WIP

flake.nix

@@ -13,6 +13,7 @@
           ''
           mkdir -p $out/bin
           ln -s ${pkgs.llvmPackages_14.clang-unwrapped}/bin/clang $out/bin/clang-irrt
+          ln -s ${pkgs.llvmPackages_14.clang}/bin/clang $out/bin/clang-irrt-test
           ln -s ${pkgs.llvmPackages_14.llvm.out}/bin/llvm-as $out/bin/llvm-as-irrt
           '';
         nac3artiq = pkgs.python3Packages.toPythonModule (
@@ -23,6 +24,7 @@
             cargoLock = {
               lockFile = ./Cargo.lock;
             };
+            cargoTestFlags = [ "--features" "test" ];
             passthru.cargoLock = cargoLock;
             nativeBuildInputs = [ pkgs.python3 pkgs.llvmPackages_14.clang llvm-tools-irrt pkgs.llvmPackages_14.llvm.out llvm-nac3 ];
             buildInputs = [ pkgs.python3 llvm-nac3 ];
@@ -161,7 +163,10 @@
           clippy
           pre-commit
           rustfmt
+          rust-analyzer
         ];
+        # https://nixos.wiki/wiki/Rust#Shell.nix_example
+        RUST_SRC_PATH = "${pkgs.rust.packages.stable.rustPlatform.rustLibSrc}";
       };
       devShells.x86_64-linux.msys2 = pkgs.mkShell {
         name = "nac3-dev-shell-msys2";

nac3core/Cargo.toml

@@ -1,3 +1,6 @@
+[features]
+test = []
+
 [package]
 name = "nac3core"
 version = "0.1.0"

nac3core/build.rs

@@ -7,8 +7,8 @@ use std::{
     process::{Command, Stdio},
 };
 
-fn main() {
-    const FILE: &str = "src/codegen/irrt/irrt.cpp";
+fn compile_irrt(irrt_dir: &Path, out_dir: &Path) {
+    let irrt_cpp_path = irrt_dir.join("irrt.cpp");
 
     /*
      * HACK: Sadly, clang doesn't let us emit generic LLVM bitcode.
@@ -16,7 +16,7 @@ fn main() {
      */
     let flags: &[&str] = &[
         "--target=wasm32",
-        FILE,
+        irrt_cpp_path.to_str().unwrap(),
         "-x",
         "c++",
         "-fno-discard-value-names",
@@ -31,13 +31,14 @@ fn main() {
         "-S",
         "-Wall",
         "-Wextra",
+        "-Werror=return-type",
+        "-I",
+        irrt_dir.to_str().unwrap(),
         "-o",
         "-",
     ];
 
-    println!("cargo:rerun-if-changed={FILE}");
-    let out_dir = env::var("OUT_DIR").unwrap();
-    let out_path = Path::new(&out_dir);
+    println!("cargo:rerun-if-changed={}", out_dir.to_str().unwrap());
 
     let output = Command::new("clang-irrt")
         .args(flags)
@@ -52,7 +53,11 @@ fn main() {
     let output = std::str::from_utf8(&output.stdout).unwrap().replace("\r\n", "\n");
     let mut filtered_output = String::with_capacity(output.len());
 
-    let regex_filter = Regex::new(r"(?ms:^define.*?\}$)|(?m:^declare.*?$)").unwrap();
+    // (?ms:^define.*?\}$) to capture `define` blocks
+    // (?m:^declare.*?$) to capture `declare` blocks
+    // (?m:^%.+?=\s*type\s*\{.+?\}$) to capture `type` declarations
+    let regex_filter =
+        Regex::new(r"(?ms:^define.*?\}$)|(?m:^declare.*?$)|(?m:^%.+?=\s*type\s*\{.+?\}$)").unwrap();
     for f in regex_filter.captures_iter(&output) {
         assert_eq!(f.len(), 1);
         filtered_output.push_str(&f[0]);
@@ -65,18 +70,65 @@ fn main() {
 
     println!("cargo:rerun-if-env-changed=DEBUG_DUMP_IRRT");
     if env::var("DEBUG_DUMP_IRRT").is_ok() {
-        let mut file = File::create(out_path.join("irrt.ll")).unwrap();
+        let mut file = File::create(out_dir.join("irrt.ll")).unwrap();
         file.write_all(output.as_bytes()).unwrap();
-        let mut file = File::create(out_path.join("irrt-filtered.ll")).unwrap();
+        let mut file = File::create(out_dir.join("irrt-filtered.ll")).unwrap();
         file.write_all(filtered_output.as_bytes()).unwrap();
     }
 
     let mut llvm_as = Command::new("llvm-as-irrt")
         .stdin(Stdio::piped())
         .arg("-o")
-        .arg(out_path.join("irrt.bc"))
+        .arg(out_dir.join("irrt.bc"))
         .spawn()
         .unwrap();
     llvm_as.stdin.as_mut().unwrap().write_all(filtered_output.as_bytes()).unwrap();
     assert!(llvm_as.wait().unwrap().success());
 }
+
+fn compile_irrt_test(irrt_dir: &Path, out_dir: &Path) {
+    let irrt_test_cpp_path = irrt_dir.join("irrt_test.cpp");
+    let exe_path = out_dir.join("irrt_test.out");
+
+    let flags: &[&str] = &[
+        irrt_test_cpp_path.to_str().unwrap(),
+        "-x",
+        "c++",
+        "-I",
+        irrt_dir.to_str().unwrap(),
+        "-g",
+        "-fno-discard-value-names",
+        "-O0",
+        "-Wall",
+        "-Wextra",
+        "-Werror=return-type",
+        "-lm", // for `tgamma()`, `lgamma()`
+        "-o",
+        exe_path.to_str().unwrap(),
+    ];
+
+    Command::new("clang-irrt-test")
+        .args(flags)
+        .output()
+        .map(|o| {
+            assert!(o.status.success(), "{}", std::str::from_utf8(&o.stderr).unwrap());
+            o
+        })
+        .unwrap();
+    println!("cargo:rerun-if-changed={}", out_dir.to_str().unwrap());
+}
+
+fn main() {
+    let out_dir = env::var("OUT_DIR").unwrap();
+    let out_dir = Path::new(&out_dir);
+
+    let irrt_dir = Path::new("./irrt");
+
+    compile_irrt(irrt_dir, out_dir);
+
+    // https://github.com/rust-lang/cargo/issues/2549
+    // `cargo test -F test` to also build `irrt_test.cpp
+    if cfg!(feature = "test") {
+        compile_irrt_test(irrt_dir, out_dir);
+    }
+}

nac3core/irrt/irrt.cpp

@@ -0,0 +1,5 @@
+#include "irrt_everything.hpp"
+
+/*
+    This file will be read by `clang-irrt` to conveniently produce LLVM IR for `nac3core/codegen`.
+*/

nac3core/irrt/irrt.hpp

@@ -0,0 +1,437 @@
+#ifndef IRRT_DONT_TYPEDEF_INTS
+typedef _BitInt(8) int8_t;
+typedef unsigned _BitInt(8) uint8_t;
+typedef _BitInt(32) int32_t;
+typedef unsigned _BitInt(32) uint32_t;
+typedef _BitInt(64) int64_t;
+typedef unsigned _BitInt(64) uint64_t;
+#endif
+
+// NDArray indices are always `uint32_t`.
+typedef uint32_t NDIndex;
+// The type of an index or a value describing the length of a range/slice is
+// always `int32_t`.
+typedef int32_t SliceIndex;
+
+template <typename T>
+static T max(T a, T b) {
+    return a > b ? a : b;
+}
+
+template <typename T>
+static T min(T a, 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>
+static 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>
+static 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>
+static 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>
+static 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>
+static 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>
+static 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];
+    }
+}
+
+template<typename SizeT>
+static void __nac3_ndarray_strides_from_shape_impl(
+    SizeT ndims,
+    SizeT *shape,
+    SizeT *dst_strides
+) {
+    SizeT stride_product = 1;
+    for (SizeT i = 0; i < ndims; i++) {
+        int dim_i = ndims - i - 1;
+        dst_strides[dim_i] = stride_product;
+        stride_product *= shape[dim_i];
+    }
+}
+
+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);
+    }
+
+    void __nac3_ndarray_strides_from_shape(uint32_t ndims, uint32_t* shape, uint32_t* dst_strides) {
+        __nac3_ndarray_strides_from_shape_impl(ndims, shape, dst_strides);
+    }
+
+    void __nac3_ndarray_strides_from_shape64(uint64_t ndims, uint64_t* shape, uint64_t* dst_strides) {
+        __nac3_ndarray_strides_from_shape_impl(ndims, shape, dst_strides);
+    }
+}

nac3core/irrt/irrt_basic.hpp

@@ -0,0 +1,216 @@
+#pragma once
+
+#include "irrt_utils.hpp"
+#include "irrt_typedefs.hpp"
+
+/*
+    This header contains IRRT implementations
+    that do not deserved to be categorized (e.g., into numpy, etc.)
+
+    Check out other *.hpp files before including them here!!
+*/
+
+// The type of an index or a value describing the length of a range/slice is
+// always `int32_t`.
+
+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;
+    }
+}
+
+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);
+    }
+}

nac3core/irrt/irrt_everything.hpp

@@ -0,0 +1,14 @@
+#pragma once
+
+#include "irrt_utils.hpp"
+#include "irrt_typedefs.hpp"
+#include "irrt_basic.hpp"
+#include "irrt_slice.hpp"
+#include "irrt_numpy_ndarray.hpp"
+
+/*
+    All IRRT implementations.
+
+    We don't have any pre-compiled objects, so we are writing all implementations in headers and
+    concatenate them with `#include` into one massive source file that contains all the IRRT stuff.
+*/

nac3core/irrt/irrt_numpy_ndarray.hpp

@@ -0,0 +1,466 @@
+#pragma once
+
+#include "irrt_utils.hpp"
+#include "irrt_typedefs.hpp"
+#include "irrt_slice.hpp"
+
+/*
+    NDArray-related implementations.
+`*/
+
+// NDArray indices are always `uint32_t`.
+using NDIndex = uint32_t;
+
+namespace {
+    namespace ndarray_util {
+        template <typename SizeT>
+        static void set_indices_by_nth(SizeT ndims, const SizeT* shape, SizeT* indices, SizeT nth) {
+            for (int32_t i = 0; i < ndims; i++) {
+                int32_t dim_i = ndims - i - 1;
+                int32_t dim = shape[dim_i];
+
+                indices[dim_i] = nth % dim;
+                nth /= dim;
+            }
+        }
+
+        // Compute the strides of an ndarray given an ndarray `shape`
+        // and assuming that the ndarray is *fully C-contagious*.
+        //
+        // You might want to read up on https://ajcr.net/stride-guide-part-1/.
+        template <typename SizeT>
+        static void set_strides_by_shape(SizeT itemsize, SizeT ndims, SizeT* dst_strides, const SizeT* shape) {
+            SizeT stride_product = 1;
+            for (SizeT i = 0; i < ndims; i++) {
+                int dim_i = ndims - i - 1;
+                dst_strides[dim_i] = stride_product * itemsize;
+                stride_product *= shape[dim_i];
+            }
+        }
+
+        // Compute the size/# of elements of an ndarray given its shape
+        template <typename SizeT>
+        static SizeT calc_size_from_shape(SizeT ndims, const SizeT* shape) {
+            SizeT size = 1;
+            for (SizeT dim_i = 0; dim_i < ndims; dim_i++) size *= shape[dim_i];
+            return size;
+        }
+
+        template <typename SizeT>
+        static bool can_broadcast_shape_to(
+            const SizeT target_ndims,
+            const SizeT *target_shape,
+            const SizeT src_ndims,
+            const SizeT *src_shape
+        ) {
+            /*
+                // See https://numpy.org/doc/stable/user/basics.broadcasting.html
+
+                This function handles this example:
+                ```
+                Image  (3d array): 256 x 256 x 3
+                Scale  (1d array):             3
+                Result (3d array): 256 x 256 x 3
+                ```
+
+                Other interesting examples to consider:
+                - `can_broadcast_shape_to([3], [1, 1, 1, 1, 3]) == true`
+                - `can_broadcast_shape_to([3], [3, 1]) == false`
+                - `can_broadcast_shape_to([256, 256, 3], [256, 1, 3]) == true`
+
+                In cases when the shapes contain zero(es):
+                - `can_broadcast_shape_to([0], [1]) == true`
+                - `can_broadcast_shape_to([0], [2]) == false`
+                - `can_broadcast_shape_to([0, 4, 0, 0], [1]) == true`
+                - `can_broadcast_shape_to([0, 4, 0, 0], [1, 1, 1, 1]) == true`
+                - `can_broadcast_shape_to([0, 4, 0, 0], [1, 4, 1, 1]) == true`
+                - `can_broadcast_shape_to([4, 3], [0, 3]) == false`
+                - `can_broadcast_shape_to([4, 3], [0, 0]) == false`
+            */
+
+            // This is essentially doing the following in Python:
+            // `for target_dim, src_dim in itertools.zip_longest(target_shape[::-1], src_shape[::-1], fillvalue=1)`
+            for (SizeT i = 0; i < max(target_ndims, src_ndims); i++) {
+                SizeT target_dim_i = target_ndims - i - 1;
+                SizeT src_dim_i = src_ndims - i - 1;
+
+                bool target_dim_exists = target_dim_i >= 0;
+                bool src_dim_exists = src_dim_i >= 0;
+
+                SizeT target_dim = target_dim_exists ? target_shape[target_dim_i] : 1;
+                SizeT src_dim = src_dim_exists ? src_shape[src_dim_i] : 1;
+
+                bool ok = src_dim == 1 || target_dim == src_dim;
+                if (!ok) return false;
+            }
+
+            return true;
+        }
+    }
+
+    typedef uint8_t NDSliceType;
+    extern "C" {
+        const NDSliceType INPUT_SLICE_TYPE_INDEX = 0;
+        const NDSliceType INPUT_SLICE_TYPE_SLICE = 1;
+    }
+
+    struct NDSlice {
+        // A poor-man's `std::variant<int, UserRange>`
+        NDSliceType type;
+
+        /*
+            if type == INPUT_SLICE_TYPE_INDEX => `slice` points to a single `SizeT`
+            if type == INPUT_SLICE_TYPE_SLICE => `slice` points to a single `UserRange`
+        */
+        uint8_t *slice;
+    };
+
+    namespace ndarray_util {
+        template<typename SizeT>
+        SizeT deduce_ndims_after_slicing(SizeT ndims, SizeT num_slices, const NDSlice *slices) {
+            irrt_assert(num_slices <= ndims);
+
+            SizeT final_ndims = ndims;
+            for (SizeT i = 0; i < num_slices; i++) {
+                if (slices[i].type == INPUT_SLICE_TYPE_INDEX) {
+                    final_ndims--; // An integer slice demotes the rank by 1
+                }
+            }
+            return final_ndims;
+        }
+    }
+
+    template <typename SizeT>
+    struct NDArrayIndicesIter {
+        SizeT ndims;
+        const SizeT *shape;
+        SizeT *indices;
+
+        void set_indices_zero() {
+            __builtin_memset(indices, 0, sizeof(SizeT) * ndims);
+        }
+
+        void next() {
+            for (SizeT i = 0; i < ndims; i++) {
+                SizeT dim_i = ndims - i - 1;
+
+                indices[dim_i]++;
+                if (indices[dim_i] < shape[dim_i]) {
+                    break;
+                } else {
+                    indices[dim_i] = 0;
+                }
+            }
+        }
+    };
+
+    // The NDArray object. `SizeT` is the *signed* size type of this ndarray.
+    //
+    // NOTE: The order of fields is IMPORTANT. DON'T TOUCH IT
+    //
+    // Some resources you might find helpful:
+    // - The official numpy implementations:
+    //   - https://github.com/numpy/numpy/blob/735a477f0bc2b5b84d0e72d92f224bde78d4e069/doc/source/reference/c-api/types-and-structures.rst
+    // - On strides (about reshaping, slicing, C-contagiousness, etc)
+    //   - https://ajcr.net/stride-guide-part-1/.
+    //   - https://ajcr.net/stride-guide-part-2/.
+    //   - https://ajcr.net/stride-guide-part-3/.
+    template <typename SizeT>
+    struct NDArray {
+        // The underlying data this `ndarray` is pointing to.
+        //
+        // NOTE: Formally this should be of type `void *`, but clang
+        // translates `void *` to `i8 *` when run with `-S -emit-llvm`,
+        // so we will put `uint8_t *` here for clarity.
+        uint8_t *data;
+
+        // The number of bytes of a single element in `data`.
+        //
+        // The `SizeT` is treated as `unsigned`.
+        SizeT itemsize;
+
+        // The number of dimensions of this shape.
+        //
+        // The `SizeT` is treated as `unsigned`.
+        SizeT ndims;
+
+        // Array shape, with length equal to `ndims`.
+        //
+        // The `SizeT` is treated as `unsigned`.
+        //
+        // NOTE: `shape` can contain 0.
+        //   (those appear when the user makes an out of bounds slice into an ndarray, e.g., `np.zeros((3, 3))[400:].shape == (0, 3)`)
+        SizeT *shape;
+
+        // Array strides (stride value is in number of bytes, NOT number of elements), with length equal to `ndims`.
+        //
+        // The `SizeT` is treated as `signed`.
+        //
+        // NOTE: `strides` can have negative numbers.
+        //   (those appear when there is a slice with a negative step, e.g., `my_array[::-1]`)
+        SizeT *strides;
+
+        // Calculate the size/# of elements of an `ndarray`.
+        // This function corresponds to `np.size(<ndarray>)` or `ndarray.size`
+        SizeT size() {
+            return ndarray_util::calc_size_from_shape(ndims, shape);
+        }
+
+        // Calculate the number of bytes of its content of an `ndarray` *in its view*.
+        // This function corresponds to `ndarray.nbytes`
+        SizeT nbytes() {
+            return this->size() * itemsize;
+        }
+
+        void set_value_at_pelement(uint8_t* pelement, const uint8_t* pvalue) {
+            __builtin_memcpy(pelement, pvalue, itemsize);
+        }
+
+        uint8_t* get_pelement(const SizeT *indices) {
+            uint8_t* element = data;
+            for (SizeT dim_i = 0; dim_i < ndims; dim_i++)
+                element += indices[dim_i] * strides[dim_i];
+            return element;
+        }
+
+        uint8_t* get_nth_pelement(SizeT nth) {
+            irrt_assert(0 <= nth);
+            irrt_assert(nth < this->size());
+
+            SizeT* indices = (SizeT*) __builtin_alloca(sizeof(SizeT) * this->ndims);
+            ndarray_util::set_indices_by_nth(this->ndims, this->shape, indices, nth);
+            return get_pelement(indices);
+        }
+
+        // Get pointer to the first element of this ndarray, assuming
+        // `this->size() > 0`, i.e., not "degenerate" due to zeroes in `this->shape`)
+        //
+        // This is particularly useful for when the ndarray is just containing a single scalar.
+        uint8_t* get_first_pelement() {
+            irrt_assert(this->size() > 0);
+            return this->data; // ...It is simply `this->data`
+        }
+
+        // Is the given `indices` valid/in-bounds?
+        bool in_bounds(const SizeT *indices) {
+            for (SizeT dim_i = 0; dim_i < ndims; dim_i++) {
+                bool dim_ok = indices[dim_i] < shape[dim_i];
+                if (!dim_ok) return false;
+            }
+            return true;
+        }
+
+        // Fill the ndarray with a value
+        void fill_generic(const uint8_t* pvalue) {
+            NDArrayIndicesIter<SizeT> iter;
+            iter.ndims = this->ndims;
+            iter.shape = this->shape;
+            iter.indices = (SizeT*) __builtin_alloca(sizeof(SizeT) * ndims);
+            iter.set_indices_zero();
+
+            for (SizeT i = 0; i < this->size(); i++, iter.next()) {
+                uint8_t* pelement = get_pelement(iter.indices);
+                set_value_at_pelement(pelement, pvalue);
+            }
+        }
+
+        // Set the strides of the ndarray with `ndarray_util::set_strides_by_shape`
+        void set_strides_by_shape() {
+            ndarray_util::set_strides_by_shape(itemsize, ndims, strides, shape);
+        }
+
+        // https://numpy.org/doc/stable/reference/generated/numpy.eye.html
+        void set_to_eye(SizeT k, const uint8_t* zero_pvalue, const uint8_t* one_pvalue) {
+            __builtin_assume(ndims == 2);
+
+            // TODO: Better implementation
+
+            fill_generic(zero_pvalue);
+            for (SizeT i = 0; i < min(shape[0], shape[1]); i++) {
+                SizeT row = i;
+                SizeT col = i + k;
+                SizeT indices[2] = { row, col };
+
+                if (!in_bounds(indices)) continue;
+
+                uint8_t* pelement = get_pelement(indices);
+                set_value_at_pelement(pelement, one_pvalue);
+            }
+        }
+
+        // To support numpy complex slices (e.g., `my_array[:50:2,4,:2:-1]`)
+        //
+        // Things assumed by this function:
+        // - `dst_ndarray` is allocated by the caller
+        // - `dst_ndarray.ndims` has the correct value (according to `ndarray_util::deduce_ndims_after_slicing`).
+        //   - ... and `dst_ndarray.shape` and `dst_ndarray.strides` have been allocated by the caller as well
+        //
+        // Other notes:
+        // - `dst_ndarray->data` does not have to be set, it will be derived.
+        // - `dst_ndarray->itemsize` does not have to be set, it will be set to `this->itemsize`
+        // - `dst_ndarray->shape` and `dst_ndarray.strides` can contain empty values
+        void slice(SizeT num_ndslices, NDSlice* ndslices, NDArray<SizeT>* dst_ndarray) {
+            // REFERENCE CODE (check out `_index_helper` in `__getitem__`):
+            //   https://github.com/wadetb/tinynumpy/blob/0d23d22e07062ffab2afa287374c7b366eebdda1/tinynumpy/tinynumpy.py#L652
+
+            irrt_assert(dst_ndarray->ndims == ndarray_util::deduce_ndims_after_slicing(this->ndims, num_ndslices, ndslices));
+
+            dst_ndarray->data = this->data;
+
+            SizeT this_axis = 0;
+            SizeT dst_axis = 0;
+
+            for (SizeT i = 0; i < num_ndslices; i++) {
+                NDSlice *ndslice = &ndslices[i];
+                if (ndslice->type == INPUT_SLICE_TYPE_INDEX) {
+                    // Handle when the ndslice is just a single (possibly negative) integer
+                    // e.g., `my_array[::2, -5, ::-1]`
+                    //                      ^^------ like this
+                    SizeT index_user = *((SizeT*) ndslice->slice);
+                    SizeT index = resolve_index_in_length(this->shape[this_axis], index_user);
+                    dst_ndarray->data += index * this->strides[this_axis]; // Add offset
+
+                    // Next
+                    this_axis++;
+                } else if (ndslice->type == INPUT_SLICE_TYPE_SLICE) {
+                    // Handle when the ndslice is a slice (represented by UserSlice in IRRT)
+                    // e.g., `my_array[::2, -5, ::-1]`
+                    //                 ^^^------^^^^----- like these
+                    UserSlice<SizeT>* user_slice = (UserSlice<SizeT>*) ndslice->slice;
+                    Slice<SizeT> slice = user_slice->indices(this->shape[this_axis]); // To resolve negative indices and other funny stuff written by the user
+
+                    // NOTE: There is no need to write special code to handle negative steps/strides.
+                    // This simple implementation meticulously handles both positive and negative steps/strides.
+                    // Check out the tinynumpy and IRRT's test cases if you are not convinced.
+                    dst_ndarray->data += slice.start * this->strides[this_axis]; // Add offset (NOTE: no need to `* itemsize`, strides count in # of bytes)
+                    dst_ndarray->strides[dst_axis] = slice.step * this->strides[this_axis]; // Determine stride
+                    dst_ndarray->shape[dst_axis] = slice.len(); // Determine shape dimension
+
+                    // Next
+                    dst_axis++;
+                    this_axis++;
+                } else {
+                    __builtin_unreachable();
+                }
+            }
+
+            irrt_assert(dst_axis == dst_ndarray->ndims); // Sanity check on the implementation
+        }
+
+        // Similar to `np.broadcast_to(<ndarray>, <target_shape>)`
+        // Assumptions:
+        //   - `this` has to be fully initialized.
+        //   - `dst_ndarray->ndims` has to be set.
+        //   - `dst_ndarray->shape` has to be set, this determines the shape `this` broadcasts to.
+        //
+        // Other notes:
+        //   - `dst_ndarray->data` does not have to be set, it will be set to `this->data`.
+        //   - `dst_ndarray->itemsize` does not have to be set, it will be set to `this->data`.
+        //   - `dst_ndarray->strides` does not have to be set, it will be overwritten.
+        //
+        // Cautions:
+        //   ```
+        //   xs = np.zeros((4,))
+        //   ys = np.zero((4, 1))
+        //   ys[:] = xs # ok
+        // 
+        //   xs = np.zeros((1, 4))
+        //   ys = np.zero((4,))
+        //   ys[:] = xs # allowed
+        //   # However `np.broadcast_to(xs, (4,))` would fails, as per numpy's broadcasting rule.
+        //   # and apparently numpy will "deprecate" this? SEE https://github.com/numpy/numpy/issues/21744
+        //   # This implementation will NOT support this assignment.
+        //   ```
+        void broadcast_to(NDArray<SizeT>* dst_ndarray) {
+            dst_ndarray->data = this->data;
+            dst_ndarray->itemsize = this->itemsize;
+
+            irrt_assert(
+                ndarray_util::can_broadcast_shape_to(
+                    dst_ndarray->ndims,
+                    dst_ndarray->shape,
+                    this->ndims,
+                    this->shape
+                )
+            );
+
+            SizeT stride_product = 1;
+            for (SizeT i = 0; i < max(this->ndims, dst_ndarray->ndims); i++) {
+                SizeT this_dim_i = this->ndims - i - 1;
+                SizeT dst_dim_i = dst_ndarray->ndims - i - 1;
+
+                bool this_dim_exists = this_dim_i >= 0;
+                bool dst_dim_exists = dst_dim_i >= 0;
+
+                // TODO: Explain how this works
+                bool c1 = this_dim_exists && this->shape[this_dim_i] == 1;
+                bool c2 = dst_dim_exists && dst_ndarray->shape[dst_dim_i] != 1;
+                if (!this_dim_exists || (c1 && c2)) {
+                    dst_ndarray->strides[dst_dim_i] = 0; // Freeze it in-place
+                } else {
+                    dst_ndarray->strides[dst_dim_i] = stride_product * this->itemsize;
+                    stride_product *= this->shape[this_dim_i]; // NOTE: this_dim_exist must be true here.
+                }
+            }
+        }
+
+        // Simulates `this_ndarray[:] = src_ndarray`, with automatic broadcasting.
+        // Caution on https://github.com/numpy/numpy/issues/21744
+        // Also see `NDArray::broadcast_to`
+        void assign_with(NDArray<SizeT>* src_ndarray) {
+            irrt_assert(
+                ndarray_util::can_broadcast_shape_to(
+                    this->ndims,
+                    this->shape,
+                    src_ndarray->ndims,
+                    src_ndarray->shape
+                )
+            );
+
+            // Broadcast the `src_ndarray` to make the reading process *much* easier
+            SizeT* broadcasted_src_ndarray_strides = __builtin_alloca(sizeof(SizeT) * this->ndims); // Remember to allocate strides beforehand
+            NDArray<SizeT> broadcasted_src_ndarray = {
+                .ndims = this->ndims,
+                .shape = this->shape,
+                .strides = broadcasted_src_ndarray_strides
+            };
+            src_ndarray->broadcast_to(&broadcasted_src_ndarray);
+
+            // Using iter instead of `get_nth_pelement` because it is slightly faster
+            SizeT* indices = __builtin_alloca(sizeof(SizeT) * this->ndims);
+            auto iter = NDArrayIndicesIter<SizeT> {
+                .ndims = this->ndims,
+                .shape = this->shape,
+                .indices = indices
+            };
+            const SizeT this_size = this->size();
+            for (SizeT i = 0; i < this_size; i++, iter.next()) {
+                uint8_t* src_pelement = broadcasted_src_ndarray_strides->get_pelement(indices);
+                uint8_t* this_pelement = this->get_pelement(indices);
+                this->set_value_at_pelement(src_pelement, src_pelement);
+            }
+        }
+    };
+}
+
+extern "C" {
+    uint32_t __nac3_ndarray_size(NDArray<int32_t>* ndarray) {
+        return ndarray->size();
+    }
+
+    uint64_t __nac3_ndarray_size64(NDArray<int64_t>* ndarray) {
+        return ndarray->size();
+    }
+
+    void __nac3_ndarray_fill_generic(NDArray<int32_t>* ndarray, uint8_t* pvalue) {
+        ndarray->fill_generic(pvalue);
+    }
+
+    void __nac3_ndarray_fill_generic64(NDArray<int64_t>* ndarray, uint8_t* pvalue) {
+        ndarray->fill_generic(pvalue);
+    }
+
+    // void __nac3_ndarray_slice(NDArray<int32_t>* ndarray, int32_t num_slices, NDSlice<int32_t> *slices, NDArray<int32_t> *dst_ndarray) {
+    //     // ndarray->slice(num_slices, slices, dst_ndarray);
+    // }
+}

nac3core/irrt/irrt_slice.hpp

@@ -0,0 +1,80 @@
+#pragma once
+
+#include "irrt_utils.hpp"
+#include "irrt_typedefs.hpp"
+
+namespace {
+    // A proper slice in IRRT, all negative indices have be resolved to absolute values.
+    // Even though nac3core's slices are always `int32_t`, we will template slice anyway
+    // since this struct is used as a general utility.
+    template <typename T>
+    struct Slice {
+        T start;
+        T stop;
+        T step;
+
+        // The length/The number of elements of the slice if it were a range,
+        // i.e., the value of `len(range(this->start, this->stop, this->end))`
+        T len() {
+            T diff = stop - 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;
+            }
+        }
+    };
+
+    template<typename T>
+    T resolve_index_in_length(T length, T index) {
+        irrt_assert(length >= 0);
+        if (index < 0) {
+            // Remember that index is negative, so do a plus here
+            return max(length + index, 0);
+        } else {
+            return min(length, index);
+        }
+    }
+
+    // NOTE: using a bitfield for the `*_defined` is better, at the
+    // cost of a more annoying implementation in nac3core inkwell
+    template <typename T>
+    struct UserSlice {
+        uint8_t start_defined;
+        T start;
+
+        uint8_t stop_defined;
+        T stop;
+
+        uint8_t step_defined;
+        T step;
+
+        // Like Python's `slice(start, stop, step).indices(length)`
+        Slice<T> indices(T length) {
+            // NOTE: This function implements Python's `slice.indices` *FAITHFULLY*.
+            // SEE: https://github.com/python/cpython/blob/f62161837e68c1c77961435f1b954412dd5c2b65/Objects/sliceobject.c#L546
+            irrt_assert(length >= 0);
+            irrt_assert(!step_defined || step != 0); // step_defined -> step != 0; step cannot be zero if specified by user
+
+            Slice<T> result;
+            result.step = step_defined ? step : 1;
+            bool step_is_negative = result.step < 0;
+
+            if (start_defined) {
+                result.start = resolve_index_in_length(length, start);
+            } else {
+                result.start = step_is_negative ? length - 1 : 0;
+            }
+            
+            if (stop_defined) {
+                result.stop = resolve_index_in_length(length, stop);
+            } else {
+                result.stop = step_is_negative ? -1 : length;
+            }
+
+            return result;
+        }
+    };
+}

nac3core/irrt/irrt_test.cpp

@@ -0,0 +1,658 @@
+// This file will be compiled like a real C++ program,
+// and we do have the luxury to use the standard libraries.
+// That is if the nix flakes do not have issues... especially on msys2...
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+
+// Set `IRRT_DONT_TYPEDEF_INTS` because `cstdint` defines them
+#define IRRT_DONT_TYPEDEF_INTS
+#include "irrt_everything.hpp"
+
+void test_fail() {
+    printf("[!] Test failed\n");
+    exit(1);
+}
+
+void __begin_test(const char* function_name, const char* file, int line) {
+    printf("######### Running %s @ %s:%d\n", function_name, file, line);
+}
+
+#define BEGIN_TEST() __begin_test(__FUNCTION__, __FILE__, __LINE__)
+
+template <typename T>
+void debug_print_array(const char* format, int len, T* as) {
+    printf("[");
+    for (int i = 0; i < len; i++) {
+        if (i != 0) printf(", ");
+        printf(format, as[i]);
+    }
+    printf("]");
+}
+
+template <typename T>
+void assert_arrays_match(const char* label, const char* format, int len, T* expected, T* got) {
+    if (!arrays_match(len, expected, got)) {
+        printf(">>>>>>> %s\n", label);
+        printf("    Expecting = ");
+        debug_print_array(format, len, expected);
+        printf("\n");
+        printf("          Got = ");
+        debug_print_array(format, len, got);
+        printf("\n");
+        test_fail();
+    }
+}
+
+template <typename T>
+void assert_values_match(const char* label, const char* format, T expected, T got) {
+    if (expected != got) {
+        printf(">>>>>>> %s\n", label);
+        printf("    Expecting = ");
+        printf(format, expected);
+        printf("\n");
+        printf("          Got = ");
+        printf(format, got);
+        printf("\n");
+        test_fail();
+    }
+}
+
+void print_repeated(const char *str, int count) {
+    for (int i = 0; i < count; i++) {
+        printf("%s", str);
+    }
+}
+
+template<typename SizeT, typename ElementT>
+void __print_ndarray_aux(const char *format, bool first, bool last, SizeT* cursor, SizeT depth, NDArray<SizeT>* ndarray) {
+    // A really lazy recursive implementation
+    
+    // Add left padding unless its the first entry (since there would be "[[[" before it)
+    if (!first) {
+        print_repeated(" ", depth);
+    }
+
+    const SizeT dim = ndarray->shape[depth];
+    if (depth + 1 == ndarray->ndims) {
+        // Recursed down to last dimension, print the values in a nice list
+        printf("[");
+
+        SizeT* indices = (SizeT*) __builtin_alloca(sizeof(SizeT) * ndarray->ndims);
+        for (SizeT i = 0; i < dim; i++) {
+            ndarray_util::set_indices_by_nth(ndarray->ndims, ndarray->shape, indices, *cursor);
+            ElementT* pelement = (ElementT*) ndarray->get_pelement(indices);
+            ElementT element = *pelement;
+
+            if (i != 0) printf(", "); // List delimiter
+            printf(format, element);
+            printf("(@");
+            debug_print_array("%d", ndarray->ndims, indices);
+            printf(")");
+
+            (*cursor)++;
+        }
+        printf("]");
+    } else {
+        printf("[");
+        for (SizeT i = 0; i < ndarray->shape[depth]; i++) {
+            __print_ndarray_aux<SizeT, ElementT>(
+                format,
+                i == 0, // first?
+                i + 1 == dim, // last?
+                cursor,
+                depth + 1,
+                ndarray
+            );
+        }
+        printf("]");
+    }
+
+    // Add newline unless its the last entry (since there will be "]]]" after it)
+    if (!last) {
+        print_repeated("\n", depth);
+    }
+}
+
+template<typename SizeT, typename ElementT>
+void print_ndarray(const char *format, NDArray<SizeT>* ndarray) {
+    if (ndarray->ndims == 0) {
+        printf("<empty ndarray>");
+    } else {
+        SizeT cursor = 0;
+        __print_ndarray_aux<SizeT, ElementT>(format, true, true, &cursor, 0, ndarray);
+    }
+    printf("\n");
+}
+
+void test_calc_size_from_shape_normal() {
+    // Test shapes with normal values
+    BEGIN_TEST();
+
+    int32_t shape[4] = { 2, 3, 5, 7 };
+    assert_values_match("size", "%d", 210, ndarray_util::calc_size_from_shape<int32_t>(4, shape));
+}
+
+void test_calc_size_from_shape_has_zero() {
+    // Test shapes with 0 in them
+    BEGIN_TEST();
+
+    int32_t shape[4] = { 2, 0, 5, 7 };
+    assert_values_match("size", "%d", 0, ndarray_util::calc_size_from_shape<int32_t>(4, shape));
+}
+
+void test_set_strides_by_shape() {
+    // Test `set_strides_by_shape()`
+    BEGIN_TEST();
+
+    int32_t shape[4] = { 99, 3, 5, 7 };
+    int32_t strides[4] = { 0 };
+    ndarray_util::set_strides_by_shape((int32_t) sizeof(int32_t), 4, strides, shape);
+
+    int32_t expected_strides[4] = {
+        105 * sizeof(int32_t),
+        35 * sizeof(int32_t),
+        7 * sizeof(int32_t),
+        1 * sizeof(int32_t)
+    };
+    assert_arrays_match("strides", "%u", 4u, expected_strides, strides);
+}
+
+void test_ndarray_indices_iter_normal() {
+    // Test NDArrayIndicesIter normal behavior
+    BEGIN_TEST();
+
+    int32_t shape[3] = { 1, 2, 3 };
+    int32_t indices[3] = { 0, 0, 0 };
+    auto iter = NDArrayIndicesIter<int32_t> {
+        .ndims = 3,
+        .shape = shape,
+        .indices = indices
+    };
+
+    assert_arrays_match("indices #0", "%u", 3u, iter.indices, (int32_t[3]) { 0, 0, 0 });
+    iter.next();
+    assert_arrays_match("indices #1", "%u", 3u, iter.indices, (int32_t[3]) { 0, 0, 1 });
+    iter.next();
+    assert_arrays_match("indices #2", "%u", 3u, iter.indices, (int32_t[3]) { 0, 0, 2 });
+    iter.next();
+    assert_arrays_match("indices #3", "%u", 3u, iter.indices, (int32_t[3]) { 0, 1, 0 });
+    iter.next();
+    assert_arrays_match("indices #4", "%u", 3u, iter.indices, (int32_t[3]) { 0, 1, 1 });
+    iter.next();
+    assert_arrays_match("indices #5", "%u", 3u, iter.indices, (int32_t[3]) { 0, 1, 2 });
+    iter.next();
+    assert_arrays_match("indices #6", "%u", 3u, iter.indices, (int32_t[3]) { 0, 0, 0 }); // Loops back
+    iter.next();
+    assert_arrays_match("indices #7", "%u", 3u, iter.indices, (int32_t[3]) { 0, 0, 1 });
+}
+
+void test_ndarray_fill_generic() {
+    // Test ndarray fill_generic
+    BEGIN_TEST();
+
+    // Choose a type that's neither int32_t nor uint64_t (candidates of SizeT) to spice it up
+    // Also make all the octets non-zero, to see if `memcpy` in `fill_generic` is working perfectly.
+    uint16_t fill_value = 0xFACE;
+
+    uint16_t in_data[6] = { 100, 101, 102, 103, 104, 105 }; // Fill `data` with values that != `999`
+    int32_t in_itemsize = sizeof(uint16_t);
+    const int32_t in_ndims = 2;
+    int32_t in_shape[in_ndims] = { 2, 3 };
+    int32_t in_strides[in_ndims] = {};
+    NDArray<int32_t> ndarray = {
+        .data = (uint8_t*) in_data,
+        .itemsize = in_itemsize,
+        .ndims = in_ndims,
+        .shape = in_shape,
+        .strides = in_strides,
+    };
+    ndarray.set_strides_by_shape();
+    ndarray.fill_generic((uint8_t*) &fill_value); // `fill_generic` here
+
+    uint16_t expected_data[6] = { fill_value, fill_value, fill_value, fill_value, fill_value, fill_value };
+    assert_arrays_match("data", "0x%hX", 6, expected_data, in_data);
+}
+
+void test_ndarray_set_to_eye() {
+    // Test `set_to_eye` behavior (helper function to implement `np.eye()`)
+    BEGIN_TEST();
+
+    double in_data[9] = { 99.0, 99.0, 99.0, 99.0, 99.0, 99.0, 99.0, 99.0, 99.0 };
+    int32_t in_itemsize = sizeof(double);
+    const int32_t in_ndims = 2;
+    int32_t in_shape[in_ndims] = { 3, 3 };
+    int32_t in_strides[in_ndims] = {};
+    NDArray<int32_t> ndarray = {
+        .data = (uint8_t*) in_data,
+        .itemsize = in_itemsize,
+        .ndims = in_ndims,
+        .shape = in_shape,
+        .strides = in_strides,
+    };
+    ndarray.set_strides_by_shape();
+
+    double zero = 0.0;
+    double one = 1.0;
+    ndarray.set_to_eye(1, (uint8_t*) &zero, (uint8_t*) &one);
+
+    assert_values_match("in_data[0]", "%f", 0.0, in_data[0]);
+    assert_values_match("in_data[1]", "%f", 1.0, in_data[1]);
+    assert_values_match("in_data[2]", "%f", 0.0, in_data[2]);
+    assert_values_match("in_data[3]", "%f", 0.0, in_data[3]);
+    assert_values_match("in_data[4]", "%f", 0.0, in_data[4]);
+    assert_values_match("in_data[5]", "%f", 1.0, in_data[5]);
+    assert_values_match("in_data[6]", "%f", 0.0, in_data[6]);
+    assert_values_match("in_data[7]", "%f", 0.0, in_data[7]);
+    assert_values_match("in_data[8]", "%f", 0.0, in_data[8]);
+}
+
+void test_slice_1() {
+    // Test `slice(5, None, None).indices(100) == slice(5, 100, 1)`
+    BEGIN_TEST();
+
+    UserSlice<int> user_slice = {
+        .start_defined = 1,
+        .start = 5,
+        .stop_defined = 0,
+        .step_defined = 0,
+    };
+
+    auto slice = user_slice.indices(100);
+    assert_values_match("start", "%d", 5, slice.start);
+    assert_values_match("stop", "%d", 100, slice.stop);
+    assert_values_match("step", "%d", 1, slice.step);
+}
+
+void test_slice_2() {
+    // Test `slice(400, 999, None).indices(100) == slice(100, 100, 1)`
+    BEGIN_TEST();
+
+    UserSlice<int> user_slice = {
+        .start_defined = 1,
+        .start = 400,
+        .stop_defined = 0,
+        .step_defined = 0,
+    };
+
+    auto slice = user_slice.indices(100);
+    assert_values_match("start", "%d", 100, slice.start);
+    assert_values_match("stop", "%d", 100, slice.stop);
+    assert_values_match("step", "%d", 1, slice.step);
+}
+
+void test_slice_3() {
+    // Test `slice(-10, -5, None).indices(100) == slice(90, 95, 1)`
+    BEGIN_TEST();
+
+    UserSlice<int> user_slice = {
+        .start_defined = 1,
+        .start = -10,
+        .stop_defined = 1,
+        .stop = -5,
+        .step_defined = 0,
+    };
+
+    auto slice = user_slice.indices(100);
+    assert_values_match("start", "%d", 90, slice.start);
+    assert_values_match("stop", "%d", 95, slice.stop);
+    assert_values_match("step", "%d", 1, slice.step);
+}
+
+void test_slice_4() {
+    // Test `slice(None, None, -5).indices(100) == (99, -1, -5)`
+    BEGIN_TEST();
+
+    UserSlice<int> user_slice = {
+        .start_defined = 0,
+        .stop_defined = 0,
+        .step_defined = 1,
+        .step = -5
+    };
+
+    auto slice = user_slice.indices(100);
+    assert_values_match("start", "%d", 99, slice.start);
+    assert_values_match("stop", "%d", -1, slice.stop);
+    assert_values_match("step", "%d", -5, slice.step);
+}
+
+void test_ndslice_1() {
+    /*
+        Reference Python code:
+        ```python
+        ndarray = np.arange(12, dtype=np.float64).reshape((3, 4));
+        # array([[ 0.,  1.,  2.,  3.],
+        #        [ 4.,  5.,  6.,  7.],
+        #        [ 8.,  9., 10., 11.]])
+
+        dst_ndarray = ndarray[-2:, 1::2]
+        # array([[ 5.,  7.],
+        #        [ 9., 11.]])
+
+        assert dst_ndarray.shape == (2, 2)
+        assert dst_ndarray.strides == (32, 16)
+        assert dst_ndarray[0, 0] == 5.0
+        assert dst_ndarray[0, 1] == 7.0
+        assert dst_ndarray[1, 0] == 9.0
+        assert dst_ndarray[1, 1] == 11.0
+        ```
+    */
+    BEGIN_TEST();
+
+    double in_data[12] = { 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0 };
+    int32_t in_itemsize = sizeof(double);
+    const int32_t in_ndims = 2;
+    int32_t in_shape[in_ndims] = { 3, 4 };
+    int32_t in_strides[in_ndims] = {};
+    NDArray<int32_t> ndarray = {
+        .data = (uint8_t*) in_data,
+        .itemsize = in_itemsize,
+        .ndims = in_ndims,
+        .shape = in_shape,
+        .strides = in_strides
+    };
+    ndarray.set_strides_by_shape();
+
+    // Destination ndarray
+    // As documented, ndims and shape & strides must be allocated and determined by the caller.
+    const int32_t dst_ndims = 2;
+    int32_t dst_shape[dst_ndims] = {999, 999}; // Empty values
+    int32_t dst_strides[dst_ndims] = {999, 999}; // Empty values
+    NDArray<int32_t> dst_ndarray = {
+        .data = nullptr,
+        .ndims = dst_ndims,
+        .shape = dst_shape,
+        .strides = dst_strides
+    };
+
+    // Create the slice in `ndarray[-2::, 1::2]`
+    UserSlice<int32_t> user_slice_1 = {
+        .start_defined = 1,
+        .start = -2,
+        .stop_defined = 0,
+        .step_defined = 0
+    };
+
+    UserSlice<int32_t> user_slice_2 = {
+        .start_defined = 1,
+        .start = 1,
+        .stop_defined = 0,
+        .step_defined = 1,
+        .step = 2
+    };
+
+    const int32_t num_ndslices = 2;
+    NDSlice ndslices[num_ndslices] = {
+        { .type = INPUT_SLICE_TYPE_SLICE, .slice = (uint8_t*) &user_slice_1 },
+        { .type = INPUT_SLICE_TYPE_SLICE, .slice = (uint8_t*) &user_slice_2 }
+    };
+
+    ndarray.slice(num_ndslices, ndslices, &dst_ndarray);
+
+    int32_t expected_shape[dst_ndims] = { 2, 2 };
+    int32_t expected_strides[dst_ndims] = { 32, 16 };
+    assert_arrays_match("shape", "%d", dst_ndims, expected_shape, dst_ndarray.shape);
+    assert_arrays_match("strides", "%d", dst_ndims, expected_strides, dst_ndarray.strides);
+
+    assert_values_match("dst_ndarray[0, 0]", "%f", 5.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 0, 0 })));
+    assert_values_match("dst_ndarray[0, 1]", "%f", 7.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 0, 1 })));
+    assert_values_match("dst_ndarray[1, 0]", "%f", 9.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 1, 0 })));
+    assert_values_match("dst_ndarray[1, 1]", "%f", 11.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 1, 1 })));
+}
+
+void test_ndslice_2() {
+    /*
+        ```python
+        ndarray = np.arange(12, dtype=np.float64).reshape((3, 4))
+        # array([[ 0.,  1.,  2.,  3.],
+        #        [ 4.,  5.,  6.,  7.],
+        #        [ 8.,  9., 10., 11.]])
+
+        dst_ndarray = ndarray[2, ::-2]
+        # array([11.,  9.])
+
+        assert dst_ndarray.shape == (2,)
+        assert dst_ndarray.strides == (-16,)
+        assert dst_ndarray[0] == 11.0
+        assert dst_ndarray[1] == 9.0
+
+        dst_ndarray[1, 0] == 99 # If you write to `dst_ndarray`
+        assert ndarray[1, 3] == 99 # `ndarray` also updates!!
+        ```
+    */
+    BEGIN_TEST();
+
+    double in_data[12] = { 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0 };
+    int32_t in_itemsize = sizeof(double);
+    const int32_t in_ndims = 2;
+    int32_t in_shape[in_ndims] = { 3, 4 };
+    int32_t in_strides[in_ndims] = {};
+    NDArray<int32_t> ndarray = {
+        .data = (uint8_t*) in_data,
+        .itemsize = in_itemsize,
+        .ndims = in_ndims,
+        .shape = in_shape,
+        .strides = in_strides
+    };
+    ndarray.set_strides_by_shape();
+
+    // Destination ndarray
+    // As documented, ndims and shape & strides must be allocated and determined by the caller.
+    const int32_t dst_ndims = 1;
+    int32_t dst_shape[dst_ndims] = {999}; // Empty values
+    int32_t dst_strides[dst_ndims] = {999}; // Empty values
+    NDArray<int32_t> dst_ndarray = {
+        .data = nullptr,
+        .ndims = dst_ndims,
+        .shape = dst_shape,
+        .strides = dst_strides
+    };
+
+    // Create the slice in `ndarray[2, ::-2]`
+    int32_t user_slice_1 = 2;
+    UserSlice<int32_t> user_slice_2 = {
+        .start_defined = 0,
+        .stop_defined = 0,
+        .step_defined = 1,
+        .step = -2
+    };
+
+    const int32_t num_ndslices = 2;
+    NDSlice ndslices[num_ndslices] = {
+        { .type = INPUT_SLICE_TYPE_INDEX, .slice = (uint8_t*) &user_slice_1 },
+        { .type = INPUT_SLICE_TYPE_SLICE, .slice = (uint8_t*) &user_slice_2 }
+    };
+
+    ndarray.slice(num_ndslices, ndslices, &dst_ndarray);
+
+    int32_t expected_shape[dst_ndims] = { 2 };
+    int32_t expected_strides[dst_ndims] = { -16 };
+    assert_arrays_match("shape", "%d", dst_ndims, expected_shape, dst_ndarray.shape);
+    assert_arrays_match("strides", "%d", dst_ndims, expected_strides, dst_ndarray.strides);
+
+    // [5.0, 3.0]
+    assert_values_match("dst_ndarray[0]", "%f", 11.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 0 })));
+    assert_values_match("dst_ndarray[1]", "%f", 9.0, *((double *) dst_ndarray.get_pelement((int32_t[dst_ndims]) { 1 })));
+}
+
+void test_can_broadcast_shape() {
+    BEGIN_TEST();
+
+    assert_values_match(
+        "can_broadcast_shape_to([3], [1, 1, 1, 1, 3]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 3 }, 5, (int32_t[]) { 1, 1, 1, 1, 3 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([3], [3, 1]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 3 }, 2, (int32_t[]) { 3, 1 }));
+    assert_values_match(
+        "can_broadcast_shape_to([3], [3]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 3 }, 1, (int32_t[]) { 3 }));
+    assert_values_match(
+        "can_broadcast_shape_to([1], [3]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 1 }, 1, (int32_t[]) { 3 }));
+    assert_values_match(
+        "can_broadcast_shape_to([1], [1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 1 }, 1, (int32_t[]) { 1 }));
+    assert_values_match(
+        "can_broadcast_shape_to([256, 256, 3], [256, 1, 3]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(3, (int32_t[]) { 256, 256, 3 }, 3, (int32_t[]) { 256, 1, 3 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([256, 256, 3], [3]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(3, (int32_t[]) { 256, 256, 3 }, 1, (int32_t[]) { 3 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([256, 256, 3], [2]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(3, (int32_t[]) { 256, 256, 3 }, 1, (int32_t[]) { 2 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([256, 256, 3], [1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(3, (int32_t[]) { 256, 256, 3 }, 1, (int32_t[]) { 1 })
+    );
+
+    // In cases when the shapes contain zero(es)
+    assert_values_match(
+        "can_broadcast_shape_to([0], [1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 0 }, 1, (int32_t[]) { 1 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([0], [2]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(1, (int32_t[]) { 0 }, 1, (int32_t[]) { 2 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([0, 4, 0, 0], [1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(4, (int32_t[]) { 0, 4, 0, 0 }, 1, (int32_t[]) { 1 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([0, 4, 0, 0], [1, 1, 1, 1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(4, (int32_t[]) { 0, 4, 0, 0 }, 4, (int32_t[]) { 1, 1, 1, 1 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([0, 4, 0, 0], [1, 4, 1, 1]) == true",
+        "%d",
+        true,
+        ndarray_util::can_broadcast_shape_to(4, (int32_t[]) { 0, 4, 0, 0 }, 4, (int32_t[]) { 1, 4, 1, 1 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([4, 3], [0, 3]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(2, (int32_t[]) { 4, 3 }, 2, (int32_t[]) { 0, 3 })
+    );
+    assert_values_match(
+        "can_broadcast_shape_to([4, 3], [0, 0]) == false",
+        "%d",
+        false,
+        ndarray_util::can_broadcast_shape_to(2, (int32_t[]) { 4, 3 }, 2, (int32_t[]) { 0, 0 })
+    );
+}
+
+void test_ndarray_broadcast_1() {
+    /*
+    # array = np.array([[19.9, 29.9, 39.9, 49.9]], dtype=np.float64)
+    # >>> [[19.9 29.9 39.9 49.9]]
+    #
+    # array = np.broadcast_to(array, (2, 3, 4))
+    # >>> [[[19.9 29.9 39.9 49.9]
+    # >>>   [19.9 29.9 39.9 49.9]
+    # >>>   [19.9 29.9 39.9 49.9]]
+    # >>>  [[19.9 29.9 39.9 49.9]
+    # >>>   [19.9 29.9 39.9 49.9]
+    # >>>   [19.9 29.9 39.9 49.9]]]
+    #
+    # assery array.strides == (0, 0, 8)
+
+    */
+    BEGIN_TEST();
+
+    double in_data[4] = { 19.9, 29.9, 39.9, 49.9 };
+    const int32_t in_ndims = 2;
+    int32_t in_shape[in_ndims] = {1, 4};
+    int32_t in_strides[in_ndims] = {};
+    NDArray<int32_t> ndarray = {
+        .data = (uint8_t*) in_data,
+        .itemsize = sizeof(double),
+        .ndims = in_ndims,
+        .shape = in_shape,
+        .strides = in_strides
+    };
+    ndarray.set_strides_by_shape();
+
+    const int32_t dst_ndims = 3;
+    int32_t dst_shape[dst_ndims] = {2, 3, 4};
+    int32_t dst_strides[dst_ndims] = {};
+    NDArray<int32_t> dst_ndarray = {
+        .ndims = dst_ndims,
+        .shape = dst_shape,
+        .strides = dst_strides
+    };
+
+    ndarray.broadcast_to(&dst_ndarray);
+
+    assert_arrays_match("dst_ndarray->strides", "%d", dst_ndims, (int32_t[]) { 0, 0, 8 }, dst_ndarray.strides);
+
+    assert_values_match("dst_ndarray[0, 0, 0]", "%f", 19.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 0, 0})));
+    assert_values_match("dst_ndarray[0, 0, 1]", "%f", 29.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 0, 1})));
+    assert_values_match("dst_ndarray[0, 0, 2]", "%f", 39.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 0, 2})));
+    assert_values_match("dst_ndarray[0, 0, 3]", "%f", 49.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 0, 3})));
+    assert_values_match("dst_ndarray[0, 1, 0]", "%f", 19.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 1, 0})));
+    assert_values_match("dst_ndarray[0, 1, 1]", "%f", 29.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 1, 1})));
+    assert_values_match("dst_ndarray[0, 1, 2]", "%f", 39.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 1, 2})));
+    assert_values_match("dst_ndarray[0, 1, 3]", "%f", 49.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {0, 1, 3})));
+    assert_values_match("dst_ndarray[1, 2, 3]", "%f", 49.9, *((double*) dst_ndarray.get_pelement((int32_t[]) {1, 2, 3})));
+}
+
+void test_assign_with() {
+    /*
+        ```
+        xs = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]], dtype=np.float64)
+        ys = xs.shape
+        ```
+    */
+}
+
+int main() {
+    test_calc_size_from_shape_normal();
+    test_calc_size_from_shape_has_zero();
+    test_set_strides_by_shape();
+    test_ndarray_indices_iter_normal();
+    test_ndarray_fill_generic();
+    test_ndarray_set_to_eye();
+    test_slice_1();
+    test_slice_2();
+    test_slice_3();
+    test_slice_4();
+    test_ndslice_1();
+    test_ndslice_2();
+    test_can_broadcast_shape();
+    test_ndarray_broadcast_1();
+    test_assign_with();
+    return 0;
+}

nac3core/irrt/irrt_typedefs.hpp

@@ -0,0 +1,14 @@
+#pragma once
+
+// This is made toggleable since `irrt_test.cpp` itself would include
+// headers that define the `int_t` family.
+#ifndef IRRT_DONT_TYPEDEF_INTS
+typedef _BitInt(8) int8_t;
+typedef unsigned _BitInt(8) uint8_t;
+typedef _BitInt(32) int32_t;
+typedef unsigned _BitInt(32) uint32_t;
+typedef _BitInt(64) int64_t;
+typedef unsigned _BitInt(64) uint64_t;
+#endif
+
+typedef int32_t SliceIndex;

nac3core/irrt/irrt_utils.hpp

@@ -0,0 +1,37 @@
+#pragma once
+
+#include "irrt_typedefs.hpp"
+
+namespace {
+    template <typename T>
+    T max(T a, T b) {
+        return a > b ? a : b;
+    }
+
+    template <typename T>
+    T min(T a, T b) {
+        return a > b ? b : a;
+    }
+
+    template <typename T>
+    bool arrays_match(int len, T *as, T *bs) {
+        for (int i = 0; i < len; i++) {
+            if (as[i] != bs[i]) return false;
+        }
+        return true;
+    }
+
+    void irrt_panic() {
+        // Crash the program for now.
+        // TODO: Don't crash the program
+        //       ... or at least produce a good message when doing testing IRRT
+            
+        uint8_t* death = nullptr;
+        *death = 0; // TODO: address 0 on hardware might be writable?
+    }
+
+    // TODO: Make this a macro and allow it to be toggled on/off (e.g., debug vs release)
+    void irrt_assert(bool condition) {
+        if (!condition) irrt_panic();
+    }
+}

nac3core/src/codegen/classes.rs

@@ -1,8 +1,6 @@
 use crate::codegen::{
-    irrt::{call_ndarray_calc_size, call_ndarray_flatten_index},
-    llvm_intrinsics::call_int_umin,
-    stmt::gen_for_callback_incrementing,
-    CodeGenContext, CodeGenerator,
+    llvm_intrinsics::call_int_umin, stmt::gen_for_callback_incrementing, CodeGenContext,
+    CodeGenerator,
 };
 use inkwell::context::Context;
 use inkwell::types::{ArrayType, BasicType, StructType};
@@ -12,6 +10,7 @@ use inkwell::{
     values::{BasicValueEnum, IntValue, PointerValue},
     AddressSpace, IntPredicate,
 };
+use itertools::Itertools;
 
 /// A LLVM type that is used to represent a non-primitive type in NAC3.
 pub trait ProxyType<'ctx>: Into<Self::Base> {
@@ -1601,7 +1600,8 @@ impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDataProxy<'ctx, '_> {
         ctx: &CodeGenContext<'ctx, '_>,
         generator: &G,
     ) -> IntValue<'ctx> {
-        call_ndarray_calc_size(generator, ctx, &self.as_slice_value(ctx, generator), (None, None))
+        todo!()
+        // call_ndarray_calc_size(generator, ctx, &self.as_slice_value(ctx, generator), (None, None))
     }
 }
 
@@ -1675,17 +1675,19 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
             indices_elem_ty.get_bit_width()
         );
 
-        let index = call_ndarray_flatten_index(generator, ctx, *self.0, indices);
+        todo!()
 
-        unsafe {
-            ctx.builder
-                .build_in_bounds_gep(
-                    self.base_ptr(ctx, generator),
-                    &[index],
-                    name.unwrap_or_default(),
-                )
-                .unwrap()
-        }
+        // let index = call_ndarray_flatten_index(generator, ctx, *self.0, indices);
+
+        // unsafe {
+        //     ctx.builder
+        //         .build_in_bounds_gep(
+        //             self.base_ptr(ctx, generator),
+        //             &[index],
+        //             name.unwrap_or_default(),
+        //         )
+        //         .unwrap()
+        // }
     }
 
     fn ptr_offset<G: CodeGenerator + ?Sized>(
@@ -1761,3 +1763,307 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> UntypedArrayLikeMutator<'ctx,
     for NDArrayDataProxy<'ctx, '_>
 {
 }
+
+#[derive(Debug, Clone, Copy)]
+pub struct StructField<'ctx> {
+    /// The GEP index of this struct field.
+    pub gep_index: u32,
+    /// Name of this struct field.
+    ///
+    /// Used for generating names.
+    pub name: &'static str,
+    /// The type of this struct field.
+    pub ty: BasicTypeEnum<'ctx>,
+}
+
+pub struct StructFields<'ctx> {
+    /// Name of the struct.
+    ///
+    /// Used for generating names.
+    pub name: &'static str,
+
+    /// All the [`StructField`]s of this struct.
+    ///
+    /// **NOTE:** The index position of a [`StructField`]
+    /// matches the element's [`StructField::index`].
+    pub fields: Vec<StructField<'ctx>>,
+}
+
+struct StructFieldsBuilder<'ctx> {
+    gep_index_counter: u32,
+    /// Name of the struct to be built.
+    name: &'static str,
+    fields: Vec<StructField<'ctx>>,
+}
+
+impl<'ctx> StructField<'ctx> {
+    pub fn gep(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        ptr: PointerValue<'ctx>,
+    ) -> PointerValue<'ctx> {
+        ctx.builder.build_struct_gep(ptr, self.gep_index, self.name).unwrap()
+    }
+
+    pub fn load(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        ptr: PointerValue<'ctx>,
+    ) -> BasicValueEnum<'ctx> {
+        ctx.builder.build_load(self.gep(ctx, ptr), self.name).unwrap()
+    }
+
+    pub fn store<V>(&self, ctx: &CodeGenContext<'ctx, '_>, ptr: PointerValue<'ctx>, value: V)
+    where
+        V: BasicValue<'ctx>,
+    {
+        ctx.builder.build_store(ptr, value).unwrap();
+    }
+}
+
+type IsInstanceError = String;
+type IsInstanceResult = Result<(), IsInstanceError>;
+
+pub fn check_basic_types_match<'ctx, A, B>(expected: A, got: B) -> IsInstanceResult
+where
+    A: BasicType<'ctx>,
+    B: BasicType<'ctx>,
+{
+    let expected = expected.as_basic_type_enum();
+    let got = got.as_basic_type_enum();
+
+    // Put those logic into here,
+    // otherwise there is always a fallback reporting on any kind of mismatch
+    match (expected, got) {
+        (BasicTypeEnum::IntType(expected), BasicTypeEnum::IntType(got)) => {
+            if expected.get_bit_width() != got.get_bit_width() {
+                return Err(format!(
+                    "Expected IntType ({expected}-bit(s)), got IntType ({got}-bit(s))"
+                ));
+            }
+        }
+        (expected, got) => {
+            if expected != got {
+                return Err(format!("Expected {expected}, got {got}"));
+            }
+        }
+    }
+    Ok(())
+}
+
+impl<'ctx> StructFields<'ctx> {
+    pub fn num_fields(&self) -> u32 {
+        self.fields.len() as u32
+    }
+
+    pub fn as_struct_type(&self, ctx: &'ctx Context) -> StructType<'ctx> {
+        let llvm_fields = self.fields.iter().map(|field| field.ty).collect_vec();
+        ctx.struct_type(llvm_fields.as_slice(), false)
+    }
+
+    pub fn is_type(&self, scrutinee: StructType<'ctx>) -> IsInstanceResult {
+        // Check scrutinee's number of struct fields
+        if scrutinee.count_fields() != self.num_fields() {
+            return Err(format!(
+                "Expected {expected_count} field(s) in `{struct_name}` type, got {got_count}",
+                struct_name = self.name,
+                expected_count = self.num_fields(),
+                got_count = scrutinee.count_fields(),
+            ));
+        }
+
+        // Check the scrutinee's field types
+        for field in self.fields.iter() {
+            let expected_field_ty = field.ty;
+            let got_field_ty = scrutinee.get_field_type_at_index(field.gep_index).unwrap();
+
+            if let Err(field_err) = check_basic_types_match(expected_field_ty, got_field_ty) {
+                return Err(format!(
+                    "Field GEP index {gep_index} does not match the expected type of ({struct_name}::{field_name}): {field_err}",
+                    gep_index = field.gep_index,
+                    struct_name = self.name,
+                    field_name = field.name,
+                ));
+            }
+        }
+
+        // Done
+        Ok(())
+    }
+}
+
+impl<'ctx> StructFieldsBuilder<'ctx> {
+    fn start(name: &'static str) -> Self {
+        StructFieldsBuilder { gep_index_counter: 0, name, fields: Vec::new() }
+    }
+
+    fn add_field(&mut self, name: &'static str, ty: BasicTypeEnum<'ctx>) -> StructField<'ctx> {
+        let index = self.gep_index_counter;
+        self.gep_index_counter += 1;
+        StructField { gep_index: index, name, ty }
+    }
+
+    fn end(self) -> StructFields<'ctx> {
+        StructFields { name: self.name, fields: self.fields }
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct NpArrayType<'ctx> {
+    pub size_type: IntType<'ctx>,
+    pub elem_type: BasicTypeEnum<'ctx>,
+}
+
+pub struct NpArrayStructFields<'ctx> {
+    pub whole_struct: StructFields<'ctx>,
+    pub data: StructField<'ctx>,
+    pub itemsize: StructField<'ctx>,
+    pub ndims: StructField<'ctx>,
+    pub shape: StructField<'ctx>,
+    pub strides: StructField<'ctx>,
+}
+
+impl<'ctx> NpArrayType<'ctx> {
+    pub fn new_opaque_elem(
+        ctx: &CodeGenContext<'ctx, '_>,
+        size_type: IntType<'ctx>,
+    ) -> NpArrayType<'ctx> {
+        NpArrayType { size_type, elem_type: ctx.ctx.i8_type().as_basic_type_enum() }
+    }
+
+    pub fn struct_type(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructType<'ctx> {
+        self.fields().whole_struct.as_struct_type(ctx.ctx)
+    }
+
+    pub fn fields(&self) -> NpArrayStructFields<'ctx> {
+        let mut builder = StructFieldsBuilder::start("NpArray");
+
+        let addrspace = AddressSpace::default();
+
+        let byte_type = self.size_type.get_context().i8_type();
+
+        // Make sure the struct matches PERFECTLY with that defined in `nac3core/irrt`.
+        let data = builder.add_field("data", byte_type.ptr_type(addrspace).into());
+        let itemsize = builder.add_field("itemsize", self.size_type.into());
+        let ndims = builder.add_field("ndims", self.size_type.into());
+        let shape = builder.add_field("shape", self.size_type.ptr_type(addrspace).into());
+        let strides = builder.add_field("strides", self.size_type.ptr_type(addrspace).into());
+
+        NpArrayStructFields { whole_struct: builder.end(), data, itemsize, ndims, shape, strides }
+    }
+
+    /// Allocate an `ndarray` on stack, with the following notes:
+    ///
+    /// - `ndarray.ndims` will be initialized to `in_ndims`.
+    /// - `ndarray.itemsize` will be initialized to the size of `self.elem_type.size_of()`.
+    /// - `ndarray.shape` and `ndarray.strides` will be allocated on the stack with number of elements being `in_ndims`,
+    /// all with empty/uninitialized values.
+    pub fn alloca(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        in_ndims: IntValue<'ctx>,
+        name: &str,
+    ) -> NpArrayValue<'ctx> {
+        let fields = self.fields();
+        let ptr =
+            ctx.builder.build_alloca(fields.whole_struct.as_struct_type(ctx.ctx), name).unwrap();
+
+        // Allocate `in_dims` number of `size_type` on the stack for `shape` and `strides`
+        let allocated_shape =
+            ctx.builder.build_array_alloca(fields.shape.ty, in_ndims, "allocated_shape").unwrap();
+        let allocated_strides = ctx
+            .builder
+            .build_array_alloca(fields.strides.ty, in_ndims, "allocated_strides")
+            .unwrap();
+
+        let value = NpArrayValue { ty: *self, ptr };
+        value.store_ndims(ctx, in_ndims);
+        value.store_itemsize(ctx, self.elem_type.size_of().unwrap());
+        value.store_shape(ctx, allocated_shape);
+        value.store_strides(ctx, allocated_strides);
+
+        return value;
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct NpArrayValue<'ctx> {
+    pub ty: NpArrayType<'ctx>,
+    pub ptr: PointerValue<'ctx>,
+}
+
+impl<'ctx> NpArrayValue<'ctx> {
+    pub fn load_ndims(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
+        let field = self.ty.fields().ndims;
+        field.load(ctx, self.ptr).into_int_value()
+    }
+
+    pub fn store_ndims(&self, ctx: &CodeGenContext<'ctx, '_>, value: IntValue<'ctx>) {
+        let field = self.ty.fields().ndims;
+        field.store(ctx, self.ptr, value);
+    }
+
+    pub fn load_itemsize(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
+        let field = self.ty.fields().itemsize;
+        field.load(ctx, self.ptr).into_int_value()
+    }
+
+    pub fn store_itemsize(&self, ctx: &CodeGenContext<'ctx, '_>, value: IntValue<'ctx>) {
+        let field = self.ty.fields().itemsize;
+        field.store(ctx, self.ptr, value);
+    }
+
+    pub fn load_shape(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
+        let field = self.ty.fields().shape;
+        field.load(ctx, self.ptr).into_pointer_value()
+    }
+
+    pub fn store_shape(&self, ctx: &CodeGenContext<'ctx, '_>, value: PointerValue<'ctx>) {
+        let field = self.ty.fields().shape;
+        field.store(ctx, self.ptr, value);
+    }
+
+    pub fn load_strides(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
+        let field = self.ty.fields().strides;
+        field.load(ctx, self.ptr).into_pointer_value()
+    }
+
+    pub fn store_strides(&self, ctx: &CodeGenContext<'ctx, '_>, value: PointerValue<'ctx>) {
+        let field = self.ty.fields().strides;
+        field.store(ctx, self.ptr, value);
+    }
+
+    /// TODO: DOCUMENT ME -- NDIMS WOULD NEVER CHANGE!!!!!
+    pub fn shape_slice(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+    ) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
+        let field = self.ty.fields().shape;
+        field.gep(ctx, self.ptr);
+
+        let ndims = self.load_ndims(ctx);
+
+        TypedArrayLikeAdapter {
+            adapted: ArraySliceValue(self.ptr, ndims, Some(field.name)),
+            downcast_fn: Box::new(|_ctx, x| x.into_int_value()),
+            upcast_fn: Box::new(|_ctx, x| x.as_basic_value_enum()),
+        }
+    }
+
+    /// TODO: DOCUMENT ME -- NDIMS WOULD NEVER CHANGE!!!!!
+    pub fn strides_slice(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+    ) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
+        let field = self.ty.fields().strides;
+        field.gep(ctx, self.ptr);
+
+        let ndims = self.load_ndims(ctx);
+
+        TypedArrayLikeAdapter {
+            adapted: ArraySliceValue(self.ptr, ndims, Some(field.name)),
+            downcast_fn: Box::new(|_ctx, x| x.into_int_value()),
+            upcast_fn: Box::new(|_ctx, x| x.as_basic_value_enum()),
+        }
+    }
+}

nac3core/src/codegen/expr.rs

@@ -1362,100 +1362,101 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
     } else if ty1.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
         || ty2.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
     {
-        let llvm_usize = generator.get_size_type(ctx.ctx);
+        todo!()
+        // let llvm_usize = generator.get_size_type(ctx.ctx);
 
-        let is_ndarray1 = ty1.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
-        let is_ndarray2 = ty2.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
+        // let is_ndarray1 = ty1.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
+        // let is_ndarray2 = ty2.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
 
-        if is_ndarray1 && is_ndarray2 {
-            let (ndarray_dtype1, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty1);
-            let (ndarray_dtype2, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty2);
+        // if is_ndarray1 && is_ndarray2 {
+        //     let (ndarray_dtype1, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty1);
+        //     let (ndarray_dtype2, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty2);
 
-            assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
+        //     assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
 
-            let left_val =
-                NDArrayValue::from_ptr_val(left_val.into_pointer_value(), llvm_usize, None);
-            let right_val =
-                NDArrayValue::from_ptr_val(right_val.into_pointer_value(), llvm_usize, None);
+        //     let left_val =
+        //         NDArrayValue::from_ptr_val(left_val.into_pointer_value(), llvm_usize, None);
+        //     let right_val =
+        //         NDArrayValue::from_ptr_val(right_val.into_pointer_value(), llvm_usize, None);
 
-            let res = if op.base == Operator::MatMult {
-                // MatMult is the only binop which is not an elementwise op
-                numpy::ndarray_matmul_2d(
-                    generator,
-                    ctx,
-                    ndarray_dtype1,
-                    match op.variant {
-                        BinopVariant::Normal => None,
-                        BinopVariant::AugAssign => Some(left_val),
-                    },
-                    left_val,
-                    right_val,
-                )?
-            } else {
-                numpy::ndarray_elementwise_binop_impl(
-                    generator,
-                    ctx,
-                    ndarray_dtype1,
-                    match op.variant {
-                        BinopVariant::Normal => None,
-                        BinopVariant::AugAssign => Some(left_val),
-                    },
-                    (left_val.as_base_value().into(), false),
-                    (right_val.as_base_value().into(), false),
-                    |generator, ctx, (lhs, rhs)| {
-                        gen_binop_expr_with_values(
-                            generator,
-                            ctx,
-                            (&Some(ndarray_dtype1), lhs),
-                            op,
-                            (&Some(ndarray_dtype2), rhs),
-                            ctx.current_loc,
-                        )?
-                        .unwrap()
-                        .to_basic_value_enum(
-                            ctx,
-                            generator,
-                            ndarray_dtype1,
-                        )
-                    },
-                )?
-            };
+        //     let res = if op.base == Operator::MatMult {
+        //         // MatMult is the only binop which is not an elementwise op
+        //         numpy::ndarray_matmul_2d(
+        //             generator,
+        //             ctx,
+        //             ndarray_dtype1,
+        //             match op.variant {
+        //                 BinopVariant::Normal => None,
+        //                 BinopVariant::AugAssign => Some(left_val),
+        //             },
+        //             left_val,
+        //             right_val,
+        //         )?
+        //     } else {
+        //         numpy::ndarray_elementwise_binop_impl(
+        //             generator,
+        //             ctx,
+        //             ndarray_dtype1,
+        //             match op.variant {
+        //                 BinopVariant::Normal => None,
+        //                 BinopVariant::AugAssign => Some(left_val),
+        //             },
+        //             (left_val.as_base_value().into(), false),
+        //             (right_val.as_base_value().into(), false),
+        //             |generator, ctx, (lhs, rhs)| {
+        //                 gen_binop_expr_with_values(
+        //                     generator,
+        //                     ctx,
+        //                     (&Some(ndarray_dtype1), lhs),
+        //                     op,
+        //                     (&Some(ndarray_dtype2), rhs),
+        //                     ctx.current_loc,
+        //                 )?
+        //                 .unwrap()
+        //                 .to_basic_value_enum(
+        //                     ctx,
+        //                     generator,
+        //                     ndarray_dtype1,
+        //                 )
+        //             },
+        //         )?
+        //     };
 
-            Ok(Some(res.as_base_value().into()))
-        } else {
-            let (ndarray_dtype, _) =
-                unpack_ndarray_var_tys(&mut ctx.unifier, if is_ndarray1 { ty1 } else { ty2 });
-            let ndarray_val = NDArrayValue::from_ptr_val(
-                if is_ndarray1 { left_val } else { right_val }.into_pointer_value(),
-                llvm_usize,
-                None,
-            );
-            let res = numpy::ndarray_elementwise_binop_impl(
-                generator,
-                ctx,
-                ndarray_dtype,
-                match op.variant {
-                    BinopVariant::Normal => None,
-                    BinopVariant::AugAssign => Some(ndarray_val),
-                },
-                (left_val, !is_ndarray1),
-                (right_val, !is_ndarray2),
-                |generator, ctx, (lhs, rhs)| {
-                    gen_binop_expr_with_values(
-                        generator,
-                        ctx,
-                        (&Some(ndarray_dtype), lhs),
-                        op,
-                        (&Some(ndarray_dtype), rhs),
-                        ctx.current_loc,
-                    )?
-                    .unwrap()
-                    .to_basic_value_enum(ctx, generator, ndarray_dtype)
-                },
-            )?;
+        //     Ok(Some(res.as_base_value().into()))
+        // } else {
+        //     let (ndarray_dtype, _) =
+        //         unpack_ndarray_var_tys(&mut ctx.unifier, if is_ndarray1 { ty1 } else { ty2 });
+        //     let ndarray_val = NDArrayValue::from_ptr_val(
+        //         if is_ndarray1 { left_val } else { right_val }.into_pointer_value(),
+        //         llvm_usize,
+        //         None,
+        //     );
+        //     let res = numpy::ndarray_elementwise_binop_impl(
+        //         generator,
+        //         ctx,
+        //         ndarray_dtype,
+        //         match op.variant {
+        //             BinopVariant::Normal => None,
+        //             BinopVariant::AugAssign => Some(ndarray_val),
+        //         },
+        //         (left_val, !is_ndarray1),
+        //         (right_val, !is_ndarray2),
+        //         |generator, ctx, (lhs, rhs)| {
+        //             gen_binop_expr_with_values(
+        //                 generator,
+        //                 ctx,
+        //                 (&Some(ndarray_dtype), lhs),
+        //                 op,
+        //                 (&Some(ndarray_dtype), rhs),
+        //                 ctx.current_loc,
+        //             )?
+        //             .unwrap()
+        //             .to_basic_value_enum(ctx, generator, ndarray_dtype)
+        //         },
+        //     )?;
 
-            Ok(Some(res.as_base_value().into()))
-        }
+        //     Ok(Some(res.as_base_value().into()))
+        // }
     } else {
         let left_ty_enum = ctx.unifier.get_ty_immutable(left_ty.unwrap());
         let TypeEnum::TObj { fields, obj_id, .. } = left_ty_enum.as_ref() else {
@@ -1612,40 +1613,41 @@ pub fn gen_unaryop_expr_with_values<'ctx, G: CodeGenerator>(
             _ => val.into(),
         }
     } else if ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) {
-        let llvm_usize = generator.get_size_type(ctx.ctx);
-        let (ndarray_dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
+        todo!()
+        // let llvm_usize = generator.get_size_type(ctx.ctx);
+        // let (ndarray_dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
 
-        let val = NDArrayValue::from_ptr_val(val.into_pointer_value(), llvm_usize, None);
+        // let val = NDArrayValue::from_ptr_val(val.into_pointer_value(), llvm_usize, None);
 
-        // ndarray uses `~` rather than `not` to perform elementwise inversion, convert it before
-        // passing it to the elementwise codegen function
-        let op = if ndarray_dtype.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::Bool.id()) {
-            if op == ast::Unaryop::Invert {
-                ast::Unaryop::Not
-            } else {
-                unreachable!(
-                    "ufunc {} not supported for ndarray[bool, N]",
-                    op.op_info().method_name,
-                )
-            }
-        } else {
-            op
-        };
+        // // ndarray uses `~` rather than `not` to perform elementwise inversion, convert it before
+        // // passing it to the elementwise codegen function
+        // let op = if ndarray_dtype.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::Bool.id()) {
+        //     if op == ast::Unaryop::Invert {
+        //         ast::Unaryop::Not
+        //     } else {
+        //         unreachable!(
+        //             "ufunc {} not supported for ndarray[bool, N]",
+        //             op.op_info().method_name,
+        //         )
+        //     }
+        // } else {
+        //     op
+        // };
 
-        let res = numpy::ndarray_elementwise_unaryop_impl(
-            generator,
-            ctx,
-            ndarray_dtype,
-            None,
-            val,
-            |generator, ctx, val| {
-                gen_unaryop_expr_with_values(generator, ctx, op, (&Some(ndarray_dtype), val))?
-                    .unwrap()
-                    .to_basic_value_enum(ctx, generator, ndarray_dtype)
-            },
-        )?;
+        // let res = numpy::ndarray_elementwise_unaryop_impl(
+        //     generator,
+        //     ctx,
+        //     ndarray_dtype,
+        //     None,
+        //     val,
+        //     |generator, ctx, val| {
+        //         gen_unaryop_expr_with_values(generator, ctx, op, (&Some(ndarray_dtype), val))?
+        //             .unwrap()
+        //             .to_basic_value_enum(ctx, generator, ndarray_dtype)
+        //     },
+        // )?;
 
-        res.as_base_value().into()
+        // res.as_base_value().into()
     } else {
         unimplemented!()
     }))
@@ -1688,85 +1690,86 @@ pub fn gen_cmpop_expr_with_values<'ctx, G: CodeGenerator>(
         if left_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
             || right_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
         {
-            let llvm_usize = generator.get_size_type(ctx.ctx);
+            todo!()
+            // let llvm_usize = generator.get_size_type(ctx.ctx);
 
-            let (Some(left_ty), lhs) = left else { unreachable!() };
-            let (Some(right_ty), rhs) = comparators[0] else { unreachable!() };
-            let op = ops[0];
+            // let (Some(left_ty), lhs) = left else { unreachable!() };
+            // let (Some(right_ty), rhs) = comparators[0] else { unreachable!() };
+            // let op = ops[0];
 
-            let is_ndarray1 =
-                left_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
-            let is_ndarray2 =
-                right_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
+            // let is_ndarray1 =
+            //     left_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
+            // let is_ndarray2 =
+            //     right_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
 
-            return if is_ndarray1 && is_ndarray2 {
-                let (ndarray_dtype1, _) = unpack_ndarray_var_tys(&mut ctx.unifier, left_ty);
-                let (ndarray_dtype2, _) = unpack_ndarray_var_tys(&mut ctx.unifier, right_ty);
+            // return if is_ndarray1 && is_ndarray2 {
+            //     let (ndarray_dtype1, _) = unpack_ndarray_var_tys(&mut ctx.unifier, left_ty);
+            //     let (ndarray_dtype2, _) = unpack_ndarray_var_tys(&mut ctx.unifier, right_ty);
 
-                assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
+            //     assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
 
-                let left_val =
-                    NDArrayValue::from_ptr_val(lhs.into_pointer_value(), llvm_usize, None);
-                let res = numpy::ndarray_elementwise_binop_impl(
-                    generator,
-                    ctx,
-                    ctx.primitives.bool,
-                    None,
-                    (left_val.as_base_value().into(), false),
-                    (rhs, false),
-                    |generator, ctx, (lhs, rhs)| {
-                        let val = gen_cmpop_expr_with_values(
-                            generator,
-                            ctx,
-                            (Some(ndarray_dtype1), lhs),
-                            &[op],
-                            &[(Some(ndarray_dtype2), rhs)],
-                        )?
-                        .unwrap()
-                        .to_basic_value_enum(
-                            ctx,
-                            generator,
-                            ctx.primitives.bool,
-                        )?;
+            //     let left_val =
+            //         NDArrayValue::from_ptr_val(lhs.into_pointer_value(), llvm_usize, None);
+            //     let res = numpy::ndarray_elementwise_binop_impl(
+            //         generator,
+            //         ctx,
+            //         ctx.primitives.bool,
+            //         None,
+            //         (left_val.as_base_value().into(), false),
+            //         (rhs, false),
+            //         |generator, ctx, (lhs, rhs)| {
+            //             let val = gen_cmpop_expr_with_values(
+            //                 generator,
+            //                 ctx,
+            //                 (Some(ndarray_dtype1), lhs),
+            //                 &[op],
+            //                 &[(Some(ndarray_dtype2), rhs)],
+            //             )?
+            //             .unwrap()
+            //             .to_basic_value_enum(
+            //                 ctx,
+            //                 generator,
+            //                 ctx.primitives.bool,
+            //             )?;
 
-                        Ok(generator.bool_to_i8(ctx, val.into_int_value()).into())
-                    },
-                )?;
+            //             Ok(generator.bool_to_i8(ctx, val.into_int_value()).into())
+            //         },
+            //     )?;
 
-                Ok(Some(res.as_base_value().into()))
-            } else {
-                let (ndarray_dtype, _) = unpack_ndarray_var_tys(
-                    &mut ctx.unifier,
-                    if is_ndarray1 { left_ty } else { right_ty },
-                );
-                let res = numpy::ndarray_elementwise_binop_impl(
-                    generator,
-                    ctx,
-                    ctx.primitives.bool,
-                    None,
-                    (lhs, !is_ndarray1),
-                    (rhs, !is_ndarray2),
-                    |generator, ctx, (lhs, rhs)| {
-                        let val = gen_cmpop_expr_with_values(
-                            generator,
-                            ctx,
-                            (Some(ndarray_dtype), lhs),
-                            &[op],
-                            &[(Some(ndarray_dtype), rhs)],
-                        )?
-                        .unwrap()
-                        .to_basic_value_enum(
-                            ctx,
-                            generator,
-                            ctx.primitives.bool,
-                        )?;
+            //     Ok(Some(res.as_base_value().into()))
+            // } else {
+            //     let (ndarray_dtype, _) = unpack_ndarray_var_tys(
+            //         &mut ctx.unifier,
+            //         if is_ndarray1 { left_ty } else { right_ty },
+            //     );
+            //     let res = numpy::ndarray_elementwise_binop_impl(
+            //         generator,
+            //         ctx,
+            //         ctx.primitives.bool,
+            //         None,
+            //         (lhs, !is_ndarray1),
+            //         (rhs, !is_ndarray2),
+            //         |generator, ctx, (lhs, rhs)| {
+            //             let val = gen_cmpop_expr_with_values(
+            //                 generator,
+            //                 ctx,
+            //                 (Some(ndarray_dtype), lhs),
+            //                 &[op],
+            //                 &[(Some(ndarray_dtype), rhs)],
+            //             )?
+            //             .unwrap()
+            //             .to_basic_value_enum(
+            //                 ctx,
+            //                 generator,
+            //                 ctx.primitives.bool,
+            //             )?;
 
-                        Ok(generator.bool_to_i8(ctx, val.into_int_value()).into())
-                    },
-                )?;
+            //             Ok(generator.bool_to_i8(ctx, val.into_int_value()).into())
+            //         },
+            //     )?;
 
-                Ok(Some(res.as_base_value().into()))
-            };
+            //     Ok(Some(res.as_base_value().into()))
+            // };
         }
     }
 
@@ -2102,310 +2105,312 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
     v: NDArrayValue<'ctx>,
     slice: &Expr<Option<Type>>,
 ) -> Result<Option<ValueEnum<'ctx>>, String> {
-    let llvm_i1 = ctx.ctx.bool_type();
-    let llvm_i32 = ctx.ctx.i32_type();
-    let llvm_usize = generator.get_size_type(ctx.ctx);
+    todo!()
 
-    let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndims) else {
-        unreachable!()
-    };
+    // let llvm_i1 = ctx.ctx.bool_type();
+    // let llvm_i32 = ctx.ctx.i32_type();
+    // let llvm_usize = generator.get_size_type(ctx.ctx);
 
-    let ndims = values
-        .iter()
-        .map(|ndim| u64::try_from(ndim.clone()).map_err(|()| ndim.clone()))
-        .collect::<Result<Vec<_>, _>>()
-        .map_err(|val| {
-            format!(
-                "Expected non-negative literal for ndarray.ndims, got {}",
-                i128::try_from(val).unwrap()
-            )
-        })?;
+    // let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndims) else {
+    //     unreachable!()
+    // };
 
-    assert!(!ndims.is_empty());
+    // let ndims = values
+    //     .iter()
+    //     .map(|ndim| u64::try_from(ndim.clone()).map_err(|()| ndim.clone()))
+    //     .collect::<Result<Vec<_>, _>>()
+    //     .map_err(|val| {
+    //         format!(
+    //             "Expected non-negative literal for ndarray.ndims, got {}",
+    //             i128::try_from(val).unwrap()
+    //         )
+    //     })?;
 
-    // The number of dimensions subscripted by the index expression.
-    // Slicing a ndarray will yield the same number of dimensions, whereas indexing into a
-    // dimension will remove a dimension.
-    let subscripted_dims = match &slice.node {
-        ExprKind::Tuple { elts, .. } => elts.iter().fold(0, |acc, value_subexpr| {
-            if let ExprKind::Slice { .. } = &value_subexpr.node {
-                acc
-            } else {
-                acc + 1
-            }
-        }),
+    // assert!(!ndims.is_empty());
 
-        ExprKind::Slice { .. } => 0,
-        _ => 1,
-    };
+    // // The number of dimensions subscripted by the index expression.
+    // // Slicing a ndarray will yield the same number of dimensions, whereas indexing into a
+    // // dimension will remove a dimension.
+    // let subscripted_dims = match &slice.node {
+    //     ExprKind::Tuple { elts, .. } => elts.iter().fold(0, |acc, value_subexpr| {
+    //         if let ExprKind::Slice { .. } = &value_subexpr.node {
+    //             acc
+    //         } else {
+    //             acc + 1
+    //         }
+    //     }),
 
-    let ndarray_ndims_ty = ctx.unifier.get_fresh_literal(
-        ndims.iter().map(|v| SymbolValue::U64(v - subscripted_dims)).collect(),
-        None,
-    );
-    let ndarray_ty =
-        make_ndarray_ty(&mut ctx.unifier, &ctx.primitives, Some(ty), Some(ndarray_ndims_ty));
-    let llvm_pndarray_t = ctx.get_llvm_type(generator, ndarray_ty).into_pointer_type();
-    let llvm_ndarray_t = llvm_pndarray_t.get_element_type().into_struct_type();
-    let llvm_ndarray_data_t = ctx.get_llvm_type(generator, ty).as_basic_type_enum();
+    //     ExprKind::Slice { .. } => 0,
+    //     _ => 1,
+    // };
 
-    // Check that len is non-zero
-    let len = v.load_ndims(ctx);
-    ctx.make_assert(
-        generator,
-        ctx.builder.build_int_compare(IntPredicate::SGT, len, llvm_usize.const_zero(), "").unwrap(),
-        "0:IndexError",
-        "too many indices for array: array is {0}-dimensional but 1 were indexed",
-        [Some(len), None, None],
-        slice.location,
-    );
+    // let ndarray_ndims_ty = ctx.unifier.get_fresh_literal(
+    //     ndims.iter().map(|v| SymbolValue::U64(v - subscripted_dims)).collect(),
+    //     None,
+    // );
+    // let ndarray_ty =
+    //     make_ndarray_ty(&mut ctx.unifier, &ctx.primitives, Some(ty), Some(ndarray_ndims_ty));
+    // let llvm_pndarray_t = ctx.get_llvm_type(generator, ndarray_ty).into_pointer_type();
+    // let llvm_ndarray_t = llvm_pndarray_t.get_element_type().into_struct_type();
+    // let llvm_ndarray_data_t = ctx.get_llvm_type(generator, ty).as_basic_type_enum();
 
-    // Normalizes a possibly-negative index to its corresponding positive index
-    let normalize_index = |generator: &mut G,
-                           ctx: &mut CodeGenContext<'ctx, '_>,
-                           index: IntValue<'ctx>,
-                           dim: u64| {
-        gen_if_else_expr_callback(
-            generator,
-            ctx,
-            |_, ctx| {
-                Ok(ctx
-                    .builder
-                    .build_int_compare(IntPredicate::SGE, index, index.get_type().const_zero(), "")
-                    .unwrap())
-            },
-            |_, _| Ok(Some(index)),
-            |generator, ctx| {
-                let llvm_i32 = ctx.ctx.i32_type();
+    // // Check that len is non-zero
+    // let len = v.load_ndims(ctx);
+    // ctx.make_assert(
+    //     generator,
+    //     ctx.builder.build_int_compare(IntPredicate::SGT, len, llvm_usize.const_zero(), "").unwrap(),
+    //     "0:IndexError",
+    //     "too many indices for array: array is {0}-dimensional but 1 were indexed",
+    //     [Some(len), None, None],
+    //     slice.location,
+    // );
 
-                let len = unsafe {
-                    v.dim_sizes().get_typed_unchecked(
-                        ctx,
-                        generator,
-                        &llvm_usize.const_int(dim, true),
-                        None,
-                    )
-                };
+    // // Normalizes a possibly-negative index to its corresponding positive index
+    // let normalize_index = |generator: &mut G,
+    //                        ctx: &mut CodeGenContext<'ctx, '_>,
+    //                        index: IntValue<'ctx>,
+    //                        dim: u64| {
+    //     gen_if_else_expr_callback(
+    //         generator,
+    //         ctx,
+    //         |_, ctx| {
+    //             Ok(ctx
+    //                 .builder
+    //                 .build_int_compare(IntPredicate::SGE, index, index.get_type().const_zero(), "")
+    //                 .unwrap())
+    //         },
+    //         |_, _| Ok(Some(index)),
+    //         |generator, ctx| {
+    //             let llvm_i32 = ctx.ctx.i32_type();
 
-                let index = ctx
-                    .builder
-                    .build_int_add(
-                        len,
-                        ctx.builder.build_int_s_extend(index, llvm_usize, "").unwrap(),
-                        "",
-                    )
-                    .unwrap();
+    //             let len = unsafe {
+    //                 v.dim_sizes().get_typed_unchecked(
+    //                     ctx,
+    //                     generator,
+    //                     &llvm_usize.const_int(dim, true),
+    //                     None,
+    //                 )
+    //             };
 
-                Ok(Some(ctx.builder.build_int_truncate(index, llvm_i32, "").unwrap()))
-            },
-        )
-        .map(|v| v.map(BasicValueEnum::into_int_value))
-    };
+    //             let index = ctx
+    //                 .builder
+    //                 .build_int_add(
+    //                     len,
+    //                     ctx.builder.build_int_s_extend(index, llvm_usize, "").unwrap(),
+    //                     "",
+    //                 )
+    //                 .unwrap();
 
-    // Converts a slice expression into a slice-range tuple
-    let expr_to_slice = |generator: &mut G,
-                         ctx: &mut CodeGenContext<'ctx, '_>,
-                         node: &ExprKind<Option<Type>>,
-                         dim: u64| {
-        match node {
-            ExprKind::Constant { value: Constant::Int(v), .. } => {
-                let Some(index) =
-                    normalize_index(generator, ctx, llvm_i32.const_int(*v as u64, true), dim)?
-                else {
-                    return Ok(None);
-                };
+    //             Ok(Some(ctx.builder.build_int_truncate(index, llvm_i32, "").unwrap()))
+    //         },
+    //     )
+    //     .map(|v| v.map(BasicValueEnum::into_int_value))
+    // };
 
-                Ok(Some((index, index, llvm_i32.const_int(1, true))))
-            }
+    // // Converts a slice expression into a slice-range tuple
+    // let expr_to_slice = |generator: &mut G,
+    //                      ctx: &mut CodeGenContext<'ctx, '_>,
+    //                      node: &ExprKind<Option<Type>>,
+    //                      dim: u64| {
+    //     match node {
+    //         ExprKind::Constant { value: Constant::Int(v), .. } => {
+    //             let Some(index) =
+    //                 normalize_index(generator, ctx, llvm_i32.const_int(*v as u64, true), dim)?
+    //             else {
+    //                 return Ok(None);
+    //             };
 
-            ExprKind::Slice { lower, upper, step } => {
-                let dim_sz = unsafe {
-                    v.dim_sizes().get_typed_unchecked(
-                        ctx,
-                        generator,
-                        &llvm_usize.const_int(dim, false),
-                        None,
-                    )
-                };
+    //             Ok(Some((index, index, llvm_i32.const_int(1, true))))
+    //         }
 
-                handle_slice_indices(lower, upper, step, ctx, generator, dim_sz)
-            }
+    //         ExprKind::Slice { lower, upper, step } => {
+    //             let dim_sz = unsafe {
+    //                 v.dim_sizes().get_typed_unchecked(
+    //                     ctx,
+    //                     generator,
+    //                     &llvm_usize.const_int(dim, false),
+    //                     None,
+    //                 )
+    //             };
 
-            _ => {
-                let Some(index) = generator.gen_expr(ctx, slice)? else { return Ok(None) };
-                let index = index
-                    .to_basic_value_enum(ctx, generator, slice.custom.unwrap())?
-                    .into_int_value();
-                let Some(index) = normalize_index(generator, ctx, index, dim)? else {
-                    return Ok(None);
-                };
+    //             handle_slice_indices(lower, upper, step, ctx, generator, dim_sz)
+    //         }
 
-                Ok(Some((index, index, llvm_i32.const_int(1, true))))
-            }
-        }
-    };
+    //         _ => {
+    //             let Some(index) = generator.gen_expr(ctx, slice)? else { return Ok(None) };
+    //             let index = index
+    //                 .to_basic_value_enum(ctx, generator, slice.custom.unwrap())?
+    //                 .into_int_value();
+    //             let Some(index) = normalize_index(generator, ctx, index, dim)? else {
+    //                 return Ok(None);
+    //             };
 
-    let make_indices_arr = |generator: &mut G,
-                            ctx: &mut CodeGenContext<'ctx, '_>|
-     -> Result<_, String> {
-        Ok(if let ExprKind::Tuple { elts, .. } = &slice.node {
-            let llvm_int_ty = ctx.get_llvm_type(generator, elts[0].custom.unwrap());
-            let index_addr = generator.gen_array_var_alloc(
-                ctx,
-                llvm_int_ty,
-                llvm_usize.const_int(elts.len() as u64, false),
-                None,
-            )?;
+    //             Ok(Some((index, index, llvm_i32.const_int(1, true))))
+    //         }
+    //     }
+    // };
 
-            for (i, elt) in elts.iter().enumerate() {
-                let Some(index) = generator.gen_expr(ctx, elt)? else {
-                    return Ok(None);
-                };
+    // let make_indices_arr = |generator: &mut G,
+    //                         ctx: &mut CodeGenContext<'ctx, '_>|
+    //  -> Result<_, String> {
+    //     Ok(if let ExprKind::Tuple { elts, .. } = &slice.node {
+    //         let llvm_int_ty = ctx.get_llvm_type(generator, elts[0].custom.unwrap());
+    //         let index_addr = generator.gen_array_var_alloc(
+    //             ctx,
+    //             llvm_int_ty,
+    //             llvm_usize.const_int(elts.len() as u64, false),
+    //             None,
+    //         )?;
 
-                let index = index
-                    .to_basic_value_enum(ctx, generator, elt.custom.unwrap())?
-                    .into_int_value();
-                let Some(index) = normalize_index(generator, ctx, index, 0)? else {
-                    return Ok(None);
-                };
+    //         for (i, elt) in elts.iter().enumerate() {
+    //             let Some(index) = generator.gen_expr(ctx, elt)? else {
+    //                 return Ok(None);
+    //             };
 
-                let store_ptr = unsafe {
-                    index_addr.ptr_offset_unchecked(
-                        ctx,
-                        generator,
-                        &llvm_usize.const_int(i as u64, false),
-                        None,
-                    )
-                };
-                ctx.builder.build_store(store_ptr, index).unwrap();
-            }
+    //             let index = index
+    //                 .to_basic_value_enum(ctx, generator, elt.custom.unwrap())?
+    //                 .into_int_value();
+    //             let Some(index) = normalize_index(generator, ctx, index, 0)? else {
+    //                 return Ok(None);
+    //             };
 
-            Some(index_addr)
-        } else if let Some(index) = generator.gen_expr(ctx, slice)? {
-            let llvm_int_ty = ctx.get_llvm_type(generator, slice.custom.unwrap());
-            let index_addr = generator.gen_array_var_alloc(
-                ctx,
-                llvm_int_ty,
-                llvm_usize.const_int(1u64, false),
-                None,
-            )?;
+    //             let store_ptr = unsafe {
+    //                 index_addr.ptr_offset_unchecked(
+    //                     ctx,
+    //                     generator,
+    //                     &llvm_usize.const_int(i as u64, false),
+    //                     None,
+    //                 )
+    //             };
+    //             ctx.builder.build_store(store_ptr, index).unwrap();
+    //         }
 
-            let index =
-                index.to_basic_value_enum(ctx, generator, slice.custom.unwrap())?.into_int_value();
-            let Some(index) = normalize_index(generator, ctx, index, 0)? else { return Ok(None) };
+    //         Some(index_addr)
+    //     } else if let Some(index) = generator.gen_expr(ctx, slice)? {
+    //         let llvm_int_ty = ctx.get_llvm_type(generator, slice.custom.unwrap());
+    //         let index_addr = generator.gen_array_var_alloc(
+    //             ctx,
+    //             llvm_int_ty,
+    //             llvm_usize.const_int(1u64, false),
+    //             None,
+    //         )?;
 
-            let store_ptr = unsafe {
-                index_addr.ptr_offset_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
-            };
-            ctx.builder.build_store(store_ptr, index).unwrap();
+    //         let index =
+    //             index.to_basic_value_enum(ctx, generator, slice.custom.unwrap())?.into_int_value();
+    //         let Some(index) = normalize_index(generator, ctx, index, 0)? else { return Ok(None) };
 
-            Some(index_addr)
-        } else {
-            None
-        })
-    };
+    //         let store_ptr = unsafe {
+    //             index_addr.ptr_offset_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+    //         };
+    //         ctx.builder.build_store(store_ptr, index).unwrap();
 
-    Ok(Some(if ndims.len() == 1 && ndims[0] - subscripted_dims == 0 {
-        let Some(index_addr) = make_indices_arr(generator, ctx)? else { return Ok(None) };
+    //         Some(index_addr)
+    //     } else {
+    //         None
+    //     })
+    // };
 
-        v.data().get(ctx, generator, &index_addr, None).into()
-    } else {
-        match &slice.node {
-            ExprKind::Tuple { elts, .. } => {
-                let slices = elts
-                    .iter()
-                    .enumerate()
-                    .map(|(dim, elt)| expr_to_slice(generator, ctx, &elt.node, dim as u64))
-                    .take_while_inclusive(|slice| slice.as_ref().is_ok_and(Option::is_some))
-                    .collect::<Result<Vec<_>, _>>()?;
-                if slices.len() < elts.len() {
-                    return Ok(None);
-                }
+    // Ok(Some(if ndims.len() == 1 && ndims[0] - subscripted_dims == 0 {
+    //     let Some(index_addr) = make_indices_arr(generator, ctx)? else { return Ok(None) };
 
-                let slices = slices.into_iter().map(Option::unwrap).collect_vec();
+    //     v.data().get(ctx, generator, &index_addr, None).into()
+    // } else {
+    //     match &slice.node {
+    //         ExprKind::Tuple { elts, .. } => {
+    //             let slices = elts
+    //                 .iter()
+    //                 .enumerate()
+    //                 .map(|(dim, elt)| expr_to_slice(generator, ctx, &elt.node, dim as u64))
+    //                 .take_while_inclusive(|slice| slice.as_ref().is_ok_and(Option::is_some))
+    //                 .collect::<Result<Vec<_>, _>>()?;
+    //             if slices.len() < elts.len() {
+    //                 return Ok(None);
+    //             }
 
-                numpy::ndarray_sliced_copy(generator, ctx, ty, v, &slices)?.as_base_value().into()
-            }
+    //             let slices = slices.into_iter().map(Option::unwrap).collect_vec();
 
-            ExprKind::Slice { .. } => {
-                let Some(slice) = expr_to_slice(generator, ctx, &slice.node, 0)? else {
-                    return Ok(None);
-                };
+    //             numpy::ndarray_sliced_copy(generator, ctx, ty, v, &slices)?.as_base_value().into()
+    //         }
 
-                numpy::ndarray_sliced_copy(generator, ctx, ty, v, &[slice])?.as_base_value().into()
-            }
+    //         ExprKind::Slice { .. } => {
+    //             let Some(slice) = expr_to_slice(generator, ctx, &slice.node, 0)? else {
+    //                 return Ok(None);
+    //             };
 
-            _ => {
-                // Accessing an element from a multi-dimensional `ndarray`
+    //             numpy::ndarray_sliced_copy(generator, ctx, ty, v, &[slice])?.as_base_value().into()
+    //         }
 
-                let Some(index_addr) = make_indices_arr(generator, ctx)? else { return Ok(None) };
+    //         _ => {
+    //             // Accessing an element from a multi-dimensional `ndarray`
 
-                // Create a new array, remove the top dimension from the dimension-size-list, and copy the
-                // elements over
-                let subscripted_ndarray =
-                    generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
-                let ndarray = NDArrayValue::from_ptr_val(subscripted_ndarray, llvm_usize, None);
+    //             let Some(index_addr) = make_indices_arr(generator, ctx)? else { return Ok(None) };
 
-                let num_dims = v.load_ndims(ctx);
-                ndarray.store_ndims(
-                    ctx,
-                    generator,
-                    ctx.builder
-                        .build_int_sub(num_dims, llvm_usize.const_int(1, false), "")
-                        .unwrap(),
-                );
+    //             // Create a new array, remove the top dimension from the dimension-size-list, and copy the
+    //             // elements over
+    //             let subscripted_ndarray =
+    //                 generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
+    //             let ndarray = NDArrayValue::from_ptr_val(subscripted_ndarray, llvm_usize, None);
 
-                let ndarray_num_dims = ndarray.load_ndims(ctx);
-                ndarray.create_dim_sizes(ctx, llvm_usize, ndarray_num_dims);
+    //             let num_dims = v.load_ndims(ctx);
+    //             ndarray.store_ndims(
+    //                 ctx,
+    //                 generator,
+    //                 ctx.builder
+    //                     .build_int_sub(num_dims, llvm_usize.const_int(1, false), "")
+    //                     .unwrap(),
+    //             );
 
-                let ndarray_num_dims = ndarray.load_ndims(ctx);
-                let v_dims_src_ptr = unsafe {
-                    v.dim_sizes().ptr_offset_unchecked(
-                        ctx,
-                        generator,
-                        &llvm_usize.const_int(1, false),
-                        None,
-                    )
-                };
-                call_memcpy_generic(
-                    ctx,
-                    ndarray.dim_sizes().base_ptr(ctx, generator),
-                    v_dims_src_ptr,
-                    ctx.builder
-                        .build_int_mul(ndarray_num_dims, llvm_usize.size_of(), "")
-                        .map(Into::into)
-                        .unwrap(),
-                    llvm_i1.const_zero(),
-                );
+    //             let ndarray_num_dims = ndarray.load_ndims(ctx);
+    //             ndarray.create_dim_sizes(ctx, llvm_usize, ndarray_num_dims);
 
-                let ndarray_num_elems = call_ndarray_calc_size(
-                    generator,
-                    ctx,
-                    &ndarray.dim_sizes().as_slice_value(ctx, generator),
-                    (None, None),
-                );
-                ndarray.create_data(ctx, llvm_ndarray_data_t, ndarray_num_elems);
+    //             let ndarray_num_dims = ndarray.load_ndims(ctx);
+    //             let v_dims_src_ptr = unsafe {
+    //                 v.dim_sizes().ptr_offset_unchecked(
+    //                     ctx,
+    //                     generator,
+    //                     &llvm_usize.const_int(1, false),
+    //                     None,
+    //                 )
+    //             };
+    //             call_memcpy_generic(
+    //                 ctx,
+    //                 ndarray.dim_sizes().base_ptr(ctx, generator),
+    //                 v_dims_src_ptr,
+    //                 ctx.builder
+    //                     .build_int_mul(ndarray_num_dims, llvm_usize.size_of(), "")
+    //                     .map(Into::into)
+    //                     .unwrap(),
+    //                 llvm_i1.const_zero(),
+    //             );
 
-                let v_data_src_ptr = v.data().ptr_offset(ctx, generator, &index_addr, None);
-                call_memcpy_generic(
-                    ctx,
-                    ndarray.data().base_ptr(ctx, generator),
-                    v_data_src_ptr,
-                    ctx.builder
-                        .build_int_mul(
-                            ndarray_num_elems,
-                            llvm_ndarray_data_t.size_of().unwrap(),
-                            "",
-                        )
-                        .map(Into::into)
-                        .unwrap(),
-                    llvm_i1.const_zero(),
-                );
+    //             let ndarray_num_elems = call_ndarray_calc_size(
+    //                 generator,
+    //                 ctx,
+    //                 &ndarray.dim_sizes().as_slice_value(ctx, generator),
+    //                 (None, None),
+    //             );
+    //             ndarray.create_data(ctx, llvm_ndarray_data_t, ndarray_num_elems);
 
-                ndarray.as_base_value().into()
-            }
-        }
-    }))
+    //             let v_data_src_ptr = v.data().ptr_offset(ctx, generator, &index_addr, None);
+    //             call_memcpy_generic(
+    //                 ctx,
+    //                 ndarray.data().base_ptr(ctx, generator),
+    //                 v_data_src_ptr,
+    //                 ctx.builder
+    //                     .build_int_mul(
+    //                         ndarray_num_elems,
+    //                         llvm_ndarray_data_t.size_of().unwrap(),
+    //                         "",
+    //                     )
+    //                     .map(Into::into)
+    //                     .unwrap(),
+    //                 llvm_i1.const_zero(),
+    //             );
+
+    //             ndarray.as_base_value().into()
+    //         }
+    //     }
+    // }))
 }
 
 /// See [`CodeGenerator::gen_expr`].

nac3core/src/codegen/irrt/irrt.cpp

@@ -1,414 +0,0 @@
-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"

nac3core/src/codegen/irrt/mod.rs

@@ -1,9 +1,11 @@
-use crate::typecheck::typedef::Type;
+use crate::{typecheck::typedef::Type, util::SizeVariant};
+
+mod test;
 
 use super::{
     classes::{
-        ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, ListValue, NDArrayValue,
-        TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
+        ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, ListValue, NDArrayValue, NpArrayType,
+        NpArrayValue, TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
     },
     llvm_intrinsics, CodeGenContext, CodeGenerator,
 };
@@ -14,8 +16,8 @@ use inkwell::{
     context::Context,
     memory_buffer::MemoryBuffer,
     module::Module,
-    types::{BasicTypeEnum, IntType},
-    values::{BasicValueEnum, CallSiteValue, FloatValue, IntValue},
+    types::{BasicType, BasicTypeEnum, FunctionType, IntType, PointerType},
+    values::{BasicValueEnum, CallSiteValue, FloatValue, FunctionValue, IntValue},
     AddressSpace, IntPredicate,
 };
 use itertools::Either;
@@ -927,3 +929,63 @@ pub fn call_ndarray_calc_broadcast_index<
         Box::new(|_, v| v.into()),
     )
 }
+
+fn get_size_variant<'ctx>(ty: IntType<'ctx>) -> SizeVariant {
+    match ty.get_bit_width() {
+        32 => SizeVariant::Bits32,
+        64 => SizeVariant::Bits64,
+        _ => unreachable!("Unsupported int type bit width {}", ty.get_bit_width()),
+    }
+}
+
+fn get_size_type_dependent_function<'ctx, BuildFuncTypeFn>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    size_type: IntType<'ctx>,
+    base_name: &str,
+    build_func_type: BuildFuncTypeFn,
+) -> FunctionValue<'ctx>
+where
+    BuildFuncTypeFn: Fn() -> FunctionType<'ctx>,
+{
+    let mut fn_name = base_name.to_owned();
+    match get_size_variant(size_type) {
+        SizeVariant::Bits32 => {
+            // The original fn_name is the correct function name
+        }
+        SizeVariant::Bits64 => {
+            // Append "64" at the end, this is the naming convention for 64-bit
+            fn_name.push_str("64");
+        }
+    }
+
+    // Get (or declare then get if does not exist) the corresponding function
+    ctx.module.get_function(&fn_name).unwrap_or_else(|| {
+        let fn_type = build_func_type();
+        ctx.module.add_function(&fn_name, fn_type, None)
+    })
+}
+
+fn get_ndarray_struct_ptr<'ctx>(ctx: &'ctx Context, size_type: IntType<'ctx>) -> PointerType<'ctx> {
+    let i8_type = ctx.i8_type();
+
+    let ndarray_ty = NpArrayType { size_type, elem_type: i8_type.as_basic_type_enum() };
+    let struct_ty = ndarray_ty.fields().whole_struct.as_struct_type(ctx);
+    struct_ty.ptr_type(AddressSpace::default())
+}
+
+pub fn call_nac3_ndarray_size<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    ndarray: NpArrayValue<'ctx>,
+) -> IntValue<'ctx> {
+    let size_type = ndarray.ty.size_type;
+    let function = get_size_type_dependent_function(ctx, size_type, "__nac3_ndarray_size", || {
+        size_type.fn_type(&[get_ndarray_struct_ptr(ctx.ctx, size_type).into()], false)
+    });
+
+    ctx.builder
+        .build_call(function, &[ndarray.ptr.into()], "size")
+        .unwrap()
+        .try_as_basic_value()
+        .unwrap_left()
+        .into_int_value()
+}

nac3core/src/codegen/irrt/test.rs

@@ -0,0 +1,26 @@
+#[cfg(test)]
+mod tests {
+    use std::{path::Path, process::Command};
+
+    #[test]
+    fn run_irrt_test() {
+        assert!(
+            cfg!(feature = "test"),
+            "Please do `cargo test -F test` to compile `irrt_test.out` and run test"
+        );
+
+        let irrt_test_out_path = Path::new(concat!(env!("OUT_DIR"), "/irrt_test.out"));
+        let output = Command::new(irrt_test_out_path.to_str().unwrap()).output().unwrap();
+
+        if !output.status.success() {
+            eprintln!("irrt_test failed with status {}:", output.status);
+            eprintln!("====== stdout ======");
+            eprintln!("{}", String::from_utf8(output.stdout).unwrap());
+            eprintln!("====== stderr ======");
+            eprintln!("{}", String::from_utf8(output.stderr).unwrap());
+            eprintln!("====================");
+
+            panic!("irrt_test failed");
+        }
+    }
+}

nac3core/src/lib.rs

@@ -23,3 +23,4 @@ pub mod codegen;
 pub mod symbol_resolver;
 pub mod toplevel;
 pub mod typecheck;
+pub mod util;

nac3core/src/toplevel/builtins.rs

@@ -1,5 +1,6 @@
 use std::iter::once;
 
+use crate::util::SizeVariant;
 use helper::{debug_assert_prim_is_allowed, make_exception_fields, PrimDefDetails};
 use indexmap::IndexMap;
 use inkwell::{
@@ -278,19 +279,10 @@ pub fn get_builtins(unifier: &mut Unifier, primitives: &PrimitiveStore) -> Built
         .collect()
 }
 
-/// A helper enum used by [`BuiltinBuilder`]
-#[derive(Clone, Copy)]
-enum SizeVariant {
-    Bits32,
-    Bits64,
-}
-
-impl SizeVariant {
-    fn of_int(self, primitives: &PrimitiveStore) -> Type {
-        match self {
-            SizeVariant::Bits32 => primitives.int32,
-            SizeVariant::Bits64 => primitives.int64,
-        }
+fn size_variant_to_int_type(variant: SizeVariant, primitives: &PrimitiveStore) -> Type {
+    match variant {
+        SizeVariant::Bits32 => primitives.int32,
+        SizeVariant::Bits64 => primitives.int64,
     }
 }
 
@@ -961,8 +953,9 @@ impl<'a> BuiltinBuilder<'a> {
                 resolver: None,
                 codegen_callback: Some(Arc::new(GenCall::new(Box::new(
                     |ctx, obj, fun, args, generator| {
-                        gen_ndarray_copy(ctx, &obj, fun, &args, generator)
-                            .map(|val| Some(val.as_basic_value_enum()))
+                        todo!()
+                        // gen_ndarray_copy(ctx, &obj, fun, &args, generator)
+                        //     .map(|val| Some(val.as_basic_value_enum()))
                     },
                 )))),
                 loc: None,
@@ -978,8 +971,9 @@ impl<'a> BuiltinBuilder<'a> {
                 resolver: None,
                 codegen_callback: Some(Arc::new(GenCall::new(Box::new(
                     |ctx, obj, fun, args, generator| {
-                        gen_ndarray_fill(ctx, &obj, fun, &args, generator)?;
-                        Ok(None)
+                        todo!()
+                        // gen_ndarray_fill(ctx, &obj, fun, &args, generator)?;
+                        // Ok(None)
                     },
                 )))),
                 loc: None,
@@ -1059,7 +1053,7 @@ impl<'a> BuiltinBuilder<'a> {
         );
 
         // The size variant of the function determines the size of the returned int.
-        let int_sized = size_variant.of_int(self.primitives);
+        let int_sized = size_variant_to_int_type(size_variant, self.primitives);
 
         let ndarray_int_sized =
             make_ndarray_ty(self.unifier, self.primitives, Some(int_sized), Some(common_ndim.ty));
@@ -1084,7 +1078,7 @@ impl<'a> BuiltinBuilder<'a> {
                 let arg_ty = fun.0.args[0].ty;
                 let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty)?;
 
-                let ret_elem_ty = size_variant.of_int(&ctx.primitives);
+                let ret_elem_ty = size_variant_to_int_type(size_variant, &ctx.primitives);
                 Ok(Some(builtin_fns::call_round(generator, ctx, (arg_ty, arg), ret_elem_ty)?))
             }),
         )
@@ -1125,7 +1119,7 @@ impl<'a> BuiltinBuilder<'a> {
             make_ndarray_ty(self.unifier, self.primitives, Some(float), Some(common_ndim.ty));
 
         // The size variant of the function determines the type of int returned
-        let int_sized = size_variant.of_int(self.primitives);
+        let int_sized = size_variant_to_int_type(size_variant, self.primitives);
         let ndarray_int_sized =
             make_ndarray_ty(self.unifier, self.primitives, Some(int_sized), Some(common_ndim.ty));
 
@@ -1148,7 +1142,7 @@ impl<'a> BuiltinBuilder<'a> {
                 let arg_ty = fun.0.args[0].ty;
                 let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty)?;
 
-                let ret_elem_ty = size_variant.of_int(&ctx.primitives);
+                let ret_elem_ty = size_variant_to_int_type(size_variant, &ctx.primitives);
                 let func = match kind {
                     Kind::Ceil => builtin_fns::call_ceil,
                     Kind::Floor => builtin_fns::call_floor,
@@ -1199,13 +1193,14 @@ impl<'a> BuiltinBuilder<'a> {
             self.ndarray_float,
             &[(self.ndarray_factory_fn_shape_arg_tvar.ty, "shape")],
             Box::new(move |ctx, obj, fun, args, generator| {
-                let func = match prim {
-                    PrimDef::FunNpNDArray | PrimDef::FunNpEmpty => gen_ndarray_empty,
-                    PrimDef::FunNpZeros => gen_ndarray_zeros,
-                    PrimDef::FunNpOnes => gen_ndarray_ones,
-                    _ => unreachable!(),
-                };
-                func(ctx, &obj, fun, &args, generator).map(|val| Some(val.as_basic_value_enum()))
+                todo!()
+                // let func = match prim {
+                //     PrimDef::FunNpNDArray | PrimDef::FunNpEmpty => gen_ndarray_empty,
+                //     PrimDef::FunNpZeros => gen_ndarray_zeros,
+                //     PrimDef::FunNpOnes => gen_ndarray_ones,
+                //     _ => unreachable!(),
+                // };
+                // func(ctx, &obj, fun, &args, generator).map(|val| Some(val.as_basic_value_enum()))
             }),
         )
     }
@@ -1251,8 +1246,9 @@ impl<'a> BuiltinBuilder<'a> {
                     resolver: None,
                     codegen_callback: Some(Arc::new(GenCall::new(Box::new(
                         |ctx, obj, fun, args, generator| {
-                            gen_ndarray_array(ctx, &obj, fun, &args, generator)
-                                .map(|val| Some(val.as_basic_value_enum()))
+                            todo!()
+                            // gen_ndarray_array(ctx, &obj, fun, &args, generator)
+                            //     .map(|val| Some(val.as_basic_value_enum()))
                         },
                     )))),
                     loc: None,
@@ -1270,8 +1266,9 @@ impl<'a> BuiltinBuilder<'a> {
                     // type variable
                     &[(self.list_int32, "shape"), (tv.ty, "fill_value")],
                     Box::new(move |ctx, obj, fun, args, generator| {
-                        gen_ndarray_full(ctx, &obj, fun, &args, generator)
-                            .map(|val| Some(val.as_basic_value_enum()))
+                        todo!()
+                        // gen_ndarray_full(ctx, &obj, fun, &args, generator)
+                        //     .map(|val| Some(val.as_basic_value_enum()))
                     }),
                 )
             }
@@ -1303,8 +1300,9 @@ impl<'a> BuiltinBuilder<'a> {
                     resolver: None,
                     codegen_callback: Some(Arc::new(GenCall::new(Box::new(
                         |ctx, obj, fun, args, generator| {
-                            gen_ndarray_eye(ctx, &obj, fun, &args, generator)
-                                .map(|val| Some(val.as_basic_value_enum()))
+                            todo!()
+                            // gen_ndarray_eye(ctx, &obj, fun, &args, generator)
+                            //     .map(|val| Some(val.as_basic_value_enum()))
                         },
                     )))),
                     loc: None,
@@ -1317,8 +1315,9 @@ impl<'a> BuiltinBuilder<'a> {
                 self.ndarray_float_2d,
                 &[(int32, "n")],
                 Box::new(|ctx, obj, fun, args, generator| {
-                    gen_ndarray_identity(ctx, &obj, fun, &args, generator)
-                        .map(|val| Some(val.as_basic_value_enum()))
+                    todo!()
+                    // gen_ndarray_identity(ctx, &obj, fun, &args, generator)
+                    //     .map(|val| Some(val.as_basic_value_enum()))
                 }),
             ),
             _ => unreachable!(),

nix/windows/default.nix

@@ -81,6 +81,7 @@ in rec {
     ''
     mkdir -p $out/bin
     ln -s ${llvm-nac3}/bin/clang.exe $out/bin/clang-irrt.exe
+    ln -s ${llvm-nac3}/bin/clang.exe $out/bin/clang-irrt-test.exe
     ln -s ${llvm-nac3}/bin/llvm-as.exe $out/bin/llvm-as-irrt.exe
     '';
   nac3artiq = pkgs.rustPlatform.buildRustPackage {