search

15knots / cmake4eclipse

Let Eclipse use CMake to generate the buildscripts for Eclipse CDT
Eclipse Public License 2.0
85 stars 25 forks source link

Recognize nvcc as a CUDA compiler and integrate with the Nsight Eclipse Plugins #65

Closed caselitz closed 6 years ago

caselitz commented 6 years ago

This refers to: https://groups.google.com/forum/#!topic/cmake4eclipse-users/u4SO8kF0nJ8

This is a...

Brief Description

CMAKE_EXPORT_COMPILE_COMMANDS Parser should work in combination with the Nsight plugin

What is the expected behavior?

The CMAKE_EXPORT_COMPILE_COMMANDS should appear for the CUDA languages (com.nvidia.cuda.toolchain.language.cu/c/c++) in Entries|Settings Entries, the Include path etc. should be parsed, and library includes can be resolved in .cu and .cc (and so forth) files.

What behavior are you observing?

It does NOT appear, include path are not parsed, and includes cannot be resolved.

Provide the steps to reproduce the issue, if applicable:

  1. Get cmake4eclipse
  2. Get Nsight plugin http://docs.nvidia.com/cuda/nsight-eclipse-plugins-guide/index.html Help->Install new software... NsightEE - jar:file:/usr/local/cuda-9.1/nsightee_plugins/com.nvidia.cuda.repo-1.0.0-SNAPSHOT.zip!/
  3. Test with some mixed C++ CUDA project that contains .cu and .cc files and that defines some include_directories for some libraries in the CMakeLists.txt file

Useful Information

cmake4eclipse version: 1.9.1 Which OS do you use: Ubuntu 16.04 Cmake version: 3.10.2

15knots commented 6 years ago

The mechanism to detect a language is solely based on the name of the compiler executable. e.g.

  1. C --> gcc, cc, cl, icl ,...
  2. C++ -> g++, c++

Please provide some more information to detect com.nvidia.cuda.toolchain.language.cu/c/c++ language. You will find that information in the compile_commands.json file which is generated by cmake.

And please confirm that com.nvidia.cuda.toolchain.language.cu/c/c++ actually are language IDs that can be fed to the CDT indexer and will expose the expected behavior. (I could not find any source code of NsightEE.)

caselitz commented 6 years ago

The mechanism to detect a language is solely based on the name of the compiler executable. e.g.

  1. C --> gcc, cc, cl, icl ,...
  2. C++ -> g++, c++

So I guess 3. com.nvidia.cuda.toolchain.language.cu/c/c++ -> nvcc

Please provide some more information to detect com.nvidia.cuda.toolchain.language.cu/c/c++ language. You will find that information in the compile_commands.json file which is generated by cmake.

In the first example of the compile_commands.json I posted in the mailinglist you can see that /usr/local/cuda-9.1/bin/nvcc is the compiler being used. I'm not sure what "more information" you need compared to what I've posted - everything else are basically duplicates of the given examples.

And please confirm that com.nvidia.cuda.toolchain.language.cu/c/c++ actually are language IDs that can be fed to the CDT indexer and will expose the expected behavior.

I do think that they are language IDs because they appear in the Entries Languages column (if I select CUDA Linux toolchain as current toolchain or add the nvcc compiler to the "Used tools" manually) like this:

[Unspecified][id=com.nvidia.cuda.toolchain.language.cuda.c]
[Unspecified][id=com.nvidia.cuda.toolchain.language.cuda.cu]
[Unspecified][id=com.nvidia.cuda.toolchain.language.cuda.c++]

I'm not familiar enough with this stuff to confirm that they "can be fed to the CDT indexer and will expose the expected behavior" (if you tell me how to check, I am happy to test).

Maybe also relevant, in the "Language Mappings" only the following seems to be added by the Plugin:

com.nvidia.cuda.ide.editor.cudac

From what I understand, for building, the current Toolchain does not matter (only current Builder "CMake Builder (GNU Make)" and CMakeLists.txt does - and I do not have any issue here). But the current Toolchain does matter for indexing, which is my problem. Is it correct that the selected current Toolchain only chooses a preset of "Used tools" and this is what actually matters? Because even when I choose the CUDA Linux toolchain and add the ''GCC C++ Compiler" to the "Used tools", the Language "GNU C++" appears in the Entries Languages column for which the CMAKE_EXPORT_COMPILE_COMMANDS Parser works and Includes etc. get properly resolved in .cc files (but not in .cu files, probably because these are not linked to the "GNU C++" Language).

(I could not find any source code of NsightEE.)

I'm afraid this is proprietary NVIDIA stuff. :( Btw., just to point out again, I am using the Nsight Eclipse Plugin. Not sure whats the difference to this, except that the URL contains "nsightee" instead of "nsight" and it is supposed to work only with "Eclipse environment (v4.4 and v4.5)" instead of "vanilla Eclipse 4.4 or later". I feel what I'm using is the newer version of their plugin and their documentation is inconsistent. Again another thing is the "Nsight Eclipse Edition" WITHOUT Plugin, which is a complete Eclipse modified and shipped by NVIDIA.

15knots commented 6 years ago

I'm not familiar enough with this stuff to confirm that they "can be fed to the CDT indexer and will expose the expected behavior" (if you tell me how to check, I am happy to test).

The indexer is consulted by the source code editor to get for example the include paths. The C-Editor, for example, queries the indexer for C-include paths by specifying a language ID (to get only the entries for the C language).. The CUDA source editor will query the indexer by passing its own language ID (possibly 'com.nvidia.cuda.toolchain.language.cu', but that's a guess). The parser has to specify the proper ID to feed the include paths for cuda to the indexer, otherwise the CUDA editor will not find them and they will not show up in the Entries Languages column and 'Jump to declaration' on a macro that originates from a header file below the include path will not work.

I already asked Nvidia support to confirm the language ID, but they did not respond yet.

But the current Toolchain does matter for indexing, which is my problem. Is it correct that the selected current Toolchain only chooses a preset of "Used tools" and this is what actually matters? Because even when I choose the

I agree, the tools specified in the toolchain should not affect editor behavior. I played around and deleted each tool entry in the toolchain settings. But after I deleted the 'GCC C Compiler' entry, the C-editor was no longer aware of any C include paths and macros. The same holds for the 'GCC C++ Compiler' entry. So it seems you have to add the NVCC compiler there, too.

I suggest to just try the cuda language ID. I added an entry for nvcc which uses com.nvidia.cuda.toolchain.language.cu as the language ID. You may try it by adding the direct update site URL https://bintray.com/15knots/p2-zip/download_file?file_path=cmake4eclipse-1.10.0.zip. If it is working, we guessed the proper language ID.

caselitz commented 6 years ago

Not sure if that is helpful here, but I looked into how the editor is chosen - apparently by File Association which can be Content Type based. The Nsight Plugin seems to modify the Content Types as follows:

+ Text
  + C Source File (*.c)
    - C Header File (*.h)
    + C++ Source File (*.C, *.c++, *.cc, *.cpp, *.cxx)
      + C++ Header File (*.cuh, *.cuinc, *.h, *.hcu, *.hh, *.hpp, *.hxx, *.inc)
        - CUDA Header Content Type (*.cuh, *.h, *.hcu)
      - CUDA C Source File (*.cu)
      - CUDA C Source File (*.cu)

By default the Cuda Editor seems to be used only for *.cu files, i.e., CUDA C Source Files (no idea why this appears two times in the list).

caselitz commented 6 years ago

I already asked Nvidia support to confirm the language ID, but they did not respond yet.

Let's hope they give you an (official) statement, but I kind of doubt it. :(

I agree, the tools specified in the toolchain should not affect editor behavior. I played around and deleted each tool entry in the toolchain settings. But after I deleted the 'GCC C Compiler' entry, the C-editor was no longer aware of any C include paths and macros. The same holds for the 'GCC C++ Compiler' entry. So it seems you have to add the NVCC compiler there, too.

Yes. So my conclusion is, what matters (at least for indexing) are the "Used tools", not the "Current toolchain" (which might be only a preset for the set of "Used tools").

I suggest to just try the cuda language ID. I added an entry for nvcc which uses com.nvidia.cuda.toolchain.language.cu as the language ID. You may try it by adding the direct update site URL https://bintray.com/15knots/p2-zip/download_file?file_path=cmake4eclipse-1.10.0.zip. If it is working, we guessed the proper language ID.

Unfortunately it does not, same behavior as before (CMAKE_EXPORT_COMPILE_COMMANDS still not showing up for the three cuda.c/cu/c++ Languages in the Settings Entries). Maybe we should try com.nvidia.cuda.toolchain.language.c or com.nvidia.cuda.toolchain.language.c++. Or why can't we take all three? I think it is possible that one file contains multiple languages, e.g., c, c++, and cuda code.

caselitz commented 6 years ago

The CUDA source editor will query the indexer by passing its own language ID (possibly 'com.nvidia.cuda.toolchain.language.cu', but that's a guess). The parser has to specify the proper ID to feed the include paths for cuda to the indexer, otherwise the CUDA editor will not find them and they will not show up in the Entries Languages column and 'Jump to declaration' on a macro that originates from a header file below the include path will not work.

       | -------- lang id -------> |         | <-------- lang id ------ |
Editor |                           | Indexer |                          | Parser
       | <--- path etc for id ---- |         | <--- path etc for id --- |

Is that summarized correctly? So we need to guess the lang id that the Cuda Editor is sending to the Indexer, so that you can put this value in the lang id that the Parser sends to the Indexer?

And can the Editor and Parser send multiple different lang ids?

What I still don't understand is the relation between Editor and file type *.* (or does this Content Type somehow matter here?). What I observed: Thanks to the CMAKE_EXPORT_COMPILE_COMMANDS Parser, includes can be resolved in a .cc file but NOT in a .cuh file, even though both are opened in the C/C++-Editor.

15knots commented 6 years ago

| -------- lang id -------> | | <-------- lang id ------ | Editor | | Indexer | | Parser | <--- path etc for id ---- | | <--- path etc for id --- | Is that summarized correctly? So we need to guess the lang id that the Cuda Editor is sending to the Indexer, so that you can put this value in the lang id that the Parser sends to the Indexer?

Yes, that's my assumption. At least the C/C++ editors work that way. But without any info from nvidia or the source code of their Builtin Specs detector or editor I cannot make this working (although it should be trivial).

What I still don't understand is the relation between Editor and file type . (or does this Content Type somehow matter here?).

The content type should not matter. It it used to open the appropriate Editor if you double-click a source file.

Thanks to the CMAKE_EXPORT_COMPILE_COMMANDS Parser, includes can be resolved in a .cc file but NOT in a .cuh file, even though both are opened in the C/C++-Editor.

I just can guess here: Maybe you have to tell the C/C++-Editor to recognize *.cuh as a C- header file (contentn type), maybe you can make it work by adding a language mapping. Consult the CDT forum for more precise info.

caselitz commented 6 years ago

Try the 1.10.0 version mentioned above. Its a guess of the lang ID, but if it works, we're done.

I did, but it didn't help. See the end of my second last post.

caselitz commented 6 years ago

But without any info from nvidia or the source code of their Builtin Specs detector or editor I cannot make this working (although it should be trivial).

I see your point. But can't we at least

try com.nvidia.cuda.toolchain.language.c or com.nvidia.cuda.toolchain.language.c++. Or why can't we take all three?

15knots commented 6 years ago

Could you open the plugin-jar that contains class com.nvidia.cuda.toolchain.internal.NVCCBuiltinSpecsDetector and send me the file plugins.xml in there per PM? It would help in reverse-engineering.

caselitz commented 6 years ago

I think there is no PM on Github? So I just post it here - I hope that's the right one.

Container: com.nvidia.cuda.toolchain_9.1.0.201711011803.jar File: plugin.xml

<?xml version="1.0" encoding="UTF-8"?>
<?eclipse version="3.4"?>
<plugin>
   <extension
         point="org.eclipse.cdt.managedbuilder.core.buildDefinitions">
      <managedBuildRevision
            fileVersion="4.0.0">
      </managedBuildRevision>
      <toolChain
            configurationEnvironmentSupplier="com.nvidia.cuda.toolchain.internal.CUDAToolkitEnvironmentSupplier"
            configurationMacroSupplier="com.nvidia.cuda.toolchain.internal.CudaToolkitMacroSupplier"
            id="com.nvidia.cuda.toolchain.baseToolchain"
            isAbstract="true"
            languageSettingsProviders="com.nvidia.cuda.toolchain.CUDACProvider"
            name="CUDA base toolchain"
            supportsManagedBuild="true"
            targetTool="nvcc.linker;nvcc.archiver">
         <builder
               id="com.nvidia.cuda.toolchain.builder"
               isAbstract="false"
               isVariableCaseSensitive="false"
               name="CUDA Toolkit Builder"
               superClass="cdt.managedbuild.target.gnu.builder">
         </builder>
         <tool
               command="${nvcc}"
               id="nvcc.linker"
               commandLineGenerator="com.nvidia.cuda.toolchain.internal.LinkerCommandLineGenerator"
               isAbstract="false"
               name="NVCC linker"
               natureFilter="both"
               outputFlag="-o">
            <supportedProperties>
               <property
                     id="org.eclipse.cdt.build.core.buildArtefactType">
                  <value
                        id="org.eclipse.cdt.build.core.buildArtefactType.exe">
                  </value>
                  <value
                        id="org.eclipse.cdt.build.core.buildArtefactType.sharedLib">
                  </value>
               </property>
            </supportedProperties>
            <inputType
                  buildVariable="OBJS"
                  id="com.nvidia.cuda.toolchain.nvcc.linker.input"
                  multipleOfType="true"
                  sources="o"
                  superClass="cdt.managedbuild.tool.gnu.cpp.linker.input">
               <additionalInput
                     kind="additionalinputdependency"
                     paths="$(USER_OBJS)">
               </additionalInput>
               <additionalInput
                     kind="additionalinput"
                     paths="$(LIBS)">
               </additionalInput>
            </inputType>
            <outputType
                outputs=""
                buildVariable="EXECUTABLES"
                id="com.nvidia.cuda.toolchain.nvcc.linker.output">
               <enablement
                     type="ALL">
                  <checkOption
                        isRegex="false"
                        optionId="com.nvidia.cuda.toolchain.linker.shared"
                        value="false">
                  </checkOption>
               </enablement>
            </outputType>
            <outputType
                outputPrefix="lib"
                buildVariable="LIBRARIES"
                id="com.nvidia.cuda.toolchain.nvcc.linker.output.so">
               <enablement
                     type="ALL">
                  <checkOption
                        isRegex="false"
                        optionId="com.nvidia.cuda.toolchain.linker.shared"
                        value="true">
                  </checkOption>
               </enablement>
            </outputType>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.linker.category.libraries"
                  name="Libraries">
            </optionCategory>
            <option
                  name="CUDA Runtime Library"
                  category="com.nvidia.cuda.toolchain.linker.category.libraries"
                  id="com.nvidia.cuda.toolchain.linker.cudart"
                  valueType="enumerated">
                  <enumeratedOptionValue
                  name="Static"
                  command="--cudart=static"
                  id="com.nvidia.cuda.toolchain.linker.cudart.static">
                  </enumeratedOptionValue>
                  <enumeratedOptionValue
                  name="Shared"
                  command="--cudart=shared"
                  id="com.nvidia.cuda.toolchain.linker.cudart.shared">
                  </enumeratedOptionValue>
            </option>
            <option superClass="gnu.cpp.link.option.libs"
                category="com.nvidia.cuda.toolchain.linker.category.libraries" id="nvcc.linker.libs">
            </option>
            <option superClass="gnu.cpp.link.option.paths"
                category="com.nvidia.cuda.toolchain.linker.category.libraries" id="nvcc.linker.paths">
            </option>
            <optionCategory
                  id="com.nvidia.cuda.toolchain.linker.category.misc"
                  name="Miscellaneous">
            </optionCategory>
            <option
                  browseType="file"
                  category="com.nvidia.cuda.toolchain.linker.category.misc"
                  command="-ccbin"
                  commandGenerator="com.nvidia.cuda.toolchain.internal.CommandSpaceValueGenerator"
                  defaultValue="${ccbin}"
                  id="com.nvidia.cuda.toolchain.linker.ccbin"
                  isAbstract="false"
                  name="Path to the host compiler"
                  valueType="string">
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.linker.category.misc"
                  commandGenerator="com.nvidia.cuda.toolchain.internal.LinkWithOpenGLOptionCommandGenerator"
                  defaultValue="false"
                  id="nvcc.linker.option.linkgl"
                  isAbstract="false"
                  name="Link with OpenGL libraries"
                  resourceFilter="all"
                  valueType="boolean">
            </option>
            <option superClass="gnu.cpp.link.option.other"
                category="com.nvidia.cuda.toolchain.linker.category.misc" id="nvcc.linker.other">
            </option>
            <option superClass="gnu.cpp.link.option.userobjs"
                category="com.nvidia.cuda.toolchain.linker.category.misc" id="nvcc.linker.userobjs">
            </option>
            <optionCategory
                  id="com.nvidia.cuda.toolchain.linker.category.shared"
                  name="Shared Library Settings">
            </optionCategory>
            <option
                  defaultValue="false"
                  id="com.nvidia.cuda.toolchain.linker.shared"
                  isAbstract="false"
                  command="--shared"
                  category="com.nvidia.cuda.toolchain.linker.category.shared"
                  name="Build shared library"
                  resourceFilter="all"
                  valueType="boolean">
               <enablement
                     attribute="defaultValue"
                     extensionAdjustment="false"
                     type="CONTAINER_ATTRIBUTE"
                     value="true">
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildArtefactType"
                        value="org.eclipse.cdt.build.core.buildArtefactType.sharedLib">
                  </checkBuildProperty>
               </enablement>
            </option>
            <enablement
                  type="ALL">
               <not>
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildArtefactType"
                        value="org.eclipse.cdt.build.core.buildArtefactType.staticLib">
                  </checkBuildProperty>
               </not>
            </enablement>
         </tool>
         <tool
               command="${nvcc}"
               id="nvcc.archiver"
               isAbstract="false"
               name="NVCC archiver"
               natureFilter="both"
               outputFlag="-lib -o">
            <supportedProperties>
               <property
                     id="org.eclipse.cdt.build.core.buildArtefactType">
                  <value
                        id="org.eclipse.cdt.build.core.buildArtefactType.staticLib"></value>
               </property>
            </supportedProperties>
            <inputType
                  buildVariable="OBJS"
                  id="com.nvidia.cuda.toolchain.nvcc.archiver.input"
                  multipleOfType="true"
                  sources="o"
                  superClass="cdt.managedbuild.tool.gnu.cpp.linker.input">
            </inputType>
            <outputType
                  buildVariable="ARCHIVES"
                  id="com.nvidia.cuda.toolchain.nvcc.archiver.output.ar"
                  outputPrefix="lib">
            </outputType>
            <enablement
                  type="ALL">
               <checkBuildProperty
                     property="org.eclipse.cdt.build.core.buildArtefactType"
                     value="org.eclipse.cdt.build.core.buildArtefactType.staticLib">
               </checkBuildProperty>
            </enablement>
        </tool>         
         <tool
               command="${nvcc}"
               errorParsers="com.nvidia.cuda.toolchain.nvccErrorParser;org.eclipse.cdt.core.GCCErrorParser"
               id="nvcc.compiler"
               isAbstract="false"
               name="NVCC Compiler"
               natureFilter="both"
               outputFlag="-c -o">
            <inputType
                  dependencyCalculator="com.nvidia.cuda.toolchain.internal.NVCCManagedDependencyGenerator"
                  id="com.nvidia.cuda.toolchain.input.c"
                  languageId="com.nvidia.cuda.toolchain.language.cuda.c"
                  sourceContentType="org.eclipse.cdt.core.cSource"
                  primaryInput="true"
                  scannerConfigDiscoveryProfileId="com.nvidia.cuda.ide.build.NVCCPerProjectProfile|org.eclipse.cdt.managedbuilder.core.GCCManagedMakePerProjectProfileC|org.eclipse.cdt.make.core.GCCStandardMakePerFileProfile"
                  sources="c">
            </inputType>
            <inputType
                  dependencyCalculator="com.nvidia.cuda.toolchain.internal.NVCCManagedDependencyGenerator"
                  dependencyContentType="com.nvidia.cuda.toolchain.CUDAHeaderContentType"
                  dependencyExtensions="h,cuh,hcu"
                  id="com.nvidia.cuda.toolchain.input.cu"
                  languageId="com.nvidia.cuda.toolchain.language.cuda.cu"
                  primaryInput="true"
                  sourceContentType="com.nvidia.cuda.toolchain.CUDAContentType"
                  scannerConfigDiscoveryProfileId="com.nvidia.cuda.ide.build.NVCCPerProjectProfile"
                  sources="cu">
            </inputType>
            <inputType
                  dependencyCalculator="com.nvidia.cuda.toolchain.internal.NVCCManagedDependencyGenerator"
                  id="com.nvidia.cuda.toolchain.input.cpp"
                  languageId="com.nvidia.cuda.toolchain.language.cuda.c++"
                  sourceContentType="org.eclipse.cdt.core.cxxSource"
                  primaryInput="true"
                  scannerConfigDiscoveryProfileId="com.nvidia.cuda.ide.build.NVCCPerProjectProfile|org.eclipse.cdt.managedbuilder.core.GCCManagedMakePerProjectProfileCPP|org.eclipse.cdt.make.core.GCCStandardMakePerFileProfile"
                  sources="cpp">
            </inputType>
            <outputType
                  buildVariable="OBJS"
                  id="com.nvidia.cuda.toolchain.output.o"
                  outputs="o"
                  primaryOutput="true">
            </outputType>
            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.dialect"
                  name="Dialect">
            </optionCategory>
            <option
                  name="Language standard"
                  category="com.nvidia.cuda.toolchain.compiler.category.dialect"
                  id="com.nvidia.cuda.toolchain.compiler.dialect"
                  valueType="enumerated">
                  <enumeratedOptionValue
                  name="Default"
                  command=""
                  id="com.nvidia.cuda.toolchain.compiler.dialect.default">
                  </enumeratedOptionValue>
                  <enumeratedOptionValue
                  name="C++11"
                  command="-std=c++11"
                  id="com.nvidia.cuda.toolchain.compiler.dialect.cpp11">
                  </enumeratedOptionValue>
                   <enumeratedOptionValue
                  name="C++14"
                  command="-std=c++14"
                  id="com.nvidia.cuda.toolchain.compiler.dialect.cpp14">
                  </enumeratedOptionValue>
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.dialect"
                  command="--expt-relaxed-constexpr"
                  id="com.nvidia.cuda.toolchain.compiler.relaxed-constexpr"
                  isAbstract="false"
                  name="Allow host code to invoke __device__ constexpr functions"
                  valueType="boolean">
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.dialect"
                  command="--expt-extended-lambda"
                  id="com.nvidia.cuda.toolchain.compiler.extended-lambda"
                  isAbstract="false"
                  name="Allow __host__, __device__ annotations in lambda declaration"
                  valueType="boolean">
            </option>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.preprocessor"
                  name="Preprocessor">
            </optionCategory>
            <option superClass="gnu.cpp.compiler.option.preprocessor.def"
                category="com.nvidia.cuda.toolchain.compiler.category.preprocessor" id="nvcc.compiler.def.symbols">
            </option>
            <option superClass="gnu.cpp.compiler.option.preprocessor.undef"
                category="com.nvidia.cuda.toolchain.compiler.category.preprocessor" id="nvcc.compiler.undef.symbol">
            </option>
            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.includes"
                  name="Includes">
            </optionCategory>
            <option superClass="gnu.cpp.compiler.option.include.paths"
                category="com.nvidia.cuda.toolchain.compiler.category.includes" id="nvcc.compiler.include.paths">
            </option>
            <option superClass="gnu.cpp.compiler.option.include.files"
                category="com.nvidia.cuda.toolchain.compiler.category.includes" id="nvcc.compiler.include.files">
            </option>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.optimization"
                  name="Optimization">
            </optionCategory>
            <option
                  name="Optimization level"
                  category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                  id="com.nvidia.cuda.toolchain.compiler.optimization"
                  valueType="enumerated">
                  <enumeratedOptionValue
                        id="com.nvidia.cuda.toolchain.compiler.optimization.default"
                        name="Default">
                  </enumeratedOptionValue>
                  <enumeratedOptionValue
                  name="O1"
                  command="-O1"
                  id="com.nvidia.cuda.toolchain.compiler.optimization.level1">
                  </enumeratedOptionValue>
                  <enumeratedOptionValue
                  name="O2"
                  command="-O2"
                  id="com.nvidia.cuda.toolchain.compiler.optimization.level2">
                  </enumeratedOptionValue>
                  <enumeratedOptionValue
                  name="O3"
                  command="-O3"
                  id="com.nvidia.cuda.toolchain.compiler.optimization.level3">
                  </enumeratedOptionValue>
                  <enablement
                        attribute="defaultValue"
                        extensionAdjustment="false"
                        type="CONTAINER_ATTRIBUTE"
                        value="com.nvidia.cuda.toolchain.compiler.optimization.level3">
                     <checkBuildProperty
                           property="org.eclipse.cdt.build.core.buildType"
                           value="org.eclipse.cdt.build.core.buildType.release">
                     </checkBuildProperty>
                  </enablement>

            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                command="-maxrregcount ${VALUE}" id="nvcc.compiler.maxregcount"
                isAbstract="false" name="Maximum number of registers (-maxrregcount)"
                tip="Specify the maximum amount of registers that GPU functions can use. Until a function-specific limit, a higher value will generally increase the performance of individual GPU threads that execute this function. However, because thread registers are allocated from a global register pool on each GPU, a higher value of this option will also reduce the maximum thread block size, thereby reducing the amount of thread parallelism. Hence, a good maxrregcount value is the result of a trade-off.&#x0A;If this option is not specified, then no maximum is assumed. Otherwise the specified value will be rounded to the next multiple of 4 registers until the GPU specific maximum of 128 registers."
                valueType="string">
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                command="--use_fast_math" defaultValue="false"
                id="nvcc.compiler.usefastmath" isAbstract="false"
                name="Make use of fast math library (-use_fast_math)" 
                tip="Make use of fast math library. -use_fast_math implies -ftz=true -prec-div=false -prec-sqrt=false."
                valueType="boolean">
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                command="-ftz true" defaultValue="false" id="nvcc.compiler.ftz"
                isAbstract="false" name="Flush denormal values to zero in single-precision FP operations (-ftz)" 
                tip="When performing single-precision floating-point operations, flush denormal values to zero or preserve denormal values. -use_fast_math implies --ftz=true."
                valueType="boolean">
                <enablement type="UI_ENABLEMENT">
                    <checkOption isRegex="false"
                        optionId="nvcc.compiler.usefastmath" value="false">
                    </checkOption>
                </enablement>
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                commandFalse="-prec-div false" defaultValue="true"
                id="nvcc.compiler.precdiv" isAbstract="false"
                name="Use IEEE round-to-nearest mode for precision FP division (-prec-div)" 
                tip="For single-precision floating-point division and reciprocals, use IEEE round-to-nearest mode or use a faster approximation. -use_fast_math implies --prec-div=false."
                valueType="boolean">
                <enablement type="UI_ENABLEMENT">
                    <checkOption isRegex="false"
                        optionId="nvcc.compiler.usefastmath" value="false">
                    </checkOption>
                </enablement>
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                commandFalse="-prec-sqrt false" defaultValue="true"
                id="nvcc.compiler.precsqrt" isAbstract="false"
                name="Use IEEE round-to-nearest mode for precision FP sqrt (-prec-sqrt)" 
                tip="For single-precision floating-point square root, use IEEE round-to-nearest mode or use a faster approximation. -use_fast_math implies --prec-sqrt=false."
                valueType="boolean">
                <enablement type="UI_ENABLEMENT">
                    <checkOption isRegex="false"
                        optionId="nvcc.compiler.usefastmath" value="false">
                    </checkOption>
                </enablement>
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.optimization"
                commandFalse="-fmad false" defaultValue="true"
                id="nvcc.compiler.fmad" isAbstract="false"
                name="Contract FP multiplies and adds/subtracts into FP multiply-add operations (-fmad)"
                tip="Enables (disables) the contraction of floating-point multiplies and adds/subtracts into floating-point multiply-add operations (FMAD, FFMA or DFMA)."
                valueType="boolean">
                <enablement type="UI_ENABLEMENT">
                    <checkOption isRegex="false"
                        optionId="nvcc.compiler.usefastmath" value="false">
                    </checkOption>
                </enablement>
            </option>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.debugging"
                  name="Debugging">
            </optionCategory>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.debugging"
                  command="--device-debug"
                  id="com.nvidia.cuda.toolchain.compiler.device-debug"
                  isAbstract="false"
                  name="Generate debug information for device code (-G)"
                  valueType="boolean">
               <enablement
                     attribute="defaultValue"
                     type="CONTAINER_ATTRIBUTE"
                     value="true">
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildType"
                        value="org.eclipse.cdt.build.core.buildType.debug"></checkBuildProperty>
               </enablement>
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.debugging"
                  command="--debug"
                  id="com.nvidia.cuda.toolchain.compiler.debug"
                  isAbstract="false"
                  name="Generate debug information for host code (-g)"
                  valueType="boolean">
               <enablement
                     attribute="defaultValue"
                     type="CONTAINER_ATTRIBUTE"
                     value="true">
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildType"
                        value="org.eclipse.cdt.build.core.buildType.debug">
                  </checkBuildProperty>
               </enablement>
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.debugging"
                  command="--generate-line-info"
                  id="com.nvidia.cuda.toolchain.compiler.lineinfo"
                  isAbstract="false"
                  name="Generate line-number information for device code (-lineinfo)"
                  valueType="boolean">
            </option>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  name="CUDA">
            </optionCategory>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_30,code=sm_30"
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_30_sass"
                  isAbstract="false"
                  name="Generate SM 3.0 SASS"
                  value="true"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_32_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_32,code=sm_32"
                  isAbstract="false"
                  name="Generate SM 3.2 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_35_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_35,code=sm_35"
                  isAbstract="false"
                  name="Generate SM 3.5 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_50_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_50,code=sm_50"
                  isAbstract="false"
                  name="Generate SM 5.0 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_53_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_53,code=sm_53"
                  isAbstract="false"
                  name="Generate SM 5.3 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_60_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_60,code=sm_60"
                  isAbstract="false"
                  name="Generate SM 6.0 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_61_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_61,code=sm_61"
                  isAbstract="false"
                  name="Generate SM 6.1 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_62_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_62,code=sm_62"
                  isAbstract="false"
                  name="Generate SM 6.2 SASS"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_70_sass"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_70,code=sm_70"
                  isAbstract="false"
                  name="Generate SM 7.0 SASS"
                  valueType="boolean">
            </option>

            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_30,code=compute_30"
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_30_ptx"
                  isAbstract="false"
                  name="Generate SM 3.0 PTX"
                  value="true"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_35_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_35,code=compute_35"
                  isAbstract="false"
                  name="Generate SM 3.5 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_50_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_50,code=compute_50"
                  isAbstract="false"
                  name="Generate SM 5.0 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_53_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_53,code=compute_53"
                  isAbstract="false"
                  name="Generate SM 5.3 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_60_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_60,code=compute_60"
                  isAbstract="false"
                  name="Generate SM 6.0 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_61_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_61,code=compute_61"
                  isAbstract="false"
                  name="Generate SM 6.1 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_62_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_62,code=compute_62"
                  isAbstract="false"
                  name="Generate SM 6.2 PTX"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.cuda.sm_70_ptx"
                  category="com.nvidia.cuda.toolchain.compiler.category.cuda"
                  command="-gencode arch=compute_70,code=compute_70"
                  isAbstract="false"
                  name="Generate SM 7.0 PTX"
                  valueType="boolean">
            </option>

            <optionCategory
                  id="com.nvidia.cuda.toolchain.compiler.category.misc"
                  name="Miscellaneous">
            </optionCategory>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.verbose"
                  category="com.nvidia.cuda.toolchain.compiler.category.misc"
                  command="--verbose"
                  isAbstract="false"
                  name="Verbose (-v)"
                  valueType="boolean">
            </option>
            <option
                  id="com.nvidia.cuda.toolchain.compiler.keep"
                  category="com.nvidia.cuda.toolchain.compiler.category.misc"
                  command="--keep"
                  isAbstract="false"
                  name="Keep all intermediate files (-keep)"
                  valueType="boolean">
            </option>
            <option
                  category="com.nvidia.cuda.toolchain.compiler.category.misc"
                  command="-Xcompiler -fPIC"
                  defaultValue="false"
                  id="nvcc.compiler.pic"
                  superClass="gnu.cpp.compiler.option.other.pic"
                  valueType="boolean">
               <enablement
                     attribute="value"
                     type="CONTAINER_ATTRIBUTE"
                     value="true">
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildArtefactType"
                        value="org.eclipse.cdt.build.core.buildArtefactType.sharedLib">
                  </checkBuildProperty>
               </enablement>
               <enablement
                     attribute="value"
                     type="CONTAINER_ATTRIBUTE"
                     value="false">
                     <not>
                  <checkBuildProperty
                        property="org.eclipse.cdt.build.core.buildArtefactType"
                        value="org.eclipse.cdt.build.core.buildArtefactType.sharedLib">
                  </checkBuildProperty>
                     </not>
               </enablement>
            </option>
            <option category="com.nvidia.cuda.toolchain.compiler.category.misc"
                command="-w" defaultValue="false"
                id="nvcc.compiler.disableWarnings" isAbstract="false"
                name="Inhibit all warning messages (-w)" 
                valueType="boolean">
            </option>
            <option
                  browseType="directory"
                  category="com.nvidia.cuda.toolchain.compiler.category.misc"
                  command="--keep-dir "
                  commandGenerator="com.nvidia.cuda.toolchain.internal.CommandSpaceValueGenerator"
                  id="com.nvidia.cuda.toolchain.compiler.keepdir"
                  isAbstract="false"
                  name="Intermediate files keep directory"
                  valueType="string">
            </option>
            <option
                  browseType="file"
                  category="com.nvidia.cuda.toolchain.compiler.category.misc"
                  command="-ccbin"
                  commandGenerator="com.nvidia.cuda.toolchain.internal.CommandSpaceValueGenerator"
                  defaultValue="${ccbin}"
                  id="com.nvidia.cuda.toolchain.compiler.ccbin"
                  isAbstract="false"
                  name="Path to the host compiler"
                  valueType="string">
            </option>
         </tool>
      </toolChain>
     <toolChain
           id="com.nvidia.cuda.toolchain.linuxToolchain"
           isAbstract="false"
           name="CUDA Linux toolchain"
           osList="linux"
           superClass="com.nvidia.cuda.toolchain.baseToolchain">
               <targetPlatform
               archList="all"
               binaryParser="org.eclipse.cdt.core.ELF"
               id="com.nvidia.cuda.toolchain.targetLinuxPlatform"
               isAbstract="false"
               osList="linux">
         </targetPlatform>
      </toolChain>
     <toolChain
           id="com.nvidia.cuda.toolchain.darwinToolchain"
           isAbstract="false"
           name="CUDA MacOS X toolchain"
           osList="macosx"
           superClass="com.nvidia.cuda.toolchain.baseToolchain">
               <targetPlatform
               archList="all"
               binaryParser="org.eclipse.cdt.core.MachO64"
               id="com.nvidia.cuda.toolchain.targetDarwinPlatform"
               isAbstract="false"
               osList="macosx">
         </targetPlatform>
      </toolChain>

      <!-- Executable project types -->
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.exe"
            id="com.nvidia.cuda.toolchain.linux.build.exe"
            isAbstract="false"
            isTest="false">
         <configuration
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.linux.configuration.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.debugToolchain"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
         <configuration
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.linux.configuration.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.releaseToolchain"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
      </projectType>
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.exe"
            id="com.nvidia.cuda.toolchain.darwin.build.exe"
            isAbstract="false"
            isTest="false">
         <configuration
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.darwin.configuration.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.debugToolchain"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
         <configuration
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.darwin.configuration.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.releaseToolchain"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
      </projectType>

      <!-- Shared library project types -->
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.sharedLib"
            id="com.nvidia.cuda.toolchain.linux.build.so"
            isAbstract="false"
            isTest="false">
         <configuration
               artifactExtension="so"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.linux.configuration.so.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.debugToolchain.so"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
         <configuration
               artifactExtension="so"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.linux.configuration.so.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.releaseToolchain.so"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
      </projectType>
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.sharedLib"
            id="com.nvidia.cuda.toolchain.darwin.build.so"
            isAbstract="false"
            isTest="false">
         <configuration
               artifactExtension="dylib"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.darwin.configuration.so.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.debugToolchain.so"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
         <configuration
               artifactExtension="dylib"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.darwin.configuration.so.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.releaseToolchain.so"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
      </projectType>

      <!-- Static library project types -->
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.staticLib"
            id="com.nvidia.cuda.toolchain.linux.build.ar"
            isAbstract="false"
            isTest="false">
         <configuration
               artifactExtension="a"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.linux.configuration.ar.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.debugToolchain.ar"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
         <configuration
               artifactExtension="a"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.linux.configuration.ar.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.linux.releaseToolchain.ar"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.linuxToolchain">
            </toolChain>
         </configuration>
      </projectType>
      <projectType
            buildArtefactType="org.eclipse.cdt.build.core.buildArtefactType.staticLib"
            id="com.nvidia.cuda.toolchain.darwin.build.ar"
            isAbstract="false"
            isTest="false">
         <configuration
               artifactExtension="a"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.debug"
               id="com.nvidia.cuda.toolchain.darwin.configuration.ar.debug"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Debug">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.debugToolchain.ar"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
         <configuration
               artifactExtension="a"
               buildProperties="org.eclipse.cdt.build.core.buildType=org.eclipse.cdt.build.core.buildType.release"
               id="com.nvidia.cuda.toolchain.darwin.configuration.ar.release"
               languageSettingsProviders="org.eclipse.cdt.ui.UserLanguageSettingsProvider;org.eclipse.cdt.core.ReferencedProjectsLanguageSettingsProvider;org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider;${Toolchain};"
               name="Release">
            <toolChain
                  id="com.nvidia.cuda.toolchain.darwin.releaseToolchain.ar"
                  isAbstract="false"
                  superClass="com.nvidia.cuda.toolchain.darwinToolchain">
            </toolChain>
         </configuration>
      </projectType>
   </extension>
   <extension
         point="org.eclipse.cdt.core.templateAssociations">
      <template
            id="org.eclipse.cdt.build.core.templates.HelloWorldCAnsiProject">
         <toolChain
               id="com.nvidia.cuda.toolchain.linuxToolchain">
         </toolChain>
         <toolChain
               id="com.nvidia.cuda.toolchain.darwinToolchain">
         </toolChain>
      </template>

      <template
            id="org.eclipse.cdt.build.core.templates.HelloWorldCCProject">
         <toolChain
               id="com.nvidia.cuda.toolchain.linuxToolchain">
         </toolChain>
         <toolChain
               id="com.nvidia.cuda.toolchain.darwinToolchain">
         </toolChain>
      </template>
   </extension>
   <extension
         id="id1"
         name="name"
         point="org.eclipse.cdt.core.ErrorParser">
      <errorparser
            id="com.nvidia.cuda.toolchain.nvccErrorParser"
            name="NVCC error parser">
         <pattern
               description-expr="$3"
               eat-processed-line="true"
               file-expr="$1"
               line-expr="$2"
               regex="(.*?)\((\d+)\): error: (identifier \&quot;(.*)\&quot; is undefined)"
               severity="Error"
               variable-expr="$4">
         </pattern>
         <pattern
               description-expr="$3"
               eat-processed-line="true"
               file-expr="$1"
               line-expr="$2"
               regex="(.*?)\((\d+)\): error: (.*)"
               severity="Error">
         </pattern>
      </errorparser>
   </extension>
   <extension
         id="com.nvidia.cuda.ide.build.NVCCPerProjectProfile"
         name="NVCC Scanner Info per project profile"
         point="org.eclipse.cdt.make.core.ScannerConfigurationDiscoveryProfile">
      <scannerInfoCollector
            class="com.nvidia.cuda.toolchain.internal.SICollector"
            scope="project"/>
      <buildOutputProvider>
         <open/>
         <scannerInfoConsoleParser class="org.eclipse.cdt.make.internal.core.scannerconfig.gnu.GCCScannerInfoConsoleParser"/>
      </buildOutputProvider>
      <scannerInfoProvider providerId="nvcc">
         <run
               command="${cuda_tk_bin}/nvcc"
               arguments="-dryrun"
               class="com.nvidia.cuda.toolchain.internal.NvccSpecsRunSIProvider"/>
         <scannerInfoConsoleParser class="com.nvidia.cuda.toolchain.internal.NvccSpecsConsoleParser"/>
      </scannerInfoProvider>
      <scannerInfoProvider providerId="specsFile">
         <run
               arguments="-E -Xcompiler -P -Xcompiler -v -Xcompiler -dD ${plugin_state_location}/${specs_file}"
               command="${cuda_tk_bin}/nvcc"
               class="com.nvidia.cuda.toolchain.internal.GccNvccRunSpecsProvider"/>
         <scannerInfoConsoleParser class="org.eclipse.cdt.make.internal.core.scannerconfig.gnu.GCCSpecsConsoleParser"/>
      </scannerInfoProvider>
   </extension>
   <extension
         point="org.eclipse.core.variables.dynamicVariables">
      <variable
            description="Path to NVCC compiler"
            name="nvcc"
            resolver="com.nvidia.cuda.toolchain.internal.DynamicVariableResolver"
            supportsArgument="false">
      </variable>
      <variable
            description="Path to the CC compiler binary"
            name="ccbin"
            resolver="com.nvidia.cuda.toolchain.internal.DynamicVariableResolver">
      </variable>
       <variable
               description="CUDA toolkit by name bin directory location"
               name="cuda_tk_bin"
               resolver="com.nvidia.cuda.toolchain.internal.DynamicVariableResolver"
               supportsArgument="true">
       </variable>
       <variable
               description="CUDA toolkit samples directory location"
               name="cuda_samples_dir"
               resolver="com.nvidia.cuda.toolchain.internal.DynamicVariableResolver"
               supportsArgument="true">
       </variable>
       <variable
               description="CUDA toolkit samples common libraries directory location"
               name="cuda_samples_common_lib_dir"
               resolver="com.nvidia.cuda.toolchain.internal.DynamicVariableResolver">
       </variable>
   </extension>
   <extension
         point="org.eclipse.core.contenttype.contentTypes">
      <content-type
            base-type="org.eclipse.cdt.core.cxxSource"
            file-extensions="cu"
            id="CUDAContentType"
            name="CUDA C Source File"
            priority="high">
      </content-type>
      <content-type
            base-type="org.eclipse.cdt.core.cxxHeader"
            file-extensions="h,cuh,hcu"
            id="CUDAHeaderContentType"
            name="CUDA Header Content Type"
            priority="high">
      </content-type>
   </extension>
   <extension
         point="org.eclipse.cdt.core.LanguageSettingsProvider">
      <provider
            class="com.nvidia.cuda.toolchain.internal.NVCCBuiltinSpecsDetector"
            id="com.nvidia.cuda.toolchain.CUDACProvider"
            name="Nvcc Builtins provider"
            parameter="${COMMAND} ${FLAGS} -E -Xcompiler -P -Xcompiler -v -Xcompiler -dD &quot;${INPUTS}&quot;"
            prefer-non-shared="true">
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.c">
         </language-scope>
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.c++">
         </language-scope>
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.cu">
         </language-scope>
      </provider>
   </extension>
   <!-- <extension
         point="org.eclipse.cdt.core.templates">
      <template
            id="com.nvidia.cuda.toolchain.saxpy.c.template"
            location="templates/saxpyc/template.xml"
            projectType="org.eclipse.cdt.build.core.buildArtefactType.exe">
      </template>
   </extension> -->
   <extension
         point="org.eclipse.ui.preferencePages">
      <page
            class="com.nvidia.cuda.toolchain.internal.CUDAToolkitPreferencePage"
            id="com.nvidia.cuda.toolchain.preferencePage"
            name="CUDA">
      </page>
   </extension>
   <extension
         point="org.eclipse.cdt.core.language">
      <language
            class="org.eclipse.cdt.core.dom.ast.gnu.cpp.GPPLanguage"
            id="com.nvidia.cuda.toolchain.CUDALangugage"
            name="CUDA">
         <contentType
               id="com.nvidia.cuda.toolchain.CUDAContentType">
         </contentType>
         <contentType
               id="com.nvidia.cuda.toolchain.CUDAHeaderContentType">
         </contentType>
      </language>
   </extension>
   <extension
         point="org.eclipse.ui.propertyPages">
      <page
            category="org.eclipse.cdt.managedbuilder.ui.properties.Page_head_build"
            class="com.nvidia.cuda.toolchain.internal.CUDAToolkitPropertyPage"
            id="com.nvidia.cuda.toolchain.propertyPage"
            name="CUDA Toolkit">
         <filter
               name="projectNature"
               value="org.eclipse.cdt.core.cnature">
         </filter>
         <enabledWhen>
            <adapt
                  type="org.eclipse.core.resources.IProject">
            </adapt>
         </enabledWhen>
      </page>
   </extension>

</plugin>
15knots commented 6 years ago 
         point="org.eclipse.cdt.core.LanguageSettingsProvider">
      <provider
            class="com.nvidia.cuda.toolchain.internal.NVCCBuiltinSpecsDetector"
            id="com.nvidia.cuda.toolchain.CUDACProvider"
            name="Nvcc Builtins provider"
            parameter="${COMMAND} ${FLAGS} -E -Xcompiler -P -Xcompiler -v -Xcompiler -dD &quot;${INPUTS}&quot;"
            prefer-non-shared="true">
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.c">
         </language-scope>
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.c++">
         </language-scope>
         <language-scope
               id="com.nvidia.cuda.toolchain.language.cuda.cu">

This is exactly the language ID I was looking for ^^^^^. The cmake4eclispe version you tried has com.nvidia.cuda.toolchain.language.cu.

Please ry the new version at URL https://bintray.com/15knots/p2-zip/download_file?file_path=cmake4eclipse-1.10.0.zip.

caselitz commented 6 years ago

You mean com.nvidia.cuda.toolchain.CUDACProvider?

Please ry the new version at URL https://bintray.com/15knots/p2-zip/download_file?file_path=cmake4eclipse-1.10.0.zip.

Unfortunately still not working, i.e., CMAKE_EXPORT_COMPILE_COMMANDS Parser only appears for GNU C and GNU C++ but not for com.nvidia.cuda.toolchain.language.cuda.c/cu/c++.

15knots commented 6 years ago

Add the nvcc compiler to the Tool Chain: used tools list.

caselitz commented 6 years ago

I have. All the used tools from current toolchain Linux GCC plus NVVC linker, compiler, archiver.

Otherwise the three entries in the language column com.nvidia.cuda.toolchain.language.cuda.c/cu/c++. would also not appear.

15knots commented 6 years ago

I try to reproduce locally and installed a recent version of NSight and cmake. But this version of NSight has no file com.nvidia.cuda.toolchain_9.1.0.201711011803.jar but com.nvidia.cuda.ide.toolchains_9.1.0.201711040225.jar, which is a different thing.

So it looks NVidia removed its own language settings provider (NVCCBuiltinSpecsDetector).

Anyway, I tried a different solution:

  1. removed all providers in the 'Providers' tab,
  2. added the 'CMake Build Output Parser'
  3. delete the build directory and build the project.

The 'CMake Build Output Parser' recognizes nvcc as a C-compiler and send the appropriate language ID to the indexer. In my case, an include directory /usr/local/cuda-9.1/include/ was showing up, although the CU source editor is still complaining about an unresolved inclusion on <cuda.h>.

caselitz commented 6 years ago

I try to reproduce locally and installed a recent version of NSight and cmake. But this version of NSight has no file com.nvidia.cuda.toolchain_9.1.0.201711011803.jar but com.nvidia.cuda.ide.toolchains_9.1.0.201711040225.jar, which is a different thing.

So it looks NVidia removed its own language settings provider (NVCCBuiltinSpecsDetector).

I don't think so. As I pointed out before, there are two different things:

  1. The Nsight Eclipse Edition, which is an complete Eclipse version modified by NVIDIA and shipped with CUDA. In this way they are providing their debugging/profiling tools for quite a while. I guess that's the one you installed?
  2. The Nsight Eclipse Plugin, which is only a Plugin that can be installed in any vanilla Eclipse version. I think that is quite new (first time provided with CUDA 9.0), but the better way in my opinion. So that's what I have.

Probably since you use 1 and I use 2, there are different files. And NVIDIA might use different ways to provide the settings because they can modify the whole source code in 1 while in 2 they need to rely on what can be realized with an Eclipse plugin. So I don't think they removed the NVCCBuiltinSpecsDetector, but it is only existent in 2 but is not (was never) in 1.

caselitz commented 6 years ago

The Nvcc Builtins provider works fine for me. CUDA headers get resolved in the CU source editor.

Whats not working is that other library includes (include path specified via cmake) get resolved, probably because the CMAKE_EXPORT_COMPILE_COMMANDS Parser does only appear for the GNU C/C++ and not for the [ Unspecified ] [com.nvidia.cuda.toolchain.language.cuda.c/cu/c++] Languages in the Entries tab.

15knots commented 6 years ago

If the 'CMake Build Output Parser' works for you as expected, I know which launguage ID to pass to the indexer.

OTOH, i noticed that if I add CUDA as a language in cmakelists.txt, cmake prints a warning Manually-specified variables were not used by the project: CMAKE_EXPORT_COMPILE_COMMANDS and no compile_commands.json file gets generated, which effectively disables the parser.

caselitz commented 6 years ago

If the 'CMake Build Output Parser' works for you as expected

It does NOT. Looks like the 'CMake Build Output Parser (deprecated)' is showing the same behavior as the 'CMAKE_EXPORT_COMPILE_COMMANDS Parser' - they

only appear for the GNU C/C++ and not for the [ Unspecified ] [com.nvidia.cuda.toolchain.language.cuda.c/cu/c++] Languages in the Entries tab.

What works for me is the 'Nvcc Builtins provider' - it appears for the [ Unspecified ] [com.nvidia.cuda.toolchain.language.cuda.c/cu/c++] Languages in the Entries tab (but not for the GNU C/C++, which should be fine as it only provides Cuda specific stuff).

OTOH, i noticed that if I add CUDA as a language in cmakelists.txt, cmake prints a warning Manually-specified variables were not used by the project: CMAKE_EXPORT_COMPILE_COMMANDS and no compile_commands.json file gets generated, which effectively disables the parser.

I do not have this behavior (having project(projectname CXX CUDA) in my CMakeLists.txt) - I don't get the error and the compile_commands.json gets properly generated.

Are you using the Nsight Eclipse Edition or the Nsight Eclipse Plugin (which I use)? If we are using different things, that might explain different behavior.

15knots commented 6 years ago

I am using Nsight Eclipse Plugin . My cmakelists alos had find_package(CUDA REQUIRED), after removing that, the compile_commands.json gets generated.

I changed the parser to recognize nvcc as a C-compiler (like gcc). Now include paths defined in CMakeLists.txt show up in the Includes folder in the Project Explorer view and in the Entries tab (as the GNU C language). Would this satisfy your needs?

caselitz commented 6 years ago

Sounds good. Can you provide a test-cmake4eclipse.zip again, so I can test it on Monday? Thanks for your effort!

15knots commented 6 years ago

BTW:

Whats not working is that other library includes (include path specified via cmake) get resolved, probably because the CMAKE_EXPORT_COMPILE_COMMANDS Parser does only appear for the GNU C/C++ and not for the [ Unspecified ] [com.nvidia.cuda.toolchain.language.cuda.c/cu/c++] Languages in the Entries tab.

From my point of view, the Nvcc Builtins provider seem to be broken somehow, it should not show these [ Unspecified ] entries but a human readable name instead. So I would not care too much, if the include path specified via cmake do not show up there. But regardless what is shown in the Entires tab, the editor should work with include paths specified via cmake if you press F3 (open declaration and it shold know about preprocessor define (no grey background in comditionally compiled code section).

I created a sample project to verify that the editor is working as expected: https://github.com/15knots/cmake4eclipse-sample-projects

15knots commented 6 years ago

Playing around with the sample project above and changing the language ID I found the following.

Language ID 'GCC'

Screenshot showing CUDA editor languageid-gcc-in_cuda_editor

Language ID 'CUDA'

Screenshot showing CUDA editor languageid-cuda-in_cuda_editor

So I guess, the CUDA language ID is what we need, regardless what the Entries tab is showing.

The plugin version on bintray uses the CUDA language ID. Please verify my findings.

caselitz commented 6 years ago

Yes, the new version on bintray works. Nice, thank you! :)

Now the CMAKE_EXPORT_COMPILE_COMMANDS Parser appears for the [ Unspecified ] [com.nvidia.cuda.toolchain.language.cuda.c/cu/c++] languages in the Entries tab and for ...cuda.cu it also displays the include path etc.

However, in my project not all include paths work. I looked at the compile_commands.json and I guess the problem is related to parsing it. While it works for -I<path>, it does not for -isystem=<path>. Looks like cmake generates -isystem <path> for c++ and -isystem=<path> for nvcc. Maybe fixing this can be combined with https://github.com/15knots/cmake4eclipse/pull/67.

caselitz commented 6 years ago

Quick question: I saw that you merged https://github.com/15knots/cmake4eclipse/pull/67, but does this directly resolve the issue I mentioned? I'm not sure how the parsing works, but I'm afraid that the = will end up in the include path?

15knots commented 6 years ago

No. Its just to allow whitespace between -isystem and /inc/path. File an issue for that (nvcc specific).