 32-bit RGBA bitmap 71 72 Sometimes DM calls these configs, sometimes sinks. Sorry. There are many 73 possible configs but generally we pay most attention to 8888 and gpu. 74 75 DM always tries to draw all sources into all sinks, which is why we multiply 76 492 by 3. The unit tests don't really fit into this source-sink model, so they 77 stand alone. A couple thousand tasks is pretty normal. Let's look at the 78 status line for one of those tasks. 79 ~~~ 80 ( 25MB 1857) 1.36ms 8888 image mandrill_132x132_12x12.astc-5-subsets 81 ~~~ 82 83 This status line tells us several things. 84 85 First, it tells us that at the time we wrote the status line, the maximum 86 amount of memory DM had ever used was 25MB. Note this is a high water mark, 87 not the current memory usage. This is mostly useful for us to track on our 88 buildbots, some of which run perilously close to the system memory limit. 89 90 Next, the status line tells us that there are 1857 unfinished tasks, either 91 currently running or waiting to run. We generally run one task per hardware 92 thread available, so on a typical laptop there are probably 4 or 8 running at 93 once. Sometimes the counts appear to show up out of order, particularly at DM 94 startup; it's harmless, and doesn't affect the correctness of the run. 95 96 Next, we see this task took 1.36 milliseconds to run. Generally, the precision 97 of this timer is around 1 microsecond. The time is purely there for 98 informational purposes, to make it easier for us to find slow tests. 99 100 Finally we see the configuration and name of the test we ran. We drew the test 101 "mandrill_132x132_12x12.astc-5-subsets", which is an "image" source, into an 102 "8888" sink. 103 104 When DM finishes running, you should find a directory with file named dm.json, 105 and some nested directories filled with lots of images. 106 ~~~ 107 $ ls dm_output 108 565 8888 dm.json gpu 109 110 $ find dm_output -name '*.png' 111 dm_output/565/gm/3x3bitmaprect.png 112 dm_output/565/gm/aaclip.png 113 dm_output/565/gm/aarectmodes.png 114 dm_output/565/gm/alphagradients.png 115 dm_output/565/gm/arcofzorro.png 116 dm_output/565/gm/arithmode.png 117 dm_output/565/gm/astcbitmap.png 118 dm_output/565/gm/bezier_conic_effects.png 119 dm_output/565/gm/bezier_cubic_effects.png 120 dm_output/565/gm/bezier_quad_effects.png 121 ... 122 ~~~ 123 124 The directories are nested first by sink type (`--config`), then by source type (`--src`). 125 The image from the task we just looked at, "8888 image mandrill_132x132_12x12.astc-5-subsets", 126 can be found at dm_output/8888/image/mandrill_132x132_12x12.astc-5-subsets.png. 127 128 dm.json is used by our automated testing system, so you can ignore it if you 129 like. It contains a listing of each test run and a checksum of the image 130 generated for that run. 131 132 ### Detail <a name="digests"></a> 133 Boring technical detail: The checksum is not a checksum of the 134 .png file, but rather a checksum of the raw pixels used to create that .png. 135 That means it is possible for two different configurations to produce 136 the same exact .png, but have their checksums differ. 137 138 Unit tests don't generally output anything but a status update when they pass. 139 If a test fails, DM will print out its assertion failures, both at the time 140 they happen and then again all together after everything is done running. 141 These failures are also included in the dm.json file. 142 143 DM has a simple facility to compare against the results of a previous run: 144 ~~~ 145 $ python bin/sync-and-gyp 146 $ ninja -C out/Debug dm 147 $ out/Debug/dm -w good 148 149 # do some work 150 151 $ python bin/sync-and-gyp 152 $ ninja -C out/Debug dm 153 $ out/Debug/dm -r good -w bad 154 ~~~ 155 When using `-r`, DM will display a failure for any test that didn't produce the 156 same image as the `good` run. 157 158 For anything fancier, I suggest using skdiff: 159 ~~~ 160 $ python bin/sync-and-gyp 161 $ ninja -C out/Debug dm 162 $ out/Debug/dm -w good 163 164 # do some work 165 166 $ python bin/sync-and-gyp 167 $ ninja -C out/Debug dm 168 $ out/Debug/dm -w bad 169 170 $ ninja -C out/Debug skdiff 171 $ mkdir diff 172 $ out/Debug/skdiff good bad diff 173 174 # open diff/index.html in your web browser 175 ~~~ 176 177 That's the basics of DM. DM supports many other modes and flags. Here are a 178 few examples you might find handy. 179 ~~~ 180 $ out/Debug/dm --help # Print all flags, their defaults, and a brief explanation of each. 181 $ out/Debug/dm --src tests # Run only unit tests. 182 $ out/Debug/dm --nocpu # Test only GPU-backed work. 183 $ out/Debug/dm --nogpu # Test only CPU-backed work. 184 $ out/Debug/dm --match blur # Run only work with "blur" in its name. 185 $ out/Debug/dm --dryRun # Don't really do anything, just print out what we'd do. 186 ~~~ 187
