feat: Add initial support for CUDA (#433)

Author: varungandhi-src

docs/Development.md

@@ -11,6 +11,7 @@
   - [Stacktraces](#stacktraces)
   - [Attaching a debugger](#attaching-a-debugger)
   - [Debugging on Linux](#debugging-on-linux)
+    - [Testing CUDA/NVCC support](#testing-cudanvcc-support)
   - [Inspecting Clang ASTs](#inspecting-clang-asts)
   - [Automated test case reduction](#automated-test-case-reduction)
   - [Debugging preprocessor issues](#debugging-preprocessor-issues)
@@ -162,6 +163,14 @@ There is a [VM setup script](/tools/vm-setup.sh) available
 to configure a GCP VM for building scip-clang.
 We recommend using Ubuntu 20.04+ with 16 cores or more.
 
+#### Testing CUDA/NVCC support
+
+There is a [CUDA-specific VM setup script](/tools/vm-setup-cuda.sh)
+which installs the CUDA SDK. Use it in a GCP VM
+which has a GPU attached.
+
+You may need to restart your shell for changes to take effect.
+
 ### Inspecting Clang ASTs
 
 Print the AST nodes:

docs/IndexingProjects.md

@@ -6,6 +6,7 @@
 - [Redpanda](#redpanda)
 - [Postgres](#postgres)
 - [Boost](#boost)
+- [Apache MXNet](#apache-mxnet)
 
 ## scip-clang
 
@@ -224,3 +225,25 @@ for d in dirs:
         print("Indexing failed for {}; skipping upload".format(d))
         continue
 ```
+
+## Apache MXNet
+
+Apache MXNet is an interesting project to test because
+it has over 100 CUDA files.
+
+The indexing steps below are based on the [official instructions](https://mxnet.apache.org/versions/1.9.1/get_started/build_from_source.html).
+
+```
+git clone --recursive https://github.com/apache/mxnet
+cd mxnet
+sudo apt-get update
+sudo apt-get install -y build-essential git ninja-build ccache libopenblas-dev libopencv-dev cmake gfortran clang++-14 clang libomp-dev
+```
+
+```
+cmake -B build -G Ninja -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DUSE_CUDA=ON
+
+cat build/compile_commands.json | jq --arg rdir "$(clang -print-resource-dir)" 'map(if .command | contains("nvcc") then .command += " -resource-dir " + $rdir else . end)' > build/modified.json
+
+scip-clang --compdb-path build/modified.json
+```

indexer/CompilationDatabase.cc

@@ -118,6 +118,38 @@ struct GccToolchainInfo : ToolchainInfo {
   }
 };
 
+struct NvccToolchainInfo : ToolchainInfo {
+  scip_clang::AbsolutePath cudaDir;
+
+  NvccToolchainInfo(scip_clang::AbsolutePath cudaDir)
+      : ToolchainInfo(), cudaDir(cudaDir) {}
+
+  virtual CompilerKind kind() const override {
+    return CompilerKind::Nvcc;
+  }
+
+  virtual bool isWellFormed() const override {
+    auto path = scip_clang::joinPath(cudaDir.asStringRef(), "include");
+    if (!std::filesystem::exists(path)) {
+      spdlog::error(
+          "directory '{}' does not exist; expected to find CUDA SDK headers"
+          " there because nvcc was found at {}",
+          path,
+          scip_clang::joinPath(cudaDir.asStringRef(),
+                               scip_clang::joinPath("bin", "nvcc")));
+      return false;
+    }
+    return true;
+  }
+
+  virtual void
+  adjustCommandLine(std::vector<std::string> &commandLine) const override {
+    commandLine.push_back(
+        fmt::format("-isystem{}{}include", this->cudaDir.asStringRef(),
+                    std::filesystem::path::preferred_separator));
+  }
+};
+
 } // namespace
 
 static CompletedProcess runProcess(std::vector<std::string> &args,
@@ -231,7 +263,20 @@ ToolchainInfo::infer(const scip_clang::AbsolutePath &compilerPath) {
                                               findSearchDirsInvocation);
   }
 
-  spdlog::warn("compiler at '{}' is not one of clang/clang++/gcc/g++",
+  std::vector<std::string> argv = {compilerPath.asStringRef(), "--version"};
+  auto compilerVersionResult = ::runProcess(argv, "checking for NVCC");
+  if (compilerVersionResult.isSuccess()
+      && !compilerVersionResult.stdoutLines.empty()
+      && absl::StrContains(compilerVersionResult.stdoutLines[0], "NVIDIA")) {
+    if (auto binDir = compilerPath.asRef().prefix()) {
+      if (auto cudaDir = binDir->prefix()) {
+        return std::make_unique<NvccToolchainInfo>(
+            scip_clang::AbsolutePath(*cudaDir));
+      }
+    }
+  }
+
+  spdlog::warn("compiler at '{}' is not one of clang/clang++/gcc/g++/nvcc",
                compilerPath.asStringRef());
   noteStdlib();
   return failure;
@@ -693,8 +738,12 @@ void ResumableParser::initialize(compdb::File compdb, ParseOptions options) {
   this->reader.IterativeParseInit();
   this->options = options;
   std::vector<std::string> extensions;
-  // Via https://stackoverflow.com/a/3223792/2682729
-  for (auto ext : {"c", "C", "cc", "cpp", "CPP", "cxx", "c++"}) {
+  // clang-format off
+  // Via https://stackoverflow.com/a/3223792/2682729 (for C and C++)
+  // For CUDA, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#basics-cdp1
+  // and https://github.com/github-linguist/linguist/blob/master/lib/linguist/languages.yml#L1342-L1346
+  // clang-format on
+  for (auto ext : {"c", "C", "cc", "cpp", "CPP", "cxx", "c++", "cu"}) {
     extensions.emplace_back(llvm::Regex::escape(fmt::format(".{}", ext)));
   };
   this->fileExtensionRegex =

indexer/CompilationDatabase.h

@@ -108,6 +108,7 @@ class CommandObjectHandler
 enum class CompilerKind {
   Gcc,
   Clang,
+  Nvcc,
 };
 
 struct ToolchainInfo {

tools/vm-setup-cuda.sh

@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+sudo dpkg -i cuda-keyring_1.1-1_all.deb
+rm -rf cuda-keyring_1.1-1_all.deb
+sudo apt-get update
+sudo apt-get -y install cuda
+
+ZSH_INIT="$HOME/.zshrc"
+if [ -f "$ZSH_INIT" ] && ! grep -q "cuda" "$ZSH_INIT"; then
+    echo 'export PATH="/usr/local/cuda/bin:$PATH"' >> "$ZSH_INIT"
+    echo "Added CUDA path to $ZSH_INIT"
+fi
+
+BASH_INIT="$HOME/.bashrc"
+if [ -f "$BASH_INIT" ] && ! grep -q "cuda" "$BASH_INIT"; then
+    echo 'export PATH="/usr/local/cuda/bin:$PATH"' >> "$BASH_INIT"
+    echo "Added CUDA path to $BASH_INIT"
+fi
+
+FISH_INIT="$HOME/.config/fish/config.fish"
+if [ -f "$FISH_INIT" ] && ! grep -q "cuda" "$FISH_INIT"; then
+    echo "fish_add_path /usr/local/cuda/bin" >> "$FISH_INIT"
+    echo "Added CUDA path to $FISH_INIT"
+fi