ed58e004e0
The client can do a test run of their application with a persistent cache set to SkSL mode. They store the key and data blobs that are produced. Ship those blobs with the application. At startup, call GrContext::precompileShader for each key/data pair. This compiles the shaders, and stores the GL program ID, plus a small amount of metadata in our runtime program cache. Caveats: * Currently only implemented for the GL backend. Other backends will require more metadata to do any useful amount of work. Metal may need a more drastic workflow change, involving offline compilation of the shaders. * Currently only implemented for cached SkSL (not GLSL or program binaries). Supporting other formats again requires more metadata, and the cached shaders become increasingly specialized to GPU and driver versions. * Reusing the cached SkSL on different hardware is not supported. Many driver workarounds are implemented in the SkSL -> GLSL transformation, but some are higher level. Limiting device variance by artificially hiding extensions may help, but there are no guarantees. * The 'gltestprecompile' DM config exercises this code similarly to 'gltestpersistentcache', ensuring that results are visually identical when precompiling, and that no cache misses occur after precompiling. Change-Id: Id314c5d5f5a58fe503a0505a613bd4a540cc3589 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/239438 Reviewed-by: Greg Daniel <egdaniel@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com> Commit-Queue: Brian Osman <brianosman@google.com> |
||
|---|---|---|
| .. | ||
| android_compile | ||
| assets | ||
| buildstats | ||
| g3_compile | ||
| gen_tasks_logic | ||
| recipe_modules | ||
| recipes | ||
| tools/luci-go | ||
| assets.isolate | ||
| build_task_drivers.sh | ||
| bundle_recipes.sh | ||
| calmbench.isolate | ||
| cfg.json | ||
| check_deps.py | ||
| compile_android_framework.isolate | ||
| compile_g3_framework.isolate | ||
| compile.isolate | ||
| empty.isolate | ||
| gen_compile_isolate.py | ||
| gen_tasks.go | ||
| git_utils.py | ||
| infra_tests.isolate | ||
| infra_tests.py | ||
| infrabots.isolate | ||
| ios_bin.isolate | ||
| isolate_android_sdk_linux.isolate | ||
| isolate_gcloud_linux.isolate | ||
| isolate_go.isolate | ||
| jobs.json | ||
| Makefile | ||
| OWNERS | ||
| perf_skia_bundled.isolate | ||
| README.md | ||
| recipes.isolate | ||
| recipes.py | ||
| resources.isolate | ||
| run_recipe.isolate | ||
| run_recipe.py | ||
| skpbench_skia_bundled.isolate | ||
| swarm_recipe.isolate | ||
| task_drivers.isolate | ||
| tasks.json | ||
| test_skia_bundled.isolate | ||
| test_utils.py | ||
| update_meta_config.py | ||
| upload_skps.py | ||
| utils.py | ||
| zip_utils_test.py | ||
| zip_utils.py | ||
Skia Infrastructure
This directory contains infrastructure elements.
Tasks and Jobs
Files in this directory define a DAG of tasks which run at every Skia commit. A task is a small, self-contained unit which runs via Swarming on a machine in the pool. Tasks may be chained together, eg. one task to compile test binaries and another to actually run them.
Jobs are collections of related tasks which help define sub-sections of the DAG, for example, to be used as try jobs. Each job is defined as an entry point into the DAG.
The tasks.json file in this directory is the master list of tasks and jobs for the repo. Note that tasks.json is NEVER edited by hand but generated via gen_task.go and the input files enumerated below. The Task Scheduler reads the tasks.json file at each commit to determine which jobs to run. For convenience, gen_tasks.go is provided to generate tasks.json and also to test it for correct syntax and detecting cycles and orphaned tasks. Always edit gen_tasks.go or one of the following input JSON files, rather than tasks.json itself:
- cfg.json - Basic configuration information for gen_tasks.go.
- jobs.json - The master list of all jobs to run. Edit this to add or remove bots.
Whenever gen_tasks.go, any of the above JSON files, or assets are changed, you need to run gen_tasks.go to regenerate tasks.json:
$ go run infra/bots/gen_tasks.go
Or:
$ cd infra/bots; make train
There is also a test mode which performs sanity-checks and verifies that tasks.json is unchanged:
$ go run infra/bots/gen_tasks.go --test
Or:
$ cd infra/bots; make test
Recipes
Recipes are the framework used by Skia's infrastructure to perform work inside of Swarming tasks. The main elements are:
- recipes.py - Used for running and testing recipes.
- recipes - These are the entry points for each type of task, eg. compiling or running tests.
- recipe_modules - Shared modules which are used by recipes.
- .recipe_deps - Recipes and modules may depend on modules from other repos. The recipes.py script automatically syncs those dependencies in this directory.
Isolate Files
These files determine which parts of the repository are transferred to the bot when a Swarming task is triggered. The Isolate tool hashes each file and will upload any new/changed files. Bots maintain a cache so that they can efficiently download only the files they don't have.
Assets
Artifacts used by the infrastructure are versioned here, along with scripts for recreating/uploading/downloading them. See the README in that directory for more information. Any time an asset used by the bots changes, you need to re-run gen_tasks.go.
Tools
Assorted other infrastructure-related tools, eg. isolate and CIPD binaries.
CT
Helpers for running Skia tasks in Cluster Telemetry.