feat: embed a statically linked LLVM execution engine

Add EMBEDDED mode to run generated modules in-process without lli or PATH
lookup. Link LLVM and support archives plus the C++ runtime statically,
while retaining the system C runtime on Linux.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

3 files changed, 61 insertions, 1 deletions

diff --git a/tools/driver/EmbeddedRunner.cc b/tools/driver/EmbeddedRunner.cc
new file mode 100644
index 0000000..81a2265
--- /dev/null
+++ b/tools/driver/EmbeddedRunner.cc
@@ -0,0 +1,45 @@
+#include "EmbeddedRunner.h"
+#include "cps/CodeGen.h"
+#include "llvm/ExecutionEngine/ExecutionEngine.h"
+#include "llvm/ExecutionEngine/MCJIT.h"
+#include "llvm/ExecutionEngine/SectionMemoryManager.h"
+#include "llvm/Support/DynamicLibrary.h"
+#include "llvm/Support/TargetSelect.h"
+#include "llvm/TargetParser/Host.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+
+int runEmbedded(const cps::CodeGen &CG) {
+    if (llvm::InitializeNativeTarget() || llvm::InitializeNativeTargetAsmPrinter()) {
+        llvm::errs() << "Error: cannot initialize the embedded LLVM target\n";
+        return 1;
+    }
+    std::string Error;
+    if (llvm::sys::DynamicLibrary::LoadLibraryPermanently(nullptr, &Error)) {
+        llvm::errs() << "Error: cannot load process symbols: " << Error << "\n";
+        return 1;
+    }
+    // Keep CodeGen's module and context alive while the engine owns its clone.
+    auto Module = llvm::CloneModule(CG.getModule());
+    Module->setTargetTriple(llvm::sys::getProcessTriple());
+    llvm::EngineBuilder Builder(std::move(Module));
+    Builder.setEngineKind(llvm::EngineKind::JIT);
+    Builder.setErrorStr(&Error);
+    Builder.setMCJITMemoryManager(std::make_unique<llvm::SectionMemoryManager>());
+    std::unique_ptr<llvm::ExecutionEngine> Engine(Builder.create());
+    if (!Engine) {
+        llvm::errs() << "Error: cannot create the embedded LLVM engine: " << Error << "\n";
+        return 1;
+    }
+    Engine->finalizeObject();
+    if (Engine->hasError()) {
+        llvm::errs() << "Error: embedded LLVM execution failed: " << Engine->getErrorMessage() << "\n";
+        return 1;
+    }
+    auto MainAddress = Engine->getFunctionAddress("main");
+    if (!MainAddress) {
+        llvm::errs() << "Error: embedded LLVM could not resolve main\n";
+        return 1;
+    }
+    auto Main = reinterpret_cast<int (*)()>(MainAddress);
+    return Main();
+}
diff --git a/tools/driver/EmbeddedRunner.h b/tools/driver/EmbeddedRunner.h
new file mode 100644
index 0000000..1651f4d
--- /dev/null
+++ b/tools/driver/EmbeddedRunner.h
@@ -0,0 +1,5 @@
+#pragma once
+
+namespace cps { class CodeGen; }
+
+int runEmbedded(const cps::CodeGen &CG);
diff --git a/tools/driver/main.cc b/tools/driver/main.cc
index f5373a0..4df8562 100644
--- a/tools/driver/main.cc
+++ b/tools/driver/main.cc
@@ -1,6 +1,9 @@
 #include "cps/Lexer.h"
 #include "cps/Parser.h"
 #include "cps/CodeGen.h"
+#ifdef CPSC_RUN_EMBEDDED
+#include "EmbeddedRunner.h"
+#endif
 #include "llvm/ADT/SmallString.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/FileUtilities.h"
@@ -48,7 +51,9 @@ int main(int argc, char **argv) {
         llvm::errs() << "Error: no source file specified\n";
         return 1;
     }
+#ifdef CPSC_RUN_PATH
     std::string LliPath;
+#endif
     if (Run) {
 #ifdef CPSC_RUN_PATH
         auto Lli = llvm::sys::findProgramByName("lli");
@@ -57,7 +62,7 @@ int main(int argc, char **argv) {
             return 1;
         }
         LliPath = *Lli;
-#else
+#elif defined(CPSC_RUN_NONE)
         llvm::errs() << "Error: -r is disabled in this build (CPSC_RUN_MODE=NONE)\n";
         return 1;
 #endif
@@ -93,6 +98,9 @@ int main(int argc, char **argv) {
 
     if (CG.hadError()) return 1;
 
+#ifdef CPSC_RUN_EMBEDDED
+    if (Run && OutputPath.empty()) return runEmbedded(CG);
+#endif
     std::error_code EC;
     int OutputFD;
     llvm::FileRemover TemporaryOutput;
@@ -137,6 +145,8 @@ int main(int argc, char **argv) {
         return 1;
     }
     return Result;
+#elif defined(CPSC_RUN_EMBEDDED)
+    return runEmbedded(CG);
 #else
     return 0;
 #endif