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>

CHANGELOG.md

@@ -10,6 +10,7 @@ Supported `polkadot-sdk` rev:`c29e72a8628835e34deb6aa7db9a78a2e4eabcee`
 - Support for solc v0.8.29
 - Decouples the solc JSON-input-output type definitions from the Solidity fronted and expose them via a dedicated crate.
 - `--supported-solc-versions` for `resolc` binary to return a `semver` range of supported `solc` versions.
+- 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.
 
 ### Changed
 - Runner `resolc` using webkit is no longer supported.

Cargo.lock

@@ -8368,6 +8368,7 @@ dependencies = [
  "hex",
  "inkwell",
  "itertools 0.14.0",
+ "libc",
  "num",
  "polkavm-common 0.21.0",
  "polkavm-disassembler",

crates/llvm-context/Cargo.toml

@@ -22,6 +22,7 @@ num = { workspace = true }
 hex = { workspace = true }
 sha3 = { workspace = true }
 inkwell = { workspace = true }
+libc = { workspace = true }
 polkavm-disassembler = { workspace = true }
 polkavm-common = { workspace = true }
 

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())
     }

crates/solidity/src/lib.rs

@@ -54,6 +54,7 @@ pub fn yul<T: Compiler>(
     optimizer_settings: revive_llvm_context::OptimizerSettings,
     include_metadata_hash: bool,
     debug_config: revive_llvm_context::DebugConfig,
+    llvm_arguments: &[String],
 ) -> anyhow::Result<Build> {
     let path = match input_files.len() {
         1 => input_files.first().expect("Always exists"),
@@ -74,7 +75,12 @@ pub fn yul<T: Compiler>(
     let solc_validator = Some(&*solc);
     let project = Project::try_from_yul_path(path, solc_validator)?;
 
-    let build = project.compile(optimizer_settings, include_metadata_hash, debug_config)?;
+    let build = project.compile(
+        optimizer_settings,
+        include_metadata_hash,
+        debug_config,
+        llvm_arguments,
+    )?;
 
     Ok(build)
 }
@@ -85,6 +91,7 @@ pub fn llvm_ir(
     optimizer_settings: revive_llvm_context::OptimizerSettings,
     include_metadata_hash: bool,
     debug_config: revive_llvm_context::DebugConfig,
+    llvm_arguments: &[String],
 ) -> anyhow::Result<Build> {
     let path = match input_files.len() {
         1 => input_files.first().expect("Always exists"),
@@ -97,7 +104,12 @@ pub fn llvm_ir(
 
     let project = Project::try_from_llvm_ir_path(path)?;
 
-    let build = project.compile(optimizer_settings, include_metadata_hash, debug_config)?;
+    let build = project.compile(
+        optimizer_settings,
+        include_metadata_hash,
+        debug_config,
+        llvm_arguments,
+    )?;
 
     Ok(build)
 }
@@ -118,6 +130,7 @@ pub fn standard_output<T: Compiler>(
     remappings: Option<BTreeSet<String>>,
     suppressed_warnings: Option<Vec<ResolcWarning>>,
     debug_config: revive_llvm_context::DebugConfig,
+    llvm_arguments: &[String],
 ) -> anyhow::Result<Build> {
     let solc_version = solc.version()?;
 
@@ -171,7 +184,12 @@ pub fn standard_output<T: Compiler>(
         &debug_config,
     )?;
 
-    let build = project.compile(optimizer_settings, include_metadata_hash, debug_config)?;
+    let build = project.compile(
+        optimizer_settings,
+        include_metadata_hash,
+        debug_config,
+        llvm_arguments,
+    )?;
 
     Ok(build)
 }
@@ -184,6 +202,7 @@ pub fn standard_json<T: Compiler>(
     include_paths: Vec<String>,
     allow_paths: Option<String>,
     debug_config: revive_llvm_context::DebugConfig,
+    llvm_arguments: &[String],
 ) -> anyhow::Result<()> {
     let solc_version = solc.version()?;
 
@@ -226,7 +245,12 @@ pub fn standard_json<T: Compiler>(
         let missing_libraries = project.get_missing_libraries();
         missing_libraries.write_to_standard_json(&mut solc_output, &solc_version)?;
     } else {
-        let build = project.compile(optimizer_settings, include_metadata_hash, debug_config)?;
+        let build = project.compile(
+            optimizer_settings,
+            include_metadata_hash,
+            debug_config,
+            llvm_arguments,
+        )?;
         build.write_to_standard_json(&mut solc_output, &solc_version)?;
     }
     serde_json::to_writer(std::io::stdout(), &solc_output)?;
@@ -252,6 +276,7 @@ pub fn combined_json<T: Compiler>(
     debug_config: revive_llvm_context::DebugConfig,
     output_directory: Option<PathBuf>,
     overwrite: bool,
+    llvm_arguments: &[String],
 ) -> anyhow::Result<()> {
     let build = standard_output(
         input_files,
@@ -267,6 +292,7 @@ pub fn combined_json<T: Compiler>(
         remappings,
         suppressed_warnings,
         debug_config,
+        llvm_arguments,
     )?;
 
     let mut combined_json = solc.combined_json(input_files, format.as_str())?;

crates/solidity/src/process/input.rs

@@ -20,6 +20,8 @@ pub struct Input {
     pub optimizer_settings: revive_llvm_context::OptimizerSettings,
     /// The debug output config.
     pub debug_config: revive_llvm_context::DebugConfig,
+    /// The extra LLVM arguments give used for manual control.
+    pub llvm_arguments: Vec<String>,
 }
 
 impl Input {
@@ -30,6 +32,7 @@ impl Input {
         include_metadata_hash: bool,
         optimizer_settings: revive_llvm_context::OptimizerSettings,
         debug_config: revive_llvm_context::DebugConfig,
+        llvm_arguments: Vec<String>,
     ) -> Self {
         Self {
             contract,
@@ -37,6 +40,7 @@ impl Input {
             include_metadata_hash,
             optimizer_settings,
             debug_config,
+            llvm_arguments,
         }
     }
 }

crates/solidity/src/process/mod.rs

@@ -37,11 +37,19 @@ pub trait Process {
         }
 
         let input: Input = revive_common::deserialize_from_slice(buffer.as_slice())?;
+
+        revive_llvm_context::initialize_llvm(
+            revive_llvm_context::Target::PVM,
+            crate::DEFAULT_EXECUTABLE_NAME,
+            &input.llvm_arguments,
+        );
+
         let result = input.contract.compile(
             input.project,
             input.optimizer_settings,
             input.include_metadata_hash,
             input.debug_config,
+            &input.llvm_arguments,
         );
 
         match result {

crates/solidity/src/project/contract/metadata.rs

@@ -18,6 +18,8 @@ pub struct Metadata {
     pub revive_version: String,
     /// The PolkaVM compiler optimizer settings.
     pub optimizer_settings: revive_llvm_context::OptimizerSettings,
+    /// The extra LLVM arguments give used for manual control.
+    pub llvm_arguments: Vec<String>,
 }
 
 impl Metadata {
@@ -27,6 +29,7 @@ impl Metadata {
         solc_version: String,
         revive_pallet_version: Option<semver::Version>,
         optimizer_settings: revive_llvm_context::OptimizerSettings,
+        llvm_arguments: Vec<String>,
     ) -> Self {
         Self {
             solc_metadata,
@@ -34,6 +37,7 @@ impl Metadata {
             revive_pallet_version,
             revive_version: ResolcVersion::default().long,
             optimizer_settings,
+            llvm_arguments,
         }
     }
 }

crates/solidity/src/project/contract/mod.rs

@@ -77,6 +77,7 @@ impl Contract {
         optimizer_settings: revive_llvm_context::OptimizerSettings,
         include_metadata_hash: bool,
         debug_config: revive_llvm_context::DebugConfig,
+        llvm_arguments: &[String],
     ) -> anyhow::Result<ContractBuild> {
         let llvm = inkwell::context::Context::create();
         let optimizer = revive_llvm_context::Optimizer::new(optimizer_settings);
@@ -89,6 +90,7 @@ impl Contract {
             version.long.clone(),
             version.l2_revision.clone(),
             optimizer.settings().to_owned(),
+            llvm_arguments.to_vec(),
         );
         let metadata_json = serde_json::to_value(&metadata).expect("Always valid");
         let metadata_hash: Option<[u8; revive_common::BYTE_LENGTH_WORD]> = if include_metadata_hash
@@ -120,6 +122,7 @@ impl Contract {
             Some(project),
             include_metadata_hash,
             debug_config,
+            llvm_arguments,
         );
         context.set_solidity_data(revive_llvm_context::PolkaVMContextSolidityData::default());
         match self.ir {

crates/solidity/src/project/mod.rs

@@ -66,6 +66,7 @@ impl Project {
         optimizer_settings: revive_llvm_context::OptimizerSettings,
         include_metadata_hash: bool,
         debug_config: revive_llvm_context::DebugConfig,
+        llvm_arguments: &[String],
     ) -> anyhow::Result<Build> {
         let project = self.clone();
         #[cfg(feature = "parallel")]
@@ -81,6 +82,7 @@ impl Project {
                     include_metadata_hash,
                     optimizer_settings.clone(),
                     debug_config.clone(),
+                    llvm_arguments.to_vec(),
                 );
                 let process_output = {
                     #[cfg(target_os = "emscripten")]
@@ -316,6 +318,7 @@ impl revive_llvm_context::PolkaVMDependency for Project {
         optimizer_settings: revive_llvm_context::OptimizerSettings,
         include_metadata_hash: bool,
         debug_config: revive_llvm_context::DebugConfig,
+        llvm_arguments: &[String],
     ) -> anyhow::Result<String> {
         let contract_path = project.resolve_path(identifier)?;
         let contract = project
@@ -335,6 +338,7 @@ impl revive_llvm_context::PolkaVMDependency for Project {
                 optimizer_settings,
                 include_metadata_hash,
                 debug_config,
+                llvm_arguments,
             )
             .map_err(|error| {
                 anyhow::anyhow!(

crates/solidity/src/resolc/arguments.rs

@@ -166,6 +166,10 @@ pub struct Arguments {
     #[cfg(debug_assertions)]
     #[arg(long = "recursive-process-input")]
     pub recursive_process_input: Option<String>,
+
+    #[arg(long = "llvm-arg")]
+    /// These are passed to LLVM as the command line to allow manual control.
+    pub llvm_arguments: Vec<String>,
 }
 
 impl Arguments {

crates/solidity/src/resolc/main.rs

@@ -64,8 +64,6 @@ fn main_inner() -> anyhow::Result<()> {
         .stack_size(RAYON_WORKER_STACK_SIZE)
         .build_global()
         .expect("Thread pool configuration failure");
-    inkwell::support::enable_llvm_pretty_stack_trace();
-    revive_llvm_context::initialize_target(revive_llvm_context::Target::PVM); // TODO: pass from CLI
 
     if arguments.recursive_process {
         #[cfg(debug_assertions)]
@@ -157,6 +155,7 @@ fn main_inner() -> anyhow::Result<()> {
             optimizer_settings,
             include_metadata_hash,
             debug_config,
+            &arguments.llvm_arguments,
         )
     } else if arguments.llvm_ir {
         revive_solidity::llvm_ir(
@@ -164,6 +163,7 @@ fn main_inner() -> anyhow::Result<()> {
             optimizer_settings,
             include_metadata_hash,
             debug_config,
+            &arguments.llvm_arguments,
         )
     } else if arguments.standard_json {
         revive_solidity::standard_json(
@@ -173,6 +173,7 @@ fn main_inner() -> anyhow::Result<()> {
             arguments.include_paths,
             arguments.allow_paths,
             debug_config,
+            &arguments.llvm_arguments,
         )?;
         return Ok(());
     } else if let Some(format) = arguments.combined_json {
@@ -193,6 +194,7 @@ fn main_inner() -> anyhow::Result<()> {
             debug_config,
             arguments.output_directory,
             arguments.overwrite,
+            &arguments.llvm_arguments,
         )?;
         return Ok(());
     } else {
@@ -210,6 +212,7 @@ fn main_inner() -> anyhow::Result<()> {
             remappings,
             suppressed_warnings,
             debug_config,
+            &arguments.llvm_arguments,
         )
     }?;
 

crates/solidity/src/test_utils.rs

@@ -73,7 +73,11 @@ pub fn build_solidity_with_options(
     check_dependencies();
 
     inkwell::support::enable_llvm_pretty_stack_trace();
-    revive_llvm_context::initialize_target(revive_llvm_context::Target::PVM);
+    revive_llvm_context::initialize_llvm(
+        revive_llvm_context::Target::PVM,
+        crate::DEFAULT_EXECUTABLE_NAME,
+        &[],
+    );
     let _ = crate::process::native_process::EXECUTABLE
         .set(PathBuf::from(crate::r#const::DEFAULT_EXECUTABLE_NAME));
 
@@ -106,7 +110,8 @@ pub fn build_solidity_with_options(
         &DEBUG_CONFIG,
     )?;
 
-    let build: crate::Build = project.compile(optimizer_settings, false, DEBUG_CONFIG)?;
+    let build: crate::Build =
+        project.compile(optimizer_settings, false, DEBUG_CONFIG, Default::default())?;
     build.write_to_standard_json(&mut output, &solc_version)?;
 
     Ok(output)
@@ -122,7 +127,11 @@ pub fn build_solidity_with_options_evm(
     check_dependencies();
 
     inkwell::support::enable_llvm_pretty_stack_trace();
-    revive_llvm_context::initialize_target(revive_llvm_context::Target::PVM);
+    revive_llvm_context::initialize_llvm(
+        revive_llvm_context::Target::PVM,
+        crate::DEFAULT_EXECUTABLE_NAME,
+        &[],
+    );
     let _ = crate::process::native_process::EXECUTABLE
         .set(PathBuf::from(crate::r#const::DEFAULT_EXECUTABLE_NAME));
 
@@ -174,7 +183,11 @@ pub fn build_solidity_and_detect_missing_libraries(
     check_dependencies();
 
     inkwell::support::enable_llvm_pretty_stack_trace();
-    revive_llvm_context::initialize_target(revive_llvm_context::Target::PVM);
+    revive_llvm_context::initialize_llvm(
+        revive_llvm_context::Target::PVM,
+        crate::DEFAULT_EXECUTABLE_NAME,
+        &[],
+    );
     let _ = crate::process::native_process::EXECUTABLE
         .set(PathBuf::from(crate::r#const::DEFAULT_EXECUTABLE_NAME));
 
@@ -213,7 +226,11 @@ pub fn build_yul(source_code: &str) -> anyhow::Result<()> {
     check_dependencies();
 
     inkwell::support::enable_llvm_pretty_stack_trace();
-    revive_llvm_context::initialize_target(revive_llvm_context::Target::PVM);
+    revive_llvm_context::initialize_llvm(
+        revive_llvm_context::Target::PVM,
+        crate::DEFAULT_EXECUTABLE_NAME,
+        &[],
+    );
     let optimizer_settings = revive_llvm_context::OptimizerSettings::none();
 
     let project = Project::try_from_yul_string::<SolcCompiler>(
@@ -221,7 +238,7 @@ pub fn build_yul(source_code: &str) -> anyhow::Result<()> {
         source_code,
         None,
     )?;
-    let _build = project.compile(optimizer_settings, false, DEBUG_CONFIG)?;
+    let _build = project.compile(optimizer_settings, false, DEBUG_CONFIG, Default::default())?;
 
     Ok(())
 }