Suport passing arbitrary llvm arguments (#271)

- Support for passing LLVM command line options via the prcoess input or
providing one or more `--llvm-arg='..'` resolc CLI flag. This allows
more fine-grained control over the LLVM backend configuration.
- Make LLVM initialization idempotent.

---------

Signed-off-by: Cyrill Leutwiler <bigcyrill@hotmail.com>

crates/llvm-context/src/lib.rs

@@ -1,9 +1,7 @@
 //! The LLVM context library.
 
-pub(crate) mod debug_config;
-pub(crate) mod optimizer;
-pub(crate) mod polkavm;
-pub(crate) mod target_machine;
+use std::ffi::CString;
+use std::sync::OnceLock;
 
 pub use self::debug_config::ir_type::IRType as DebugConfigIR;
 pub use self::debug_config::DebugConfig;
@@ -74,9 +72,41 @@ pub use self::polkavm::WriteLLVM as PolkaVMWriteLLVM;
 pub use self::target_machine::target::Target;
 pub use self::target_machine::TargetMachine;
 
-/// Initializes the target machine.
-pub fn initialize_target(target: Target) {
+pub(crate) mod debug_config;
+pub(crate) mod optimizer;
+pub(crate) mod polkavm;
+pub(crate) mod target_machine;
+
+static DID_INITIALIZE: OnceLock<()> = OnceLock::new();
+
+/// Initializes the LLVM compiler backend.
+///
+/// This is a no-op if called subsequentially.
+///
+/// `llvm_arguments` are passed as-is to the LLVM CL options parser.
+pub fn initialize_llvm(target: Target, name: &str, llvm_arguments: &[String]) {
+    let Ok(_) = DID_INITIALIZE.set(()) else {
+        return; // Tests don't go through a recursive process
+    };
+
+    let argv = [name.to_string()]
+        .iter()
+        .chain(llvm_arguments)
+        .map(|arg| CString::new(arg.as_bytes()).unwrap())
+        .collect::<Vec<_>>();
+    let argv: Vec<*const libc::c_char> = argv.iter().map(|arg| arg.as_ptr()).collect();
+    let overview = CString::new("").unwrap();
+    unsafe {
+        inkwell::llvm_sys::support::LLVMParseCommandLineOptions(
+            argv.len() as i32,
+            argv.as_ptr(),
+            overview.as_ptr(),
+        );
+    }
+
+    inkwell::support::enable_llvm_pretty_stack_trace();
+
     match target {
-        Target::PVM => self::polkavm::initialize_target(),
+        Target::PVM => inkwell::targets::Target::initialize_riscv(&Default::default()),
     }
 }

crates/llvm-context/src/polkavm/context/mod.rs

@@ -83,6 +83,8 @@ where
     current_function: Option<Rc<RefCell<Function<'ctx>>>>,
     /// The loop context stack.
     loop_stack: Vec<Loop<'ctx>>,
+    /// The extra LLVM arguments that were used during target initialization.
+    llvm_arguments: &'ctx [String],
 
     /// The project dependency manager. It can be any entity implementing the trait.
     /// The manager is used to get information about contracts and their dependencies during
@@ -223,6 +225,7 @@ where
         dependency_manager: Option<D>,
         include_metadata_hash: bool,
         debug_config: DebugConfig,
+        llvm_arguments: &'ctx [String],
     ) -> Self {
         Self::set_data_layout(llvm, &module);
         Self::link_stdlib_module(llvm, &module);
@@ -250,6 +253,7 @@ where
             functions: HashMap::with_capacity(Self::FUNCTIONS_HASHMAP_INITIAL_CAPACITY),
             current_function: None,
             loop_stack: Vec::with_capacity(Self::LOOP_STACK_INITIAL_CAPACITY),
+            llvm_arguments,
 
             dependency_manager,
             include_metadata_hash,
@@ -639,6 +643,7 @@ where
                     self.optimizer.settings().to_owned(),
                     self.include_metadata_hash,
                     self.debug_config.clone(),
+                    self.llvm_arguments,
                 )
             })
     }

crates/llvm-context/src/polkavm/context/tests.rs

@@ -10,12 +10,20 @@ pub fn create_context(
     llvm: &inkwell::context::Context,
     optimizer_settings: OptimizerSettings,
 ) -> Context<DummyDependency> {
-    crate::polkavm::initialize_target();
+    crate::initialize_llvm(crate::Target::PVM, "resolc", Default::default());
 
     let module = llvm.create_module("test");
     let optimizer = Optimizer::new(optimizer_settings);
 
-    Context::<DummyDependency>::new(llvm, module, optimizer, None, true, Default::default())
+    Context::<DummyDependency>::new(
+        llvm,
+        module,
+        optimizer,
+        None,
+        true,
+        Default::default(),
+        Default::default(),
+    )
 }
 
 #[test]

crates/llvm-context/src/polkavm/mod.rs

@@ -17,11 +17,6 @@ use sha3::Digest;
 use self::context::build::Build;
 use self::context::Context;
 
-/// Initializes the PolkaVM target machine.
-pub fn initialize_target() {
-    inkwell::targets::Target::initialize_riscv(&Default::default());
-}
-
 /// Builds PolkaVM assembly text.
 pub fn build_assembly_text(
     contract_path: &str,
@@ -94,6 +89,7 @@ pub trait Dependency {
         optimizer_settings: OptimizerSettings,
         include_metadata_hash: bool,
         debug_config: DebugConfig,
+        llvm_arguments: &[String],
     ) -> anyhow::Result<String>;
 
     /// Resolves a full contract path.
@@ -114,6 +110,7 @@ impl Dependency for DummyDependency {
         _optimizer_settings: OptimizerSettings,
         _include_metadata_hash: bool,
         _debug_config: DebugConfig,
+        _llvm_arguments: &[String],
     ) -> anyhow::Result<String> {
         Ok(String::new())
     }