initial subkernels commit

This reverts commit 308edb8237.

nac3artiq/demo/embedding_map.py

@@ -1,11 +1,18 @@
 class EmbeddingMap:
-    def __init__(self):
+    def __init__(self, old_embedding_map=None):
         self.object_inverse_map = {}
         self.object_map = {}
         self.string_map = {}
         self.string_reverse_map = {}
         self.function_map = {}
         self.attributes_writeback = []
+        self.subkernel_message_map = {}
+
+        if not old_embedding_map is None:
+            for key, obj_ref in old_embedding_map.subkernels().items():
+                self.object_map[key] = obj_ref
+                obj_id = id(obj_ref)
+                self.object_inverse_map[obj_id] = key
 
         # preallocate exception names
         self.preallocate_runtime_exception_names(["RuntimeError",
@@ -64,3 +71,23 @@ class EmbeddingMap:
     def retrieve_str(self, key):
         return self.string_map[key]
 
+    def subkernels(self):
+        subkernels = {}
+        for k, v in self.object_map.items():
+            if getattr(v, "__artiq_destination__") is not None:
+                subkernels[k] = v
+        return subkernels
+    
+    def subkernel_messages(self):
+        messages = {}
+        for msg_id in self.subkernel_message_map.values():
+            messages[msg_id] = self.retrieve_object(msg_id)
+        return messages
+
+    def subkernel_messages_unpaired(self):
+        unpaired = []
+        for msg_id in self.subkernel_message_map.values():
+            msg_obj = self.retrieve_object(msg_id)
+            if msg_obj.send_loc is None or msg_obj.recv_loc is None:
+                unpaired.append(msg_obj)
+        return unpaired

nac3artiq/demo/min_artiq.py

@@ -130,8 +130,19 @@ def kernel(function_or_method):
         @wraps(function_or_method)
         def run_on_core(*args, **kwargs):
             raise RuntimeError("Kernel functions need explicit core.run()")
+    run_on_core.__artiq_kernel__ = True
+    run_on_core.__artiq_destination__ = None
     return run_on_core
 
+def subkernel(function_or_method, destination=0):
+    """Decorates a function or method to be executed on a satellite core device."""
+    _register_function(function_or_method)
+    @wraps(function_or_method)
+    def run_on_core(*args, **kwargs):
+        raise RuntimeError("Subkernels cannot be called by the host")
+    run_on_core.__artiq_kernel__ = True
+    run_on_core.__artiq_destination__ = destination
+    return function_or_method
 
 def portable(function):
     """Decorates a function or method to be executed on the same device (host/core device) as the caller."""

nac3artiq/src/lib.rs

@@ -196,6 +196,7 @@ impl Nac3 {
                             decorator_list.iter().any(|decorator| {
                                 if let ExprKind::Name { id, .. } = decorator.node {
                                     id.to_string() == "kernel"
+                                        || id.to_string() == "subkernel"
                                         || id.to_string() == "portable"
                                         || id.to_string() == "rpc"
                                 } else {
@@ -212,7 +213,7 @@ impl Nac3 {
                     decorator_list.iter().any(|decorator| {
                         if let ExprKind::Name { id, .. } = decorator.node {
                             let id = id.to_string();
-                            id == "extern" || id == "portable" || id == "kernel" || id == "rpc"
+                            id == "extern" || id == "portable" || id == "kernel" || id == "subkernel" || id == "rpc"
                         } else {
                             false
                         }
@@ -448,7 +449,6 @@ impl Nac3 {
                         pyid_to_type: pyid_to_type.clone(),
                         primitive_ids: self.primitive_ids.clone(),
                         global_value_ids: global_value_ids.clone(),
-                        class_names: Mutex::default(),
                         name_to_pyid: name_to_pyid.clone(),
                         module: module.clone(),
                         id_to_pyval: RwLock::default(),
@@ -540,7 +540,6 @@ impl Nac3 {
             pyid_to_type: pyid_to_type.clone(),
             primitive_ids: self.primitive_ids.clone(),
             global_value_ids: global_value_ids.clone(),
-            class_names: Mutex::default(),
             id_to_pyval: RwLock::default(),
             id_to_primitive: RwLock::default(),
             field_to_val: RwLock::default(),
@@ -557,6 +556,10 @@ impl Nac3 {
             .register_top_level(synthesized.pop().unwrap(), Some(resolver.clone()), "", false)
             .unwrap();
 
+        // Process IRRT
+        let context = inkwell::context::Context::create();
+        let irrt = load_irrt(&context, resolver.as_ref());
+
         let fun_signature =
             FunSignature { args: vec![], ret: self.primitive.none, vars: VarMap::new() };
         let mut store = ConcreteTypeStore::new();
@@ -727,7 +730,7 @@ impl Nac3 {
             membuffer.lock().push(buffer);
         });
 
-        let context = inkwell::context::Context::create();
+        // Link all modules into `main`.
         let buffers = membuffers.lock();
         let main = context
             .create_module_from_ir(MemoryBuffer::create_from_memory_range(&buffers[0], "main"))
@@ -756,8 +759,7 @@ impl Nac3 {
             )
             .unwrap();
 
-        main.link_in_module(load_irrt(&context))
-            .map_err(|err| CompileError::new_err(err.to_string()))?;
+        main.link_in_module(irrt).map_err(|err| CompileError::new_err(err.to_string()))?;
 
         let mut function_iter = main.get_first_function();
         while let Some(func) = function_iter {

nac3artiq/src/subkernels.rs

@@ -0,0 +1,91 @@
+pub struct Subkernels {}
+
+impl Subkernels {
+    fn get_subkernel_builtins() -> Vec<Box<BuiltinFuncSpec>> {
+        vec![
+                Box::new(|primitives, unifier| {
+                    let arg_ty = unifier.get_fresh_var(Some("T".into()), None);
+                    (
+                        "subkernel_await".into(),
+                        FunSignature {
+                            args: vec![FuncArg {
+                                name: "subkernel".into(),
+                                ty: arg_ty.ty,
+                                default_value: None,
+                                is_vararg: false,
+                            },
+                            FuncArg {
+                                name: "timeout".into(),
+                                ty: primitives.int32,
+                                default_value: 0,
+                                is_vararg: false
+                            }],
+                            // todo: figure out how to return the type function returns 
+                            // or catch it in the type inferencer
+                            ret: primitives.none,
+                            vars: into_var_map([arg_ty]),
+                        },
+                        Arc::new(GenCall::new(Box::new(move |ctx, obj, fun, args, generator| {
+                            gen_subkernel_await(ctx, &obj, fun, &args, generator)?;
+    
+                            Ok(None)
+                        }))),
+                    )
+                }),
+                Box::new(|primitives, unifier| {
+                    let arg_ty = unifier.get_fresh_var(Some("T".into()), None);
+                    (
+                        "subkernel_preload".into(),
+                        FunSignature {
+                            args: vec![
+                                FuncArg {
+                                    name: "subkernel".into(),
+                                    ty: arg_ty.ty,
+                                    default_value: None,
+                                    is_vararg: false,
+                                }
+                            ],
+                            ret: primitives.none,
+                            vars: into_var_map([arg_ty]),
+                        },
+                        Arc::new(GenCall::new(Box::new(move |ctx, obj, fun, args, generator| {
+                            gen_subkernel_preload(ctx, &obj, fun, &args, generator)?;
+    
+                            Ok(None)
+                        }))),
+                    )
+                }),
+            ]
+    }
+
+    fn gen_subkernel_await<'ctx>(ctx: &mut CodeGenContext<'ctx, '_>, awaited: BasicValueEnum<'ctx>, timeout: BasicValueEnum<'ctx>) {
+        let sid_type = ctx.ctx.i32_type();
+        // how to deal with optional arguments?
+        let timeout_type = ctx.ctx.i64_type();
+        let subkernel_await_finish = ctx.module.get_function("subkernel_await_finish").unwrap_or_else(|| {
+            ctx.module.add_function(
+                "subkernel_await_finish",
+                ctx.ctx.void_type().fn_type(&[sid_type.into(), timeout_type.into()], false),
+                None,
+            )
+        });
+
+        // call or invoke
+        // generate RPC for receiving return value depending on fun ret
+    }
+
+    fn gen_subkernel_preload<'ctx>(ctx: &mut CodeGenContext<'ctx, '_>, preloaded: BasicValueEnum<'ctx>) {
+        let sid_type = ctx.ctx.i32_type();
+        let dest_type = ctx.ctx.i8_type();
+        let run_type = ctx.ctx.i1_type();
+        let subkernel_load_run = ctx.module.get_function("subkernel_load_run").unwrap_or_else(|| {
+            ctx.module.add_function(
+                "subkernel_load_run",
+                ctx.ctx.void_type().fn_type(&[sid_type.into(), dest_type, run_type], false),
+                None,
+            )
+        });
+        // retrieve destination and sid from the fn (?)
+        // call or invoke
+    }
+}

nac3artiq/src/symbol_resolver.rs

@@ -23,7 +23,7 @@ use nac3core::{
     },
 };
 use nac3parser::ast::{self, StrRef};
-use parking_lot::{Mutex, RwLock};
+use parking_lot::RwLock;
 use pyo3::{
     types::{PyDict, PyTuple},
     PyAny, PyObject, PyResult, Python,
@@ -79,7 +79,6 @@ pub struct InnerResolver {
     pub id_to_primitive: RwLock<HashMap<u64, PrimitiveValue>>,
     pub field_to_val: RwLock<HashMap<ResolverField, Option<PyFieldHandle>>>,
     pub global_value_ids: Arc<RwLock<HashMap<u64, PyObject>>>,
-    pub class_names: Mutex<HashMap<StrRef, Type>>,
     pub pyid_to_def: Arc<RwLock<HashMap<u64, DefinitionId>>>,
     pub pyid_to_type: Arc<RwLock<HashMap<u64, Type>>>,
     pub primitive_ids: PrimitivePythonId,

nac3core/build.rs

@@ -18,27 +18,36 @@ fn main() {
      * HACK: Sadly, clang doesn't let us emit generic LLVM bitcode.
      * Compiling for WASM32 and filtering the output with regex is the closest we can get.
      */
-    let flags: &[&str] = &[
+    let mut flags: Vec<&str> = vec![
         "--target=wasm32",
         "-x",
         "c++",
+        "-std=c++20",
         "-fno-discard-value-names",
         "-fno-exceptions",
         "-fno-rtti",
-        match env::var("PROFILE").as_deref() {
-            Ok("debug") => "-O0",
-            Ok("release") => "-O3",
-            flavor => panic!("Unknown or missing build flavor {flavor:?}"),
-        },
         "-emit-llvm",
         "-S",
         "-Wall",
         "-Wextra",
         "-o",
         "-",
+        "-I",
+        irrt_dir.to_str().unwrap(),
         irrt_cpp_path.to_str().unwrap(),
     ];
 
+    match env::var("PROFILE").as_deref() {
+        Ok("debug") => {
+            flags.push("-O0");
+            flags.push("-DIRRT_DEBUG_ASSERT");
+        }
+        Ok("release") => {
+            flags.push("-O3");
+        }
+        flavor => panic!("Unknown or missing build flavor {flavor:?}"),
+    }
+
     // Tell Cargo to rerun if any file under `irrt_dir` (recursive) changes
     println!("cargo:rerun-if-changed={}", irrt_dir.to_str().unwrap());
 
@@ -61,7 +70,12 @@ fn main() {
     // Regex:
     // - `(?ms:^define.*?\}$)` captures LLVM `define` blocks
     // - `(?m:^declare.*?$)` captures LLVM `declare` lines
-    let regex_filter = Regex::new(r"(?ms:^define.*?\}$)|(?m:^declare.*?$)").unwrap();
+    // - `(?m:^%.+?=\s*type\s*\{.+?\}$)` captures LLVM `type` declarations
+    // - `(?m:^@.+?=.+$)` captures global constants
+    let regex_filter = Regex::new(
+        r"(?ms:^define.*?\}$)|(?m:^declare.*?$)|(?m:^%.+?=\s*type\s*\{.+?\}$)|(?m:^@.+?=.+$)",
+    )
+    .unwrap();
     for f in regex_filter.captures_iter(&output) {
         assert_eq!(f.len(), 1);
         filtered_output.push_str(&f[0]);

nac3core/irrt/irrt.cpp

@@ -1,336 +1,6 @@
-using int8_t = _BitInt(8);
-using uint8_t = unsigned _BitInt(8);
-using int32_t = _BitInt(32);
-using uint32_t = unsigned _BitInt(32);
-using int64_t = _BitInt(64);
-using uint64_t = unsigned _BitInt(64);
-
-// NDArray indices are always `uint32_t`.
-using NDIndex = uint32_t;
-// The type of an index or a value describing the length of a range/slice is always `int32_t`.
-using SliceIndex = int32_t;
-
-namespace {
-template<typename T>
-const T& max(const T& a, const T& b) {
-    return a > b ? a : b;
-}
-
-template<typename T>
-const T& min(const T& a, const T& b) {
-    return a > b ? b : a;
-}
-
-// adapted from GNU Scientific Library: https://git.savannah.gnu.org/cgit/gsl.git/tree/sys/pow_int.c
-// need to make sure `exp >= 0` before calling this function
-template<typename T>
-T __nac3_int_exp_impl(T base, T exp) {
-    T res = 1;
-    /* repeated squaring method */
-    do {
-        if (exp & 1) {
-            res *= base; /* for n odd */
-        }
-        exp >>= 1;
-        base *= base;
-    } while (exp);
-    return res;
-}
-
-template<typename SizeT>
-SizeT __nac3_ndarray_calc_size_impl(const SizeT* list_data, SizeT list_len, SizeT begin_idx, SizeT end_idx) {
-    __builtin_assume(end_idx <= list_len);
-
-    SizeT num_elems = 1;
-    for (SizeT i = begin_idx; i < end_idx; ++i) {
-        SizeT val = list_data[i];
-        __builtin_assume(val > 0);
-        num_elems *= val;
-    }
-    return num_elems;
-}
-
-template<typename SizeT>
-void __nac3_ndarray_calc_nd_indices_impl(SizeT index, const SizeT* dims, SizeT num_dims, NDIndex* idxs) {
-    SizeT stride = 1;
-    for (SizeT dim = 0; dim < num_dims; dim++) {
-        SizeT i = num_dims - dim - 1;
-        __builtin_assume(dims[i] > 0);
-        idxs[i] = (index / stride) % dims[i];
-        stride *= dims[i];
-    }
-}
-
-template<typename SizeT>
-SizeT __nac3_ndarray_flatten_index_impl(const SizeT* dims, SizeT num_dims, const NDIndex* indices, SizeT num_indices) {
-    SizeT idx = 0;
-    SizeT stride = 1;
-    for (SizeT i = 0; i < num_dims; ++i) {
-        SizeT ri = num_dims - i - 1;
-        if (ri < num_indices) {
-            idx += stride * indices[ri];
-        }
-
-        __builtin_assume(dims[i] > 0);
-        stride *= dims[ri];
-    }
-    return idx;
-}
-
-template<typename SizeT>
-void __nac3_ndarray_calc_broadcast_impl(const SizeT* lhs_dims,
-                                        SizeT lhs_ndims,
-                                        const SizeT* rhs_dims,
-                                        SizeT rhs_ndims,
-                                        SizeT* out_dims) {
-    SizeT max_ndims = lhs_ndims > rhs_ndims ? lhs_ndims : rhs_ndims;
-
-    for (SizeT i = 0; i < max_ndims; ++i) {
-        const SizeT* lhs_dim_sz = i < lhs_ndims ? &lhs_dims[lhs_ndims - i - 1] : nullptr;
-        const SizeT* rhs_dim_sz = i < rhs_ndims ? &rhs_dims[rhs_ndims - i - 1] : nullptr;
-        SizeT* out_dim = &out_dims[max_ndims - i - 1];
-
-        if (lhs_dim_sz == nullptr) {
-            *out_dim = *rhs_dim_sz;
-        } else if (rhs_dim_sz == nullptr) {
-            *out_dim = *lhs_dim_sz;
-        } else if (*lhs_dim_sz == 1) {
-            *out_dim = *rhs_dim_sz;
-        } else if (*rhs_dim_sz == 1) {
-            *out_dim = *lhs_dim_sz;
-        } else if (*lhs_dim_sz == *rhs_dim_sz) {
-            *out_dim = *lhs_dim_sz;
-        } else {
-            __builtin_unreachable();
-        }
-    }
-}
-
-template<typename SizeT>
-void __nac3_ndarray_calc_broadcast_idx_impl(const SizeT* src_dims,
-                                            SizeT src_ndims,
-                                            const NDIndex* in_idx,
-                                            NDIndex* out_idx) {
-    for (SizeT i = 0; i < src_ndims; ++i) {
-        SizeT src_i = src_ndims - i - 1;
-        out_idx[src_i] = src_dims[src_i] == 1 ? 0 : in_idx[src_i];
-    }
-}
-}  // namespace
-
-extern "C" {
-#define DEF_nac3_int_exp_(T)                   \
-    T __nac3_int_exp_##T(T base, T exp) {      \
-        return __nac3_int_exp_impl(base, exp); \
-    }
-
-DEF_nac3_int_exp_(int32_t) DEF_nac3_int_exp_(int64_t) DEF_nac3_int_exp_(uint32_t) DEF_nac3_int_exp_(uint64_t)
-
-    SliceIndex __nac3_slice_index_bound(SliceIndex i, const SliceIndex len) {
-    if (i < 0) {
-        i = len + i;
-    }
-    if (i < 0) {
-        return 0;
-    } else if (i > len) {
-        return len;
-    }
-    return i;
-}
-
-SliceIndex __nac3_range_slice_len(const SliceIndex start, const SliceIndex end, const SliceIndex step) {
-    SliceIndex diff = end - start;
-    if (diff > 0 && step > 0) {
-        return ((diff - 1) / step) + 1;
-    } else if (diff < 0 && step < 0) {
-        return ((diff + 1) / step) + 1;
-    } else {
-        return 0;
-    }
-}
-
-// Handle list assignment and dropping part of the list when
-// both dest_step and src_step are +1.
-// - All the index must *not* be out-of-bound or negative,
-// - The end index is *inclusive*,
-// - The length of src and dest slice size should already
-// be checked: if dest.step == 1 then len(src) <= len(dest) else len(src) == len(dest)
-SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
-                                             SliceIndex dest_end,
-                                             SliceIndex dest_step,
-                                             uint8_t* dest_arr,
-                                             SliceIndex dest_arr_len,
-                                             SliceIndex src_start,
-                                             SliceIndex src_end,
-                                             SliceIndex src_step,
-                                             uint8_t* src_arr,
-                                             SliceIndex src_arr_len,
-                                             const SliceIndex size) {
-    /* if dest_arr_len == 0, do nothing since we do not support extending list */
-    if (dest_arr_len == 0)
-        return dest_arr_len;
-    /* if both step is 1, memmove directly, handle the dropping of the list, and shrink size */
-    if (src_step == dest_step && dest_step == 1) {
-        const SliceIndex src_len = (src_end >= src_start) ? (src_end - src_start + 1) : 0;
-        const SliceIndex dest_len = (dest_end >= dest_start) ? (dest_end - dest_start + 1) : 0;
-        if (src_len > 0) {
-            __builtin_memmove(dest_arr + dest_start * size, src_arr + src_start * size, src_len * size);
-        }
-        if (dest_len > 0) {
-            /* dropping */
-            __builtin_memmove(dest_arr + (dest_start + src_len) * size, dest_arr + (dest_end + 1) * size,
-                              (dest_arr_len - dest_end - 1) * size);
-        }
-        /* shrink size */
-        return dest_arr_len - (dest_len - src_len);
-    }
-    /* if two range overlaps, need alloca */
-    uint8_t need_alloca = (dest_arr == src_arr)
-                          && !(max(dest_start, dest_end) < min(src_start, src_end)
-                               || max(src_start, src_end) < min(dest_start, dest_end));
-    if (need_alloca) {
-        uint8_t* tmp = reinterpret_cast<uint8_t*>(__builtin_alloca(src_arr_len * size));
-        __builtin_memcpy(tmp, src_arr, src_arr_len * size);
-        src_arr = tmp;
-    }
-    SliceIndex src_ind = src_start;
-    SliceIndex dest_ind = dest_start;
-    for (; (src_step > 0) ? (src_ind <= src_end) : (src_ind >= src_end); src_ind += src_step, dest_ind += dest_step) {
-        /* for constant optimization */
-        if (size == 1) {
-            __builtin_memcpy(dest_arr + dest_ind, src_arr + src_ind, 1);
-        } else if (size == 4) {
-            __builtin_memcpy(dest_arr + dest_ind * 4, src_arr + src_ind * 4, 4);
-        } else if (size == 8) {
-            __builtin_memcpy(dest_arr + dest_ind * 8, src_arr + src_ind * 8, 8);
-        } else {
-            /* memcpy for var size, cannot overlap after previous alloca */
-            __builtin_memcpy(dest_arr + dest_ind * size, src_arr + src_ind * size, size);
-        }
-    }
-    /* only dest_step == 1 can we shrink the dest list. */
-    /* size should be ensured prior to calling this function */
-    if (dest_step == 1 && dest_end >= dest_start) {
-        __builtin_memmove(dest_arr + dest_ind * size, dest_arr + (dest_end + 1) * size,
-                          (dest_arr_len - dest_end - 1) * size);
-        return dest_arr_len - (dest_end - dest_ind) - 1;
-    }
-    return dest_arr_len;
-}
-
-int32_t __nac3_isinf(double x) {
-    return __builtin_isinf(x);
-}
-
-int32_t __nac3_isnan(double x) {
-    return __builtin_isnan(x);
-}
-
-double tgamma(double arg);
-
-double __nac3_gamma(double z) {
-    // Handling for denormals
-    //     | x                 | Python gamma(x) | C tgamma(x) |
-    // --- | ----------------- | --------------- | ----------- |
-    // (1) | nan               | nan             | nan         |
-    // (2) | -inf              | -inf            | inf         |
-    // (3) | inf               | inf             | inf         |
-    // (4) | 0.0               | inf             | inf         |
-    // (5) | {-1.0, -2.0, ...} | inf             | nan         |
-
-    // (1)-(3)
-    if (__builtin_isinf(z) || __builtin_isnan(z)) {
-        return z;
-    }
-
-    double v = tgamma(z);
-
-    // (4)-(5)
-    return __builtin_isinf(v) || __builtin_isnan(v) ? __builtin_inf() : v;
-}
-
-double lgamma(double arg);
-
-double __nac3_gammaln(double x) {
-    // libm's handling of value overflows differs from scipy:
-    // - scipy: gammaln(-inf) -> -inf
-    // - libm : lgamma(-inf) -> inf
-
-    if (__builtin_isinf(x)) {
-        return x;
-    }
-
-    return lgamma(x);
-}
-
-double j0(double x);
-
-double __nac3_j0(double x) {
-    // libm's handling of value overflows differs from scipy:
-    // - scipy: j0(inf) -> nan
-    // - libm : j0(inf) -> 0.0
-
-    if (__builtin_isinf(x)) {
-        return __builtin_nan("");
-    }
-
-    return j0(x);
-}
-
-uint32_t __nac3_ndarray_calc_size(const uint32_t* list_data, uint32_t list_len, uint32_t begin_idx, uint32_t end_idx) {
-    return __nac3_ndarray_calc_size_impl(list_data, list_len, begin_idx, end_idx);
-}
-
-uint64_t
-__nac3_ndarray_calc_size64(const uint64_t* list_data, uint64_t list_len, uint64_t begin_idx, uint64_t end_idx) {
-    return __nac3_ndarray_calc_size_impl(list_data, list_len, begin_idx, end_idx);
-}
-
-void __nac3_ndarray_calc_nd_indices(uint32_t index, const uint32_t* dims, uint32_t num_dims, NDIndex* idxs) {
-    __nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
-}
-
-void __nac3_ndarray_calc_nd_indices64(uint64_t index, const uint64_t* dims, uint64_t num_dims, NDIndex* idxs) {
-    __nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
-}
-
-uint32_t
-__nac3_ndarray_flatten_index(const uint32_t* dims, uint32_t num_dims, const NDIndex* indices, uint32_t num_indices) {
-    return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
-}
-
-uint64_t
-__nac3_ndarray_flatten_index64(const uint64_t* dims, uint64_t num_dims, const NDIndex* indices, uint64_t num_indices) {
-    return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
-}
-
-void __nac3_ndarray_calc_broadcast(const uint32_t* lhs_dims,
-                                   uint32_t lhs_ndims,
-                                   const uint32_t* rhs_dims,
-                                   uint32_t rhs_ndims,
-                                   uint32_t* out_dims) {
-    return __nac3_ndarray_calc_broadcast_impl(lhs_dims, lhs_ndims, rhs_dims, rhs_ndims, out_dims);
-}
-
-void __nac3_ndarray_calc_broadcast64(const uint64_t* lhs_dims,
-                                     uint64_t lhs_ndims,
-                                     const uint64_t* rhs_dims,
-                                     uint64_t rhs_ndims,
-                                     uint64_t* out_dims) {
-    return __nac3_ndarray_calc_broadcast_impl(lhs_dims, lhs_ndims, rhs_dims, rhs_ndims, out_dims);
-}
-
-void __nac3_ndarray_calc_broadcast_idx(const uint32_t* src_dims,
-                                       uint32_t src_ndims,
-                                       const NDIndex* in_idx,
-                                       NDIndex* out_idx) {
-    __nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
-}
-
-void __nac3_ndarray_calc_broadcast_idx64(const uint64_t* src_dims,
-                                         uint64_t src_ndims,
-                                         const NDIndex* in_idx,
-                                         NDIndex* out_idx) {
-    __nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
-}
-}  // extern "C"
+#include <irrt/exception.hpp>
+#include <irrt/int_types.hpp>
+#include <irrt/list.hpp>
+#include <irrt/math.hpp>
+#include <irrt/ndarray.hpp>
+#include <irrt/slice.hpp>

nac3core/irrt/irrt/cslice.hpp

@@ -0,0 +1,9 @@
+#pragma once
+
+#include <irrt/int_types.hpp>
+
+template<typename SizeT>
+struct CSlice {
+    uint8_t* base;
+    SizeT len;
+};

nac3core/irrt/irrt/debug.hpp

@@ -0,0 +1,25 @@
+#pragma once
+
+// Set in nac3core/build.rs
+#ifdef IRRT_DEBUG_ASSERT
+#define IRRT_DEBUG_ASSERT_BOOL true
+#else
+#define IRRT_DEBUG_ASSERT_BOOL false
+#endif
+
+#define raise_debug_assert(SizeT, msg, param1, param2, param3) \
+    raise_exception(SizeT, EXN_ASSERTION_ERROR, "IRRT debug assert failed: " msg, param1, param2, param3)
+
+#define debug_assert_eq(SizeT, lhs, rhs)                                           \
+    if constexpr (IRRT_DEBUG_ASSERT_BOOL) {                                        \
+        if ((lhs) != (rhs)) {                                                      \
+            raise_debug_assert(SizeT, "LHS = {0}. RHS = {1}", lhs, rhs, NO_PARAM); \
+        }                                                                          \
+    }
+
+#define debug_assert(SizeT, expr)                                                  \
+    if constexpr (IRRT_DEBUG_ASSERT_BOOL) {                                        \
+        if (!(expr)) {                                                             \
+            raise_debug_assert(SizeT, "Got false.", NO_PARAM, NO_PARAM, NO_PARAM); \
+        }                                                                          \
+    }

nac3core/irrt/irrt/exception.hpp

@@ -0,0 +1,82 @@
+#pragma once
+
+#include <irrt/cslice.hpp>
+#include <irrt/int_types.hpp>
+
+/**
+ * @brief The int type of ARTIQ exception IDs.
+ */
+typedef int32_t ExceptionId;
+
+/*
+ * Set of exceptions C++ IRRT can use.
+ * Must be synchronized with `setup_irrt_exceptions` in `nac3core/src/codegen/irrt/mod.rs`.
+ */
+extern "C" {
+ExceptionId EXN_INDEX_ERROR;
+ExceptionId EXN_VALUE_ERROR;
+ExceptionId EXN_ASSERTION_ERROR;
+ExceptionId EXN_TYPE_ERROR;
+}
+
+/**
+ * @brief Extern function to `__nac3_raise`
+ *
+ * The parameter `err` could be `Exception<int32_t>` or `Exception<int64_t>`. The caller
+ * must make sure to pass `Exception`s with the correct `SizeT` depending on the `size_t` of the runtime.
+ */
+extern "C" void __nac3_raise(void* err);
+
+namespace {
+/**
+ * @brief NAC3's Exception struct
+ */
+template<typename SizeT>
+struct Exception {
+    ExceptionId id;
+    CSlice<SizeT> filename;
+    int32_t line;
+    int32_t column;
+    CSlice<SizeT> function;
+    CSlice<SizeT> msg;
+    int64_t params[3];
+};
+
+constexpr int64_t NO_PARAM = 0;
+
+template<typename SizeT>
+void _raise_exception_helper(ExceptionId id,
+                             const char* filename,
+                             int32_t line,
+                             const char* function,
+                             const char* msg,
+                             int64_t param0,
+                             int64_t param1,
+                             int64_t param2) {
+    Exception<SizeT> e = {
+        .id = id,
+        .filename = {.base = reinterpret_cast<const uint8_t*>(filename), .len = __builtin_strlen(filename)},
+        .line = line,
+        .column = 0,
+        .function = {.base = reinterpret_cast<const uint8_t*>(function), .len = __builtin_strlen(function)},
+        .msg = {.base = reinterpret_cast<const uint8_t*>(msg), .len = __builtin_strlen(msg)},
+    };
+    e.params[0] = param0;
+    e.params[1] = param1;
+    e.params[2] = param2;
+    __nac3_raise(reinterpret_cast<void*>(&e));
+    __builtin_unreachable();
+}
+
+/**
+ * @brief Raise an exception with location details (location in the IRRT source files).
+ * @param SizeT The runtime `size_t` type.
+ * @param id The ID of the exception to raise.
+ * @param msg A global constant C-string of the error message.
+ *
+ * `param0` to `param2` are optional format arguments of `msg`. They should be set to
+ * `NO_PARAM` to indicate they are unused.
+ */
+#define raise_exception(SizeT, id, msg, param0, param1, param2) \
+    _raise_exception_helper<SizeT>(id, __FILE__, __LINE__, __FUNCTION__, msg, param0, param1, param2)
+}  // namespace

nac3core/irrt/irrt/int_types.hpp

@@ -0,0 +1,13 @@
+#pragma once
+
+using int8_t = _BitInt(8);
+using uint8_t = unsigned _BitInt(8);
+using int32_t = _BitInt(32);
+using uint32_t = unsigned _BitInt(32);
+using int64_t = _BitInt(64);
+using uint64_t = unsigned _BitInt(64);
+
+// NDArray indices are always `uint32_t`.
+using NDIndex = uint32_t;
+// The type of an index or a value describing the length of a range/slice is always `int32_t`.
+using SliceIndex = int32_t;

nac3core/irrt/irrt/list.hpp

@@ -0,0 +1,75 @@
+#pragma once
+
+#include <irrt/int_types.hpp>
+#include <irrt/math_util.hpp>
+
+extern "C" {
+// Handle list assignment and dropping part of the list when
+// both dest_step and src_step are +1.
+// - All the index must *not* be out-of-bound or negative,
+// - The end index is *inclusive*,
+// - The length of src and dest slice size should already
+// be checked: if dest.step == 1 then len(src) <= len(dest) else len(src) == len(dest)
+SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
+                                             SliceIndex dest_end,
+                                             SliceIndex dest_step,
+                                             uint8_t* dest_arr,
+                                             SliceIndex dest_arr_len,
+                                             SliceIndex src_start,
+                                             SliceIndex src_end,
+                                             SliceIndex src_step,
+                                             uint8_t* src_arr,
+                                             SliceIndex src_arr_len,
+                                             const SliceIndex size) {
+    /* if dest_arr_len == 0, do nothing since we do not support extending list */
+    if (dest_arr_len == 0)
+        return dest_arr_len;
+    /* if both step is 1, memmove directly, handle the dropping of the list, and shrink size */
+    if (src_step == dest_step && dest_step == 1) {
+        const SliceIndex src_len = (src_end >= src_start) ? (src_end - src_start + 1) : 0;
+        const SliceIndex dest_len = (dest_end >= dest_start) ? (dest_end - dest_start + 1) : 0;
+        if (src_len > 0) {
+            __builtin_memmove(dest_arr + dest_start * size, src_arr + src_start * size, src_len * size);
+        }
+        if (dest_len > 0) {
+            /* dropping */
+            __builtin_memmove(dest_arr + (dest_start + src_len) * size, dest_arr + (dest_end + 1) * size,
+                              (dest_arr_len - dest_end - 1) * size);
+        }
+        /* shrink size */
+        return dest_arr_len - (dest_len - src_len);
+    }
+    /* if two range overlaps, need alloca */
+    uint8_t need_alloca = (dest_arr == src_arr)
+                          && !(max(dest_start, dest_end) < min(src_start, src_end)
+                               || max(src_start, src_end) < min(dest_start, dest_end));
+    if (need_alloca) {
+        uint8_t* tmp = reinterpret_cast<uint8_t*>(__builtin_alloca(src_arr_len * size));
+        __builtin_memcpy(tmp, src_arr, src_arr_len * size);
+        src_arr = tmp;
+    }
+    SliceIndex src_ind = src_start;
+    SliceIndex dest_ind = dest_start;
+    for (; (src_step > 0) ? (src_ind <= src_end) : (src_ind >= src_end); src_ind += src_step, dest_ind += dest_step) {
+        /* for constant optimization */
+        if (size == 1) {
+            __builtin_memcpy(dest_arr + dest_ind, src_arr + src_ind, 1);
+        } else if (size == 4) {
+            __builtin_memcpy(dest_arr + dest_ind * 4, src_arr + src_ind * 4, 4);
+        } else if (size == 8) {
+            __builtin_memcpy(dest_arr + dest_ind * 8, src_arr + src_ind * 8, 8);
+        } else {
+            /* memcpy for var size, cannot overlap after previous alloca */
+            __builtin_memcpy(dest_arr + dest_ind * size, src_arr + src_ind * size, size);
+        }
+    }
+    /* only dest_step == 1 can we shrink the dest list. */
+    /* size should be ensured prior to calling this function */
+    if (dest_step == 1 && dest_end >= dest_start) {
+        __builtin_memmove(dest_arr + dest_ind * size, dest_arr + (dest_end + 1) * size,
+                          (dest_arr_len - dest_end - 1) * size);
+        return dest_arr_len - (dest_end - dest_ind) - 1;
+    }
+    return dest_arr_len;
+}
+}  // extern "C"

nac3core/irrt/irrt/math.hpp

@@ -0,0 +1,93 @@
+#pragma once
+
+namespace {
+// adapted from GNU Scientific Library: https://git.savannah.gnu.org/cgit/gsl.git/tree/sys/pow_int.c
+// need to make sure `exp >= 0` before calling this function
+template<typename T>
+T __nac3_int_exp_impl(T base, T exp) {
+    T res = 1;
+    /* repeated squaring method */
+    do {
+        if (exp & 1) {
+            res *= base; /* for n odd */
+        }
+        exp >>= 1;
+        base *= base;
+    } while (exp);
+    return res;
+}
+}  // namespace
+
+#define DEF_nac3_int_exp_(T)                   \
+    T __nac3_int_exp_##T(T base, T exp) {      \
+        return __nac3_int_exp_impl(base, exp); \
+    }
+
+extern "C" {
+
+// Putting semicolons here to make clang-format not reformat this into
+// a stair shape.
+DEF_nac3_int_exp_(int32_t);
+DEF_nac3_int_exp_(int64_t);
+DEF_nac3_int_exp_(uint32_t);
+DEF_nac3_int_exp_(uint64_t);
+
+int32_t __nac3_isinf(double x) {
+    return __builtin_isinf(x);
+}
+
+int32_t __nac3_isnan(double x) {
+    return __builtin_isnan(x);
+}
+
+double tgamma(double arg);
+
+double __nac3_gamma(double z) {
+    // Handling for denormals
+    //     | x                 | Python gamma(x) | C tgamma(x) |
+    // --- | ----------------- | --------------- | ----------- |
+    // (1) | nan               | nan             | nan         |
+    // (2) | -inf              | -inf            | inf         |
+    // (3) | inf               | inf             | inf         |
+    // (4) | 0.0               | inf             | inf         |
+    // (5) | {-1.0, -2.0, ...} | inf             | nan         |
+
+    // (1)-(3)
+    if (__builtin_isinf(z) || __builtin_isnan(z)) {
+        return z;
+    }
+
+    double v = tgamma(z);
+
+    // (4)-(5)
+    return __builtin_isinf(v) || __builtin_isnan(v) ? __builtin_inf() : v;
+}
+
+double lgamma(double arg);
+
+double __nac3_gammaln(double x) {
+    // libm's handling of value overflows differs from scipy:
+    // - scipy: gammaln(-inf) -> -inf
+    // - libm : lgamma(-inf) -> inf
+
+    if (__builtin_isinf(x)) {
+        return x;
+    }
+
+    return lgamma(x);
+}
+
+double j0(double x);
+
+double __nac3_j0(double x) {
+    // libm's handling of value overflows differs from scipy:
+    // - scipy: j0(inf) -> nan
+    // - libm : j0(inf) -> 0.0
+
+    if (__builtin_isinf(x)) {
+        return __builtin_nan("");
+    }
+
+    return j0(x);
+}
+}

nac3core/irrt/irrt/math_util.hpp

@@ -0,0 +1,13 @@
+#pragma once
+
+namespace {
+template<typename T>
+const T& max(const T& a, const T& b) {
+    return a > b ? a : b;
+}
+
+template<typename T>
+const T& min(const T& a, const T& b) {
+    return a > b ? b : a;
+}
+}  // namespace

nac3core/irrt/irrt/ndarray.hpp

@@ -0,0 +1,144 @@
+#pragma once
+
+#include <irrt/int_types.hpp>
+
+namespace {
+template<typename SizeT>
+SizeT __nac3_ndarray_calc_size_impl(const SizeT* list_data, SizeT list_len, SizeT begin_idx, SizeT end_idx) {
+    __builtin_assume(end_idx <= list_len);
+
+    SizeT num_elems = 1;
+    for (SizeT i = begin_idx; i < end_idx; ++i) {
+        SizeT val = list_data[i];
+        __builtin_assume(val > 0);
+        num_elems *= val;
+    }
+    return num_elems;
+}
+
+template<typename SizeT>
+void __nac3_ndarray_calc_nd_indices_impl(SizeT index, const SizeT* dims, SizeT num_dims, NDIndex* idxs) {
+    SizeT stride = 1;
+    for (SizeT dim = 0; dim < num_dims; dim++) {
+        SizeT i = num_dims - dim - 1;
+        __builtin_assume(dims[i] > 0);
+        idxs[i] = (index / stride) % dims[i];
+        stride *= dims[i];
+    }
+}
+
+template<typename SizeT>
+SizeT __nac3_ndarray_flatten_index_impl(const SizeT* dims, SizeT num_dims, const NDIndex* indices, SizeT num_indices) {
+    SizeT idx = 0;
+    SizeT stride = 1;
+    for (SizeT i = 0; i < num_dims; ++i) {
+        SizeT ri = num_dims - i - 1;
+        if (ri < num_indices) {
+            idx += stride * indices[ri];
+        }
+
+        __builtin_assume(dims[i] > 0);
+        stride *= dims[ri];
+    }
+    return idx;
+}
+
+template<typename SizeT>
+void __nac3_ndarray_calc_broadcast_impl(const SizeT* lhs_dims,
+                                        SizeT lhs_ndims,
+                                        const SizeT* rhs_dims,
+                                        SizeT rhs_ndims,
+                                        SizeT* out_dims) {
+    SizeT max_ndims = lhs_ndims > rhs_ndims ? lhs_ndims : rhs_ndims;
+
+    for (SizeT i = 0; i < max_ndims; ++i) {
+        const SizeT* lhs_dim_sz = i < lhs_ndims ? &lhs_dims[lhs_ndims - i - 1] : nullptr;
+        const SizeT* rhs_dim_sz = i < rhs_ndims ? &rhs_dims[rhs_ndims - i - 1] : nullptr;
+        SizeT* out_dim = &out_dims[max_ndims - i - 1];
+
+        if (lhs_dim_sz == nullptr) {
+            *out_dim = *rhs_dim_sz;
+        } else if (rhs_dim_sz == nullptr) {
+            *out_dim = *lhs_dim_sz;
+        } else if (*lhs_dim_sz == 1) {
+            *out_dim = *rhs_dim_sz;
+        } else if (*rhs_dim_sz == 1) {
+            *out_dim = *lhs_dim_sz;
+        } else if (*lhs_dim_sz == *rhs_dim_sz) {
+            *out_dim = *lhs_dim_sz;
+        } else {
+            __builtin_unreachable();
+        }
+    }
+}
+
+template<typename SizeT>
+void __nac3_ndarray_calc_broadcast_idx_impl(const SizeT* src_dims,
+                                            SizeT src_ndims,
+                                            const NDIndex* in_idx,
+                                            NDIndex* out_idx) {
+    for (SizeT i = 0; i < src_ndims; ++i) {
+        SizeT src_i = src_ndims - i - 1;
+        out_idx[src_i] = src_dims[src_i] == 1 ? 0 : in_idx[src_i];
+    }
+}
+}  // namespace
+
+extern "C" {
+uint32_t __nac3_ndarray_calc_size(const uint32_t* list_data, uint32_t list_len, uint32_t begin_idx, uint32_t end_idx) {
+    return __nac3_ndarray_calc_size_impl(list_data, list_len, begin_idx, end_idx);
+}
+
+uint64_t
+__nac3_ndarray_calc_size64(const uint64_t* list_data, uint64_t list_len, uint64_t begin_idx, uint64_t end_idx) {
+    return __nac3_ndarray_calc_size_impl(list_data, list_len, begin_idx, end_idx);
+}
+
+void __nac3_ndarray_calc_nd_indices(uint32_t index, const uint32_t* dims, uint32_t num_dims, NDIndex* idxs) {
+    __nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
+}
+
+void __nac3_ndarray_calc_nd_indices64(uint64_t index, const uint64_t* dims, uint64_t num_dims, NDIndex* idxs) {
+    __nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
+}
+
+uint32_t
+__nac3_ndarray_flatten_index(const uint32_t* dims, uint32_t num_dims, const NDIndex* indices, uint32_t num_indices) {
+    return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
+}
+
+uint64_t
+__nac3_ndarray_flatten_index64(const uint64_t* dims, uint64_t num_dims, const NDIndex* indices, uint64_t num_indices) {
+    return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
+}
+
+void __nac3_ndarray_calc_broadcast(const uint32_t* lhs_dims,
+                                   uint32_t lhs_ndims,
+                                   const uint32_t* rhs_dims,
+                                   uint32_t rhs_ndims,
+                                   uint32_t* out_dims) {
+    return __nac3_ndarray_calc_broadcast_impl(lhs_dims, lhs_ndims, rhs_dims, rhs_ndims, out_dims);
+}
+
+void __nac3_ndarray_calc_broadcast64(const uint64_t* lhs_dims,
+                                     uint64_t lhs_ndims,
+                                     const uint64_t* rhs_dims,
+                                     uint64_t rhs_ndims,
+                                     uint64_t* out_dims) {
+    return __nac3_ndarray_calc_broadcast_impl(lhs_dims, lhs_ndims, rhs_dims, rhs_ndims, out_dims);
+}
+
+void __nac3_ndarray_calc_broadcast_idx(const uint32_t* src_dims,
+                                       uint32_t src_ndims,
+                                       const NDIndex* in_idx,
+                                       NDIndex* out_idx) {
+    __nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
+}
+
+void __nac3_ndarray_calc_broadcast_idx64(const uint64_t* src_dims,
+                                         uint64_t src_ndims,
+                                         const NDIndex* in_idx,
+                                         NDIndex* out_idx) {
+    __nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
+}
+}

nac3core/irrt/irrt/slice.hpp

@@ -0,0 +1,28 @@
+#pragma once
+
+#include <irrt/int_types.hpp>
+
+extern "C" {
+SliceIndex __nac3_slice_index_bound(SliceIndex i, const SliceIndex len) {
+    if (i < 0) {
+        i = len + i;
+    }
+    if (i < 0) {
+        return 0;
+    } else if (i > len) {
+        return len;
+    }
+    return i;
+}
+
+SliceIndex __nac3_range_slice_len(const SliceIndex start, const SliceIndex end, const SliceIndex step) {
+    SliceIndex diff = end - start;
+    if (diff > 0 && step > 0) {
+        return ((diff - 1) / step) + 1;
+    } else if (diff < 0 && step < 0) {
+        return ((diff + 1) / step) + 1;
+    } else {
+        return 0;
+    }
+}
+}

nac3core/src/codegen/irrt/mod.rs

@@ -1,4 +1,4 @@
-use crate::typecheck::typedef::Type;
+use crate::{symbol_resolver::SymbolResolver, typecheck::typedef::Type};
 
 use super::{
     classes::{
@@ -16,14 +16,14 @@ use inkwell::{
     memory_buffer::MemoryBuffer,
     module::Module,
     types::{BasicTypeEnum, IntType},
-    values::{BasicValueEnum, CallSiteValue, FloatValue, IntValue},
+    values::{BasicValue, BasicValueEnum, CallSiteValue, FloatValue, IntValue},
     AddressSpace, IntPredicate,
 };
 use itertools::Either;
 use nac3parser::ast::Expr;
 
 #[must_use]
-pub fn load_irrt(ctx: &Context) -> Module {
+pub fn load_irrt<'ctx>(ctx: &'ctx Context, symbol_resolver: &dyn SymbolResolver) -> Module<'ctx> {
     let bitcode_buf = MemoryBuffer::create_from_memory_range(
         include_bytes!(concat!(env!("OUT_DIR"), "/irrt.bc")),
         "irrt_bitcode_buffer",
@@ -39,6 +39,25 @@ pub fn load_irrt(ctx: &Context) -> Module {
         let function = irrt_mod.get_function(symbol).unwrap();
         function.add_attribute(AttributeLoc::Function, ctx.create_enum_attribute(inline_attr, 0));
     }
+
+    // Initialize all global `EXN_*` exception IDs in IRRT with the [`SymbolResolver`].
+    let exn_id_type = ctx.i32_type();
+    let errors = &[
+        ("EXN_INDEX_ERROR", "0:IndexError"),
+        ("EXN_VALUE_ERROR", "0:ValueError"),
+        ("EXN_ASSERTION_ERROR", "0:AssertionError"),
+        ("EXN_TYPE_ERROR", "0:TypeError"),
+    ];
+    for (irrt_name, symbol_name) in errors {
+        let exn_id = symbol_resolver.get_string_id(symbol_name);
+        let exn_id = exn_id_type.const_int(exn_id as u64, false).as_basic_value_enum();
+
+        let global = irrt_mod.get_global(irrt_name).unwrap_or_else(|| {
+            panic!("Exception symbol name '{irrt_name}' should exist in the IRRT LLVM module")
+        });
+        global.set_initializer(&exn_id);
+    }
+
     irrt_mod
 }
 

nac3core/src/typecheck/magic_methods.rs

@@ -520,6 +520,23 @@ pub fn typeof_binop(
         }
 
         Operator::MatMult => {
+            // NOTE: NumPy matmul's LHS and RHS must both be ndarrays. Scalars are not allowed.
+            match (&*unifier.get_ty(lhs), &*unifier.get_ty(rhs)) {
+                (
+                    TypeEnum::TObj { obj_id: lhs_obj_id, .. },
+                    TypeEnum::TObj { obj_id: rhs_obj_id, .. },
+                ) if *lhs_obj_id == primitives.ndarray.obj_id(unifier).unwrap()
+                    && *rhs_obj_id == primitives.ndarray.obj_id(unifier).unwrap() =>
+                {
+                    // LHS and RHS have valid types
+                }
+                _ => {
+                    let lhs_str = unifier.stringify(lhs);
+                    let rhs_str = unifier.stringify(rhs);
+                    return Err(format!("ndarray.__matmul__ only accepts ndarray operands, but left operand has type {lhs_str}, and right operand has type {rhs_str}"));
+                }
+            }
+
             let (_, lhs_ndims) = unpack_ndarray_var_tys(unifier, lhs);
             let lhs_ndims = match &*unifier.get_ty_immutable(lhs_ndims) {
                 TypeEnum::TLiteral { values, .. } => {

nac3standalone/src/basic_symbol_resolver.rs

@@ -15,7 +15,6 @@ use std::{collections::HashMap, sync::Arc};
 pub struct ResolverInternal {
     pub id_to_type: Mutex<HashMap<StrRef, Type>>,
     pub id_to_def: Mutex<HashMap<StrRef, DefinitionId>>,
-    pub class_names: Mutex<HashMap<StrRef, Type>>,
     pub module_globals: Mutex<HashMap<StrRef, SymbolValue>>,
     pub str_store: Mutex<HashMap<String, i32>>,
 }

nac3standalone/src/main.rs

@@ -306,7 +306,6 @@ fn main() {
     let internal_resolver: Arc<ResolverInternal> = ResolverInternal {
         id_to_type: builtins_ty.into(),
         id_to_def: builtins_def.into(),
-        class_names: Mutex::default(),
         module_globals: Mutex::default(),
         str_store: Mutex::default(),
     }
@@ -314,6 +313,15 @@ fn main() {
     let resolver =
         Arc::new(Resolver(internal_resolver.clone())) as Arc<dyn SymbolResolver + Send + Sync>;
 
+    let context = inkwell::context::Context::create();
+
+    // Process IRRT
+    let irrt = load_irrt(&context, resolver.as_ref());
+    if emit_llvm {
+        irrt.write_bitcode_to_path(Path::new("irrt.bc"));
+    }
+
+    // Process the Python script
     let parser_result = parser::parse_program(&program, file_name.into()).unwrap();
 
     for stmt in parser_result {
@@ -418,8 +426,8 @@ fn main() {
     registry.add_task(task);
     registry.wait_tasks_complete(handles);
 
+    // Link all modules together into `main`
     let buffers = membuffers.lock();
-    let context = inkwell::context::Context::create();
     let main = context
         .create_module_from_ir(MemoryBuffer::create_from_memory_range(&buffers[0], "main"))
         .unwrap();
@@ -439,12 +447,9 @@ fn main() {
         main.link_in_module(other).unwrap();
     }
 
-    let irrt = load_irrt(&context);
-    if emit_llvm {
-        irrt.write_bitcode_to_path(Path::new("irrt.bc"));
-    }
     main.link_in_module(irrt).unwrap();
 
+    // Private all functions except "run"
     let mut function_iter = main.get_first_function();
     while let Some(func) = function_iter {
         if func.count_basic_blocks() > 0 && func.get_name().to_str().unwrap() != "run" {
@@ -453,6 +458,7 @@ fn main() {
         function_iter = func.get_next_function();
     }
 
+    // Optimize `main`
     let target_machine = llvm_options
         .target
         .create_target_machine(llvm_options.opt_level)
@@ -466,6 +472,7 @@ fn main() {
         panic!("Failed to run optimization for module `main`: {}", err.to_string());
     }
 
+    // Write output
     target_machine
         .write_to_file(&main, FileType::Object, Path::new("module.o"))
         .expect("couldn't write module to file");