WIP 6

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

@@ -3,20 +3,34 @@ use std::{
     env,
     fs::File,
     io::Write,
-    path::Path,
+    path::{Path, PathBuf},
     process::{Command, Stdio},
 };
 
-fn main() {
+const CMD_IRRT_CLANG: &str = "clang-irrt";
-    const FILE: &str = "src/codegen/irrt/irrt.cpp";
+const CMD_IRRT_CLANG_TEST: &str = "clang-irrt-test";
+const CMD_IRRT_LLVM_AS: &str = "llvm-as-irrt";
+
+fn get_out_dir() -> PathBuf {
+    PathBuf::from(env::var("OUT_DIR").unwrap())
+}
+
+fn get_irrt_dir() -> &'static Path {
+    Path::new("irrt")
+}
+
+/// Compile `irrt.cpp` for use in `src/codegen`
+fn compile_irrt_cpp() {
+    let out_dir = get_out_dir();
+    let irrt_dir = get_irrt_dir();
 
     /*
      * HACK: Sadly, clang doesn't let us emit generic LLVM bitcode.
      * Compiling for WASM32 and filtering the output with regex is the closest we can get.
      */
+    let irrt_cpp_path = irrt_dir.join("irrt.cpp");
     let flags: &[&str] = &[
         "--target=wasm32",
-        FILE,
         "-x",
         "c++",
         "-fno-discard-value-names",
@@ -31,15 +45,19 @@ fn main() {
         "-S",
         "-Wall",
         "-Wextra",
+        "-Werror=return-type",
+        "-I",
+        irrt_dir.to_str().unwrap(),
         "-o",
         "-",
+        irrt_cpp_path.to_str().unwrap(),
     ];
 
-    println!("cargo:rerun-if-changed={FILE}");
+    // Tell Cargo to rerun if any file under `irrt_dir` (recursive) changes
-    let out_dir = env::var("OUT_DIR").unwrap();
+    println!("cargo:rerun-if-changed={}", irrt_dir.to_str().unwrap());
-    let out_path = Path::new(&out_dir);
 
-    let output = Command::new("clang-irrt")
+    // Compile IRRT and capture the LLVM IR output
+    let output = Command::new(CMD_IRRT_CLANG)
         .args(flags)
         .output()
         .map(|o| {
@@ -52,7 +70,17 @@ 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();
+    // Filter out irrelevant IR
+    //
+    // Regex:
+    // - `(?ms:^define.*?\}$)` captures LLVM `define` blocks
+    // - `(?m:^declare.*?$)` captures LLVM `declare` lines
+    // - `(?m:^%.+?=\s*type\s*\{.+?\}$)` captures LLVM `type` declarations
+    // - `(?m:^@.+?=.+$)` captures global constants
+    let regex_filter = Regex::new(
+        r"(?ms:^define.*?\}$)|(?m:^declare.*?$)|(?m:^%.+?=\s*type\s*\{.+?\}$)|(?m:^@.+?=.+$)",
+    )
+    .unwrap();
     for f in regex_filter.captures_iter(&output) {
         assert_eq!(f.len(), 1);
         filtered_output.push_str(&f[0]);
@@ -63,20 +91,73 @@ fn main() {
         .unwrap()
         .replace_all(&filtered_output, "");
 
-    println!("cargo:rerun-if-env-changed=DEBUG_DUMP_IRRT");
+    // For debugging
-    if env::var("DEBUG_DUMP_IRRT").is_ok() {
+    // Doing `DEBUG_DUMP_IRRT=1 cargo build -p nac3core` dumps the LLVM IR generated
-        let mut file = File::create(out_path.join("irrt.ll")).unwrap();
+    const DEBUG_DUMP_IRRT: &str = "DEBUG_DUMP_IRRT";
+    println!("cargo:rerun-if-env-changed={DEBUG_DUMP_IRRT}");
+    if env::var(DEBUG_DUMP_IRRT).is_ok() {
+        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")
+    // Assemble the emitted and filtered IR to .bc
+    // That .bc will be integrated into nac3core's codegen
+    let mut llvm_as = Command::new(CMD_IRRT_LLVM_AS)
         .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());
 }
+
+/// Compile `irrt_test.cpp` for testing
+fn compile_irrt_test_cpp() {
+    let out_dir = get_out_dir();
+    let irrt_dir = get_irrt_dir();
+
+    let exe_path = out_dir.join("irrt_test.out"); // Output path of the compiled test executable
+    let irrt_test_cpp_path = irrt_dir.join("irrt_test.cpp");
+    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()`
+        "-I",
+        irrt_dir.to_str().unwrap(),
+        "-o",
+        exe_path.to_str().unwrap(),
+    ];
+
+    Command::new(CMD_IRRT_CLANG_TEST)
+        .args(flags)
+        .output()
+        .map(|o| {
+            assert!(o.status.success(), "{}", std::str::from_utf8(&o.stderr).unwrap());
+            o
+        })
+        .unwrap();
+    println!("cargo:rerun-if-changed={}", irrt_dir.to_str().unwrap());
+}
+
+fn main() {
+    compile_irrt_cpp();
+
+    // 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_cpp();
+    }
+}

nac3core/irrt/irrt.cpp

@@ -0,0 +1,9 @@
+#define IRRT_DEFINE_TYPEDEF_INTS
+#include <irrt_everything.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/core.hpp

@@ -1,9 +1,7 @@
-using int8_t = _BitInt(8);
+#pragma once
-using uint8_t = unsigned _BitInt(8);
+
-using int32_t = _BitInt(32);
+#include <irrt/utils.hpp>
-using uint32_t = unsigned _BitInt(32);
+#include <irrt/int_defs.hpp>
-using int64_t = _BitInt(64);
-using uint64_t = unsigned _BitInt(64);
 
 // NDArray indices are always `uint32_t`.
 using NDIndex = uint32_t;
@@ -11,16 +9,6 @@ using NDIndex = uint32_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>

nac3core/irrt/irrt/error_context.hpp

@@ -0,0 +1,85 @@
+#pragma once
+
+#include <irrt/int_defs.hpp>
+#include <irrt/utils.hpp>
+
+namespace {
+// nac3core's "str" struct type definition
+template <typename SizeT>
+struct Str {
+    const char* content;
+    SizeT length;
+};
+
+// A limited set of errors IRRT could use.
+typedef uint32_t ErrorId;
+struct ErrorIds {
+    ErrorId index_error;
+    ErrorId value_error;
+    ErrorId assertion_error;
+    ErrorId runtime_error;
+};
+
+struct ErrorContext {
+    // Context
+    ErrorIds* error_ids;
+
+    // Error thrown by IRRT
+    ErrorId error_id;
+    const char* message_template; // MUST BE `&'static`
+    uint64_t param1;
+    uint64_t param2;
+    uint64_t param3;
+
+    void initialize(ErrorIds* error_ids) {
+        this->error_ids = error_ids;
+        clear_error();
+    }
+
+    void clear_error() {
+        // Point the message_template to an empty str. Don't set it to nullptr as a sentinel
+        this->message_template = "";
+    }
+
+    void set_error(ErrorId error_id, const char* message, uint64_t param1 = 0, uint64_t param2 = 0, uint64_t param3 = 0) {
+        this->error_id = error_id;
+        this->message_template = message;
+        this->param1 = param1;
+        this->param2 = param2;
+        this->param3 = param3;
+    }
+
+    bool has_error() {
+        return !cstr_utils::is_empty(message_template);
+    }
+
+    template <typename SizeT>
+    void set_error_str(Str<SizeT> *dst_str) {
+        dst_str->content = message_template;
+        dst_str->length = (SizeT) cstr_utils::length(message_template);
+    }
+};
+}
+
+extern "C" {
+
+void __nac3_error_context_initialize(ErrorContext* errctx, ErrorIds* error_ids) {
+    errctx->initialize(error_ids);
+}
+
+uint8_t __nac3_error_context_has_no_error(ErrorContext* errctx) {
+    return !errctx->has_error();
+}
+
+void __nac3_error_context_get_error_str(ErrorContext* errctx, Str<int32_t> *dst_str) {
+    errctx->set_error_str<int32_t>(dst_str);
+}
+
+void __nac3_error_context_get_error_str64(ErrorContext* errctx, Str<int64_t> *dst_str) {
+    errctx->set_error_str<int64_t>(dst_str);
+}
+
+void __nac3_error_dummy_raise(ErrorContext* errctx) {
+    errctx->set_error(errctx->error_ids->runtime_error, "THROWN FROM __nac3_error_dummy_raise!!!!!!");
+}
+}

nac3core/irrt/irrt/int_defs.hpp

@@ -0,0 +1,12 @@
+#pragma once
+
+// This is made toggleable since `irrt_test.cpp` itself would include
+// headers that define these typedefs
+#ifdef IRRT_DEFINE_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

nac3core/irrt/irrt/ndarray/ndarray.hpp

@@ -0,0 +1,155 @@
+#pragma once
+
+#include <irrt/int_defs.hpp>
+#include <irrt/ndarray/ndarray_util.hpp>
+
+namespace {
+
+// 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.
+    //
+    // This pointer should point to the first element of the ndarray directly
+    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;
+    }
+
+    // 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);
+    }
+
+    uint8_t* get_pelement_by_indices(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) {
+        SizeT* indices = (SizeT*) __builtin_alloca(sizeof(SizeT) * this->ndims);
+        ndarray_util::set_indices_by_nth(this->ndims, this->shape, indices, nth);
+        return get_pelement_by_indices(indices);
+    }
+
+    // Get the pointer to the nth element of the ndarray as if it were flattened.
+    uint8_t* checked_get_nth_pelement(ErrorContext* errctx, SizeT nth) {
+        SizeT arr_size = this->size();
+        if (!(0 <= nth && nth < arr_size)) {
+            errctx->set_error(
+                errctx->error_ids->index_error,
+                "index {0} is out of bounds, valid range is {1} <= index < {2}",
+                nth, 0, arr_size
+            );
+            return 0;
+        }
+        return get_nth_pelement(nth);
+    }
+
+    void set_pelement_value(uint8_t* pelement, const uint8_t* pvalue) {
+        __builtin_memcpy(pelement, pvalue, itemsize);
+    }
+
+    // Fill the ndarray with a value
+    void fill_generic(const uint8_t* pvalue) {
+        const SizeT size = this->size();
+        for (SizeT i = 0; i < size; i++) {
+            uint8_t* pelement = get_nth_pelement(i); // No need for checked_get_nth_pelement
+            set_pelement_value(pelement, pvalue);
+        }
+    }
+};
+}
+
+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();
+}
+
+uint32_t __nac3_ndarray_nbytes(NDArray<int32_t>* ndarray) {
+    return ndarray->nbytes();
+}
+
+uint64_t __nac3_ndarray_nbytes64(NDArray<int64_t>* ndarray) {
+    return ndarray->nbytes();
+}
+
+void __nac3_ndarray_util_assert_shape_no_negative(ErrorContext* errctx, int32_t ndims, int32_t* shape) {
+    ndarray_util::assert_shape_no_negative(errctx, ndims, shape);
+}
+
+void __nac3_ndarray_util_assert_shape_no_negative64(ErrorContext* errctx, int64_t ndims, int64_t* shape) {
+    ndarray_util::assert_shape_no_negative(errctx, ndims, shape);
+}
+
+void __nac3_ndarray_set_strides_by_shape(NDArray<int32_t>* ndarray) {
+    ndarray->set_strides_by_shape();
+}
+
+void __nac3_ndarray_set_strides_by_shape64(NDArray<int64_t>* ndarray) {
+    ndarray->set_strides_by_shape();
+}
+
+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);
+}
+}

nac3core/irrt/irrt/ndarray/ndarray_util.hpp

@@ -0,0 +1,107 @@
+#pragma once
+
+#include <irrt/int_defs.hpp>
+
+namespace {
+namespace ndarray_util {
+
+// Throw an error if there is an axis with negative dimension
+template <typename SizeT>
+void assert_shape_no_negative(ErrorContext* errctx, SizeT ndims, const SizeT* shape) {
+    for (SizeT axis = 0; axis < ndims; axis++) {
+        if (shape[axis] < 0) {
+            errctx->set_error(
+                errctx->error_ids->value_error,
+                "negative dimensions are not allowed; axis {0} has dimension {1}",
+                axis, shape[axis]
+            );
+            return;
+        }
+    }
+}
+
+// Compute the size/# of elements of an ndarray given its shape
+template <typename SizeT>
+SizeT calc_size_from_shape(SizeT ndims, const SizeT* shape) {
+    SizeT size = 1;
+    for (SizeT axis = 0; axis < ndims; axis++) size *= shape[axis];
+    return size;
+}
+
+// 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>
+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 axis = ndims - i - 1;
+        dst_strides[axis] = stride_product * itemsize;
+        stride_product *= shape[axis];
+    }
+}
+
+template <typename SizeT>
+void set_indices_by_nth(SizeT ndims, const SizeT* shape, SizeT* indices, SizeT nth) {
+    for (int32_t i = 0; i < ndims; i++) {
+        int32_t axis = ndims - i - 1;
+        int32_t dim = shape[axis];
+
+        indices[axis] = nth % dim;
+        nth /= dim;
+    }
+}
+
+template <typename SizeT>
+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_axis = target_ndims - i - 1;
+        SizeT src_axis = src_ndims - i - 1;
+
+        bool target_dim_exists = target_axis >= 0;
+        bool src_dim_exists = src_axis >= 0;
+
+        SizeT target_dim = target_dim_exists ? target_shape[target_axis] : 1;
+        SizeT src_dim = src_dim_exists ? src_shape[src_axis] : 1;
+
+        bool ok = src_dim == 1 || target_dim == src_dim;
+        if (!ok) return false;
+    }
+
+    return true;
+}
+}
+}

nac3core/irrt/irrt/string_utils.hpp

@@ -0,0 +1,60 @@
+#pragma once
+
+#include <irrt/int_defs.hpp>
+
+namespace {
+namespace string {
+    bool is_empty(const char* str) {
+        return str[0] == '\0';
+    }
+
+    int8_t compare(const char* a, const char* b) {
+        uint32_t i = 0;
+        while (true) {
+            if (a[i] < b[i]) {
+                return -1;
+            } else if (a[i] > b[i]) {
+                return 1;
+            } else { // a[i] == b[i]
+                if (a[i] == '\0') {
+                    return 0;
+                } else {
+                    continue;
+                }
+            }
+        }
+    }
+
+    int8_t equal(const char* a, const char* b) {
+        return compare(a, b) == 0;
+    }
+
+    uint32_t length(const char* str) {
+        uint32_t length = 0;
+        while (*str != '\0') {
+            length++;
+            str++;
+        }
+        return length;
+    }
+
+    bool copy(const char* src, char* dst, uint32_t dst_max_size) {
+        for (uint32_t i = 0; i < dst_max_size; i++) {
+            bool is_last = i + 1 == dst_max_size;
+            if (is_last && src[i] != '\0') {
+                dst[i] = '\0';
+                return false;
+            }
+
+            if (src[i] == '\0') {
+                dst[i] = '\0';
+                return true;
+            }
+
+            dst[i] = src[i];
+        }
+
+        __builtin_unreachable();
+    }
+}
+}

nac3core/irrt/irrt/utils.hpp

@@ -0,0 +1,88 @@
+#pragma once
+
+#include <irrt/int_defs.hpp>
+
+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;
+}
+
+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;
+}
+
+template<typename T>
+uint32_t int_log_floor(T value, T base) {
+    uint32_t result = 0;
+    while (value >= base) {
+        result++;
+        value /= base;
+    }
+    return result;
+}
+
+namespace cstr_utils {
+    bool is_empty(const char* str) {
+        return str[0] == '\0';
+    }
+
+    int8_t compare(const char* a, const char* b) {
+        uint32_t i = 0;
+        while (true) {
+            if (a[i] < b[i]) {
+                return -1;
+            } else if (a[i] > b[i]) {
+                return 1;
+            } else { // a[i] == b[i] 
+                if (a[i] == '\0') {
+                    return 0;
+                } else {
+                    i++;
+                }
+            }
+        }
+    }
+
+    int8_t equal(const char* a, const char* b) {
+        return compare(a, b) == 0;
+    }
+
+    uint32_t length(const char* str) {
+        uint32_t length = 0;
+        while (*str != '\0') {
+            length++;
+            str++;
+        }
+        return length;
+    }
+
+    bool copy(const char* src, char* dst, uint32_t dst_max_size) {
+        for (uint32_t i = 0; i < dst_max_size; i++) {
+            bool is_last = i + 1 == dst_max_size;
+            if (is_last && src[i] != '\0') {
+                dst[i] = '\0';
+                return false;
+            }
+
+            if (src[i] == '\0') {
+                dst[i] = '\0';
+                return true;
+            }
+
+            dst[i] = src[i];
+        }
+
+        __builtin_unreachable();
+    }
+}
+}

nac3core/irrt/irrt_everything.hpp

@@ -0,0 +1,7 @@
+#pragma once
+
+#include <irrt/core.hpp>
+#include <irrt/error_context.hpp>
+#include <irrt/int_defs.hpp>
+#include <irrt/utils.hpp>
+#include <irrt/ndarray/ndarray.hpp>

nac3core/irrt/irrt_test.cpp

@@ -0,0 +1,18 @@
+// 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>
+
+#include <irrt_everything.hpp>
+
+#include <test/core.hpp>
+#include <test/test_core.hpp>
+#include <test/test_utils.hpp>
+
+int main() {
+    run_test_core();
+    run_test_utils();
+    return 0;
+}

nac3core/irrt/test/core.hpp

@@ -0,0 +1,88 @@
+#pragma once
+
+// Include this header for every test_*.cpp
+
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+
+#include <test/print.hpp>
+
+// Some utils can be used here
+#include "../irrt/utils.hpp"
+
+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__)
+
+void test_fail() {
+    printf("[!] Test failed. Exiting with status code 1.\n");
+    exit(1);
+}
+
+template <typename T>
+void debug_print_array(int len, const T* as) {
+    printf("[");
+    for (int i = 0; i < len; i++) {
+        if (i != 0) printf(", ");
+        print_value(as[i]);
+    }
+    printf("]");
+}
+
+void print_assertion_passed(const char* file, int line) {
+    printf("[*] Assertion passed on %s:%d\n", file, line);
+}
+
+void print_assertion_failed(const char* file, int line) {
+    printf("[!] Assertion failed on %s:%d\n", file, line);
+}
+
+void __assert_true(const char* file, int line, bool cond) {
+    if (cond) {
+        print_assertion_passed(file, line);
+    } else {
+        print_assertion_failed(file, line);
+        test_fail();
+    }
+}
+
+#define assert_true(cond) __assert_true(__FILE__, __LINE__, cond)
+
+template <typename T>
+void __assert_arrays_match(const char* file, int line, int len, const T* expected, const T* got) {
+    if (arrays_match(len, expected, got)) {
+        print_assertion_passed(file, line);
+    } else {
+        print_assertion_failed(file, line);
+        printf("Expect = ");
+        debug_print_array(len, expected);
+        printf("\n");
+        printf("   Got = ");
+        debug_print_array(len, got);
+        printf("\n");
+        test_fail();
+    }
+}
+
+#define assert_arrays_match(len, expected, got) __assert_arrays_match(__FILE__, __LINE__, len, expected, got)
+
+template <typename T>
+void __assert_values_match(const char* file, int line, T expected, T got) {
+    if (expected == got) {
+        print_assertion_passed(file, line);
+    } else {
+        print_assertion_failed(file, line);
+        printf("Expect = ");
+        print_value(expected);
+        printf("\n");
+        printf("   Got = ");
+        print_value(got);
+        printf("\n");
+        test_fail();
+    }
+}
+
+#define assert_values_match(expected, got) __assert_values_match(__FILE__, __LINE__, expected, got)

nac3core/irrt/test/print.hpp

@@ -0,0 +1,42 @@
+#pragma once
+
+#include <cstdlib>
+#include <cstdio>
+
+template <class T>
+void print_value(T value);
+
+template <>
+void print_value(char value) {
+    printf("'%c' (ord=%d)", value, value);
+}
+
+template <>
+void print_value(int8_t value) {
+    printf("%d", value);
+}
+
+template <>
+void print_value(int32_t value) {
+    printf("%d", value);
+}
+
+template <>
+void print_value(uint8_t value) {
+    printf("%u", value);
+}
+
+template <>
+void print_value(uint32_t value) {
+    printf("%u", value);
+}
+
+template <>
+void print_value(double value) {
+    printf("%f", value);
+}
+
+template <>
+void print_value(char* value) {
+    printf("%s", value);
+}

nac3core/irrt/test/test_core.hpp

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <test/core.hpp>
+#include <irrt/core.hpp>
+
+void test_int_exp() {
+    BEGIN_TEST();
+
+    assert_values_match(125, __nac3_int_exp_impl<int32_t>(5, 3));
+    assert_values_match(3125, __nac3_int_exp_impl<int32_t>(5, 5));
+}
+
+void run_test_core() {
+    test_int_exp();
+}

nac3core/irrt/test/test_utils.hpp

@@ -0,0 +1,27 @@
+#pragma once
+
+#include <test/core.hpp>
+#include <irrt/utils.hpp>
+
+void test_int_log_10() {
+    BEGIN_TEST();
+
+    assert_values_match((uint32_t) 0, int_log_floor(0, 10));
+    assert_values_match((uint32_t) 0, int_log_floor(9, 10));
+    assert_values_match((uint32_t) 1, int_log_floor(10, 10));
+    assert_values_match((uint32_t) 1, int_log_floor(11, 10));
+    assert_values_match((uint32_t) 1, int_log_floor(99, 10));
+    assert_values_match((uint32_t) 2, int_log_floor(100, 10));
+    assert_values_match((uint32_t) 2, int_log_floor(101, 10));
+}
+
+void test_cstr_utils() {
+    BEGIN_TEST();
+
+    assert_values_match((uint32_t) 42, (uint32_t) cstr_utils::length("THROWN FROM __nac3_error_dummy_raise!!!!!!"));
+}
+
+void run_test_utils() {
+    test_int_log_10();
+    test_cstr_utils();
+}

nac3core/src/codegen/builtin_fns.rs

@@ -661,90 +661,6 @@ pub fn call_min<'ctx>(
     }
 }
 
-/// Invokes the `np_min` builtin function.
-pub fn call_numpy_min<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    a: (Type, BasicValueEnum<'ctx>),
-) -> Result<BasicValueEnum<'ctx>, String> {
-    const FN_NAME: &str = "np_min";
-
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-
-    let (a_ty, a) = a;
-
-    Ok(match a {
-        BasicValueEnum::IntValue(_) | BasicValueEnum::FloatValue(_) => {
-            debug_assert!([
-                ctx.primitives.bool,
-                ctx.primitives.int32,
-                ctx.primitives.uint32,
-                ctx.primitives.int64,
-                ctx.primitives.uint64,
-                ctx.primitives.float,
-            ]
-            .iter()
-            .any(|ty| ctx.unifier.unioned(a_ty, *ty)));
-
-            a
-        }
-
-        BasicValueEnum::PointerValue(n)
-            if a_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
-        {
-            let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, a_ty);
-            let llvm_ndarray_ty = ctx.get_llvm_type(generator, elem_ty);
-
-            let n = NDArrayValue::from_ptr_val(n, llvm_usize, None);
-            let n_sz = irrt::call_ndarray_calc_size(generator, ctx, &n.dim_sizes(), (None, None));
-            if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
-                let n_sz_eqz = ctx
-                    .builder
-                    .build_int_compare(IntPredicate::NE, n_sz, n_sz.get_type().const_zero(), "")
-                    .unwrap();
-
-                ctx.make_assert(
-                    generator,
-                    n_sz_eqz,
-                    "0:ValueError",
-                    "zero-size array to reduction operation minimum which has no identity",
-                    [None, None, None],
-                    ctx.current_loc,
-                );
-            }
-
-            let accumulator_addr = generator.gen_var_alloc(ctx, llvm_ndarray_ty, None)?;
-            unsafe {
-                let identity =
-                    n.data().get_unchecked(ctx, generator, &llvm_usize.const_zero(), None);
-                ctx.builder.build_store(accumulator_addr, identity).unwrap();
-            }
-
-            gen_for_callback_incrementing(
-                generator,
-                ctx,
-                llvm_usize.const_int(1, false),
-                (n_sz, false),
-                |generator, ctx, _, idx| {
-                    let elem = unsafe { n.data().get_unchecked(ctx, generator, &idx, None) };
-
-                    let accumulator = ctx.builder.build_load(accumulator_addr, "").unwrap();
-                    let result = call_min(ctx, (elem_ty, accumulator), (elem_ty, elem));
-                    ctx.builder.build_store(accumulator_addr, result).unwrap();
-
-                    Ok(())
-                },
-                llvm_usize.const_int(1, false),
-            )?;
-
-            let accumulator = ctx.builder.build_load(accumulator_addr, "").unwrap();
-            accumulator
-        }
-
-        _ => unsupported_type(ctx, FN_NAME, &[a_ty]),
-    })
-}
-
 /// Invokes the `np_minimum` builtin function.
 pub fn call_numpy_minimum<'ctx, G: CodeGenerator + ?Sized>(
     generator: &mut G,
@@ -877,18 +793,20 @@ pub fn call_max<'ctx>(
     }
 }
 
-/// Invokes the `np_max` builtin function.
+/// Invokes the `np_max`, `np_min`, `np_argmax`, `np_argmin` functions
-pub fn call_numpy_max<'ctx, G: CodeGenerator + ?Sized>(
+/// * `fn_name`: Can be one of `"np_argmin"`, `"np_argmax"`, `"np_max"`, `"np_min"`
+pub fn call_numpy_max_min<'ctx, G: CodeGenerator + ?Sized>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, '_>,
     a: (Type, BasicValueEnum<'ctx>),
+    fn_name: &str,
 ) -> Result<BasicValueEnum<'ctx>, String> {
-    const FN_NAME: &str = "np_max";
+    debug_assert!(["np_argmin", "np_argmax", "np_max", "np_min"].iter().any(|f| *f == fn_name));
 
+    let llvm_int64 = ctx.ctx.i64_type();
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
     let (a_ty, a) = a;
-
     Ok(match a {
         BasicValueEnum::IntValue(_) | BasicValueEnum::FloatValue(_) => {
             debug_assert!([
@@ -902,9 +820,12 @@ pub fn call_numpy_max<'ctx, G: CodeGenerator + ?Sized>(
             .iter()
             .any(|ty| ctx.unifier.unioned(a_ty, *ty)));
 
-            a
+            match fn_name {
+                "np_argmin" | "np_argmax" => llvm_int64.const_zero().into(),
+                "np_max" | "np_min" => a,
+                _ => unreachable!(),
+            }
         }
-
         BasicValueEnum::PointerValue(n)
             if a_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
@@ -923,41 +844,81 @@ pub fn call_numpy_max<'ctx, G: CodeGenerator + ?Sized>(
                     generator,
                     n_sz_eqz,
                     "0:ValueError",
-                    "zero-size array to reduction operation minimum which has no identity",
+                    format!("zero-size array to reduction operation {fn_name}").as_str(),
                     [None, None, None],
                     ctx.current_loc,
                 );
             }
 
             let accumulator_addr = generator.gen_var_alloc(ctx, llvm_ndarray_ty, None)?;
+            let res_idx = generator.gen_var_alloc(ctx, llvm_int64.into(), None)?;
+
             unsafe {
                 let identity =
                     n.data().get_unchecked(ctx, generator, &llvm_usize.const_zero(), None);
                 ctx.builder.build_store(accumulator_addr, identity).unwrap();
+                ctx.builder.build_store(res_idx, llvm_int64.const_zero()).unwrap();
             }
 
             gen_for_callback_incrementing(
                 generator,
                 ctx,
-                llvm_usize.const_int(1, false),
+                llvm_int64.const_int(1, false),
                 (n_sz, false),
                 |generator, ctx, _, idx| {
                     let elem = unsafe { n.data().get_unchecked(ctx, generator, &idx, None) };
-
                     let accumulator = ctx.builder.build_load(accumulator_addr, "").unwrap();
-                    let result = call_max(ctx, (elem_ty, accumulator), (elem_ty, elem));
+                    let cur_idx = ctx.builder.build_load(res_idx, "").unwrap();
+
+                    let result = match fn_name {
+                        "np_argmin" | "np_min" => {
+                            call_min(ctx, (elem_ty, accumulator), (elem_ty, elem))
+                        }
+                        "np_argmax" | "np_max" => {
+                            call_max(ctx, (elem_ty, accumulator), (elem_ty, elem))
+                        }
+                        _ => unreachable!(),
+                    };
+
+                    let updated_idx = match (accumulator, result) {
+                        (BasicValueEnum::IntValue(m), BasicValueEnum::IntValue(n)) => ctx
+                            .builder
+                            .build_select(
+                                ctx.builder.build_int_compare(IntPredicate::NE, m, n, "").unwrap(),
+                                idx.into(),
+                                cur_idx,
+                                "",
+                            )
+                            .unwrap(),
+                        (BasicValueEnum::FloatValue(m), BasicValueEnum::FloatValue(n)) => ctx
+                            .builder
+                            .build_select(
+                                ctx.builder
+                                    .build_float_compare(FloatPredicate::ONE, m, n, "")
+                                    .unwrap(),
+                                idx.into(),
+                                cur_idx,
+                                "",
+                            )
+                            .unwrap(),
+                        _ => unsupported_type(ctx, fn_name, &[elem_ty, elem_ty]),
+                    };
+                    ctx.builder.build_store(res_idx, updated_idx).unwrap();
                     ctx.builder.build_store(accumulator_addr, result).unwrap();
 
                     Ok(())
                 },
-                llvm_usize.const_int(1, false),
+                llvm_int64.const_int(1, false),
             )?;
 
-            let accumulator = ctx.builder.build_load(accumulator_addr, "").unwrap();
+            match fn_name {
-            accumulator
+                "np_argmin" | "np_argmax" => ctx.builder.build_load(res_idx, "").unwrap(),
+                "np_max" | "np_min" => ctx.builder.build_load(accumulator_addr, "").unwrap(),
+                _ => unreachable!(),
+            }
         }
 
-        _ => unsupported_type(ctx, FN_NAME, &[a_ty]),
+        _ => unsupported_type(ctx, fn_name, &[a_ty]),
     })
 }
 

nac3core/src/codegen/expr.rs

@@ -576,6 +576,21 @@ impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
         params: [Option<IntValue<'ctx>>; 3],
         loc: Location,
     ) {
+        let error_id = self.resolver.get_string_id(name);
+        let error_id = self.ctx.i32_type().const_int(error_id as u64, false);
+        self.raise_exn_by_id(generator, error_id, msg, params, loc);
+    }
+
+    pub fn raise_exn_by_id<G: CodeGenerator + ?Sized>(
+        &mut self,
+        generator: &mut G,
+        error_id: IntValue<'ctx>,
+        msg: BasicValueEnum<'ctx>,
+        params: [Option<IntValue<'ctx>>; 3],
+        loc: Location,
+    ) {
+        assert_eq!(error_id.get_type().get_bit_width(), 32);
+
         let zelf = if let Some(exception_val) = self.exception_val {
             exception_val
         } else {
@@ -588,8 +603,7 @@ impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
         let zero = int32.const_zero();
         unsafe {
             let id_ptr = self.builder.build_in_bounds_gep(zelf, &[zero, zero], "exn.id").unwrap();
-            let id = self.resolver.get_string_id(name);
+            self.builder.build_store(id_ptr, error_id).unwrap();
-            self.builder.build_store(id_ptr, int32.const_int(id as u64, false)).unwrap();
             let ptr = self
                 .builder
                 .build_in_bounds_gep(zelf, &[zero, int32.const_int(5, false)], "exn.msg")
@@ -652,6 +666,32 @@ impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
         self.raise_exn(generator, err_name, err_msg, params, loc);
         self.builder.position_at_end(then_block);
     }
+
+    pub fn make_assert_impl_by_id<G: CodeGenerator + ?Sized>(
+        &mut self,
+        generator: &mut G,
+        cond: IntValue<'ctx>,
+        err_id: IntValue<'ctx>,
+        err_msg: BasicValueEnum<'ctx>,
+        params: [Option<IntValue<'ctx>>; 3],
+        loc: Location,
+    ) {
+        let i1 = self.ctx.bool_type();
+        let i1_true = i1.const_all_ones();
+        // we assume that the condition is most probably true, so the normal path is the most
+        // probable path
+        // even if this assumption is violated, it does not matter as exception unwinding is
+        // slow anyway...
+        let cond = call_expect(self, cond, i1_true, Some("expect"));
+        let current_bb = self.builder.get_insert_block().unwrap();
+        let current_fun = current_bb.get_parent().unwrap();
+        let then_block = self.ctx.insert_basic_block_after(current_bb, "succ");
+        let exn_block = self.ctx.append_basic_block(current_fun, "fail");
+        self.builder.build_conditional_branch(cond, then_block, exn_block).unwrap();
+        self.builder.position_at_end(exn_block);
+        self.raise_exn_by_id(generator, err_id, err_msg, params, loc);
+        self.builder.position_at_end(then_block);
+    }
 }
 
 /// See [`CodeGenerator::gen_constructor`].
@@ -2457,7 +2497,29 @@ pub fn gen_expr<'ctx, G: CodeGenerator>(
             Some((_, Some(static_value), _)) => ValueEnum::Static(static_value.clone()),
             None => {
                 let resolver = ctx.resolver.clone();
-                resolver.get_symbol_value(*id, ctx).unwrap()
+                if let Some(res) = resolver.get_symbol_value(*id, ctx) {
+                    res
+                } else {
+                    // Allow "raise Exception" short form
+                    let def_id = resolver.get_identifier_def(*id).map_err(|e| {
+                        format!("{} (at {})", e.iter().next().unwrap(), expr.location)
+                    })?;
+                    let def = ctx.top_level.definitions.read();
+                    if let TopLevelDef::Class { constructor, .. } = *def[def_id.0].read() {
+                        let TypeEnum::TFunc(signature) =
+                            ctx.unifier.get_ty(constructor.unwrap()).as_ref().clone()
+                        else {
+                            return Err(format!(
+                                "Failed to resolve symbol {} (at {})",
+                                id, expr.location
+                            ));
+                        };
+                        return Ok(generator
+                            .gen_call(ctx, None, (&signature, def_id), Vec::default())?
+                            .map(Into::into));
+                    }
+                    return Err(format!("Failed to resolve symbol {} (at {})", id, expr.location));
+                }
             }
         },
         ExprKind::List { elts, .. } => {

nac3core/src/codegen/irrt/error_context.rs

@@ -0,0 +1,196 @@
+use inkwell::types::IntType;
+use inkwell::values::IntValue;
+
+use crate::codegen::optics::*;
+use crate::codegen::CodeGenContext;
+use crate::codegen::CodeGenerator;
+
+use super::util::get_sized_dependent_function_name;
+use super::util::FunctionBuilder;
+
+#[derive(Debug, Clone)]
+pub struct StrLens<'ctx> {
+    pub size_type: IntType<'ctx>,
+}
+
+// TODO: nac3core has hardcoded a lot of "str"
+pub struct StrFields<'ctx> {
+    pub content: GepGetter<AddressLens<IntLens<'ctx>>>,
+    pub length: GepGetter<IntLens<'ctx>>,
+}
+
+impl<'ctx> StructureOptic<'ctx> for StrLens<'ctx> {
+    type Fields = StrFields<'ctx>;
+
+    fn struct_name(&self) -> &'static str {
+        "str"
+    }
+
+    fn build_fields(&self, builder: &mut FieldBuilder<'ctx>) -> Self::Fields {
+        StrFields {
+            content: builder.add_field("content", AddressLens(IntLens(builder.ctx.i8_type()))),
+            length: builder.add_field("length", IntLens(self.size_type)),
+        }
+    }
+}
+
+pub struct ErrorIdsFields<'ctx> {
+    pub index_error: GepGetter<IntLens<'ctx>>,
+    pub value_error: GepGetter<IntLens<'ctx>>,
+    pub assertion_error: GepGetter<IntLens<'ctx>>,
+    pub runtime_error: GepGetter<IntLens<'ctx>>,
+}
+
+#[derive(Debug, Clone)]
+pub struct ErrorIdsLens;
+
+impl<'ctx> StructureOptic<'ctx> for ErrorIdsLens {
+    type Fields = ErrorIdsFields<'ctx>;
+
+    fn struct_name(&self) -> &'static str {
+        "ErrorIds"
+    }
+
+    fn build_fields(&self, builder: &mut FieldBuilder<'ctx>) -> Self::Fields {
+        let i32_lens = IntLens(builder.ctx.i32_type());
+        ErrorIdsFields {
+            index_error: builder.add_field("index_error", i32_lens),
+            value_error: builder.add_field("value_error", i32_lens),
+            assertion_error: builder.add_field("assertion_error", i32_lens),
+            runtime_error: builder.add_field("runtime_error", i32_lens),
+        }
+    }
+}
+
+pub struct ErrorContextFields<'ctx> {
+    pub error_ids: GepGetter<AddressLens<ErrorIdsLens>>,
+    pub error_id: GepGetter<IntLens<'ctx>>,
+    pub message_template: GepGetter<AddressLens<IntLens<'ctx>>>,
+    pub param1: GepGetter<IntLens<'ctx>>,
+    pub param2: GepGetter<IntLens<'ctx>>,
+    pub param3: GepGetter<IntLens<'ctx>>,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct ErrorContextLens;
+
+impl<'ctx> StructureOptic<'ctx> for ErrorContextLens {
+    type Fields = ErrorContextFields<'ctx>;
+
+    fn struct_name(&self) -> &'static str {
+        "ErrorContext"
+    }
+
+    fn build_fields(&self, builder: &mut FieldBuilder<'ctx>) -> Self::Fields {
+        ErrorContextFields {
+            error_ids: builder.add_field("error_ids", AddressLens(ErrorIdsLens)),
+            error_id: builder.add_field("error_id", IntLens(builder.ctx.i32_type())),
+            message_template: builder
+                .add_field("message_template", AddressLens(IntLens(builder.ctx.i8_type()))),
+            param1: builder.add_field("param1", IntLens(builder.ctx.i64_type())),
+            param2: builder.add_field("param2", IntLens(builder.ctx.i64_type())),
+            param3: builder.add_field("param3", IntLens(builder.ctx.i64_type())),
+        }
+    }
+}
+
+fn build_error_ids<'ctx>(ctx: &CodeGenContext<'ctx, '_>) -> Address<'ctx, ErrorIdsLens> {
+    // ErrorIdsLens.get_fields(ctx.ctx).assertion_error.
+    let error_ids = ErrorIdsLens.alloca(ctx, "error_ids");
+    let llvm_i32 = ctx.ctx.i32_type();
+
+    let get_string_id =
+        |string_id| llvm_i32.const_int(ctx.resolver.get_string_id(string_id) as u64, false);
+
+    error_ids.focus(ctx, |fields| &fields.index_error).store(ctx, &get_string_id("0:IndexError"));
+    error_ids.focus(ctx, |fields| &fields.value_error).store(ctx, &get_string_id("0:ValueError"));
+    error_ids
+        .focus(ctx, |fields| &fields.assertion_error)
+        .store(ctx, &get_string_id("0:AssertionError"));
+    error_ids
+        .focus(ctx, |fields| &fields.runtime_error)
+        .store(ctx, &get_string_id("0:RuntimeError"));
+
+    error_ids
+}
+
+pub fn call_nac3_error_context_initialize<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    errctx: &Address<'ctx, ErrorContextLens>,
+    error_ids: &Address<'ctx, ErrorIdsLens>,
+) {
+    FunctionBuilder::begin(ctx, "__nac3_error_context_initialize")
+        .arg("errctx", &AddressLens(ErrorContextLens), errctx)
+        .arg("error_ids", &AddressLens(ErrorIdsLens), error_ids)
+        .returning_void();
+}
+
+pub fn call_nac3_error_context_has_no_error<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    errctx: &Address<'ctx, ErrorContextLens>,
+) -> IntValue<'ctx> {
+    FunctionBuilder::begin(ctx, "__nac3_error_context_has_no_error")
+        .arg("errctx", &AddressLens(ErrorContextLens), errctx)
+        .returning("has_error", &IntLens(ctx.ctx.bool_type()))
+}
+
+pub fn call_nac3_error_context_get_error_str<'ctx>(
+    size_type: IntType<'ctx>,
+    ctx: &CodeGenContext<'ctx, '_>,
+    errctx: &Address<'ctx, ErrorContextLens>,
+    dst_str: &Address<'ctx, StrLens<'ctx>>,
+) -> IntValue<'ctx> {
+    FunctionBuilder::begin(
+        ctx,
+        &get_sized_dependent_function_name(size_type, "__nac3_error_context_get_error_str"),
+    )
+    .arg("errctx", &AddressLens(ErrorContextLens), errctx)
+    .arg("dst_str", &AddressLens(StrLens { size_type }), dst_str)
+    .returning("has_error", &IntLens(ctx.ctx.bool_type()))
+}
+
+pub fn prepare_error_context<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+) -> Address<'ctx, ErrorContextLens> {
+    let error_ids = build_error_ids(ctx);
+    let errctx_ptr = ErrorContextLens.alloca(ctx, "errctx");
+    call_nac3_error_context_initialize(ctx, &errctx_ptr, &error_ids);
+    errctx_ptr
+}
+
+pub fn check_error_context<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    errctx_ptr: &Address<'ctx, ErrorContextLens>,
+) {
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    let has_error = call_nac3_error_context_has_no_error(ctx, errctx_ptr);
+    let error_str_ptr = StrLens { size_type }.alloca(ctx, "error_str");
+    call_nac3_error_context_get_error_str(size_type, ctx, errctx_ptr, &error_str_ptr);
+
+    let error_id = errctx_ptr.focus(ctx, |fields| &fields.error_id).load(ctx, "error_id");
+    let error_str = error_str_ptr.load(ctx, "error_str");
+    let param1 = errctx_ptr.focus(ctx, |fields| &fields.param1).load(ctx, "param1");
+    let param2 = errctx_ptr.focus(ctx, |fields| &fields.param2).load(ctx, "param2");
+    let param3 = errctx_ptr.focus(ctx, |fields| &fields.param3).load(ctx, "param3");
+    ctx.make_assert_impl_by_id(
+        generator,
+        has_error,
+        error_id,
+        error_str.get_llvm_value(),
+        [Some(param1), Some(param2), Some(param3)],
+        ctx.current_loc,
+    );
+}
+
+pub fn call_nac3_dummy_raise<G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext,
+) {
+    let errctx = prepare_error_context(ctx);
+    FunctionBuilder::begin(ctx, "__nac3_error_dummy_raise")
+        .arg("errctx", &AddressLens(ErrorContextLens), &errctx)
+        .returning_void();
+    check_error_context(generator, ctx, &errctx);
+}

nac3core/src/codegen/irrt/mod.rs

@@ -21,6 +21,11 @@ use inkwell::{
 use itertools::Either;
 use nac3parser::ast::Expr;
 
+pub mod error_context;
+pub mod numpy;
+pub mod test;
+pub mod util;
+
 #[must_use]
 pub fn load_irrt(ctx: &Context) -> Module {
     let bitcode_buf = MemoryBuffer::create_from_memory_range(

nac3core/src/codegen/irrt/numpy.rs

@@ -0,0 +1,415 @@
+use std::marker::PhantomData;
+
+use inkwell::{
+    types::{BasicType, BasicTypeEnum, IntType},
+    values::{BasicValueEnum, IntValue, PointerValue},
+};
+
+use crate::codegen::optics::*;
+use crate::{
+    codegen::{
+        classes::{ListValue, UntypedArrayLikeAccessor},
+        stmt::gen_for_callback_incrementing,
+        CodeGenContext, CodeGenerator,
+    },
+    typecheck::typedef::{Type, TypeEnum},
+};
+
+use super::{
+    error_context::{check_error_context, prepare_error_context, ErrorContextLens},
+    util::{get_sized_dependent_function_name, FunctionBuilder},
+};
+
+pub struct NpArrayFields<'ctx> {
+    pub data: GepGetter<AddressLens<IntLens<'ctx>>>,
+    pub itemsize: GepGetter<IntLens<'ctx>>,
+    pub ndims: GepGetter<IntLens<'ctx>>,
+    pub shape: GepGetter<AddressLens<IntLens<'ctx>>>,
+    pub strides: GepGetter<AddressLens<IntLens<'ctx>>>,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct NpArrayLens<'ctx> {
+    pub size_type: IntType<'ctx>,
+}
+
+impl<'ctx> StructureOptic<'ctx> for NpArrayLens<'ctx> {
+    type Fields = NpArrayFields<'ctx>;
+
+    fn struct_name(&self) -> &'static str {
+        "NDArray"
+    }
+
+    fn build_fields(&self, builder: &mut FieldBuilder<'ctx>) -> Self::Fields {
+        NpArrayFields {
+            data: builder.add_field("data", AddressLens(IntLens(builder.ctx.i8_type()))),
+            itemsize: builder.add_field("itemsize", IntLens(builder.ctx.i8_type())),
+            ndims: builder.add_field("ndims", IntLens(builder.ctx.i8_type())),
+            shape: builder.add_field("shape", AddressLens(IntLens(self.size_type))),
+            strides: builder.add_field("strides", AddressLens(IntLens(self.size_type))),
+        }
+    }
+}
+
+// Other convenient utilities for NpArray
+impl<'ctx> Address<'ctx, NpArrayLens<'ctx>> {
+    pub fn shape_array(&self, ctx: &CodeGenContext<'ctx, '_>) -> ArraySlice<'ctx, IntLens<'ctx>> {
+        let ndims = self.focus(ctx, |fields| &fields.ndims).load(ctx, "ndims");
+        let shape_base_ptr = self.focus(ctx, |fields| &fields.shape).load(ctx, "shape");
+        ArraySlice { num_elements: ndims, base: shape_base_ptr }
+    }
+
+    pub fn strides_array(&self, ctx: &CodeGenContext<'ctx, '_>) -> ArraySlice<'ctx, IntLens<'ctx>> {
+        let ndims = self.focus(ctx, |fields| &fields.ndims).load(ctx, "ndims");
+        let strides_base_ptr = self.focus(ctx, |fields| &fields.strides).load(ctx, "strides");
+        ArraySlice { num_elements: ndims, base: strides_base_ptr }
+    }
+}
+
+type ProducerWriteToArray<'ctx, G, ElementOptic> = Box<
+    dyn Fn(
+            &mut G,
+            &mut CodeGenContext<'ctx, '_>,
+            &ArraySlice<'ctx, ElementOptic>,
+        ) -> Result<(), String>
+        + 'ctx,
+>;
+
+struct Producer<'ctx, G: CodeGenerator + ?Sized, ElementOptic> {
+    pub count: IntValue<'ctx>,
+    pub write_to_array: ProducerWriteToArray<'ctx, G, ElementOptic>,
+}
+
+/// TODO: UPDATE DOCUMENTATION
+/// LLVM-typed implementation for generating a [`Producer`] that sets a list of ints.
+///
+/// * `elem_ty` - The element type of the `NDArray`.
+/// * `shape` - The `shape` parameter used to construct the `NDArray`.
+///
+/// ### Notes on `shape`
+///
+/// Just like numpy, the `shape` argument can be:
+///   1. A list of `int32`;   e.g., `np.empty([600, 800, 3])`
+///   2. A tuple of `int32`;  e.g., `np.empty((600, 800, 3))`
+///   3. A scalar `int32`;     e.g., `np.empty(3)`, this is functionally equivalent to `np.empty([3])`
+///
+/// See also [`typecheck::type_inferencer::fold_numpy_function_call_shape_argument`] to
+/// learn how `shape` gets from being a Python user expression to here.
+pub fn parse_input_shape_arg<'ctx, G>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    shape: BasicValueEnum<'ctx>,
+    shape_ty: Type,
+) -> Producer<'ctx, G, IntLens<'ctx>>
+where
+    G: CodeGenerator + ?Sized,
+{
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    match &*ctx.unifier.get_ty(shape_ty) {
+        TypeEnum::TObj { obj_id, .. }
+            if *obj_id == ctx.primitives.list.obj_id(&ctx.unifier).unwrap() =>
+        {
+            // 1. A list of ints; e.g., `np.empty([600, 800, 3])`
+
+            // A list has to be a PointerValue
+            let shape_list = ListValue::from_ptr_val(shape.into_pointer_value(), size_type, None);
+
+            // Create `Producer`
+            let ndims = shape_list.load_size(ctx, Some("count"));
+            Producer {
+                count: ndims,
+                write_to_array: Box::new(move |ctx, generator, dst_array| {
+                    // Basically iterate through the list and write to `dst_slice` accordingly
+                    let init_val = size_type.const_zero();
+                    let max_val = (ndims, false);
+                    let incr_val = size_type.const_int(1, false);
+                    gen_for_callback_incrementing(
+                        ctx,
+                        generator,
+                        init_val,
+                        max_val,
+                        |generator, ctx, _hooks, axis| {
+                            // Get the dimension at `axis`
+                            let dim =
+                                shape_list.data().get(ctx, generator, &axis, None).into_int_value();
+
+                            // Cast `dim` to SizeT
+                            let dim = ctx
+                                .builder
+                                .build_int_s_extend_or_bit_cast(dim, size_type, "dim_casted")
+                                .unwrap();
+
+                            // Write
+                            dst_array.ix(ctx, axis, "dim").store(ctx, &dim);
+                            Ok(())
+                        },
+                        incr_val,
+                    )
+                }),
+            }
+        }
+        TypeEnum::TTuple { ty: tuple_types } => {
+            // 2. A tuple of ints; e.g., `np.empty((600, 800, 3))`
+
+            // Get the length/size of the tuple, which also happens to be the value of `ndims`.
+            let ndims = tuple_types.len();
+
+            // A tuple has to be a StructValue
+            // Read [`codegen::expr::gen_expr`] to see how `nac3core` translates a Python tuple into LLVM.
+            let shape_tuple = shape.into_struct_value();
+
+            Producer {
+                count: size_type.const_int(ndims as u64, false),
+                write_to_array: Box::new(move |_generator, ctx, dst_array| {
+                    for axis in 0..ndims {
+                        // Get the dimension at `axis`
+                        let dim = ctx
+                            .builder
+                            .build_extract_value(
+                                shape_tuple,
+                                axis as u32,
+                                format!("dim{axis}").as_str(),
+                            )
+                            .unwrap()
+                            .into_int_value();
+
+                        // Cast `dim` to SizeT
+                        let dim = ctx
+                            .builder
+                            .build_int_s_extend_or_bit_cast(dim, size_type, "dim_casted")
+                            .unwrap();
+
+                        // Write
+                        dst_array
+                            .ix(ctx, size_type.const_int(axis as u64, false), "dim")
+                            .store(ctx, &dim);
+                    }
+                    Ok(())
+                }),
+            }
+        }
+        TypeEnum::TObj { obj_id, .. }
+            if *obj_id == ctx.primitives.int32.obj_id(&ctx.unifier).unwrap() =>
+        {
+            // 3. A scalar int; e.g., `np.empty(3)`, this is functionally equivalent to `np.empty([3])`
+
+            // The value has to be an integer
+            let shape_int = shape.into_int_value();
+
+            Producer {
+                count: size_type.const_int(1, false),
+                write_to_array: Box::new(move |_generator, ctx, dst_array| {
+                    // Cast `shape_int` to SizeT
+                    let dim = ctx
+                        .builder
+                        .build_int_s_extend_or_bit_cast(shape_int, size_type, "dim_casted")
+                        .unwrap();
+
+                    // Write
+                    dst_array
+                        .ix(ctx, size_type.const_zero() /* Only index 0 is set */, "dim")
+                        .store(ctx, &dim);
+
+                    Ok(())
+                }),
+            }
+        }
+        _ => panic!("parse_input_shape_arg encountered unknown type"),
+    }
+}
+
+pub fn alloca_ndarray<'ctx, G>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    elem_type: BasicTypeEnum<'ctx>,
+    ndims: IntValue<'ctx>,
+    name: &str,
+) -> Result<Address<'ctx, NpArrayLens<'ctx>>, String>
+where
+    G: CodeGenerator + ?Sized,
+{
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    // Allocate ndarray
+    let ndarray_ptr = NpArrayLens { size_type }.alloca(ctx, name);
+
+    // Set ndims
+    ndarray_ptr.focus(ctx, |fields| &fields.ndims).store(ctx, &ndims);
+
+    // Set itemsize
+    let itemsize = elem_type.size_of().unwrap();
+    let itemsize =
+        ctx.builder.build_int_s_extend_or_bit_cast(itemsize, size_type, "itemsize").unwrap();
+    ndarray_ptr.focus(ctx, |fields| &fields.itemsize).store(ctx, &itemsize);
+
+    // Allocate and set shape
+    let shape_ptr = ctx.builder.build_array_alloca(size_type, ndims, "shape").unwrap();
+    ndarray_ptr
+        .focus(ctx, |fields| &fields.shape)
+        .store(ctx, &Address { addressee_optic: IntLens(size_type), address: shape_ptr });
+
+    // Allocate and set strides
+    let strides_ptr = ctx.builder.build_array_alloca(size_type, ndims, "strides").unwrap();
+    ndarray_ptr
+        .focus(ctx, |fields| &fields.strides)
+        .store(ctx, &Address { addressee_optic: IntLens(size_type), address: strides_ptr });
+
+    Ok(ndarray_ptr)
+}
+
+pub enum NDArrayInitMode<'ctx, G: CodeGenerator + ?Sized> {
+    NDim { ndim: IntValue<'ctx> },
+    Shape { shape: Producer<'ctx, G, IntLens<'ctx>> },
+    ShapeAndAllocaData { shape: Producer<'ctx, G, IntLens<'ctx>> },
+}
+
+/// TODO: DOCUMENT ME
+pub fn alloca_ndarray_and_init<'ctx, G>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    elem_type: BasicTypeEnum<'ctx>,
+    init_mode: NDArrayInitMode<'ctx, G>,
+    name: &str,
+) -> Result<Address<'ctx, NpArrayLens<'ctx>>, String>
+where
+    G: CodeGenerator + ?Sized,
+{
+    // It is implemented verbosely in order to make the initialization modes super clear in their intent.
+    match init_mode {
+        NDArrayInitMode::NDim { ndim } => {
+            let ndarray = alloca_ndarray(generator, ctx, elem_type, ndims, name)?;
+            Ok(ndarray)
+        }
+        NDArrayInitMode::Shape { shape } => {
+            let ndims = shape.count;
+            let ndarray_ptr = alloca_ndarray(generator, ctx, elem_type, ndims, name)?;
+
+            // Fill `ndarray.shape`
+            (shape.write_to_array)(generator, ctx, &ndarray_ptr.shape_array(ctx))?;
+
+            // Check if `shape` has bad inputs
+            call_nac3_ndarray_util_assert_shape_no_negative(
+                generator,
+                ctx,
+                ndims,
+                &ndarray_ptr.focus(ctx, |fields| &fields.shape).load(ctx, "shape"),
+            );
+
+            // NOTE: DO NOT DO `set_strides_by_shape` HERE.
+            // Simply this is because we specified that `SetShape` wouldn't do `set_strides_by_shape`
+
+            Ok(ndarray_ptr)
+        }
+        NDArrayInitMode::ShapeAndAllocaData { shape } => {
+            let ndims = shape.count;
+            let ndarray_ptr = alloca_ndarray(generator, ctx, elem_type, ndims, name)?;
+
+            // Fill `ndarray.shape`
+            (shape.write_to_array)(generator, ctx, &ndarray_ptr.shape_array(ctx))?;
+
+            // Check if `shape` has bad inputs
+            call_nac3_ndarray_util_assert_shape_no_negative(
+                generator,
+                ctx,
+                ndims,
+                &ndarray_ptr.focus(ctx, |fields| &fields.shape).load(ctx, "shape"),
+            );
+
+            // Now we populate `ndarray.data` by alloca-ing.
+            // But first, we need to know the size of the ndarray to know how many elements to alloca,
+            // since calculating nbytes of an ndarray requires `ndarray.shape` to be set.
+            let ndarray_nbytes = call_nac3_ndarray_nbytes(generator, ctx, &ndarray_ptr);
+
+            // Alloca `data` and assign it to `ndarray.data`
+            let data_ptr =
+                ctx.builder.build_array_alloca(ctx.ctx.i8_type(), ndarray_nbytes, "data").unwrap();
+            ndarray_ptr.focus(ctx, |fields| &fields.data).store(
+                ctx,
+                &Address { addressee_optic: IntLens::int8(ctx.ctx), address: data_ptr },
+            );
+
+            // Finally, do `set_strides_by_shape`
+            // Check out https://ajcr.net/stride-guide-part-1/ to see what numpy "strides" are.
+            call_nac3_ndarray_set_strides_by_shape(generator, ctx, &ndarray_ptr);
+
+            Ok(ndarray_ptr)
+        }
+    }
+}
+
+fn call_nac3_ndarray_util_assert_shape_no_negative<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ndims: IntValue<'ctx>,
+    shape_ptr: &Address<'ctx, IntLens<'ctx>>,
+) {
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    let errctx = prepare_error_context(ctx);
+    FunctionBuilder::begin(
+        ctx,
+        &get_sized_dependent_function_name(
+            size_type,
+            "__nac3_ndarray_util_assert_shape_no_negative",
+        ),
+    )
+    .arg("errctx", &AddressLens(ErrorContextLens), &errctx)
+    .arg("ndims", &IntLens(size_type), &ndims)
+    .arg("shape", &AddressLens(IntLens(size_type)), shape_ptr)
+    .returning_void();
+    check_error_context(generator, ctx, &errctx);
+}
+
+fn call_nac3_ndarray_set_strides_by_shape<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ndarray_ptr: &Address<'ctx, NpArrayLens<'ctx>>,
+) {
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    FunctionBuilder::begin(
+        ctx,
+        &get_sized_dependent_function_name(
+            size_type,
+            "__nac3_ndarray_util_assert_shape_no_negative",
+        ),
+    )
+    .arg("ndarray", &AddressLens(NpArrayLens { size_type }), ndarray_ptr)
+    .returning_void();
+}
+
+fn call_nac3_ndarray_nbytes<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ndarray_ptr: &Address<'ctx, NpArrayLens<'ctx>>,
+) -> IntValue<'ctx> {
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    FunctionBuilder::begin(
+        ctx,
+        &get_sized_dependent_function_name(
+            size_type,
+            "__nac3_ndarray_util_assert_shape_no_negative",
+        ),
+    )
+    .arg("ndarray", &AddressLens(NpArrayLens { size_type }), ndarray_ptr)
+    .returning("nbytes", &IntLens(size_type))
+}
+
+pub fn call_nac3_ndarray_fill_generic<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ndarray_ptr: &Address<'ctx, NpArrayLens<'ctx>>,
+    fill_value_ptr: PointerValue<'ctx>,
+) {
+    let size_type = generator.get_size_type(ctx.ctx);
+
+    FunctionBuilder::begin(
+        ctx,
+        &get_sized_dependent_function_name(size_type, "__nac3_ndarray_fill_generic"),
+    )
+    .arg("ndarray", &AddressLens(NpArrayLens { size_type }), ndarray_ptr)
+    .arg("pvalue", &OpaqueAddressLens, fill_value_ptr)
+    .returning_void();
+}

nac3core/src/codegen/irrt/test.rs

@@ -0,0 +1,25 @@
+#[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/codegen/irrt/util.rs

@@ -0,0 +1,77 @@
+use inkwell::{
+    types::{BasicMetadataTypeEnum, BasicType, IntType},
+    values::{AnyValue, BasicMetadataValueEnum},
+};
+
+use crate::codegen::optics::*;
+use crate::{codegen::CodeGenContext, util::SizeVariant};
+
+fn get_size_variant(ty: IntType) -> SizeVariant {
+    match ty.get_bit_width() {
+        32 => SizeVariant::Bits32,
+        64 => SizeVariant::Bits64,
+        _ => unreachable!("Unsupported int type bit width {}", ty.get_bit_width()),
+    }
+}
+
+#[must_use]
+pub fn get_sized_dependent_function_name(ty: IntType, fn_name: &str) -> String {
+    let mut fn_name = fn_name.to_owned();
+    match get_size_variant(ty) {
+        SizeVariant::Bits32 => {
+            // Do nothing, `fn_name` already has the correct name
+        }
+        SizeVariant::Bits64 => {
+            // Append "64", this is the naming convention
+            fn_name.push_str("64");
+        }
+    }
+    fn_name
+}
+
+// TODO: Variadic argument?
+pub struct FunctionBuilder<'ctx, 'a> {
+    ctx: &'a CodeGenContext<'ctx, 'a>,
+    fn_name: &'a str,
+    arguments: Vec<(BasicMetadataTypeEnum<'ctx>, BasicMetadataValueEnum<'ctx>)>,
+}
+
+impl<'ctx, 'a> FunctionBuilder<'ctx, 'a> {
+    pub fn begin(ctx: &'a CodeGenContext<'ctx, 'a>, fn_name: &'a str) -> Self {
+        FunctionBuilder { ctx, fn_name, arguments: Vec::new() }
+    }
+
+    // The name is for self-documentation
+    #[must_use]
+    pub fn arg<S: MemoryOptic<'ctx>>(mut self, _name: &'static str, optic: &S, arg: &S::MemoryValue) -> Self {
+        self.arguments
+            .push((optic.get_llvm_type(self.ctx.ctx).into(), arg.get_llvm_value().into()));
+        self
+    }
+
+    pub fn returning<S: Prism<'ctx>>(self, name: &'static str, return_prism: &S) -> S::MemoryValue {
+        let (param_tys, param_vals): (Vec<_>, Vec<_>) = self.arguments.into_iter().unzip();
+
+        let function = self.ctx.module.get_function(self.fn_name).unwrap_or_else(|| {
+            let return_type = return_prism.get_llvm_type(self.ctx.ctx);
+            let fn_type = return_type.fn_type(&param_tys, false);
+            self.ctx.module.add_function(self.fn_name, fn_type, None)
+        });
+
+        let ret = self.ctx.builder.build_call(function, &param_vals, name).unwrap();
+        return_prism.review(ret.as_any_value_enum())
+    }
+
+    // TODO: Code duplication, but otherwise returning<S: Optic<'ctx>> cannot resolve S if return_optic = None
+    pub fn returning_void(self) {
+        let (param_tys, param_vals): (Vec<_>, Vec<_>) = self.arguments.into_iter().unzip();
+
+        let function = self.ctx.module.get_function(self.fn_name).unwrap_or_else(|| {
+            let return_type = self.ctx.ctx.void_type();
+            let fn_type = return_type.fn_type(&param_tys, false);
+            self.ctx.module.add_function(self.fn_name, fn_type, None)
+        });
+
+        self.ctx.builder.build_call(function, &param_vals, "").unwrap();
+    }
+}

nac3core/src/codegen/llvm_intrinsics.rs

@@ -35,6 +35,54 @@ fn get_float_intrinsic_repr(ctx: &Context, ft: FloatType) -> &'static str {
     unreachable!()
 }
 
+/// Invokes the [`llvm.lifetime.start`](https://releases.llvm.org/14.0.0/docs/LangRef.html#llvm-lifetime-start-intrinsic)
+/// intrinsic.
+pub fn call_lifetime_start<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    size: IntValue<'ctx>,
+    ptr: PointerValue<'ctx>,
+) {
+    const FN_NAME: &str = "llvm.lifetime.start";
+    // NOTE: inkwell temporary workaround, see [`call_stackrestore`] for details
+    let intrinsic_fn = ctx.module.get_function(FN_NAME).unwrap_or_else(|| {
+        let llvm_void = ctx.ctx.void_type();
+        let llvm_i64 = ctx.ctx.i64_type();
+        let llvm_p0i8 = ctx.ctx.i8_type().ptr_type(AddressSpace::default());
+        let fn_type = llvm_void.fn_type(&[llvm_i64.into(), llvm_p0i8.into()], false);
+
+        ctx.module.add_function(FN_NAME, fn_type, None)
+    });
+
+    ctx.builder
+        .build_call(intrinsic_fn, &[size.into(), ptr.into()], "")
+        .map(CallSiteValue::try_as_basic_value)
+        .unwrap();
+}
+
+/// Invokes the [`llvm.lifetime.end`](https://releases.llvm.org/14.0.0/docs/LangRef.html#llvm-lifetime-end-intrinsic)
+/// intrinsic.
+pub fn call_lifetime_end<'ctx>(
+    ctx: &CodeGenContext<'ctx, '_>,
+    size: IntValue<'ctx>,
+    ptr: PointerValue<'ctx>,
+) {
+    const FN_NAME: &str = "llvm.lifetime.end";
+    // NOTE: inkwell temporary workaround, see [`call_stackrestore`] for details
+    let intrinsic_fn = ctx.module.get_function(FN_NAME).unwrap_or_else(|| {
+        let llvm_void = ctx.ctx.void_type();
+        let llvm_i64 = ctx.ctx.i64_type();
+        let llvm_p0i8 = ctx.ctx.i8_type().ptr_type(AddressSpace::default());
+        let fn_type = llvm_void.fn_type(&[llvm_i64.into(), llvm_p0i8.into()], false);
+
+        ctx.module.add_function(FN_NAME, fn_type, None)
+    });
+
+    ctx.builder
+        .build_call(intrinsic_fn, &[size.into(), ptr.into()], "")
+        .map(CallSiteValue::try_as_basic_value)
+        .unwrap();
+}
+
 /// Invokes the [`llvm.stacksave`](https://llvm.org/docs/LangRef.html#llvm-stacksave-intrinsic)
 /// intrinsic.
 pub fn call_stacksave<'ctx>(

nac3core/src/codegen/mod.rs

@@ -23,8 +23,10 @@ use inkwell::{
     values::{BasicValueEnum, FunctionValue, IntValue, PhiValue, PointerValue},
     AddressSpace, IntPredicate, OptimizationLevel,
 };
+use irrt::error_context::StrLens;
 use itertools::Itertools;
 use nac3parser::ast::{Location, Stmt, StrRef};
+use optics::MemoryOptic as _;
 use parking_lot::{Condvar, Mutex};
 use std::collections::{HashMap, HashSet};
 use std::sync::{
@@ -42,6 +44,8 @@ mod generator;
 pub mod irrt;
 pub mod llvm_intrinsics;
 pub mod numpy;
+pub mod numpy_new;
+pub mod optics;
 pub mod stmt;
 
 #[cfg(test)]
@@ -646,6 +650,8 @@ pub fn gen_func_impl<
         ..primitives
     };
 
+    let llvm_str_ty =
+        StrLens { size_type: generator.get_size_type(context) }.get_llvm_type(context);
     let mut type_cache: HashMap<_, _> = [
         (primitives.int32, context.i32_type().into()),
         (primitives.int64, context.i64_type().into()),
@@ -653,21 +659,7 @@ pub fn gen_func_impl<
         (primitives.uint64, context.i64_type().into()),
         (primitives.float, context.f64_type().into()),
         (primitives.bool, context.i8_type().into()),
-        (primitives.str, {
+        (primitives.str, llvm_str_ty),
-            let name = "str";
-            match module.get_struct_type(name) {
-                None => {
-                    let str_type = context.opaque_struct_type("str");
-                    let fields = [
-                        context.i8_type().ptr_type(AddressSpace::default()).into(),
-                        generator.get_size_type(context).into(),
-                    ];
-                    str_type.set_body(&fields, false);
-                    str_type.into()
-                }
-                Some(t) => t.as_basic_type_enum(),
-            }
-        }),
         (primitives.range, RangeType::new(context).as_base_type().into()),
         (primitives.exception, {
             let name = "Exception";
@@ -677,7 +669,7 @@ pub fn gen_func_impl<
                 let exception = context.opaque_struct_type("Exception");
                 let int32 = context.i32_type().into();
                 let int64 = context.i64_type().into();
-                let str_ty = module.get_struct_type("str").unwrap().as_basic_type_enum();
+                let str_ty = llvm_str_ty;
                 let fields = [int32, str_ty, int32, int32, str_ty, str_ty, int64, int64, int64];
                 exception.set_body(&fields, false);
                 exception.ptr_type(AddressSpace::default()).as_basic_type_enum()

nac3core/src/codegen/numpy_new.rs

@@ -0,0 +1,91 @@
+use inkwell::values::{BasicValueEnum, PointerValue};
+use nac3parser::ast::StrRef;
+
+use crate::{
+    codegen::optics::build_opaque_alloca, symbol_resolver::ValueEnum, toplevel::DefinitionId, typecheck::typedef::{FunSignature, Type}
+};
+
+use super::{
+    irrt::{
+        self,
+        numpy::{alloca_ndarray_and_init, parse_input_shape_arg, NDArrayInitMode, NpArrayLens},
+    },
+    optics::Address,
+    CodeGenContext, CodeGenerator,
+};
+
+/// LLVM-typed implementation for generating the implementation for constructing an empty `NDArray`.
+fn call_ndarray_empty_impl<'ctx, G>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    elem_ty: Type,
+    shape: BasicValueEnum<'ctx>,
+    shape_ty: Type,
+    name: &str,
+) -> Result<Address<'ctx, NpArrayLens<'ctx>>, String>
+where
+    G: CodeGenerator + ?Sized,
+{
+    let elem_type = ctx.get_llvm_type(generator, elem_ty);
+    let shape = parse_input_shape_arg(generator, ctx, shape, shape_ty);
+    let ndarray_ptr = alloca_ndarray_and_init(
+        generator,
+        ctx,
+        elem_type,
+        NDArrayInitMode::ShapeAndAllocaData { shape },
+        name,
+    )?;
+    Ok(ndarray_ptr)
+}
+
+fn call_ndarray_fill_impl<'ctx, G>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    elem_ty: Type,
+    shape: BasicValueEnum<'ctx>,
+    shape_ty: Type,
+    fill_value: BasicValueEnum<'ctx>,
+    name: &str,
+) -> Result<Address<'ctx, NpArrayLens<'ctx>>, String>
+where
+    G: CodeGenerator + ?Sized,
+{
+    let ndarray_ptr = call_ndarray_empty_impl(generator, ctx, elem_ty, shape, shape_ty, name)?;
+
+    // NOTE: fill_value's type is not checked!!
+    let fill_value_ptr = build_opaque_alloca(ctx, fill_value.get_type(), name);
+    fill_value_ptr.store(ctx, );
+    // let fill_value_ptr = ctx.builder.build_alloca(, "fill_value_ptr").unwrap();
+    // ctx.builder.build_store(fill_value_ptr, fill_value);
+
+    // let ok = irrt::numpy::call_nac3_ndarray_fill_generic(generator, ctx, ndarray_ptr, Address { fill_value_ptr } );
+    todo!()
+    Ok(ndarray_ptr)
+}
+
+/// Generates LLVM IR for `np.empty`.
+pub fn gen_ndarray_empty<'ctx, G>(
+    context: &mut CodeGenContext<'ctx, '_>,
+    obj: &Option<(Type, ValueEnum<'ctx>)>,
+    fun: (&FunSignature, DefinitionId),
+    args: &[(Option<StrRef>, ValueEnum<'ctx>)],
+    generator: &mut dyn CodeGenerator,
+) -> Result<PointerValue<'ctx>, String> {
+    assert!(obj.is_none());
+    assert_eq!(args.len(), 1);
+
+    // Parse arguments
+    let shape_ty = fun.0.args[0].ty;
+    let shape = args[0].1.clone().to_basic_value_enum(context, generator, shape_ty)?;
+
+    // Implementation
+    let ndarray_ptr = call_ndarray_empty_impl(
+        generator,
+        context,
+        context.primitives.float,
+        shape,
+        shape_ty,
+        "empty_ndarray",
+    )?;
+    Ok(ndarray_ptr.address)
+}

nac3core/src/codegen/optics/address.rs

@@ -0,0 +1,130 @@
+use inkwell::{
+    context::Context,
+    types::{BasicType, BasicTypeEnum},
+    values::{AnyValue, BasicValue, BasicValueEnum, PointerValue},
+    AddressSpace,
+};
+
+use crate::codegen::CodeGenContext;
+
+use super::core::{MemoryGetter, MemoryOptic, MemorySetter, OpticValue, Prism};
+
+#[derive(Debug, Clone)]
+pub struct Address<'ctx, AddresseeOptic> {
+    pub addressee_optic: AddresseeOptic,
+    pub address: PointerValue<'ctx>,
+}
+
+impl<'ctx, AddresseeOptic> Address<'ctx, AddresseeOptic> {
+    pub fn cast_to<S: MemoryOptic<'ctx>>(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        new_optic: S,
+    ) -> Address<'ctx, S> {
+        let to_ptr_type = new_optic.get_llvm_type(ctx.ctx).ptr_type(AddressSpace::default());
+        let casted_address =
+            ctx.builder.build_pointer_cast(self.address, to_ptr_type, "ptr_casted").unwrap();
+        Address { addressee_optic: new_optic, address: casted_address }
+    }
+
+    pub fn cast_to_opaque(&self) -> OpaqueAddress<'ctx> {
+        OpaqueAddress(self.address)
+    }
+}
+
+impl<'ctx, AddresseeOptic> OpticValue<'ctx> for Address<'ctx, AddresseeOptic> {
+    fn get_llvm_value(&self) -> BasicValueEnum<'ctx> {
+        self.address.as_basic_value_enum()
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct AddressLens<AddresseeOptic>(pub AddresseeOptic);
+
+impl<'ctx, AddresseeOptic: MemoryOptic<'ctx>> MemoryOptic<'ctx> for AddressLens<AddresseeOptic> {
+    type MemoryValue = Address<'ctx, AddresseeOptic>;
+
+    fn get_llvm_type(&self, ctx: &'ctx Context) -> BasicTypeEnum<'ctx> {
+        self.0.get_llvm_type(ctx).ptr_type(AddressSpace::default()).as_basic_type_enum()
+    }
+}
+
+impl<'ctx, AddresseeOptic: MemoryOptic<'ctx>> Prism<'ctx> for AddressLens<AddresseeOptic> {
+    fn review<V: AnyValue<'ctx>>(&self, value: V) -> Self::MemoryValue {
+        Address {
+            addressee_optic: self.0.clone(),
+            address: value.as_any_value_enum().into_pointer_value(),
+        }
+    }
+}
+
+impl<'ctx, AddressesOptic: MemoryOptic<'ctx>> MemoryGetter<'ctx> for AddressLens<AddressesOptic> {
+    fn get(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        name: &str,
+    ) -> Self::MemoryValue {
+        self.review(ctx.builder.build_load(pointer, name).unwrap())
+    }
+}
+
+impl<'ctx, AddressesOptic: MemoryOptic<'ctx>> MemorySetter<'ctx> for AddressLens<AddressesOptic> {
+    fn set(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        value: &Self::MemoryValue,
+    ) {
+        ctx.builder.build_store(pointer, value.address).unwrap();
+    }
+}
+
+// To make [`Address`] convenient to use
+impl<'ctx, AddresseeOptic: MemoryGetter<'ctx>> Address<'ctx, AddresseeOptic> {
+    pub fn load(&self, ctx: &CodeGenContext<'ctx, '_>, name: &str) -> AddresseeOptic::MemoryValue {
+        self.addressee_optic.get(ctx, self.address, name)
+    }
+}
+
+// To make [`Address`] convenient to use
+impl<'ctx, AddresseeOptic: MemorySetter<'ctx>> Address<'ctx, AddresseeOptic> {
+    pub fn store(&self, ctx: &CodeGenContext<'ctx, '_>, value: &AddresseeOptic::MemoryValue) {
+        self.addressee_optic.set(ctx, self.address, value);
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct OpaqueAddress<'ctx>(pub PointerValue<'ctx>);
+
+impl<'ctx> OpaqueAddress<'ctx> {
+    pub fn cast_to<AddresseeOptic: MemoryOptic<'ctx>>(
+        &self,
+        ctx: &'ctx CodeGenContext,
+        addressee_optic: AddresseeOptic,
+        name: &str,
+    ) -> Address<'ctx, AddresseeOptic> {
+        let ptr = ctx.builder.build_pointer_cast(
+            self.0,
+            addressee_optic.get_llvm_type(ctx.ctx).ptr_type(AddressSpace::default()),
+            name,
+        );
+        Address { addressee_optic, address: ptr }
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct OpaqueAddressLens;
+
+impl<'ctx> MemoryOptic<'ctx> for OpaqueAddressLens {
+    type MemoryValue = BasicValueEnum<'ctx>;
+
+    fn get_llvm_type(&self, ctx: &'ctx Context) -> BasicTypeEnum<'ctx> {
+        ctx.i8_type().ptr_type(AddressSpace::default()).as_basic_type_enum()
+    }
+}
+
+impl<'ctx> OpaqueAddress<'ctx> {
+    pub fn store(&self, ctx: &CodeGenContext<'ctx, '_>, value: BasicValueEnum<'ctx>) {
+    }
+}

nac3core/src/codegen/optics/core.rs

@@ -0,0 +1,51 @@
+use inkwell::{
+    context::Context,
+    types::BasicTypeEnum,
+    values::{AnyValue, BasicValue, BasicValueEnum, PointerValue},
+};
+
+use crate::codegen::CodeGenContext;
+
+use super::address::Address;
+
+// TODO: Write a taxonomy
+
+pub trait OpticValue<'ctx> {
+    fn get_llvm_value(&self) -> BasicValueEnum<'ctx>;
+}
+
+impl<'ctx, T: BasicValue<'ctx>> OpticValue<'ctx> for T {
+    fn get_llvm_value(&self) -> BasicValueEnum<'ctx> {
+        self.as_basic_value_enum()
+    }
+}
+
+// TODO: The interface is unintuitive
+pub trait MemoryOptic<'ctx>: Clone {
+    type MemoryValue: OpticValue<'ctx>;
+
+    fn get_llvm_type(&self, ctx: &'ctx Context) -> BasicTypeEnum<'ctx>;
+
+    fn alloca(&self, ctx: &CodeGenContext<'ctx, '_>, name: &str) -> Address<'ctx, Self> {
+        let ptr = ctx.builder.build_alloca(self.get_llvm_type(ctx.ctx), name).unwrap();
+        Address { addressee_optic: self.clone(), address: ptr }
+    }
+}
+
+pub trait Prism<'ctx>: MemoryOptic<'ctx> {
+    // TODO: Return error if `review` fails
+    fn review<V: AnyValue<'ctx>>(&self, value: V) -> Self::MemoryValue;
+}
+
+pub trait MemoryGetter<'ctx>: MemoryOptic<'ctx> {
+    fn get(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        name: &str,
+    ) -> Self::MemoryValue;
+}
+
+pub trait MemorySetter<'ctx>: MemoryOptic<'ctx> {
+    fn set(&self, ctx: &CodeGenContext<'ctx, '_>, pointer: PointerValue<'ctx>, value: &Self::MemoryValue);
+}

nac3core/src/codegen/optics/gep.rs

@@ -0,0 +1,53 @@
+use inkwell::{
+    context::Context,
+    types::{BasicType, BasicTypeEnum},
+    values::PointerValue,
+    AddressSpace,
+};
+
+use crate::codegen::CodeGenContext;
+
+use super::{
+    address::Address,
+    core::{MemoryGetter, MemoryOptic},
+};
+
+// ((Memory, Pointer) -> ElementOptic::Value*)
+#[derive(Debug, Clone)]
+pub struct GepGetter<ElementOptic> {
+    /// The LLVM GEP index
+    pub gep_index: u64,
+    /// Element (or field in the context of `struct`s) name. Used for cosmetics.
+    pub name: &'static str,
+    /// The lens to view the actual value after applying this [`FieldLens<T>`]
+    pub element_optic: ElementOptic,
+}
+
+impl<'ctx, ElementOptic: MemoryOptic<'ctx>> MemoryOptic<'ctx> for GepGetter<ElementOptic> {
+    type MemoryValue = Address<'ctx, ElementOptic>;
+
+    fn get_llvm_type(&self, ctx: &'ctx Context) -> BasicTypeEnum<'ctx> {
+        self.element_optic.get_llvm_type(ctx).ptr_type(AddressSpace::default()).as_basic_type_enum()
+    }
+}
+
+impl<'ctx, ElementOptic: MemoryOptic<'ctx>> MemoryGetter<'ctx> for GepGetter<ElementOptic> {
+    fn get(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        name: &str,
+    ) -> Self::MemoryValue {
+        let llvm_i32 = ctx.ctx.i32_type(); // TODO: I think I'm not supposed to *just* use i32 for GEP like that
+        let element_ptr = unsafe {
+            ctx.builder
+                .build_in_bounds_gep(
+                    pointer,
+                    &[llvm_i32.const_zero(), llvm_i32.const_int(self.gep_index, false)],
+                    name,
+                )
+                .unwrap()
+        };
+        Address { address: element_ptr, addressee_optic: self.element_optic.clone() }
+    }
+}

nac3core/src/codegen/optics/int.rs

@@ -0,0 +1,66 @@
+use inkwell::{
+    context::Context,
+    types::{BasicType, BasicTypeEnum, IntType},
+    values::{AnyValue, BasicValue, IntValue, PointerValue},
+};
+
+use crate::codegen::CodeGenContext;
+
+use super::core::{MemoryGetter, MemorySetter, MemoryOptic, Prism};
+
+// NOTE: I wanted to make Int8Lens, Int16Lens, Int32Lens, with all
+// having the trait IsIntLens, and implement `impl <S: IsIntLens> Optic<S> for T`,
+// but that clashes with StructureOptic!!
+#[derive(Debug, Clone, Copy)]
+pub struct IntLens<'ctx>(pub IntType<'ctx>);
+
+impl<'ctx> IntLens<'ctx> {
+    #[must_use]
+    pub fn int8(ctx: &'ctx Context) -> IntLens<'ctx> {
+        IntLens(ctx.i8_type())
+    }
+
+    #[must_use]
+    pub fn int32(ctx: &'ctx Context) -> IntLens<'ctx> {
+        IntLens(ctx.i32_type())
+    }
+
+    #[must_use]
+    pub fn int64(ctx: &'ctx Context) -> IntLens<'ctx> {
+        IntLens(ctx.i64_type())
+    }
+}
+
+impl<'ctx> MemoryOptic<'ctx> for IntLens<'ctx> {
+    type MemoryValue = IntValue<'ctx>;
+
+    fn get_llvm_type(&self, _ctx: &'ctx Context) -> BasicTypeEnum<'ctx> {
+        self.0.as_basic_type_enum()
+    }
+}
+
+impl<'ctx> Prism<'ctx> for IntLens<'ctx> {
+    fn review<V: AnyValue<'ctx>>(&self, value: V) -> Self::MemoryValue {
+        let int = value.as_any_value_enum().into_int_value();
+        debug_assert_eq!(int.get_type().get_bit_width(), self.0.get_bit_width());
+        int
+    }
+}
+
+impl<'ctx> MemoryGetter<'ctx> for IntLens<'ctx> {
+    fn get(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        name: &str,
+    ) -> Self::MemoryValue {
+        self.review(ctx.builder.build_load(pointer, name).unwrap())
+    }
+}
+
+impl<'ctx> MemorySetter<'ctx> for IntLens<'ctx> {
+    fn set(&self, ctx: &CodeGenContext<'ctx, '_>, pointer: PointerValue<'ctx>, int: &Self::MemoryValue) {
+        debug_assert_eq!(int.get_type().get_bit_width(), self.0.get_bit_width());
+        ctx.builder.build_store(pointer, int.as_basic_value_enum()).unwrap();
+    }
+}

nac3core/src/codegen/optics/ixed.rs

@@ -0,0 +1,65 @@
+use inkwell::values::IntValue;
+
+use crate::codegen::{CodeGenContext, CodeGenerator};
+
+use super::address::Address;
+
+// Name inspired by https://hackage.haskell.org/package/lens-5.3.2/docs/Control-Lens-At.html#t:Ixed
+pub trait Ixed<'ctx, ElementOptic> {
+    // TODO: Interface/Method to expose the IntType of index?
+    // or even make index itself parameterized? (probably no)
+
+    fn ix(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        index: IntValue<'ctx>,
+        name: &str,
+    ) -> Address<'ctx, ElementOptic>;
+}
+
+// TODO: Can do interface seggregation
+pub trait BoundedIxed<'ctx, ElementOptic>: Ixed<'ctx, ElementOptic> {
+    fn num_elements(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx>;
+
+    // Check if 0 <= index < self.num_elements()
+    fn ix_bounds_checked<G: CodeGenerator + ?Sized>(
+        &self,
+        generator: &mut G,
+        ctx: &mut CodeGenContext<'ctx, '_>,
+        index: IntValue<'ctx>,
+        name: &str,
+    ) -> Address<'ctx, ElementOptic> {
+        let num_elements = self.num_elements(ctx);
+        let int_type = num_elements.get_type(); // NOTE: Weird get_type(), see comment under `trait Ixed`
+
+        assert_eq!(int_type.get_bit_width(), index.get_type().get_bit_width()); // Might as well check bit width to catch bugs
+
+        // TODO: SGE or UGE? or make it defined by the implementee?
+
+        // Check `0 <= index`
+        let lower_bounded = ctx
+            .builder
+            .build_int_compare(
+                inkwell::IntPredicate::SLE,
+                int_type.const_zero(),
+                index,
+                "lower_bounded",
+            )
+            .unwrap();
+
+        // Check `index < num_elements`
+        let upper_bounded = ctx
+            .builder
+            .build_int_compare(inkwell::IntPredicate::SLT, index, num_elements, "upper_bounded")
+            .unwrap();
+
+        // Compute `0 <= index && index < num_elements`
+        let bounded = ctx.builder.build_and(lower_bounded, upper_bounded, "bounded").unwrap();
+
+        // Assert `bounded`
+        ctx.make_assert(generator, bounded, "0:IndexError", "nac3core LLVM codegen attempting to access out of bounds array index {0}. Must satisfy 0 <= index < {2}", [Some(index), Some(num_elements), None], ctx.current_loc);
+
+        // ...and finally do indexing
+        self.ix(ctx, index, name)
+    }
+}

nac3core/src/codegen/optics/mod.rs

@@ -0,0 +1,15 @@
+pub mod address;
+pub mod core;
+pub mod gep;
+pub mod int;
+pub mod ixed;
+pub mod slice;
+pub mod structure;
+
+pub use address::*;
+pub use core::*;
+pub use gep::*;
+pub use int::*;
+pub use ixed::*;
+pub use slice::*;
+pub use structure::*;

nac3core/src/codegen/optics/slice.rs

@@ -0,0 +1,36 @@
+use super::{
+    core::MemoryOptic,
+    ixed::{BoundedIxed, Ixed},
+};
+
+use inkwell::values::IntValue;
+
+use crate::codegen::CodeGenContext;
+
+use super::address::Address;
+
+pub struct ArraySlice<'ctx, ElementOptic> {
+    pub num_elements: IntValue<'ctx>,
+    pub base: Address<'ctx, ElementOptic>,
+}
+
+impl<'ctx, ElementOptic: MemoryOptic<'ctx>> Ixed<'ctx, ElementOptic> for ArraySlice<'ctx, ElementOptic> {
+    fn ix(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        index: IntValue<'ctx>,
+        name: &str,
+    ) -> Address<'ctx, ElementOptic> {
+        let element_addr =
+            unsafe { ctx.builder.build_in_bounds_gep(self.base.address, &[index], name).unwrap() };
+        Address { address: element_addr, addressee_optic: self.base.addressee_optic.clone() }
+    }
+}
+
+impl<'ctx, ElementOptic: MemoryOptic<'ctx>> BoundedIxed<'ctx, ElementOptic>
+    for ArraySlice<'ctx, ElementOptic>
+{
+    fn num_elements(&self, _ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
+        self.num_elements
+    }
+}

nac3core/src/codegen/optics/structure.rs

@@ -0,0 +1,139 @@
+use inkwell::{
+    context::Context,
+    types::{BasicType, BasicTypeEnum},
+    values::{BasicValue, BasicValueEnum, PointerValue, StructValue},
+};
+use itertools::Itertools;
+
+use crate::codegen::CodeGenContext;
+
+use super::{
+    address::Address,
+    core::{MemoryGetter, MemorySetter, MemoryOptic, OpticValue},
+    gep::GepGetter,
+};
+
+pub trait StructureOptic<'ctx>: Clone {
+    // Fields of optics
+    type Fields;
+
+    // TODO: Make it an associated function instead?
+    fn struct_name(&self) -> &'static str;
+
+    fn build_fields(&self, builder: &mut FieldBuilder<'ctx>) -> Self::Fields;
+
+    fn get_fields(&self, ctx: &'ctx Context) -> Self::Fields {
+        let mut builder = FieldBuilder::new(ctx, self.struct_name());
+        self.build_fields(&mut builder)
+    }
+}
+
+pub struct OpticalStructValue<'ctx, StructOptic> {
+    optic: StructOptic,
+    llvm: StructValue<'ctx>,
+}
+
+impl<'ctx, StructOptic> OpticValue<'ctx> for OpticalStructValue<'ctx, StructOptic> {
+    fn get_llvm_value(&self) -> BasicValueEnum<'ctx> {
+        self.llvm.as_basic_value_enum()
+    }
+}
+
+// TODO: check StructType
+impl<'ctx, T: StructureOptic<'ctx>> MemoryOptic<'ctx> for T {
+    type MemoryValue = OpticalStructValue<'ctx, Self>;
+
+    fn get_llvm_type(&self, ctx: &'ctx Context) -> BasicTypeEnum<'ctx> {
+        let mut builder = FieldBuilder::new(ctx, self.struct_name());
+        self.build_fields(&mut builder); // Self::Fields is discarded
+
+        let field_types =
+            builder.fields.iter().map(|field_info| field_info.llvm_type).collect_vec();
+        ctx.struct_type(&field_types, false).as_basic_type_enum()
+    }
+}
+
+impl<'ctx, T: StructureOptic<'ctx>> MemoryGetter<'ctx> for T {
+    fn get(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        name: &str,
+    ) -> Self::MemoryValue {
+        OpticalStructValue {
+            optic: self.clone(),
+            llvm: ctx.builder.build_load(pointer, name).unwrap().into_struct_value(),
+        }
+    }
+}
+
+impl<'ctx, T: StructureOptic<'ctx>> MemorySetter<'ctx> for T {
+    fn set(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        pointer: PointerValue<'ctx>,
+        value: &Self::MemoryValue,
+    ) {
+        ctx.builder.build_store(pointer, value.llvm).unwrap();
+    }
+}
+
+impl<'ctx, AddresseeOptic: StructureOptic<'ctx>> Address<'ctx, AddresseeOptic> {
+    pub fn focus<GetFieldGepFn, FieldElementOptic: MemoryOptic<'ctx>>(
+        &self,
+        ctx: &CodeGenContext<'ctx, '_>,
+        get_field_gep_fn: GetFieldGepFn,
+    ) -> Address<'ctx, FieldElementOptic>
+    where
+        GetFieldGepFn: FnOnce(&AddresseeOptic::Fields) -> &GepGetter<FieldElementOptic>,
+    {
+        let fields = self.addressee_optic.get_fields(ctx.ctx);
+        let field = get_field_gep_fn(&fields);
+        field.get(ctx, self.address, field.name)
+    }
+}
+
+// Only used by [`FieldBuilder`]
+#[derive(Debug)]
+struct FieldInfo<'ctx> {
+    gep_index: u64,
+    name: &'ctx str,
+    llvm_type: BasicTypeEnum<'ctx>,
+}
+
+#[derive(Debug)]
+pub struct FieldBuilder<'ctx> {
+    pub ctx: &'ctx Context,
+    gep_index_counter: u64,
+    struct_name: &'ctx str,
+    fields: Vec<FieldInfo<'ctx>>,
+}
+
+impl<'ctx> FieldBuilder<'ctx> {
+    #[must_use]
+    pub fn new(ctx: &'ctx Context, struct_name: &'ctx str) -> Self {
+        FieldBuilder { ctx, gep_index_counter: 0, struct_name, fields: Vec::new() }
+    }
+
+    fn next_gep_index(&mut self) -> u64 {
+        let index = self.gep_index_counter;
+        self.gep_index_counter += 1;
+        index
+    }
+
+    pub fn add_field<ElementOptic: MemoryOptic<'ctx>>(
+        &mut self,
+        name: &'static str,
+        element_optic: ElementOptic,
+    ) -> GepGetter<ElementOptic> {
+        let gep_index = self.next_gep_index();
+
+        self.fields.push(FieldInfo {
+            gep_index,
+            name,
+            llvm_type: element_optic.get_llvm_type(self.ctx),
+        });
+
+        GepGetter { gep_index, name, element_optic }
+    }
+}

nac3core/src/lib.rs

@@ -23,3 +23,4 @@ pub mod codegen;
 pub mod symbol_resolver;
 pub mod toplevel;
 pub mod typecheck;
+pub(crate) 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::{
@@ -278,21 +279,12 @@ pub fn get_builtins(unifier: &mut Unifier, primitives: &PrimitiveStore) -> Built
         .collect()
 }
 
-/// A helper enum used by [`BuiltinBuilder`]
+fn get_size_variant_of_int(size_variant: SizeVariant, primitives: &PrimitiveStore) -> Type {
-#[derive(Clone, Copy)]
+    match size_variant {
-enum SizeVariant {
-    Bits32,
-    Bits64,
-}
-
-impl SizeVariant {
-    fn of_int(self, primitives: &PrimitiveStore) -> Type {
-        match self {
         SizeVariant::Bits32 => primitives.int32,
         SizeVariant::Bits64 => primitives.int64,
     }
 }
-}
 
 struct BuiltinBuilder<'a> {
     unifier: &'a mut Unifier,
@@ -510,7 +502,9 @@ impl<'a> BuiltinBuilder<'a> {
 
             PrimDef::FunMin | PrimDef::FunMax => self.build_min_max_function(prim),
 
-            PrimDef::FunNpMin | PrimDef::FunNpMax => self.build_np_min_max_function(prim),
+            PrimDef::FunNpArgmin | PrimDef::FunNpArgmax | PrimDef::FunNpMin | PrimDef::FunNpMax => {
+                self.build_np_max_min_function(prim)
+            }
 
             PrimDef::FunNpMinimum | PrimDef::FunNpMaximum => {
                 self.build_np_minimum_maximum_function(prim)
@@ -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 = get_size_variant_of_int(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 = get_size_variant_of_int(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 = get_size_variant_of_int(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 = get_size_variant_of_int(size_variant, &ctx.primitives);
                 let func = match kind {
                     Kind::Ceil => builtin_fns::call_ceil,
                     Kind::Floor => builtin_fns::call_floor,
@@ -1555,10 +1549,19 @@ impl<'a> BuiltinBuilder<'a> {
         }
     }
 
-    /// Build the functions `np_min()` and `np_max()`.
+    /// Build the functions `np_max()`, `np_min()`, `np_argmax()` and `np_argmin()`
-    fn build_np_min_max_function(&mut self, prim: PrimDef) -> TopLevelDef {
+    /// Calls `call_numpy_max_min` with the function name
-        debug_assert_prim_is_allowed(prim, &[PrimDef::FunNpMin, PrimDef::FunNpMax]);
+    fn build_np_max_min_function(&mut self, prim: PrimDef) -> TopLevelDef {
+        debug_assert_prim_is_allowed(
+            prim,
+            &[PrimDef::FunNpArgmin, PrimDef::FunNpArgmax, PrimDef::FunNpMin, PrimDef::FunNpMax],
+        );
 
+        let (var_map, ret_ty) = match prim {
+            PrimDef::FunNpArgmax | PrimDef::FunNpArgmin => {
+                (self.num_or_ndarray_var_map.clone(), self.primitives.int64)
+            }
+            PrimDef::FunNpMax | PrimDef::FunNpMin => {
                 let ret_ty = self.unifier.get_fresh_var(Some("R".into()), None);
                 let var_map = self
                     .num_or_ndarray_var_map
@@ -1566,28 +1569,25 @@ impl<'a> BuiltinBuilder<'a> {
                     .into_iter()
                     .chain(once((ret_ty.id, ret_ty.ty)))
                     .collect::<IndexMap<_, _>>();
+                (var_map, ret_ty.ty)
+            }
+            _ => unreachable!(),
+        };
 
         create_fn_by_codegen(
             self.unifier,
             &var_map,
             prim.name(),
-            ret_ty.ty,
+            ret_ty,
-            &[(self.float_or_ndarray_ty.ty, "a")],
+            &[(self.num_or_ndarray_ty.ty, "a")],
             Box::new(move |ctx, _, fun, args, generator| {
                 let a_ty = fun.0.args[0].ty;
                 let a = args[0].1.clone().to_basic_value_enum(ctx, generator, a_ty)?;
 
-                let func = match prim {
+                Ok(Some(builtin_fns::call_numpy_max_min(generator, ctx, (a_ty, a), prim.name())?))
-                    PrimDef::FunNpMin => builtin_fns::call_numpy_min,
-                    PrimDef::FunNpMax => builtin_fns::call_numpy_max,
-                    _ => unreachable!(),
-                };
-
-                Ok(Some(func(generator, ctx, (a_ty, a))?))
             }),
         )
     }
-
     /// Build the functions `np_minimum()` and `np_maximum()`.
     fn build_np_minimum_maximum_function(&mut self, prim: PrimDef) -> TopLevelDef {
         debug_assert_prim_is_allowed(prim, &[PrimDef::FunNpMinimum, PrimDef::FunNpMaximum]);

nac3core/src/toplevel/helper.rs

@@ -62,9 +62,11 @@ pub enum PrimDef {
     FunMin,
     FunNpMin,
     FunNpMinimum,
+    FunNpArgmin,
     FunMax,
     FunNpMax,
     FunNpMaximum,
+    FunNpArgmax,
     FunAbs,
     FunNpIsNan,
     FunNpIsInf,
@@ -216,9 +218,11 @@ impl PrimDef {
             PrimDef::FunMin => fun("min", None),
             PrimDef::FunNpMin => fun("np_min", None),
             PrimDef::FunNpMinimum => fun("np_minimum", None),
+            PrimDef::FunNpArgmin => fun("np_argmin", None),
             PrimDef::FunMax => fun("max", None),
             PrimDef::FunNpMax => fun("np_max", None),
             PrimDef::FunNpMaximum => fun("np_maximum", None),
+            PrimDef::FunNpArgmax => fun("np_argmax", None),
             PrimDef::FunAbs => fun("abs", None),
             PrimDef::FunNpIsNan => fun("np_isnan", None),
             PrimDef::FunNpIsInf => fun("np_isinf", None),

nac3core/src/typecheck/type_inferencer/mod.rs

@@ -398,7 +398,10 @@ impl<'a> Fold<()> for Inferencer<'a> {
                 }
                 if let Some(exc) = exc {
                     self.virtual_checks.push((
-                        exc.custom.unwrap(),
+                        match &*self.unifier.get_ty(exc.custom.unwrap()) {
+                            TypeEnum::TFunc(sign) => sign.ret,
+                            _ => exc.custom.unwrap(),
+                        },
                         self.primitives.exception,
                         exc.location,
                     ));

nac3core/src/util.rs

@@ -0,0 +1,6 @@
+/// A helper enum used by [`BuiltinBuilder`]
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum SizeVariant {
+    Bits32,
+    Bits64,
+}

nac3standalone/demo/interpret_demo.py

@@ -183,8 +183,10 @@ def patch(module):
     module.np_isinf = np.isinf
     module.np_min = np.min
     module.np_minimum = np.minimum
+    module.np_argmin = np.argmin
     module.np_max = np.max
     module.np_maximum = np.maximum
+    module.np_argmax = np.argmax
     module.np_sin = np.sin
     module.np_cos = np.cos
     module.np_exp = np.exp

nac3standalone/demo/src/ndarray.py

@@ -867,6 +867,13 @@ def test_ndarray_minimum_broadcast_rhs_scalar():
     output_ndarray_float_2(min_x_zeros)
     output_ndarray_float_2(min_x_ones)
 
+def test_ndarray_argmin():
+    x = np_array([[1., 2.], [3., 4.]])
+    y = np_argmin(x)
+
+    output_ndarray_float_2(x)
+    output_int64(y)
+
 def test_ndarray_max():
     x = np_identity(2)
     y = np_max(x)
@@ -910,6 +917,13 @@ def test_ndarray_maximum_broadcast_rhs_scalar():
     output_ndarray_float_2(max_x_zeros)
     output_ndarray_float_2(max_x_ones)
 
+def test_ndarray_argmax():
+    x = np_array([[1., 2.], [3., 4.]])
+    y = np_argmax(x)
+
+    output_ndarray_float_2(x)
+    output_int64(y)
+
 def test_ndarray_abs():
     x = np_identity(2)
     y = abs(x)
@@ -1524,11 +1538,13 @@ def run() -> int32:
     test_ndarray_minimum_broadcast()
     test_ndarray_minimum_broadcast_lhs_scalar()
     test_ndarray_minimum_broadcast_rhs_scalar()
+    test_ndarray_argmin()
     test_ndarray_max()
     test_ndarray_maximum()
     test_ndarray_maximum_broadcast()
     test_ndarray_maximum_broadcast_lhs_scalar()
     test_ndarray_maximum_broadcast_rhs_scalar()
+    test_ndarray_argmax()
     test_ndarray_abs()
     test_ndarray_isnan()
     test_ndarray_isinf()