Vulkan-Docs/xml/README.adoc

203 lines
8.2 KiB
Plaintext

= Vulkan^(R)^ API Registry Build Instructions and Notes
Jon Leech
* <<intro,Introduction>>
* <<files,Files>>
* <<targets,Makefile Targets>>
* <<linux,Linux Software Dependencies>>
* <<windows,Windows Software Dependencies>>
* <<history,Revision History>>
[[intro]]
== Introduction
This is the Vulkan XML API Registry. It is used to generate the canonical
`vulkan_core.h` and the API Asciidoc include files used by the Vulkan
Specification and Reference Pages.
When changes to the header or the includes are needed, follow this workflow.
Normally changes are needed only when defining a new extension or core
version of the API.
* Create a git branch to work in locally
* Edit `vk.xml`
* `make ; make test`
** This generates headers in `../include/vulkan` including `vulkan_core.h`
and a set of platform-dependent headers `vulkan_<platform>.h`.
* `(cd .. && make generated)`
** This generates asciidoc includes for the spec. There are many ways to
invoke the Makefile in the spec directory; this simple recipe only
generates includes for the core Vulkan API without any extensions.
* Repeat until the headers and/or includes are correct.
* Commit your changes to your local git branch, push to your upstream git
server (your personal repository clone of KhronosGroup/Vulkan-Docs on
Github, for people outside Khronos; the Khronos member gitlab server,
otherwise), and create a pull or merge request against the specification
branch (`master`) or other appropriate target.
For a detailed description of the schema, go to `..` and `make registry`,
which generates $(OUTDIR)/registry.html. This includes some examples of how
to make simple changes in the API via the XML.
The generator scripts are written in Python 3, using the `etree` package for
processing XML.
[[files]]
== Files
* Makefile - generates headers and extension loader sources from XML (see
<<targets,Makefile Targets>> below).
* vk.xml - XML API description.
* registry.rnc - RelaxNG compact schema for validating XML against the
schema.
* Python scripts in ../scripts/ which operate on vk.xml.
** ../scripts/genvk.py - Python script to generate vulkan_core.h and other
targets.
** ../scripts/reg.py - Python tools to read XML file and convert it into C
headers.
** ../scripts/generator.py - output generator base class.
** ../scripts/cgenerator.py - C header output generator.
** ../scripts/docgenerator.py - Asciidoc interface language include
generator.
** ../scripts/extensionmetadocgenerator.py - Generator for Asciidoc
extension descriptions in spec appendices.
** ../scripts/hostsyncgenerator.py - Asciidoc host sync table generator.
** ../scripts/pygenerator.py - Generates python encoding of the API
description.
** ../scripts/validitygenerator.py - Asciidoc validity language generator.
* ../include/vulkan/vulkan_core.h - Generated Vulkan non-platform API
header.
* ../include/vulkan/vulkan_<platform>.h - Generated Vulkan platform API
headers.
* ../scripts/indexExt.py - generate HTML index of all extensions for
inclusion into the Vulkan registry index page.
* ../scripts/extDependency.py - generate extension dependencies in Bash
and Python form for use when building the specification.
[[targets]]
== Makefile Targets
* `install` (default target) - regenerate header files in
`../include/vulkan`.
* `test` - make sure `../include/vulkan/vulkan_core.h` compiles.
*Important!* Can also be used to test if platform headers compile by
specifying `make TESTDEFS=-DVK_USE_PLATFORM_<PLATFORM>_<AUTHORID> test`.
* `validate` - validate `vk.xml` against the schema. Requires installing
`jing` (see <<linux,Software Dependencies>> below). Also important!
* `clean_dirt` - remove intermediate files.
* `clean` - remove generated files. Usually done when preparing to merge
to `master` branch via ```make clean ; make install```.
If you have trouble running the Makefile on your platform, the following
steps will build `vulkan_core.h` and test that it compiles:
[source,sh]
----
# Regenerate header from XML
python3 ../scripts/genvk.py -o ../include/vulkan vulkan_core.h
# Verify that the resulting header compiles
gcc -Wall -c -I../include testcore.c
g++ -Wall -c -std=c++98 -I../include testcore.c
g++ -Wall -c -std=c++11 -I../include testcore.c
----
[[linux]]
== Linux Software Depencies
These are the minimum versions of required tools in a Debian 8 development
environment. Earlier versions *may* work, but unless they are verified by
someone else, there's no way to know that:
* Python 3 (`python3`, version: 3.4.2)
* pass:[g++] and gcc (`g++-4.9` / `gcc-4.9`, version: 4.9.2-10 - gcc 4.8
also reported to work)
* GNU make (`make` version: 4.0.8-1; older versions probably OK)
* Jing (`jing` version: 20131210+dfsg+1-1; needed only for optional XML
validation)
[[windows]]
== Windows Software Dependencies
Using the Windows 10 Ubuntu subsystem, if available, is probably the most
pleasant way of building. Cygwin64 is also a viable approach.
On native Windows without a Linux emulation environment, one way to build on
Windows is:
* Install python (32-bit works great): https://www.python.org/downloads/
* Ensure the pip module is installed (should be by default)
* Run the `genvk.py` script in C:\PathToVulkan\src\specfile
** ```C:\PathToPython\python3.exe genvk.py vulkan_core.h```
[[history]]
== Revision History
* 2019/03/10 -
Update for script reorganization.
* 2018/05/21 -
Don't generate vulkan_ext.[ch] from the `install` target. Add a new
shortcut `extloader` target for people still using this code and needing
to regenerate it.
* 2018/03/13 -
Update for new directory structure.
* 2018/03/06 -
Update for Vulkan 1.1 release and `master` branch.
* 2015/09/18 -
Split platform-specific headers into their own vulkan_<platform>.h
files, move vulkan.h to vulkan_core.h, and add a new (static) vulkan.h
which includes appropriate combinations of the other headers.
* 2015/06/01 -
The header that is generated has been improved relative to the first
version. Function arguments are indented like the hand-generated header,
enumerant BEGIN/END_RANGE enums are named the same, etc. The ordering of
declarations is unlike the hand-generated header, and probably always
will because it results from a type/enum/function dependency analysis.
Some of this can be forced by being more explicit about it, if that is a
big deal.
* 2015/06/02 -
Per WG signoff, converted hex constant values to decimal (for
non-bitmasks) and VK_BIT macros to 'bitpos' attributes in the XML and
hex constants in the header. Updated schema to match. Changed <ptype>
tag to <type>.
* 2015/06/03 -
Moved into new 'vulkan' tree (did not bother preserving history in
previous repo). Added semantic knowledge about structs and unions to
<type> tags instead of just imbedding C struct definitions. Improved
registry.rnc schema a bit.
* 2015/06/07 -
Incorporate feedback from F2F including Python 3 and Windows fixes to
the scripts. Add documentation to readme.pdf. Fold in multiple merge
requests resulting from action items agreed at the F2F, to prepare
for everyone moving to XML instead of directly editing the header.
* 2015/06/20 -
Add vulkan-docs target and instructions for installing python3 and
python-lxml for Windows.
* 2015/08/13 -
Bring documentation up to date with Makefile targets (default is now
../include/vulkan.h).
* 2015/09/02 -
Update README with required (or known working) versions of toolchain
components.
* 2015/09/02 -
Move include/vulkan.h to vulkan/vulkan.h so #include "vulkan/vulkan.h"
is the normal usage (Bug 14576).
* 2016/02/12 -
Update README and remove old files to stage for public release.
* 2016/05/31 -
Remove dependency on lxml.
* 2016/07/27 -
Update documentation for changes to schema and generator scripts.
* 2016/08/26 -
Move README to an asciidoc file and update for the single-branch model.
Use 'clean' target to remove generated files in both spec source and
registry Makefiles.
* 2017/02/20 -
Move registry.txt (schema documentation) to the Vulkan spec source
directory and update the README here.