Khronos reference front-end for GLSL and ESSL, and sample SPIR-V generator
Перейти к файлу
steve-lunarg ef33ec0925 HLSL: add intrinsic function implicit promotions
This PR handles implicit promotions for intrinsics when there is no exact match,
such as for example clamp(int, bool, float).  In this case the int and bool will
be promoted to a float, and the clamp(float, float, float) form used.

These promotions can be mixed with shape conversions, e.g, clamp(int, bool2, float2).

Output conversions are handled either via the existing addOutputArgumentConversion
function, which this PR generalizes to handle either aggregates or unaries, or by
intrinsic decomposition.  If there are methods or intrinsics to be decomposed,
then decomposition is responsible for any output conversions, which turns out to
happen automatically in all current cases.  This can be revisited once inout
conversions are in place.

Some cases of actual ambiguity were fixed in several tests, e.g, spv.register.autoassign.*

Some intrinsics with only uint versions were expanded to signed ints natively, where the
underlying AST and SPIR-V supports that.  E.g, countbits.  This avoids extraneous
conversion nodes.

A new function promoteAggregate is added, and used by findFunction.  This is essentially
a generalization of the "promote 1st or 2nd arg" algorithm in promoteBinary.

The actual selection proceeds in three steps, as described in the comments in
hlslParseContext::findFunction:

1. Attempt an exact match.  If found, use it.
2. If not, obtain the operator from step 1, and promote arguments.
3. Re-select the intrinsic overload from the results of step 2.
2016-11-23 10:36:34 -07:00
External Updated cmake to better organize folders and options. 2016-05-05 14:45:53 -04:00
OGLCompilersDLL Build: C++ headers: Replace PR #366 with a more directed version. 2016-08-05 14:04:23 -06:00
SPIRV GLSL: The execution scope for barriers should be Workgroup. 2016-11-14 10:10:05 +01:00
StandAlone Add UAV (image) binding offset and HLSL register support 2016-11-01 14:44:54 -06:00
Test HLSL: add intrinsic function implicit promotions 2016-11-23 10:36:34 -07:00
glslang HLSL: add intrinsic function implicit promotions 2016-11-23 10:36:34 -07:00
gtests HLSL: add intrinsic function implicit promotions 2016-11-23 10:36:34 -07:00
hlsl HLSL: add intrinsic function implicit promotions 2016-11-23 10:36:34 -07:00
.appveyor.yml Add test configuration for Appveyor. 2016-05-10 10:52:32 -04:00
.clang-format Increase column limit to 120. 2016-06-05 20:45:34 -04:00
.gitattributes Tests: Move to just linefeed to simply cross-platform test changes. 2015-12-10 18:01:54 -07:00
.gitignore add reflection queries to return a TType. Fix minor issue with interface names. 2016-10-02 16:57:58 -06:00
.travis.yml Add configuration for Travis to test on Linux and Mac OS X. 2016-05-04 11:55:10 -04:00
CMakeLists.txt #517: Enable AMD extensions by default 2016-09-27 10:30:14 +02:00
ChooseMSVCCRT.cmake Allow choice of DLL or static CRT libraries through CMake options. 2014-03-12 02:34:44 +00:00
README-spirv-remap.txt WIP: SPV Remapper: add remapper test framework 2016-08-25 10:46:51 -06:00
README.md Updated README.md 2016-09-29 16:29:58 -07:00
make-revision Revisioning: Include GLSL.std.450 version. Also, the revision.h script now includes redirection to revision.h. 2015-08-03 12:04:56 -06:00

README.md

Also see the Khronos landing page for glslang as a reference front end:

https://www.khronos.org/opengles/sdk/tools/Reference-Compiler/

The above page includes where to get binaries, and is kept up to date regarding the feature level of glslang.

glslang

Build Status Build status

An OpenGL and OpenGL ES shader front end and validator.

There are several components:

  1. A GLSL/ESSL front-end for reference validation and translation of GLSL/ESSL into an AST.

  2. An HLSL front-end for translation of a broad generic HLL into the AST.

  3. A SPIR-V back end for translating the AST to SPIR-V.

  4. A standalone wrapper, glslangValidator, that can be used as a command-line tool for the above.

How to add a feature protected by a version/extension/stage/profile: See the comment in glslang/MachineIndependent/Versions.cpp.

Tasks waiting to be done are documented as GitHub issues.

Execution of Standalone Wrapper

To use the standalone binary form, execute glslangValidator, and it will print a usage statement. Basic operation is to give it a file containing a shader, and it will print out warnings/errors and optionally an AST.

The applied stage-specific rules are based on the file extension:

  • .vert for a vertex shader
  • .tesc for a tessellation control shader
  • .tese for a tessellation evaluation shader
  • .geom for a geometry shader
  • .frag for a fragment shader
  • .comp for a compute shader

There is also a non-shader extension

  • .conf for a configuration file of limits, see usage statement for example

Building

Dependencies

  • CMake: for generating compilation targets.
  • bison: optional, but needed when changing the grammar (glslang.y).
  • googletest: optional, but should use if making any changes to glslang.

Build steps

1) Check-Out this project

cd <parent of where you want glslang to be>
# If using SSH
git clone git@github.com:KhronosGroup/glslang.git
# Or if using HTTPS
git clone https://github.com/KhronosGroup/glslang.git

2) Check-Out External Projects

cd <the directory glslang was cloned to, "External" will be a subdirectory>
git clone https://github.com/google/googletest.git External/googletest

3) Configure

Assume the source directory is $SOURCE_DIR and the build directory is $BUILD_DIR:

For building on Linux (assuming using the Ninja generator):

cd $BUILD_DIR

cmake -GNinja -DCMAKE_BUILD_TYPE={Debug|Release|RelWithDebInfo} \
      -DCMAKE_INSTALL_PREFIX=`pwd`/install $SOURCE_DIR

For building on Windows:

cmake $SOURCE_DIR -DCMAKE_INSTALL_PREFIX=`pwd`/install
# The CMAKE_INSTALL_PREFIX part is for testing (explained later).

The CMake GUI also works for Windows (version 3.4.1 tested).

4) Build and Install

# for Linux:
ninja install

# for Windows:
cmake --build . --config {Release|Debug|MinSizeRel|RelWithDebInfo} \
      --target install

If using MSVC, after running CMake to configure, use the Configuration Manager to check the INSTALL project.

If you need to change the GLSL grammar

The grammar in glslang/MachineIndependent/glslang.y has to be recompiled with bison if it changes, the output files are committed to the repo to avoid every developer needing to have bison configured to compile the project when grammar changes are quite infrequent. For windows you can get binaries from GnuWin32.

The command to rebuild is:

bison --defines=MachineIndependent/glslang_tab.cpp.h
      -t MachineIndependent/glslang.y
      -o MachineIndependent/glslang_tab.cpp

The above command is also available in the bash script at glslang/updateGrammar.

Testing

Right now, there are two test harnesses existing in glslang: one is Google Test, one is the runtests script. The former runs unit tests and single-shader single-threaded integration tests, while the latter runs multiple-shader linking tests and multi-threaded tests.

Running tests

The runtests script requires compiled binaries to be installed into $BUILD_DIR/install. Please make sure you have supplied the correct configuration to CMake (using -DCMAKE_INSTALL_PREFIX) when building; otherwise, you may want to modify the path in the runtests script.

Running Google Test-backed tests:

cd $BUILD_DIR

# for Linux:
ctest

# for Windows:
ctest -C {Debug|Release|RelWithDebInfo|MinSizeRel}

# or, run the test binary directly
# (which gives more fine-grained control like filtering):
<dir-to-glslangtests-in-build-dir>/glslangtests

Running runtests script-backed tests:

cd $SOURCE_DIR/Test && ./runtests

Contributing tests

Test results should always be included with a pull request that modifies functionality.

If you are writing unit tests, please use the Google Test framework and place the tests under the gtests/ directory.

Integration tests are placed in the Test/ directory. It contains test input and a subdirectory baseResults/ that contains the expected results of the tests. Both the tests and baseResults/ are under source-code control.

Google Test runs those integration tests by reading the test input, compiling them, and then compare against the expected results in baseResults/. The integration tests to run via Google Test is registered in various gtests/*.FromFile.cpp source files. glslangtests provides a command-line option --update-mode, which, if supplied, will overwrite the golden files under the baseResults/ directory with real output from that invocation. For more information, please check gtests/ directory's README.

For the runtests script, it will generate current results in the localResults/ directory and diff them against the baseResults/. When you want to update the tracked test results, they need to be copied from localResults/ to baseResults/. This can be done by the bump shell script.

You can add your own private list of tests, not tracked publicly, by using localtestlist to list non-tracked tests. This is automatically read by runtests and included in the diff and bump process.

Programmatic Interfaces

Another piece of software can programmatically translate shaders to an AST using one of two different interfaces:

  • A new C++ class-oriented interface, or
  • The original C functional interface

The main() in StandAlone/StandAlone.cpp shows examples using both styles.

C++ Class Interface (new, preferred)

This interface is in roughly the last 1/3 of ShaderLang.h. It is in the glslang namespace and contains the following.

const char* GetEsslVersionString();
const char* GetGlslVersionString();
bool InitializeProcess();
void FinalizeProcess();

class TShader
    bool parse(...);
    void setStrings(...);
    const char* getInfoLog();

class TProgram
    void addShader(...);
    bool link(...);
    const char* getInfoLog();
    Reflection queries

See ShaderLang.h and the usage of it in StandAlone/StandAlone.cpp for more details.

C Functional Interface (orignal)

This interface is in roughly the first 2/3 of ShaderLang.h, and referred to as the Sh*() interface, as all the entry points start Sh.

The Sh*() interface takes a "compiler" call-back object, which it calls after building call back that is passed the AST and can then execute a backend on it.

The following is a simplified resulting run-time call stack:

ShCompile(shader, compiler) -> compiler(AST) -> <back end>

In practice, ShCompile() takes shader strings, default version, and warning/error and other options for controlling compilation.

Basic Internal Operation

  • Initial lexical analysis is done by the preprocessor in MachineIndependent/Preprocessor, and then refined by a GLSL scanner in MachineIndependent/Scan.cpp. There is currently no use of flex.

  • Code is parsed using bison on MachineIndependent/glslang.y with the aid of a symbol table and an AST. The symbol table is not passed on to the back-end; the intermediate representation stands on its own. The tree is built by the grammar productions, many of which are offloaded into ParseHelper.cpp, and by Intermediate.cpp.

  • The intermediate representation is very high-level, and represented as an in-memory tree. This serves to lose no information from the original program, and to have efficient transfer of the result from parsing to the back-end. In the AST, constants are propogated and folded, and a very small amount of dead code is eliminated.

    To aid linking and reflection, the last top-level branch in the AST lists all global symbols.

  • The primary algorithm of the back-end compiler is to traverse the tree (high-level intermediate representation), and create an internal object code representation. There is an example of how to do this in MachineIndependent/intermOut.cpp.

  • Reduction of the tree to a linear byte-code style low-level intermediate representation is likely a good way to generate fully optimized code.

  • There is currently some dead old-style linker-type code still lying around.

  • Memory pool: parsing uses types derived from C++ std types, using a custom allocator that puts them in a memory pool. This makes allocation of individual container/contents just few cycles and deallocation free. This pool is popped after the AST is made and processed.

    The use is simple: if you are going to call new, there are three cases:

    • the object comes from the pool (its base class has the macro POOL_ALLOCATOR_NEW_DELETE in it) and you do not have to call delete

    • it is a TString, in which case call NewPoolTString(), which gets it from the pool, and there is no corresponding delete

    • the object does not come from the pool, and you have to do normal C++ memory management of what you new