[core] coregen/types: Implement StructFields for NDArray

Also rename some fields to better align with their naming in numpy.

nac3artiq/src/codegen.rs

@@ -461,8 +461,7 @@ fn format_rpc_arg<'ctx>(
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, arg_ty);
             let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
             let llvm_arg_ty = NDArrayType::new(generator, ctx.ctx, llvm_elem_ty);
-            let llvm_arg =
+            let llvm_arg = llvm_arg_ty.map_value(arg.into_pointer_value(), None);
-                NDArrayValue::from_pointer_value(arg.into_pointer_value(), llvm_usize, None);
 
             let llvm_usize_sizeof = ctx
                 .builder
@@ -499,7 +498,7 @@ fn format_rpc_arg<'ctx>(
             call_memcpy_generic(
                 ctx,
                 pbuffer_dims_begin,
-                llvm_arg.dim_sizes().base_ptr(ctx, generator),
+                llvm_arg.shape().base_ptr(ctx, generator),
                 dims_buf_sz,
                 llvm_i1.const_zero(),
             );
@@ -613,7 +612,7 @@ fn format_rpc_ret<'ctx>(
             // Set `ndarray.ndims`
             ndarray.store_ndims(ctx, generator, llvm_usize.const_int(ndims, false));
             // Allocate `ndarray.shape` [size_t; ndims]
-            ndarray.create_dim_sizes(ctx, llvm_usize, ndarray.load_ndims(ctx));
+            ndarray.create_shape(ctx, llvm_usize, ndarray.load_ndims(ctx));
 
             /*
                 ndarray now:
@@ -703,7 +702,7 @@ fn format_rpc_ret<'ctx>(
 
             call_memcpy_generic(
                 ctx,
-                ndarray.dim_sizes().base_ptr(ctx, generator),
+                ndarray.shape().base_ptr(ctx, generator),
                 pbuffer_dims,
                 sizeof_dims,
                 llvm_i1.const_zero(),
@@ -715,7 +714,7 @@ fn format_rpc_ret<'ctx>(
             // `ndarray.shape` must be initialized beforehand in this implementation
             //   (for ndarray.create_data() to know how many elements to allocate)
             let num_elements =
-                call_ndarray_calc_size(generator, ctx, &ndarray.dim_sizes(), (None, None));
+                call_ndarray_calc_size(generator, ctx, &ndarray.shape(), (None, None));
 
             // debug_assert(nelems * sizeof(T) >= ndarray_nbytes)
             if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
@@ -1369,13 +1368,18 @@ fn polymorphic_print<'ctx>(
 
             TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
                 let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
+                let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
                 fmt.push_str("array([");
                 flush(ctx, generator, &mut fmt, &mut args);
 
-                let val =
+                let val = NDArrayValue::from_pointer_value(
-                    NDArrayValue::from_pointer_value(value.into_pointer_value(), llvm_usize, None);
+                    value.into_pointer_value(),
-                let len = call_ndarray_calc_size(generator, ctx, &val.dim_sizes(), (None, None));
+                    llvm_elem_ty,
+                    llvm_usize,
+                    None,
+                );
+                let len = call_ndarray_calc_size(generator, ctx, &val.shape(), (None, None));
                 let last =
                     ctx.builder.build_int_sub(len, llvm_usize.const_int(1, false), "").unwrap();
 

nac3core/irrt/irrt.cpp

@@ -1,5 +1,4 @@
 #include "irrt/exception.hpp"
-#include "irrt/int_types.hpp"
 #include "irrt/list.hpp"
 #include "irrt/math.hpp"
 #include "irrt/ndarray.hpp"

nac3core/irrt/irrt/cslice.hpp

@@ -4,6 +4,6 @@
 
 template<typename SizeT>
 struct CSlice {
-    uint8_t* base;
+    void* base;
     SizeT len;
 };

nac3core/irrt/irrt/exception.hpp

@@ -6,7 +6,7 @@
 /**
  * @brief The int type of ARTIQ exception IDs.
  */
-typedef int32_t ExceptionId;
+using ExceptionId = int32_t;
 
 /*
  * Set of exceptions C++ IRRT can use.
@@ -55,14 +55,14 @@ void _raise_exception_helper(ExceptionId id,
                              int64_t param2) {
     Exception<SizeT> e = {
         .id = id,
-        .filename = {.base = reinterpret_cast<uint8_t*>(const_cast<char*>(filename)),
+        .filename = {.base = reinterpret_cast<void*>(const_cast<char*>(filename)),
-                     .len = static_cast<int32_t>(__builtin_strlen(filename))},
+                     .len = static_cast<SizeT>(__builtin_strlen(filename))},
         .line = line,
         .column = 0,
-        .function = {.base = reinterpret_cast<uint8_t*>(const_cast<char*>(function)),
+        .function = {.base = reinterpret_cast<void*>(const_cast<char*>(function)),
-                     .len = static_cast<int32_t>(__builtin_strlen(function))},
+                     .len = static_cast<SizeT>(__builtin_strlen(function))},
-        .msg = {.base = reinterpret_cast<uint8_t*>(const_cast<char*>(msg)),
+        .msg = {.base = reinterpret_cast<void*>(const_cast<char*>(msg)),
-                .len = static_cast<int32_t>(__builtin_strlen(msg))},
+                .len = static_cast<SizeT>(__builtin_strlen(msg))},
     };
     e.params[0] = param0;
     e.params[1] = param1;
@@ -70,6 +70,7 @@ void _raise_exception_helper(ExceptionId id,
     __nac3_raise(reinterpret_cast<void*>(&e));
     __builtin_unreachable();
 }
+}  // namespace
 
 /**
  * @brief Raise an exception with location details (location in the IRRT source files).
@@ -82,4 +83,3 @@ void _raise_exception_helper(ExceptionId id,
  */
 #define raise_exception(SizeT, id, msg, param0, param1, param2) \
     _raise_exception_helper<SizeT>(id, __FILE__, __LINE__, __FUNCTION__, msg, param0, param1, param2)
-}  // namespace

nac3core/irrt/irrt/int_types.hpp

@@ -8,12 +8,17 @@ using uint32_t = unsigned _BitInt(32);
 using int64_t = _BitInt(64);
 using uint64_t = unsigned _BitInt(64);
 #else
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-type"
 using int8_t = _ExtInt(8);
 using uint8_t = unsigned _ExtInt(8);
 using int32_t = _ExtInt(32);
 using uint32_t = unsigned _ExtInt(32);
 using int64_t = _ExtInt(64);
 using uint64_t = unsigned _ExtInt(64);
+#pragma clang diagnostic pop
+
 #endif
 
 // NDArray indices are always `uint32_t`.

nac3core/irrt/irrt/list.hpp

@@ -13,12 +13,12 @@ extern "C" {
 SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
                                              SliceIndex dest_end,
                                              SliceIndex dest_step,
-                                             uint8_t* dest_arr,
+                                             void* dest_arr,
                                              SliceIndex dest_arr_len,
                                              SliceIndex src_start,
                                              SliceIndex src_end,
                                              SliceIndex src_step,
-                                             uint8_t* src_arr,
+                                             void* src_arr,
                                              SliceIndex src_arr_len,
                                              const SliceIndex size) {
     /* if dest_arr_len == 0, do nothing since we do not support extending list */
@@ -29,11 +29,13 @@ SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
         const SliceIndex src_len = (src_end >= src_start) ? (src_end - src_start + 1) : 0;
         const SliceIndex dest_len = (dest_end >= dest_start) ? (dest_end - dest_start + 1) : 0;
         if (src_len > 0) {
-            __builtin_memmove(dest_arr + dest_start * size, src_arr + src_start * size, src_len * size);
+            __builtin_memmove(static_cast<uint8_t*>(dest_arr) + dest_start * size,
+                              static_cast<uint8_t*>(src_arr) + src_start * size, src_len * size);
         }
         if (dest_len > 0) {
             /* dropping */
-            __builtin_memmove(dest_arr + (dest_start + src_len) * size, dest_arr + (dest_end + 1) * size,
+            __builtin_memmove(static_cast<uint8_t*>(dest_arr) + (dest_start + src_len) * size,
+                              static_cast<uint8_t*>(dest_arr) + (dest_end + 1) * size,
                               (dest_arr_len - dest_end - 1) * size);
         }
         /* shrink size */
@@ -44,7 +46,7 @@ SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
                           && !(max(dest_start, dest_end) < min(src_start, src_end)
                                || max(src_start, src_end) < min(dest_start, dest_end));
     if (need_alloca) {
-        uint8_t* tmp = reinterpret_cast<uint8_t*>(__builtin_alloca(src_arr_len * size));
+        void* tmp = __builtin_alloca(src_arr_len * size);
         __builtin_memcpy(tmp, src_arr, src_arr_len * size);
         src_arr = tmp;
     }
@@ -53,20 +55,24 @@ SliceIndex __nac3_list_slice_assign_var_size(SliceIndex dest_start,
     for (; (src_step > 0) ? (src_ind <= src_end) : (src_ind >= src_end); src_ind += src_step, dest_ind += dest_step) {
         /* for constant optimization */
         if (size == 1) {
-            __builtin_memcpy(dest_arr + dest_ind, src_arr + src_ind, 1);
+            __builtin_memcpy(static_cast<uint8_t*>(dest_arr) + dest_ind, static_cast<uint8_t*>(src_arr) + src_ind, 1);
         } else if (size == 4) {
-            __builtin_memcpy(dest_arr + dest_ind * 4, src_arr + src_ind * 4, 4);
+            __builtin_memcpy(static_cast<uint8_t*>(dest_arr) + dest_ind * 4,
+                             static_cast<uint8_t*>(src_arr) + src_ind * 4, 4);
         } else if (size == 8) {
-            __builtin_memcpy(dest_arr + dest_ind * 8, src_arr + src_ind * 8, 8);
+            __builtin_memcpy(static_cast<uint8_t*>(dest_arr) + dest_ind * 8,
+                             static_cast<uint8_t*>(src_arr) + src_ind * 8, 8);
         } else {
             /* memcpy for var size, cannot overlap after previous alloca */
-            __builtin_memcpy(dest_arr + dest_ind * size, src_arr + src_ind * size, size);
+            __builtin_memcpy(static_cast<uint8_t*>(dest_arr) + dest_ind * size,
+                             static_cast<uint8_t*>(src_arr) + src_ind * size, size);
         }
     }
     /* only dest_step == 1 can we shrink the dest list. */
     /* size should be ensured prior to calling this function */
     if (dest_step == 1 && dest_end >= dest_start) {
-        __builtin_memmove(dest_arr + dest_ind * size, dest_arr + (dest_end + 1) * size,
+        __builtin_memmove(static_cast<uint8_t*>(dest_arr) + dest_ind * size,
+                          static_cast<uint8_t*>(dest_arr) + (dest_end + 1) * size,
                           (dest_arr_len - dest_end - 1) * size);
         return dest_arr_len - (dest_end - dest_ind) - 1;
     }

nac3core/irrt/irrt/math.hpp

@@ -90,4 +90,4 @@ double __nac3_j0(double x) {
 
     return j0(x);
 }
-}
+}  // namespace

nac3core/irrt/irrt/ndarray.hpp

@@ -141,4 +141,4 @@ void __nac3_ndarray_calc_broadcast_idx64(const uint64_t* src_dims,
                                          NDIndex* out_idx) {
     __nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
 }
-}
+}  // namespace

nac3core/irrt/irrt/slice.hpp

@@ -25,4 +25,4 @@ SliceIndex __nac3_range_slice_len(const SliceIndex start, const SliceIndex end,
         return 0;
     }
 }
-}
+}  // namespace

nac3core/src/codegen/builtin_fns.rs

@@ -21,7 +21,10 @@ use super::{
     CodeGenContext, CodeGenerator,
 };
 use crate::{
-    toplevel::{helper::PrimDef, numpy::unpack_ndarray_var_tys},
+    toplevel::{
+        helper::{arraylike_flatten_element_type, PrimDef},
+        numpy::unpack_ndarray_var_tys,
+    },
     typecheck::typedef::{Type, TypeEnum},
 };
 
@@ -65,12 +68,17 @@ pub fn call_len<'ctx, G: CodeGenerator + ?Sized>(
                 ctx.builder.build_int_truncate_or_bit_cast(len, llvm_i32, "len").unwrap()
             }
             TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
+                let elem_ty = arraylike_flatten_element_type(&mut ctx.unifier, arg_ty);
                 let llvm_usize = generator.get_size_type(ctx.ctx);
 
-                let arg =
+                let arg = NDArrayValue::from_pointer_value(
-                    NDArrayValue::from_pointer_value(arg.into_pointer_value(), llvm_usize, None);
+                    arg.into_pointer_value(),
+                    ctx.get_llvm_type(generator, elem_ty),
+                    llvm_usize,
+                    None,
+                );
 
-                let ndims = arg.dim_sizes().size(ctx, generator);
+                let ndims = arg.shape().size(ctx, generator);
                 ctx.make_assert(
                     generator,
                     ctx.builder
@@ -83,12 +91,7 @@ pub fn call_len<'ctx, G: CodeGenerator + ?Sized>(
                 );
 
                 let len = unsafe {
-                    arg.dim_sizes().get_typed_unchecked(
+                    arg.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
-                        ctx,
-                        generator,
-                        &llvm_usize.const_zero(),
-                        None,
-                    )
                 };
 
                 ctx.builder.build_int_truncate_or_bit_cast(len, llvm_i32, "len").unwrap()
@@ -143,13 +146,14 @@ pub fn call_int32<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.int32,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_int32(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -205,13 +209,14 @@ pub fn call_int64<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.int64,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_int64(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -283,13 +288,14 @@ pub fn call_uint32<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.uint32,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_uint32(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -350,13 +356,14 @@ pub fn call_uint64<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.uint64,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_uint64(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -416,13 +423,14 @@ pub fn call_float<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.float,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_float(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -462,13 +470,14 @@ pub fn call_round<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ret_elem_ty,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_round(generator, ctx, (elem_ty, val), ret_elem_ty),
             )?;
 
@@ -502,13 +511,14 @@ pub fn call_numpy_round<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.float,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_numpy_round(generator, ctx, (elem_ty, val)),
             )?;
 
@@ -567,13 +577,14 @@ pub fn call_bool<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ctx.primitives.bool,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| {
                     let elem = call_bool(generator, ctx, (elem_ty, val))?;
 
@@ -621,13 +632,14 @@ pub fn call_floor<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ret_elem_ty,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_floor(generator, ctx, (elem_ty, val), ret_elem_ty),
             )?;
 
@@ -671,13 +683,14 @@ pub fn call_ceil<'ctx, G: CodeGenerator + ?Sized>(
             if n_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) =>
         {
             let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, n_ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
                 generator,
                 ctx,
                 ret_elem_ty,
                 None,
-                NDArrayValue::from_pointer_value(n, llvm_usize, None),
+                NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None),
                 |generator, ctx, val| call_ceil(generator, ctx, (elem_ty, val), ret_elem_ty),
             )?;
 
@@ -806,8 +819,8 @@ pub fn call_numpy_minimum<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_minimum(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -906,10 +919,10 @@ pub fn call_numpy_max_min<'ctx, G: CodeGenerator + ?Sized>(
             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 llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
 
-            let n = NDArrayValue::from_pointer_value(n, llvm_usize, None);
+            let n = NDArrayValue::from_pointer_value(n, llvm_elem_ty, llvm_usize, None);
-            let n_sz = irrt::call_ndarray_calc_size(generator, ctx, &n.dim_sizes(), (None, None));
+            let n_sz = irrt::call_ndarray_calc_size(generator, ctx, &n.shape(), (None, None));
             if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
                 let n_sz_eqz = ctx
                     .builder
@@ -926,7 +939,7 @@ pub fn call_numpy_max_min<'ctx, G: CodeGenerator + ?Sized>(
                 );
             }
 
-            let accumulator_addr = generator.gen_var_alloc(ctx, llvm_ndarray_ty, None)?;
+            let accumulator_addr = generator.gen_var_alloc(ctx, llvm_elem_ty, None)?;
             let res_idx = generator.gen_var_alloc(ctx, llvm_int64.into(), None)?;
 
             unsafe {
@@ -1068,8 +1081,8 @@ pub fn call_numpy_maximum<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_maximum(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1114,6 +1127,7 @@ where
         {
             let llvm_usize = generator.get_size_type(ctx.ctx);
             let (arg_elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, arg_ty);
+            let llvm_arg_elem_ty = ctx.get_llvm_type(generator, arg_elem_ty);
             let ret_elem_ty = get_ret_elem_type(ctx, arg_elem_ty);
 
             let ndarray = ndarray_elementwise_unaryop_impl(
@@ -1121,7 +1135,7 @@ where
                 ctx,
                 ret_elem_ty,
                 None,
-                NDArrayValue::from_pointer_value(x, llvm_usize, None),
+                NDArrayValue::from_pointer_value(x, llvm_arg_elem_ty, llvm_usize, None),
                 |generator, ctx, elem_val| {
                     helper_call_numpy_unary_elementwise(
                         generator,
@@ -1508,8 +1522,8 @@ pub fn call_numpy_arctan2<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_arctan2(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1575,8 +1589,8 @@ pub fn call_numpy_copysign<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_copysign(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1642,8 +1656,8 @@ pub fn call_numpy_fmax<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_fmax(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1709,8 +1723,8 @@ pub fn call_numpy_fmin<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_fmin(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1765,8 +1779,8 @@ pub fn call_numpy_ldexp<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_ldexp(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1832,8 +1846,8 @@ pub fn call_numpy_hypot<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_hypot(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1899,8 +1913,8 @@ pub fn call_numpy_nextafter<'ctx, G: CodeGenerator + ?Sized>(
                 ctx,
                 dtype,
                 None,
-                (x1, !is_ndarray1),
+                (x1_ty, x1, !is_ndarray1),
-                (x2, !is_ndarray2),
+                (x2_ty, x2, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     call_numpy_nextafter(generator, ctx, (x1_scalar_ty, lhs), (x2_scalar_ty, rhs))
                 },
@@ -1960,14 +1974,14 @@ pub fn call_np_linalg_cholesky<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2002,14 +2016,14 @@ pub fn call_np_linalg_qr<'ctx, G: CodeGenerator + ?Sized>(
             unimplemented!("{FN_NAME} operates on float type NdArrays only");
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2052,15 +2066,15 @@ pub fn call_np_linalg_svd<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
 
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2107,14 +2121,14 @@ pub fn call_np_linalg_inv<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2149,15 +2163,15 @@ pub fn call_np_linalg_pinv<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
 
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2192,15 +2206,15 @@ pub fn call_sp_linalg_lu<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
 
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let dim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2245,7 +2259,7 @@ pub fn call_np_linalg_matrix_power<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty, x2_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
         // Changing second parameter to a `NDArray` for uniformity in function call
         let n2_array = numpy::create_ndarray_const_shape(
             generator,
@@ -2265,12 +2279,12 @@ pub fn call_np_linalg_matrix_power<'ctx, G: CodeGenerator + ?Sized>(
         let n2_array = n2_array.as_base_value().as_basic_value_enum();
 
         let outdim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
         let outdim1 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
                 .into_int_value()
         };
@@ -2340,10 +2354,10 @@ pub fn call_sp_linalg_schur<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
 
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };
@@ -2383,10 +2397,10 @@ pub fn call_sp_linalg_hessenberg<'ctx, G: CodeGenerator + ?Sized>(
             unsupported_type(ctx, FN_NAME, &[x1_ty]);
         };
 
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let n1 = NDArrayValue::from_pointer_value(n1, n1_elem_ty, llvm_usize, None);
 
         let dim0 = unsafe {
-            n1.dim_sizes()
+            n1.shape()
                 .get_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
                 .into_int_value()
         };

nac3core/src/codegen/expr.rs

@@ -1564,10 +1564,21 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
 
             assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
 
-            let left_val =
+            let llvm_ndarray_dtype1 = ctx.get_llvm_type(generator, ndarray_dtype1);
-                NDArrayValue::from_pointer_value(left_val.into_pointer_value(), llvm_usize, None);
+            let llvm_ndarray_dtype2 = ctx.get_llvm_type(generator, ndarray_dtype2);
-            let right_val =
+
-                NDArrayValue::from_pointer_value(right_val.into_pointer_value(), llvm_usize, None);
+            let left_val = NDArrayValue::from_pointer_value(
+                left_val.into_pointer_value(),
+                llvm_ndarray_dtype1,
+                llvm_usize,
+                None,
+            );
+            let right_val = NDArrayValue::from_pointer_value(
+                right_val.into_pointer_value(),
+                llvm_ndarray_dtype2,
+                llvm_usize,
+                None,
+            );
 
             let res = if op.base == Operator::MatMult {
                 // MatMult is the only binop which is not an elementwise op
@@ -1591,8 +1602,8 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
                         BinopVariant::Normal => None,
                         BinopVariant::AugAssign => Some(left_val),
                     },
-                    (left_val.as_base_value().into(), false),
+                    (ty1, left_val.as_base_value().into(), false),
-                    (right_val.as_base_value().into(), false),
+                    (ty2, right_val.as_base_value().into(), false),
                     |generator, ctx, (lhs, rhs)| {
                         gen_binop_expr_with_values(
                             generator,
@@ -1616,8 +1627,10 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
         } else {
             let (ndarray_dtype, _) =
                 unpack_ndarray_var_tys(&mut ctx.unifier, if is_ndarray1 { ty1 } else { ty2 });
+            let llvm_ndarray_dtype = ctx.get_llvm_type(generator, ndarray_dtype);
             let ndarray_val = NDArrayValue::from_pointer_value(
                 if is_ndarray1 { left_val } else { right_val }.into_pointer_value(),
+                llvm_ndarray_dtype,
                 llvm_usize,
                 None,
             );
@@ -1629,8 +1642,8 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
                     BinopVariant::Normal => None,
                     BinopVariant::AugAssign => Some(ndarray_val),
                 },
-                (left_val, !is_ndarray1),
+                (ty1, left_val, !is_ndarray1),
-                (right_val, !is_ndarray2),
+                (ty2, right_val, !is_ndarray2),
                 |generator, ctx, (lhs, rhs)| {
                     gen_binop_expr_with_values(
                         generator,
@@ -1810,8 +1823,14 @@ pub fn gen_unaryop_expr_with_values<'ctx, G: CodeGenerator>(
     } else if ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) {
         let llvm_usize = generator.get_size_type(ctx.ctx);
         let (ndarray_dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
+        let llvm_ndarray_dtype = ctx.get_llvm_type(generator, ndarray_dtype);
 
-        let val = NDArrayValue::from_pointer_value(val.into_pointer_value(), llvm_usize, None);
+        let val = NDArrayValue::from_pointer_value(
+            val.into_pointer_value(),
+            llvm_ndarray_dtype,
+            llvm_usize,
+            None,
+        );
 
         // ndarray uses `~` rather than `not` to perform elementwise inversion, convert it before
         // passing it to the elementwise codegen function
@@ -1902,15 +1921,21 @@ pub fn gen_cmpop_expr_with_values<'ctx, G: CodeGenerator>(
 
                 assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
 
-                let left_val =
+                let llvm_ndarray_dtype1 = ctx.get_llvm_type(generator, ndarray_dtype1);
-                    NDArrayValue::from_pointer_value(lhs.into_pointer_value(), llvm_usize, None);
+
+                let left_val = NDArrayValue::from_pointer_value(
+                    lhs.into_pointer_value(),
+                    llvm_ndarray_dtype1,
+                    llvm_usize,
+                    None,
+                );
                 let res = numpy::ndarray_elementwise_binop_impl(
                     generator,
                     ctx,
                     ctx.primitives.bool,
                     None,
-                    (left_val.as_base_value().into(), false),
+                    (left_ty, left_val.as_base_value().into(), false),
-                    (rhs, false),
+                    (right_ty, rhs, false),
                     |generator, ctx, (lhs, rhs)| {
                         let val = gen_cmpop_expr_with_values(
                             generator,
@@ -1941,8 +1966,8 @@ pub fn gen_cmpop_expr_with_values<'ctx, G: CodeGenerator>(
                     ctx,
                     ctx.primitives.bool,
                     None,
-                    (lhs, !is_ndarray1),
+                    (left_ty, lhs, !is_ndarray1),
-                    (rhs, !is_ndarray2),
+                    (right_ty, rhs, !is_ndarray2),
                     |generator, ctx, (lhs, rhs)| {
                         let val = gen_cmpop_expr_with_values(
                             generator,
@@ -2606,7 +2631,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                 let llvm_i32 = ctx.ctx.i32_type();
 
                 let len = unsafe {
-                    v.dim_sizes().get_typed_unchecked(
+                    v.shape().get_typed_unchecked(
                         ctx,
                         generator,
                         &llvm_usize.const_int(dim, true),
@@ -2647,7 +2672,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
 
             ExprKind::Slice { lower, upper, step } => {
                 let dim_sz = unsafe {
-                    v.dim_sizes().get_typed_unchecked(
+                    v.shape().get_typed_unchecked(
                         ctx,
                         generator,
                         &llvm_usize.const_int(dim, false),
@@ -2771,8 +2796,12 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                 // elements over
                 let subscripted_ndarray =
                     generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
-                let ndarray =
+                let ndarray = NDArrayValue::from_pointer_value(
-                    NDArrayValue::from_pointer_value(subscripted_ndarray, llvm_usize, None);
+                    subscripted_ndarray,
+                    llvm_ndarray_data_t,
+                    llvm_usize,
+                    None,
+                );
 
                 let num_dims = v.load_ndims(ctx);
                 ndarray.store_ndims(
@@ -2784,7 +2813,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                 );
 
                 let ndarray_num_dims = ndarray.load_ndims(ctx);
-                ndarray.create_dim_sizes(ctx, llvm_usize, ndarray_num_dims);
+                ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
 
                 let ndarray_num_dims = ctx
                     .builder
@@ -2795,7 +2824,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                     )
                     .unwrap();
                 let v_dims_src_ptr = unsafe {
-                    v.dim_sizes().ptr_offset_unchecked(
+                    v.shape().ptr_offset_unchecked(
                         ctx,
                         generator,
                         &llvm_usize.const_int(1, false),
@@ -2804,7 +2833,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                 };
                 call_memcpy_generic(
                     ctx,
-                    ndarray.dim_sizes().base_ptr(ctx, generator),
+                    ndarray.shape().base_ptr(ctx, generator),
                     v_dims_src_ptr,
                     ctx.builder
                         .build_int_mul(ndarray_num_dims, llvm_usize.size_of(), "")
@@ -2816,7 +2845,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
                 let ndarray_num_elems = call_ndarray_calc_size(
                     generator,
                     ctx,
-                    &ndarray.dim_sizes().as_slice_value(ctx, generator),
+                    &ndarray.shape().as_slice_value(ctx, generator),
                     (None, None),
                 );
                 let ndarray_num_elems = ctx
@@ -3510,6 +3539,7 @@ pub fn gen_expr<'ctx, G: CodeGenerator>(
                 }
                 TypeEnum::TObj { obj_id, params, .. } if *obj_id == PrimDef::NDArray.id() => {
                     let (ty, ndims) = params.iter().map(|(_, ty)| ty).collect_tuple().unwrap();
+                    let llvm_ty = ctx.get_llvm_type(generator, *ty);
 
                     let v = if let Some(v) = generator.gen_expr(ctx, value)? {
                         v.to_basic_value_enum(ctx, generator, value.custom.unwrap())?
@@ -3517,7 +3547,7 @@ pub fn gen_expr<'ctx, G: CodeGenerator>(
                     } else {
                         return Ok(None);
                     };
-                    let v = NDArrayValue::from_pointer_value(v, usize, None);
+                    let v = NDArrayValue::from_pointer_value(v, llvm_ty, usize, None);
 
                     return gen_ndarray_subscript_expr(generator, ctx, *ty, *ndims, v, slice);
                 }

nac3core/src/codegen/irrt/list.rs

@@ -0,0 +1,162 @@
+use inkwell::{
+    types::BasicTypeEnum,
+    values::{BasicValueEnum, CallSiteValue, IntValue},
+    AddressSpace, IntPredicate,
+};
+use itertools::Either;
+
+use super::calculate_len_for_slice_range;
+use crate::codegen::{
+    macros::codegen_unreachable,
+    values::{ArrayLikeValue, ListValue},
+    CodeGenContext, CodeGenerator,
+};
+
+/// This function handles 'end' **inclusively**.
+/// Order of tuples `assign_idx` and `value_idx` is ('start', 'end', 'step').
+/// Negative index should be handled before entering this function
+pub fn list_slice_assignment<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ty: BasicTypeEnum<'ctx>,
+    dest_arr: ListValue<'ctx>,
+    dest_idx: (IntValue<'ctx>, IntValue<'ctx>, IntValue<'ctx>),
+    src_arr: ListValue<'ctx>,
+    src_idx: (IntValue<'ctx>, IntValue<'ctx>, IntValue<'ctx>),
+) {
+    let size_ty = generator.get_size_type(ctx.ctx);
+    let int8_ptr = ctx.ctx.i8_type().ptr_type(AddressSpace::default());
+    let int32 = ctx.ctx.i32_type();
+    let (fun_symbol, elem_ptr_type) = ("__nac3_list_slice_assign_var_size", int8_ptr);
+    let slice_assign_fun = {
+        let ty_vec = vec![
+            int32.into(),         // dest start idx
+            int32.into(),         // dest end idx
+            int32.into(),         // dest step
+            elem_ptr_type.into(), // dest arr ptr
+            int32.into(),         // dest arr len
+            int32.into(),         // src start idx
+            int32.into(),         // src end idx
+            int32.into(),         // src step
+            elem_ptr_type.into(), // src arr ptr
+            int32.into(),         // src arr len
+            int32.into(),         // size
+        ];
+        ctx.module.get_function(fun_symbol).unwrap_or_else(|| {
+            let fn_t = int32.fn_type(ty_vec.as_slice(), false);
+            ctx.module.add_function(fun_symbol, fn_t, None)
+        })
+    };
+
+    let zero = int32.const_zero();
+    let one = int32.const_int(1, false);
+    let dest_arr_ptr = dest_arr.data().base_ptr(ctx, generator);
+    let dest_arr_ptr =
+        ctx.builder.build_pointer_cast(dest_arr_ptr, elem_ptr_type, "dest_arr_ptr_cast").unwrap();
+    let dest_len = dest_arr.load_size(ctx, Some("dest.len"));
+    let dest_len = ctx.builder.build_int_truncate_or_bit_cast(dest_len, int32, "srclen32").unwrap();
+    let src_arr_ptr = src_arr.data().base_ptr(ctx, generator);
+    let src_arr_ptr =
+        ctx.builder.build_pointer_cast(src_arr_ptr, elem_ptr_type, "src_arr_ptr_cast").unwrap();
+    let src_len = src_arr.load_size(ctx, Some("src.len"));
+    let src_len = ctx.builder.build_int_truncate_or_bit_cast(src_len, int32, "srclen32").unwrap();
+
+    // index in bound and positive should be done
+    // assert if dest.step == 1 then len(src) <= len(dest) else len(src) == len(dest), and
+    // throw exception if not satisfied
+    let src_end = ctx
+        .builder
+        .build_select(
+            ctx.builder.build_int_compare(IntPredicate::SLT, src_idx.2, zero, "is_neg").unwrap(),
+            ctx.builder.build_int_sub(src_idx.1, one, "e_min_one").unwrap(),
+            ctx.builder.build_int_add(src_idx.1, one, "e_add_one").unwrap(),
+            "final_e",
+        )
+        .map(BasicValueEnum::into_int_value)
+        .unwrap();
+    let dest_end = ctx
+        .builder
+        .build_select(
+            ctx.builder.build_int_compare(IntPredicate::SLT, dest_idx.2, zero, "is_neg").unwrap(),
+            ctx.builder.build_int_sub(dest_idx.1, one, "e_min_one").unwrap(),
+            ctx.builder.build_int_add(dest_idx.1, one, "e_add_one").unwrap(),
+            "final_e",
+        )
+        .map(BasicValueEnum::into_int_value)
+        .unwrap();
+    let src_slice_len =
+        calculate_len_for_slice_range(generator, ctx, src_idx.0, src_end, src_idx.2);
+    let dest_slice_len =
+        calculate_len_for_slice_range(generator, ctx, dest_idx.0, dest_end, dest_idx.2);
+    let src_eq_dest = ctx
+        .builder
+        .build_int_compare(IntPredicate::EQ, src_slice_len, dest_slice_len, "slice_src_eq_dest")
+        .unwrap();
+    let src_slt_dest = ctx
+        .builder
+        .build_int_compare(IntPredicate::SLT, src_slice_len, dest_slice_len, "slice_src_slt_dest")
+        .unwrap();
+    let dest_step_eq_one = ctx
+        .builder
+        .build_int_compare(
+            IntPredicate::EQ,
+            dest_idx.2,
+            dest_idx.2.get_type().const_int(1, false),
+            "slice_dest_step_eq_one",
+        )
+        .unwrap();
+    let cond_1 = ctx.builder.build_and(dest_step_eq_one, src_slt_dest, "slice_cond_1").unwrap();
+    let cond = ctx.builder.build_or(src_eq_dest, cond_1, "slice_cond").unwrap();
+    ctx.make_assert(
+        generator,
+        cond,
+        "0:ValueError",
+        "attempt to assign sequence of size {0} to slice of size {1} with step size {2}",
+        [Some(src_slice_len), Some(dest_slice_len), Some(dest_idx.2)],
+        ctx.current_loc,
+    );
+
+    let new_len = {
+        let args = vec![
+            dest_idx.0.into(),   // dest start idx
+            dest_idx.1.into(),   // dest end idx
+            dest_idx.2.into(),   // dest step
+            dest_arr_ptr.into(), // dest arr ptr
+            dest_len.into(),     // dest arr len
+            src_idx.0.into(),    // src start idx
+            src_idx.1.into(),    // src end idx
+            src_idx.2.into(),    // src step
+            src_arr_ptr.into(),  // src arr ptr
+            src_len.into(),      // src arr len
+            {
+                let s = match ty {
+                    BasicTypeEnum::FloatType(t) => t.size_of(),
+                    BasicTypeEnum::IntType(t) => t.size_of(),
+                    BasicTypeEnum::PointerType(t) => t.size_of(),
+                    BasicTypeEnum::StructType(t) => t.size_of().unwrap(),
+                    _ => codegen_unreachable!(ctx),
+                };
+                ctx.builder.build_int_truncate_or_bit_cast(s, int32, "size").unwrap()
+            }
+            .into(),
+        ];
+        ctx.builder
+            .build_call(slice_assign_fun, args.as_slice(), "slice_assign")
+            .map(CallSiteValue::try_as_basic_value)
+            .map(|v| v.map_left(BasicValueEnum::into_int_value))
+            .map(Either::unwrap_left)
+            .unwrap()
+    };
+    // update length
+    let need_update =
+        ctx.builder.build_int_compare(IntPredicate::NE, new_len, dest_len, "need_update").unwrap();
+    let current = ctx.builder.get_insert_block().unwrap().get_parent().unwrap();
+    let update_bb = ctx.ctx.append_basic_block(current, "update");
+    let cont_bb = ctx.ctx.append_basic_block(current, "cont");
+    ctx.builder.build_conditional_branch(need_update, update_bb, cont_bb).unwrap();
+    ctx.builder.position_at_end(update_bb);
+    let new_len = ctx.builder.build_int_z_extend_or_bit_cast(new_len, size_ty, "new_len").unwrap();
+    dest_arr.store_size(ctx, generator, new_len);
+    ctx.builder.build_unconditional_branch(cont_bb).unwrap();
+    ctx.builder.position_at_end(cont_bb);
+}

nac3core/src/codegen/irrt/math.rs

@@ -0,0 +1,152 @@
+use inkwell::{
+    values::{BasicValueEnum, CallSiteValue, FloatValue, IntValue},
+    IntPredicate,
+};
+use itertools::Either;
+
+use crate::codegen::{
+    macros::codegen_unreachable,
+    {CodeGenContext, CodeGenerator},
+};
+
+// repeated squaring method adapted from GNU Scientific Library:
+// https://git.savannah.gnu.org/cgit/gsl.git/tree/sys/pow_int.c
+pub fn integer_power<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    base: IntValue<'ctx>,
+    exp: IntValue<'ctx>,
+    signed: bool,
+) -> IntValue<'ctx> {
+    let symbol = match (base.get_type().get_bit_width(), exp.get_type().get_bit_width(), signed) {
+        (32, 32, true) => "__nac3_int_exp_int32_t",
+        (64, 64, true) => "__nac3_int_exp_int64_t",
+        (32, 32, false) => "__nac3_int_exp_uint32_t",
+        (64, 64, false) => "__nac3_int_exp_uint64_t",
+        _ => codegen_unreachable!(ctx),
+    };
+    let base_type = base.get_type();
+    let pow_fun = ctx.module.get_function(symbol).unwrap_or_else(|| {
+        let fn_type = base_type.fn_type(&[base_type.into(), base_type.into()], false);
+        ctx.module.add_function(symbol, fn_type, None)
+    });
+    // throw exception when exp < 0
+    let ge_zero = ctx
+        .builder
+        .build_int_compare(
+            IntPredicate::SGE,
+            exp,
+            exp.get_type().const_zero(),
+            "assert_int_pow_ge_0",
+        )
+        .unwrap();
+    ctx.make_assert(
+        generator,
+        ge_zero,
+        "0:ValueError",
+        "integer power must be positive or zero",
+        [None, None, None],
+        ctx.current_loc,
+    );
+    ctx.builder
+        .build_call(pow_fun, &[base.into(), exp.into()], "call_int_pow")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}
+
+/// Generates a call to `isinf` in IR. Returns an `i1` representing the result.
+pub fn call_isinf<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &CodeGenContext<'ctx, '_>,
+    v: FloatValue<'ctx>,
+) -> IntValue<'ctx> {
+    let intrinsic_fn = ctx.module.get_function("__nac3_isinf").unwrap_or_else(|| {
+        let fn_type = ctx.ctx.i32_type().fn_type(&[ctx.ctx.f64_type().into()], false);
+        ctx.module.add_function("__nac3_isinf", fn_type, None)
+    });
+
+    let ret = ctx
+        .builder
+        .build_call(intrinsic_fn, &[v.into()], "isinf")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap();
+
+    generator.bool_to_i1(ctx, ret)
+}
+
+/// Generates a call to `isnan` in IR. Returns an `i1` representing the result.
+pub fn call_isnan<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &CodeGenContext<'ctx, '_>,
+    v: FloatValue<'ctx>,
+) -> IntValue<'ctx> {
+    let intrinsic_fn = ctx.module.get_function("__nac3_isnan").unwrap_or_else(|| {
+        let fn_type = ctx.ctx.i32_type().fn_type(&[ctx.ctx.f64_type().into()], false);
+        ctx.module.add_function("__nac3_isnan", fn_type, None)
+    });
+
+    let ret = ctx
+        .builder
+        .build_call(intrinsic_fn, &[v.into()], "isnan")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap();
+
+    generator.bool_to_i1(ctx, ret)
+}
+
+/// Generates a call to `gamma` in IR. Returns an `f64` representing the result.
+pub fn call_gamma<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
+    let llvm_f64 = ctx.ctx.f64_type();
+
+    let intrinsic_fn = ctx.module.get_function("__nac3_gamma").unwrap_or_else(|| {
+        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
+        ctx.module.add_function("__nac3_gamma", fn_type, None)
+    });
+
+    ctx.builder
+        .build_call(intrinsic_fn, &[v.into()], "gamma")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_float_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}
+
+/// Generates a call to `gammaln` in IR. Returns an `f64` representing the result.
+pub fn call_gammaln<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
+    let llvm_f64 = ctx.ctx.f64_type();
+
+    let intrinsic_fn = ctx.module.get_function("__nac3_gammaln").unwrap_or_else(|| {
+        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
+        ctx.module.add_function("__nac3_gammaln", fn_type, None)
+    });
+
+    ctx.builder
+        .build_call(intrinsic_fn, &[v.into()], "gammaln")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_float_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}
+
+/// Generates a call to `j0` in IR. Returns an `f64` representing the result.
+pub fn call_j0<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
+    let llvm_f64 = ctx.ctx.f64_type();
+
+    let intrinsic_fn = ctx.module.get_function("__nac3_j0").unwrap_or_else(|| {
+        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
+        ctx.module.add_function("__nac3_j0", fn_type, None)
+    });
+
+    ctx.builder
+        .build_call(intrinsic_fn, &[v.into()], "j0")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_float_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}

nac3core/src/codegen/irrt/mod.rs

@@ -3,25 +3,23 @@ use inkwell::{
     context::Context,
     memory_buffer::MemoryBuffer,
     module::Module,
-    types::{BasicTypeEnum, IntType},
+    values::{BasicValue, BasicValueEnum, IntValue},
-    values::{BasicValue, BasicValueEnum, CallSiteValue, FloatValue, IntValue},
+    IntPredicate,
-    AddressSpace, IntPredicate,
 };
-use itertools::Either;
 
 use nac3parser::ast::Expr;
 
-use super::{
+use super::{CodeGenContext, CodeGenerator};
-    llvm_intrinsics,
-    macros::codegen_unreachable,
-    stmt::gen_for_callback_incrementing,
-    values::{
-        ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, ListValue, NDArrayValue,
-        TypedArrayLikeAccessor, TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
-    },
-    CodeGenContext, CodeGenerator,
-};
 use crate::{symbol_resolver::SymbolResolver, typecheck::typedef::Type};
+pub use list::*;
+pub use math::*;
+pub use ndarray::*;
+pub use slice::*;
+
+mod list;
+mod math;
+mod ndarray;
+mod slice;
 
 #[must_use]
 pub fn load_irrt<'ctx>(ctx: &'ctx Context, symbol_resolver: &dyn SymbolResolver) -> Module<'ctx> {
@@ -62,88 +60,6 @@ pub fn load_irrt<'ctx>(ctx: &'ctx Context, symbol_resolver: &dyn SymbolResolver)
     irrt_mod
 }
 
-// repeated squaring method adapted from GNU Scientific Library:
-// https://git.savannah.gnu.org/cgit/gsl.git/tree/sys/pow_int.c
-pub fn integer_power<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    base: IntValue<'ctx>,
-    exp: IntValue<'ctx>,
-    signed: bool,
-) -> IntValue<'ctx> {
-    let symbol = match (base.get_type().get_bit_width(), exp.get_type().get_bit_width(), signed) {
-        (32, 32, true) => "__nac3_int_exp_int32_t",
-        (64, 64, true) => "__nac3_int_exp_int64_t",
-        (32, 32, false) => "__nac3_int_exp_uint32_t",
-        (64, 64, false) => "__nac3_int_exp_uint64_t",
-        _ => codegen_unreachable!(ctx),
-    };
-    let base_type = base.get_type();
-    let pow_fun = ctx.module.get_function(symbol).unwrap_or_else(|| {
-        let fn_type = base_type.fn_type(&[base_type.into(), base_type.into()], false);
-        ctx.module.add_function(symbol, fn_type, None)
-    });
-    // throw exception when exp < 0
-    let ge_zero = ctx
-        .builder
-        .build_int_compare(
-            IntPredicate::SGE,
-            exp,
-            exp.get_type().const_zero(),
-            "assert_int_pow_ge_0",
-        )
-        .unwrap();
-    ctx.make_assert(
-        generator,
-        ge_zero,
-        "0:ValueError",
-        "integer power must be positive or zero",
-        [None, None, None],
-        ctx.current_loc,
-    );
-    ctx.builder
-        .build_call(pow_fun, &[base.into(), exp.into()], "call_int_pow")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
-pub fn calculate_len_for_slice_range<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    start: IntValue<'ctx>,
-    end: IntValue<'ctx>,
-    step: IntValue<'ctx>,
-) -> IntValue<'ctx> {
-    const SYMBOL: &str = "__nac3_range_slice_len";
-    let len_func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
-        let i32_t = ctx.ctx.i32_type();
-        let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into(), i32_t.into()], false);
-        ctx.module.add_function(SYMBOL, fn_t, None)
-    });
-
-    // assert step != 0, throw exception if not
-    let not_zero = ctx
-        .builder
-        .build_int_compare(IntPredicate::NE, step, step.get_type().const_zero(), "range_step_ne")
-        .unwrap();
-    ctx.make_assert(
-        generator,
-        not_zero,
-        "0:ValueError",
-        "step must not be zero",
-        [None, None, None],
-        ctx.current_loc,
-    );
-    ctx.builder
-        .build_call(len_func, &[start.into(), end.into(), step.into()], "calc_len")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
 /// NOTE: the output value of the end index of this function should be compared ***inclusively***,
 /// because python allows `a[2::-1]`, whose semantic is `[a[2], a[1], a[0]]`, which is equivalent to
 /// NO numeric slice in python.
@@ -309,644 +225,3 @@ pub fn handle_slice_indices<'ctx, G: CodeGenerator>(
         }
     }))
 }
-
-/// this function allows index out of range, since python
-/// allows index out of range in slice (`a = [1,2,3]; a[1:10] == [2,3]`).
-pub fn handle_slice_index_bound<'ctx, G: CodeGenerator>(
-    i: &Expr<Option<Type>>,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    generator: &mut G,
-    length: IntValue<'ctx>,
-) -> Result<Option<IntValue<'ctx>>, String> {
-    const SYMBOL: &str = "__nac3_slice_index_bound";
-    let func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
-        let i32_t = ctx.ctx.i32_type();
-        let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into()], false);
-        ctx.module.add_function(SYMBOL, fn_t, None)
-    });
-
-    let i = if let Some(v) = generator.gen_expr(ctx, i)? {
-        v.to_basic_value_enum(ctx, generator, i.custom.unwrap())?
-    } else {
-        return Ok(None);
-    };
-    Ok(Some(
-        ctx.builder
-            .build_call(func, &[i.into(), length.into()], "bounded_ind")
-            .map(CallSiteValue::try_as_basic_value)
-            .map(|v| v.map_left(BasicValueEnum::into_int_value))
-            .map(Either::unwrap_left)
-            .unwrap(),
-    ))
-}
-
-/// This function handles 'end' **inclusively**.
-/// Order of tuples `assign_idx` and `value_idx` is ('start', 'end', 'step').
-/// Negative index should be handled before entering this function
-pub fn list_slice_assignment<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    ty: BasicTypeEnum<'ctx>,
-    dest_arr: ListValue<'ctx>,
-    dest_idx: (IntValue<'ctx>, IntValue<'ctx>, IntValue<'ctx>),
-    src_arr: ListValue<'ctx>,
-    src_idx: (IntValue<'ctx>, IntValue<'ctx>, IntValue<'ctx>),
-) {
-    let size_ty = generator.get_size_type(ctx.ctx);
-    let int8_ptr = ctx.ctx.i8_type().ptr_type(AddressSpace::default());
-    let int32 = ctx.ctx.i32_type();
-    let (fun_symbol, elem_ptr_type) = ("__nac3_list_slice_assign_var_size", int8_ptr);
-    let slice_assign_fun = {
-        let ty_vec = vec![
-            int32.into(),         // dest start idx
-            int32.into(),         // dest end idx
-            int32.into(),         // dest step
-            elem_ptr_type.into(), // dest arr ptr
-            int32.into(),         // dest arr len
-            int32.into(),         // src start idx
-            int32.into(),         // src end idx
-            int32.into(),         // src step
-            elem_ptr_type.into(), // src arr ptr
-            int32.into(),         // src arr len
-            int32.into(),         // size
-        ];
-        ctx.module.get_function(fun_symbol).unwrap_or_else(|| {
-            let fn_t = int32.fn_type(ty_vec.as_slice(), false);
-            ctx.module.add_function(fun_symbol, fn_t, None)
-        })
-    };
-
-    let zero = int32.const_zero();
-    let one = int32.const_int(1, false);
-    let dest_arr_ptr = dest_arr.data().base_ptr(ctx, generator);
-    let dest_arr_ptr =
-        ctx.builder.build_pointer_cast(dest_arr_ptr, elem_ptr_type, "dest_arr_ptr_cast").unwrap();
-    let dest_len = dest_arr.load_size(ctx, Some("dest.len"));
-    let dest_len = ctx.builder.build_int_truncate_or_bit_cast(dest_len, int32, "srclen32").unwrap();
-    let src_arr_ptr = src_arr.data().base_ptr(ctx, generator);
-    let src_arr_ptr =
-        ctx.builder.build_pointer_cast(src_arr_ptr, elem_ptr_type, "src_arr_ptr_cast").unwrap();
-    let src_len = src_arr.load_size(ctx, Some("src.len"));
-    let src_len = ctx.builder.build_int_truncate_or_bit_cast(src_len, int32, "srclen32").unwrap();
-
-    // index in bound and positive should be done
-    // assert if dest.step == 1 then len(src) <= len(dest) else len(src) == len(dest), and
-    // throw exception if not satisfied
-    let src_end = ctx
-        .builder
-        .build_select(
-            ctx.builder.build_int_compare(IntPredicate::SLT, src_idx.2, zero, "is_neg").unwrap(),
-            ctx.builder.build_int_sub(src_idx.1, one, "e_min_one").unwrap(),
-            ctx.builder.build_int_add(src_idx.1, one, "e_add_one").unwrap(),
-            "final_e",
-        )
-        .map(BasicValueEnum::into_int_value)
-        .unwrap();
-    let dest_end = ctx
-        .builder
-        .build_select(
-            ctx.builder.build_int_compare(IntPredicate::SLT, dest_idx.2, zero, "is_neg").unwrap(),
-            ctx.builder.build_int_sub(dest_idx.1, one, "e_min_one").unwrap(),
-            ctx.builder.build_int_add(dest_idx.1, one, "e_add_one").unwrap(),
-            "final_e",
-        )
-        .map(BasicValueEnum::into_int_value)
-        .unwrap();
-    let src_slice_len =
-        calculate_len_for_slice_range(generator, ctx, src_idx.0, src_end, src_idx.2);
-    let dest_slice_len =
-        calculate_len_for_slice_range(generator, ctx, dest_idx.0, dest_end, dest_idx.2);
-    let src_eq_dest = ctx
-        .builder
-        .build_int_compare(IntPredicate::EQ, src_slice_len, dest_slice_len, "slice_src_eq_dest")
-        .unwrap();
-    let src_slt_dest = ctx
-        .builder
-        .build_int_compare(IntPredicate::SLT, src_slice_len, dest_slice_len, "slice_src_slt_dest")
-        .unwrap();
-    let dest_step_eq_one = ctx
-        .builder
-        .build_int_compare(
-            IntPredicate::EQ,
-            dest_idx.2,
-            dest_idx.2.get_type().const_int(1, false),
-            "slice_dest_step_eq_one",
-        )
-        .unwrap();
-    let cond_1 = ctx.builder.build_and(dest_step_eq_one, src_slt_dest, "slice_cond_1").unwrap();
-    let cond = ctx.builder.build_or(src_eq_dest, cond_1, "slice_cond").unwrap();
-    ctx.make_assert(
-        generator,
-        cond,
-        "0:ValueError",
-        "attempt to assign sequence of size {0} to slice of size {1} with step size {2}",
-        [Some(src_slice_len), Some(dest_slice_len), Some(dest_idx.2)],
-        ctx.current_loc,
-    );
-
-    let new_len = {
-        let args = vec![
-            dest_idx.0.into(),   // dest start idx
-            dest_idx.1.into(),   // dest end idx
-            dest_idx.2.into(),   // dest step
-            dest_arr_ptr.into(), // dest arr ptr
-            dest_len.into(),     // dest arr len
-            src_idx.0.into(),    // src start idx
-            src_idx.1.into(),    // src end idx
-            src_idx.2.into(),    // src step
-            src_arr_ptr.into(),  // src arr ptr
-            src_len.into(),      // src arr len
-            {
-                let s = match ty {
-                    BasicTypeEnum::FloatType(t) => t.size_of(),
-                    BasicTypeEnum::IntType(t) => t.size_of(),
-                    BasicTypeEnum::PointerType(t) => t.size_of(),
-                    BasicTypeEnum::StructType(t) => t.size_of().unwrap(),
-                    _ => codegen_unreachable!(ctx),
-                };
-                ctx.builder.build_int_truncate_or_bit_cast(s, int32, "size").unwrap()
-            }
-            .into(),
-        ];
-        ctx.builder
-            .build_call(slice_assign_fun, args.as_slice(), "slice_assign")
-            .map(CallSiteValue::try_as_basic_value)
-            .map(|v| v.map_left(BasicValueEnum::into_int_value))
-            .map(Either::unwrap_left)
-            .unwrap()
-    };
-    // update length
-    let need_update =
-        ctx.builder.build_int_compare(IntPredicate::NE, new_len, dest_len, "need_update").unwrap();
-    let current = ctx.builder.get_insert_block().unwrap().get_parent().unwrap();
-    let update_bb = ctx.ctx.append_basic_block(current, "update");
-    let cont_bb = ctx.ctx.append_basic_block(current, "cont");
-    ctx.builder.build_conditional_branch(need_update, update_bb, cont_bb).unwrap();
-    ctx.builder.position_at_end(update_bb);
-    let new_len = ctx.builder.build_int_z_extend_or_bit_cast(new_len, size_ty, "new_len").unwrap();
-    dest_arr.store_size(ctx, generator, new_len);
-    ctx.builder.build_unconditional_branch(cont_bb).unwrap();
-    ctx.builder.position_at_end(cont_bb);
-}
-
-/// Generates a call to `isinf` in IR. Returns an `i1` representing the result.
-pub fn call_isinf<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &CodeGenContext<'ctx, '_>,
-    v: FloatValue<'ctx>,
-) -> IntValue<'ctx> {
-    let intrinsic_fn = ctx.module.get_function("__nac3_isinf").unwrap_or_else(|| {
-        let fn_type = ctx.ctx.i32_type().fn_type(&[ctx.ctx.f64_type().into()], false);
-        ctx.module.add_function("__nac3_isinf", fn_type, None)
-    });
-
-    let ret = ctx
-        .builder
-        .build_call(intrinsic_fn, &[v.into()], "isinf")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap();
-
-    generator.bool_to_i1(ctx, ret)
-}
-
-/// Generates a call to `isnan` in IR. Returns an `i1` representing the result.
-pub fn call_isnan<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &CodeGenContext<'ctx, '_>,
-    v: FloatValue<'ctx>,
-) -> IntValue<'ctx> {
-    let intrinsic_fn = ctx.module.get_function("__nac3_isnan").unwrap_or_else(|| {
-        let fn_type = ctx.ctx.i32_type().fn_type(&[ctx.ctx.f64_type().into()], false);
-        ctx.module.add_function("__nac3_isnan", fn_type, None)
-    });
-
-    let ret = ctx
-        .builder
-        .build_call(intrinsic_fn, &[v.into()], "isnan")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap();
-
-    generator.bool_to_i1(ctx, ret)
-}
-
-/// Generates a call to `gamma` in IR. Returns an `f64` representing the result.
-pub fn call_gamma<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
-    let llvm_f64 = ctx.ctx.f64_type();
-
-    let intrinsic_fn = ctx.module.get_function("__nac3_gamma").unwrap_or_else(|| {
-        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
-        ctx.module.add_function("__nac3_gamma", fn_type, None)
-    });
-
-    ctx.builder
-        .build_call(intrinsic_fn, &[v.into()], "gamma")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_float_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
-/// Generates a call to `gammaln` in IR. Returns an `f64` representing the result.
-pub fn call_gammaln<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
-    let llvm_f64 = ctx.ctx.f64_type();
-
-    let intrinsic_fn = ctx.module.get_function("__nac3_gammaln").unwrap_or_else(|| {
-        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
-        ctx.module.add_function("__nac3_gammaln", fn_type, None)
-    });
-
-    ctx.builder
-        .build_call(intrinsic_fn, &[v.into()], "gammaln")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_float_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
-/// Generates a call to `j0` in IR. Returns an `f64` representing the result.
-pub fn call_j0<'ctx>(ctx: &CodeGenContext<'ctx, '_>, v: FloatValue<'ctx>) -> FloatValue<'ctx> {
-    let llvm_f64 = ctx.ctx.f64_type();
-
-    let intrinsic_fn = ctx.module.get_function("__nac3_j0").unwrap_or_else(|| {
-        let fn_type = llvm_f64.fn_type(&[llvm_f64.into()], false);
-        ctx.module.add_function("__nac3_j0", fn_type, None)
-    });
-
-    ctx.builder
-        .build_call(intrinsic_fn, &[v.into()], "j0")
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_float_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
-/// Generates a call to `__nac3_ndarray_calc_size`. Returns an [`IntValue`] representing the
-/// calculated total size.
-///
-/// * `dims` - An [`ArrayLikeIndexer`] containing the size of each dimension.
-/// * `range` - The dimension index to begin and end (exclusively) calculating the dimensions for,
-///   or [`None`] if starting from the first dimension and ending at the last dimension
-///   respectively.
-pub fn call_ndarray_calc_size<'ctx, G, Dims>(
-    generator: &G,
-    ctx: &CodeGenContext<'ctx, '_>,
-    dims: &Dims,
-    (begin, end): (Option<IntValue<'ctx>>, Option<IntValue<'ctx>>),
-) -> IntValue<'ctx>
-where
-    G: CodeGenerator + ?Sized,
-    Dims: ArrayLikeIndexer<'ctx>,
-{
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
-
-    let ndarray_calc_size_fn_name = match llvm_usize.get_bit_width() {
-        32 => "__nac3_ndarray_calc_size",
-        64 => "__nac3_ndarray_calc_size64",
-        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
-    };
-    let ndarray_calc_size_fn_t = llvm_usize.fn_type(
-        &[llvm_pusize.into(), llvm_usize.into(), llvm_usize.into(), llvm_usize.into()],
-        false,
-    );
-    let ndarray_calc_size_fn =
-        ctx.module.get_function(ndarray_calc_size_fn_name).unwrap_or_else(|| {
-            ctx.module.add_function(ndarray_calc_size_fn_name, ndarray_calc_size_fn_t, None)
-        });
-
-    let begin = begin.unwrap_or_else(|| llvm_usize.const_zero());
-    let end = end.unwrap_or_else(|| dims.size(ctx, generator));
-    ctx.builder
-        .build_call(
-            ndarray_calc_size_fn,
-            &[
-                dims.base_ptr(ctx, generator).into(),
-                dims.size(ctx, generator).into(),
-                begin.into(),
-                end.into(),
-            ],
-            "",
-        )
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap()
-}
-
-/// Generates a call to `__nac3_ndarray_calc_nd_indices`. Returns a [`TypeArrayLikeAdpater`]
-/// containing `i32` indices of the flattened index.
-///
-/// * `index` - The index to compute the multidimensional index for.
-/// * `ndarray` - LLVM pointer to the `NDArray`. This value must be the LLVM representation of an
-///   `NDArray`.
-pub fn call_ndarray_calc_nd_indices<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    index: IntValue<'ctx>,
-    ndarray: NDArrayValue<'ctx>,
-) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
-    let llvm_void = ctx.ctx.void_type();
-    let llvm_i32 = ctx.ctx.i32_type();
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
-    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
-
-    let ndarray_calc_nd_indices_fn_name = match llvm_usize.get_bit_width() {
-        32 => "__nac3_ndarray_calc_nd_indices",
-        64 => "__nac3_ndarray_calc_nd_indices64",
-        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
-    };
-    let ndarray_calc_nd_indices_fn =
-        ctx.module.get_function(ndarray_calc_nd_indices_fn_name).unwrap_or_else(|| {
-            let fn_type = llvm_void.fn_type(
-                &[llvm_usize.into(), llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into()],
-                false,
-            );
-
-            ctx.module.add_function(ndarray_calc_nd_indices_fn_name, fn_type, None)
-        });
-
-    let ndarray_num_dims = ndarray.load_ndims(ctx);
-    let ndarray_dims = ndarray.dim_sizes();
-
-    let indices = ctx.builder.build_array_alloca(llvm_i32, ndarray_num_dims, "").unwrap();
-
-    ctx.builder
-        .build_call(
-            ndarray_calc_nd_indices_fn,
-            &[
-                index.into(),
-                ndarray_dims.base_ptr(ctx, generator).into(),
-                ndarray_num_dims.into(),
-                indices.into(),
-            ],
-            "",
-        )
-        .unwrap();
-
-    TypedArrayLikeAdapter::from(
-        ArraySliceValue::from_ptr_val(indices, ndarray_num_dims, None),
-        Box::new(|_, v| v.into_int_value()),
-        Box::new(|_, v| v.into()),
-    )
-}
-
-fn call_ndarray_flatten_index_impl<'ctx, G, Indices>(
-    generator: &G,
-    ctx: &CodeGenContext<'ctx, '_>,
-    ndarray: NDArrayValue<'ctx>,
-    indices: &Indices,
-) -> IntValue<'ctx>
-where
-    G: CodeGenerator + ?Sized,
-    Indices: ArrayLikeIndexer<'ctx>,
-{
-    let llvm_i32 = ctx.ctx.i32_type();
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-
-    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
-    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
-
-    debug_assert_eq!(
-        IntType::try_from(indices.element_type(ctx, generator))
-            .map(IntType::get_bit_width)
-            .unwrap_or_default(),
-        llvm_i32.get_bit_width(),
-        "Expected i32 value for argument `indices` to `call_ndarray_flatten_index_impl`"
-    );
-    debug_assert_eq!(
-        indices.size(ctx, generator).get_type().get_bit_width(),
-        llvm_usize.get_bit_width(),
-        "Expected usize integer value for argument `indices_size` to `call_ndarray_flatten_index_impl`"
-    );
-
-    let ndarray_flatten_index_fn_name = match llvm_usize.get_bit_width() {
-        32 => "__nac3_ndarray_flatten_index",
-        64 => "__nac3_ndarray_flatten_index64",
-        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
-    };
-    let ndarray_flatten_index_fn =
-        ctx.module.get_function(ndarray_flatten_index_fn_name).unwrap_or_else(|| {
-            let fn_type = llvm_usize.fn_type(
-                &[llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into(), llvm_usize.into()],
-                false,
-            );
-
-            ctx.module.add_function(ndarray_flatten_index_fn_name, fn_type, None)
-        });
-
-    let ndarray_num_dims = ndarray.load_ndims(ctx);
-    let ndarray_dims = ndarray.dim_sizes();
-
-    let index = ctx
-        .builder
-        .build_call(
-            ndarray_flatten_index_fn,
-            &[
-                ndarray_dims.base_ptr(ctx, generator).into(),
-                ndarray_num_dims.into(),
-                indices.base_ptr(ctx, generator).into(),
-                indices.size(ctx, generator).into(),
-            ],
-            "",
-        )
-        .map(CallSiteValue::try_as_basic_value)
-        .map(|v| v.map_left(BasicValueEnum::into_int_value))
-        .map(Either::unwrap_left)
-        .unwrap();
-
-    index
-}
-
-/// Generates a call to `__nac3_ndarray_flatten_index`. Returns the flattened index for the
-/// multidimensional index.
-///
-/// * `ndarray` - LLVM pointer to the `NDArray`. This value must be the LLVM representation of an
-///   `NDArray`.
-/// * `indices` - The multidimensional index to compute the flattened index for.
-pub fn call_ndarray_flatten_index<'ctx, G, Index>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    ndarray: NDArrayValue<'ctx>,
-    indices: &Index,
-) -> IntValue<'ctx>
-where
-    G: CodeGenerator + ?Sized,
-    Index: ArrayLikeIndexer<'ctx>,
-{
-    call_ndarray_flatten_index_impl(generator, ctx, ndarray, indices)
-}
-
-/// Generates a call to `__nac3_ndarray_calc_broadcast`. Returns a tuple containing the number of
-/// dimension and size of each dimension of the resultant `ndarray`.
-pub fn call_ndarray_calc_broadcast<'ctx, G: CodeGenerator + ?Sized>(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    lhs: NDArrayValue<'ctx>,
-    rhs: NDArrayValue<'ctx>,
-) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
-
-    let ndarray_calc_broadcast_fn_name = match llvm_usize.get_bit_width() {
-        32 => "__nac3_ndarray_calc_broadcast",
-        64 => "__nac3_ndarray_calc_broadcast64",
-        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
-    };
-    let ndarray_calc_broadcast_fn =
-        ctx.module.get_function(ndarray_calc_broadcast_fn_name).unwrap_or_else(|| {
-            let fn_type = llvm_usize.fn_type(
-                &[
-                    llvm_pusize.into(),
-                    llvm_usize.into(),
-                    llvm_pusize.into(),
-                    llvm_usize.into(),
-                    llvm_pusize.into(),
-                ],
-                false,
-            );
-
-            ctx.module.add_function(ndarray_calc_broadcast_fn_name, fn_type, None)
-        });
-
-    let lhs_ndims = lhs.load_ndims(ctx);
-    let rhs_ndims = rhs.load_ndims(ctx);
-    let min_ndims = llvm_intrinsics::call_int_umin(ctx, lhs_ndims, rhs_ndims, None);
-
-    gen_for_callback_incrementing(
-        generator,
-        ctx,
-        None,
-        llvm_usize.const_zero(),
-        (min_ndims, false),
-        |generator, ctx, _, idx| {
-            let idx = ctx.builder.build_int_sub(min_ndims, idx, "").unwrap();
-            let (lhs_dim_sz, rhs_dim_sz) = unsafe {
-                (
-                    lhs.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None),
-                    rhs.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None),
-                )
-            };
-
-            let llvm_usize_const_one = llvm_usize.const_int(1, false);
-            let lhs_eqz = ctx
-                .builder
-                .build_int_compare(IntPredicate::EQ, lhs_dim_sz, llvm_usize_const_one, "")
-                .unwrap();
-            let rhs_eqz = ctx
-                .builder
-                .build_int_compare(IntPredicate::EQ, rhs_dim_sz, llvm_usize_const_one, "")
-                .unwrap();
-            let lhs_or_rhs_eqz = ctx.builder.build_or(lhs_eqz, rhs_eqz, "").unwrap();
-
-            let lhs_eq_rhs = ctx
-                .builder
-                .build_int_compare(IntPredicate::EQ, lhs_dim_sz, rhs_dim_sz, "")
-                .unwrap();
-
-            let is_compatible = ctx.builder.build_or(lhs_or_rhs_eqz, lhs_eq_rhs, "").unwrap();
-
-            ctx.make_assert(
-                generator,
-                is_compatible,
-                "0:ValueError",
-                "operands could not be broadcast together",
-                [None, None, None],
-                ctx.current_loc,
-            );
-
-            Ok(())
-        },
-        llvm_usize.const_int(1, false),
-    )
-    .unwrap();
-
-    let max_ndims = llvm_intrinsics::call_int_umax(ctx, lhs_ndims, rhs_ndims, None);
-    let lhs_dims = lhs.dim_sizes().base_ptr(ctx, generator);
-    let lhs_ndims = lhs.load_ndims(ctx);
-    let rhs_dims = rhs.dim_sizes().base_ptr(ctx, generator);
-    let rhs_ndims = rhs.load_ndims(ctx);
-    let out_dims = ctx.builder.build_array_alloca(llvm_usize, max_ndims, "").unwrap();
-    let out_dims = ArraySliceValue::from_ptr_val(out_dims, max_ndims, None);
-
-    ctx.builder
-        .build_call(
-            ndarray_calc_broadcast_fn,
-            &[
-                lhs_dims.into(),
-                lhs_ndims.into(),
-                rhs_dims.into(),
-                rhs_ndims.into(),
-                out_dims.base_ptr(ctx, generator).into(),
-            ],
-            "",
-        )
-        .unwrap();
-
-    TypedArrayLikeAdapter::from(
-        out_dims,
-        Box::new(|_, v| v.into_int_value()),
-        Box::new(|_, v| v.into()),
-    )
-}
-
-/// Generates a call to `__nac3_ndarray_calc_broadcast_idx`. Returns an [`ArrayAllocaValue`]
-/// containing the indices used for accessing `array` corresponding to the index of the broadcasted
-/// array `broadcast_idx`.
-pub fn call_ndarray_calc_broadcast_index<
-    'ctx,
-    G: CodeGenerator + ?Sized,
-    BroadcastIdx: UntypedArrayLikeAccessor<'ctx>,
->(
-    generator: &mut G,
-    ctx: &mut CodeGenContext<'ctx, '_>,
-    array: NDArrayValue<'ctx>,
-    broadcast_idx: &BroadcastIdx,
-) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
-    let llvm_i32 = ctx.ctx.i32_type();
-    let llvm_usize = generator.get_size_type(ctx.ctx);
-    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
-    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
-
-    let ndarray_calc_broadcast_fn_name = match llvm_usize.get_bit_width() {
-        32 => "__nac3_ndarray_calc_broadcast_idx",
-        64 => "__nac3_ndarray_calc_broadcast_idx64",
-        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
-    };
-    let ndarray_calc_broadcast_fn =
-        ctx.module.get_function(ndarray_calc_broadcast_fn_name).unwrap_or_else(|| {
-            let fn_type = llvm_usize.fn_type(
-                &[llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into(), llvm_pi32.into()],
-                false,
-            );
-
-            ctx.module.add_function(ndarray_calc_broadcast_fn_name, fn_type, None)
-        });
-
-    let broadcast_size = broadcast_idx.size(ctx, generator);
-    let out_idx = ctx.builder.build_array_alloca(llvm_i32, broadcast_size, "").unwrap();
-
-    let array_dims = array.dim_sizes().base_ptr(ctx, generator);
-    let array_ndims = array.load_ndims(ctx);
-    let broadcast_idx_ptr = unsafe {
-        broadcast_idx.ptr_offset_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
-    };
-
-    ctx.builder
-        .build_call(
-            ndarray_calc_broadcast_fn,
-            &[array_dims.into(), array_ndims.into(), broadcast_idx_ptr.into(), out_idx.into()],
-            "",
-        )
-        .unwrap();
-
-    TypedArrayLikeAdapter::from(
-        ArraySliceValue::from_ptr_val(out_idx, broadcast_size, None),
-        Box::new(|_, v| v.into_int_value()),
-        Box::new(|_, v| v.into()),
-    )
-}

nac3core/src/codegen/irrt/ndarray.rs

@@ -0,0 +1,384 @@
+use inkwell::{
+    types::IntType,
+    values::{BasicValueEnum, CallSiteValue, IntValue},
+    AddressSpace, IntPredicate,
+};
+use itertools::Either;
+
+use crate::codegen::{
+    llvm_intrinsics,
+    macros::codegen_unreachable,
+    stmt::gen_for_callback_incrementing,
+    values::{
+        ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, NDArrayValue, TypedArrayLikeAccessor,
+        TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
+    },
+    CodeGenContext, CodeGenerator,
+};
+
+/// Generates a call to `__nac3_ndarray_calc_size`. Returns an [`IntValue`] representing the
+/// calculated total size.
+///
+/// * `dims` - An [`ArrayLikeIndexer`] containing the size of each dimension.
+/// * `range` - The dimension index to begin and end (exclusively) calculating the dimensions for,
+///   or [`None`] if starting from the first dimension and ending at the last dimension
+///   respectively.
+pub fn call_ndarray_calc_size<'ctx, G, Dims>(
+    generator: &G,
+    ctx: &CodeGenContext<'ctx, '_>,
+    dims: &Dims,
+    (begin, end): (Option<IntValue<'ctx>>, Option<IntValue<'ctx>>),
+) -> IntValue<'ctx>
+where
+    G: CodeGenerator + ?Sized,
+    Dims: ArrayLikeIndexer<'ctx>,
+{
+    let llvm_usize = generator.get_size_type(ctx.ctx);
+    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
+
+    let ndarray_calc_size_fn_name = match llvm_usize.get_bit_width() {
+        32 => "__nac3_ndarray_calc_size",
+        64 => "__nac3_ndarray_calc_size64",
+        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
+    };
+    let ndarray_calc_size_fn_t = llvm_usize.fn_type(
+        &[llvm_pusize.into(), llvm_usize.into(), llvm_usize.into(), llvm_usize.into()],
+        false,
+    );
+    let ndarray_calc_size_fn =
+        ctx.module.get_function(ndarray_calc_size_fn_name).unwrap_or_else(|| {
+            ctx.module.add_function(ndarray_calc_size_fn_name, ndarray_calc_size_fn_t, None)
+        });
+
+    let begin = begin.unwrap_or_else(|| llvm_usize.const_zero());
+    let end = end.unwrap_or_else(|| dims.size(ctx, generator));
+    ctx.builder
+        .build_call(
+            ndarray_calc_size_fn,
+            &[
+                dims.base_ptr(ctx, generator).into(),
+                dims.size(ctx, generator).into(),
+                begin.into(),
+                end.into(),
+            ],
+            "",
+        )
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}
+
+/// Generates a call to `__nac3_ndarray_calc_nd_indices`. Returns a [`TypeArrayLikeAdpater`]
+/// containing `i32` indices of the flattened index.
+///
+/// * `index` - The index to compute the multidimensional index for.
+/// * `ndarray` - LLVM pointer to the `NDArray`. This value must be the LLVM representation of an
+///   `NDArray`.
+pub fn call_ndarray_calc_nd_indices<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    index: IntValue<'ctx>,
+    ndarray: NDArrayValue<'ctx>,
+) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
+    let llvm_void = ctx.ctx.void_type();
+    let llvm_i32 = ctx.ctx.i32_type();
+    let llvm_usize = generator.get_size_type(ctx.ctx);
+    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
+    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
+
+    let ndarray_calc_nd_indices_fn_name = match llvm_usize.get_bit_width() {
+        32 => "__nac3_ndarray_calc_nd_indices",
+        64 => "__nac3_ndarray_calc_nd_indices64",
+        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
+    };
+    let ndarray_calc_nd_indices_fn =
+        ctx.module.get_function(ndarray_calc_nd_indices_fn_name).unwrap_or_else(|| {
+            let fn_type = llvm_void.fn_type(
+                &[llvm_usize.into(), llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into()],
+                false,
+            );
+
+            ctx.module.add_function(ndarray_calc_nd_indices_fn_name, fn_type, None)
+        });
+
+    let ndarray_num_dims = ndarray.load_ndims(ctx);
+    let ndarray_dims = ndarray.shape();
+
+    let indices = ctx.builder.build_array_alloca(llvm_i32, ndarray_num_dims, "").unwrap();
+
+    ctx.builder
+        .build_call(
+            ndarray_calc_nd_indices_fn,
+            &[
+                index.into(),
+                ndarray_dims.base_ptr(ctx, generator).into(),
+                ndarray_num_dims.into(),
+                indices.into(),
+            ],
+            "",
+        )
+        .unwrap();
+
+    TypedArrayLikeAdapter::from(
+        ArraySliceValue::from_ptr_val(indices, ndarray_num_dims, None),
+        Box::new(|_, v| v.into_int_value()),
+        Box::new(|_, v| v.into()),
+    )
+}
+
+fn call_ndarray_flatten_index_impl<'ctx, G, Indices>(
+    generator: &G,
+    ctx: &CodeGenContext<'ctx, '_>,
+    ndarray: NDArrayValue<'ctx>,
+    indices: &Indices,
+) -> IntValue<'ctx>
+where
+    G: CodeGenerator + ?Sized,
+    Indices: ArrayLikeIndexer<'ctx>,
+{
+    let llvm_i32 = ctx.ctx.i32_type();
+    let llvm_usize = generator.get_size_type(ctx.ctx);
+
+    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
+    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
+
+    debug_assert_eq!(
+        IntType::try_from(indices.element_type(ctx, generator))
+            .map(IntType::get_bit_width)
+            .unwrap_or_default(),
+        llvm_i32.get_bit_width(),
+        "Expected i32 value for argument `indices` to `call_ndarray_flatten_index_impl`"
+    );
+    debug_assert_eq!(
+        indices.size(ctx, generator).get_type().get_bit_width(),
+        llvm_usize.get_bit_width(),
+        "Expected usize integer value for argument `indices_size` to `call_ndarray_flatten_index_impl`"
+    );
+
+    let ndarray_flatten_index_fn_name = match llvm_usize.get_bit_width() {
+        32 => "__nac3_ndarray_flatten_index",
+        64 => "__nac3_ndarray_flatten_index64",
+        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
+    };
+    let ndarray_flatten_index_fn =
+        ctx.module.get_function(ndarray_flatten_index_fn_name).unwrap_or_else(|| {
+            let fn_type = llvm_usize.fn_type(
+                &[llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into(), llvm_usize.into()],
+                false,
+            );
+
+            ctx.module.add_function(ndarray_flatten_index_fn_name, fn_type, None)
+        });
+
+    let ndarray_num_dims = ndarray.load_ndims(ctx);
+    let ndarray_dims = ndarray.shape();
+
+    let index = ctx
+        .builder
+        .build_call(
+            ndarray_flatten_index_fn,
+            &[
+                ndarray_dims.base_ptr(ctx, generator).into(),
+                ndarray_num_dims.into(),
+                indices.base_ptr(ctx, generator).into(),
+                indices.size(ctx, generator).into(),
+            ],
+            "",
+        )
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap();
+
+    index
+}
+
+/// Generates a call to `__nac3_ndarray_flatten_index`. Returns the flattened index for the
+/// multidimensional index.
+///
+/// * `ndarray` - LLVM pointer to the `NDArray`. This value must be the LLVM representation of an
+///   `NDArray`.
+/// * `indices` - The multidimensional index to compute the flattened index for.
+pub fn call_ndarray_flatten_index<'ctx, G, Index>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    ndarray: NDArrayValue<'ctx>,
+    indices: &Index,
+) -> IntValue<'ctx>
+where
+    G: CodeGenerator + ?Sized,
+    Index: ArrayLikeIndexer<'ctx>,
+{
+    call_ndarray_flatten_index_impl(generator, ctx, ndarray, indices)
+}
+
+/// Generates a call to `__nac3_ndarray_calc_broadcast`. Returns a tuple containing the number of
+/// dimension and size of each dimension of the resultant `ndarray`.
+pub fn call_ndarray_calc_broadcast<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    lhs: NDArrayValue<'ctx>,
+    rhs: NDArrayValue<'ctx>,
+) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
+    let llvm_usize = generator.get_size_type(ctx.ctx);
+    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
+
+    let ndarray_calc_broadcast_fn_name = match llvm_usize.get_bit_width() {
+        32 => "__nac3_ndarray_calc_broadcast",
+        64 => "__nac3_ndarray_calc_broadcast64",
+        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
+    };
+    let ndarray_calc_broadcast_fn =
+        ctx.module.get_function(ndarray_calc_broadcast_fn_name).unwrap_or_else(|| {
+            let fn_type = llvm_usize.fn_type(
+                &[
+                    llvm_pusize.into(),
+                    llvm_usize.into(),
+                    llvm_pusize.into(),
+                    llvm_usize.into(),
+                    llvm_pusize.into(),
+                ],
+                false,
+            );
+
+            ctx.module.add_function(ndarray_calc_broadcast_fn_name, fn_type, None)
+        });
+
+    let lhs_ndims = lhs.load_ndims(ctx);
+    let rhs_ndims = rhs.load_ndims(ctx);
+    let min_ndims = llvm_intrinsics::call_int_umin(ctx, lhs_ndims, rhs_ndims, None);
+
+    gen_for_callback_incrementing(
+        generator,
+        ctx,
+        None,
+        llvm_usize.const_zero(),
+        (min_ndims, false),
+        |generator, ctx, _, idx| {
+            let idx = ctx.builder.build_int_sub(min_ndims, idx, "").unwrap();
+            let (lhs_dim_sz, rhs_dim_sz) = unsafe {
+                (
+                    lhs.shape().get_typed_unchecked(ctx, generator, &idx, None),
+                    rhs.shape().get_typed_unchecked(ctx, generator, &idx, None),
+                )
+            };
+
+            let llvm_usize_const_one = llvm_usize.const_int(1, false);
+            let lhs_eqz = ctx
+                .builder
+                .build_int_compare(IntPredicate::EQ, lhs_dim_sz, llvm_usize_const_one, "")
+                .unwrap();
+            let rhs_eqz = ctx
+                .builder
+                .build_int_compare(IntPredicate::EQ, rhs_dim_sz, llvm_usize_const_one, "")
+                .unwrap();
+            let lhs_or_rhs_eqz = ctx.builder.build_or(lhs_eqz, rhs_eqz, "").unwrap();
+
+            let lhs_eq_rhs = ctx
+                .builder
+                .build_int_compare(IntPredicate::EQ, lhs_dim_sz, rhs_dim_sz, "")
+                .unwrap();
+
+            let is_compatible = ctx.builder.build_or(lhs_or_rhs_eqz, lhs_eq_rhs, "").unwrap();
+
+            ctx.make_assert(
+                generator,
+                is_compatible,
+                "0:ValueError",
+                "operands could not be broadcast together",
+                [None, None, None],
+                ctx.current_loc,
+            );
+
+            Ok(())
+        },
+        llvm_usize.const_int(1, false),
+    )
+    .unwrap();
+
+    let max_ndims = llvm_intrinsics::call_int_umax(ctx, lhs_ndims, rhs_ndims, None);
+    let lhs_dims = lhs.shape().base_ptr(ctx, generator);
+    let lhs_ndims = lhs.load_ndims(ctx);
+    let rhs_dims = rhs.shape().base_ptr(ctx, generator);
+    let rhs_ndims = rhs.load_ndims(ctx);
+    let out_dims = ctx.builder.build_array_alloca(llvm_usize, max_ndims, "").unwrap();
+    let out_dims = ArraySliceValue::from_ptr_val(out_dims, max_ndims, None);
+
+    ctx.builder
+        .build_call(
+            ndarray_calc_broadcast_fn,
+            &[
+                lhs_dims.into(),
+                lhs_ndims.into(),
+                rhs_dims.into(),
+                rhs_ndims.into(),
+                out_dims.base_ptr(ctx, generator).into(),
+            ],
+            "",
+        )
+        .unwrap();
+
+    TypedArrayLikeAdapter::from(
+        out_dims,
+        Box::new(|_, v| v.into_int_value()),
+        Box::new(|_, v| v.into()),
+    )
+}
+
+/// Generates a call to `__nac3_ndarray_calc_broadcast_idx`. Returns an [`ArrayAllocaValue`]
+/// containing the indices used for accessing `array` corresponding to the index of the broadcasted
+/// array `broadcast_idx`.
+pub fn call_ndarray_calc_broadcast_index<
+    'ctx,
+    G: CodeGenerator + ?Sized,
+    BroadcastIdx: UntypedArrayLikeAccessor<'ctx>,
+>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    array: NDArrayValue<'ctx>,
+    broadcast_idx: &BroadcastIdx,
+) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
+    let llvm_i32 = ctx.ctx.i32_type();
+    let llvm_usize = generator.get_size_type(ctx.ctx);
+    let llvm_pi32 = llvm_i32.ptr_type(AddressSpace::default());
+    let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
+
+    let ndarray_calc_broadcast_fn_name = match llvm_usize.get_bit_width() {
+        32 => "__nac3_ndarray_calc_broadcast_idx",
+        64 => "__nac3_ndarray_calc_broadcast_idx64",
+        bw => codegen_unreachable!(ctx, "Unsupported size type bit width: {}", bw),
+    };
+    let ndarray_calc_broadcast_fn =
+        ctx.module.get_function(ndarray_calc_broadcast_fn_name).unwrap_or_else(|| {
+            let fn_type = llvm_usize.fn_type(
+                &[llvm_pusize.into(), llvm_usize.into(), llvm_pi32.into(), llvm_pi32.into()],
+                false,
+            );
+
+            ctx.module.add_function(ndarray_calc_broadcast_fn_name, fn_type, None)
+        });
+
+    let broadcast_size = broadcast_idx.size(ctx, generator);
+    let out_idx = ctx.builder.build_array_alloca(llvm_i32, broadcast_size, "").unwrap();
+
+    let array_dims = array.shape().base_ptr(ctx, generator);
+    let array_ndims = array.load_ndims(ctx);
+    let broadcast_idx_ptr = unsafe {
+        broadcast_idx.ptr_offset_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+    };
+
+    ctx.builder
+        .build_call(
+            ndarray_calc_broadcast_fn,
+            &[array_dims.into(), array_ndims.into(), broadcast_idx_ptr.into(), out_idx.into()],
+            "",
+        )
+        .unwrap();
+
+    TypedArrayLikeAdapter::from(
+        ArraySliceValue::from_ptr_val(out_idx, broadcast_size, None),
+        Box::new(|_, v| v.into_int_value()),
+        Box::new(|_, v| v.into()),
+    )
+}

nac3core/src/codegen/irrt/slice.rs

@@ -0,0 +1,76 @@
+use inkwell::{
+    values::{BasicValueEnum, CallSiteValue, IntValue},
+    IntPredicate,
+};
+use itertools::Either;
+use nac3parser::ast::Expr;
+
+use crate::{
+    codegen::{CodeGenContext, CodeGenerator},
+    typecheck::typedef::Type,
+};
+
+/// this function allows index out of range, since python
+/// allows index out of range in slice (`a = [1,2,3]; a[1:10] == [2,3]`).
+pub fn handle_slice_index_bound<'ctx, G: CodeGenerator>(
+    i: &Expr<Option<Type>>,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    generator: &mut G,
+    length: IntValue<'ctx>,
+) -> Result<Option<IntValue<'ctx>>, String> {
+    const SYMBOL: &str = "__nac3_slice_index_bound";
+    let func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
+        let i32_t = ctx.ctx.i32_type();
+        let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into()], false);
+        ctx.module.add_function(SYMBOL, fn_t, None)
+    });
+
+    let i = if let Some(v) = generator.gen_expr(ctx, i)? {
+        v.to_basic_value_enum(ctx, generator, i.custom.unwrap())?
+    } else {
+        return Ok(None);
+    };
+    Ok(Some(
+        ctx.builder
+            .build_call(func, &[i.into(), length.into()], "bounded_ind")
+            .map(CallSiteValue::try_as_basic_value)
+            .map(|v| v.map_left(BasicValueEnum::into_int_value))
+            .map(Either::unwrap_left)
+            .unwrap(),
+    ))
+}
+
+pub fn calculate_len_for_slice_range<'ctx, G: CodeGenerator + ?Sized>(
+    generator: &mut G,
+    ctx: &mut CodeGenContext<'ctx, '_>,
+    start: IntValue<'ctx>,
+    end: IntValue<'ctx>,
+    step: IntValue<'ctx>,
+) -> IntValue<'ctx> {
+    const SYMBOL: &str = "__nac3_range_slice_len";
+    let len_func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
+        let i32_t = ctx.ctx.i32_type();
+        let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into(), i32_t.into()], false);
+        ctx.module.add_function(SYMBOL, fn_t, None)
+    });
+
+    // assert step != 0, throw exception if not
+    let not_zero = ctx
+        .builder
+        .build_int_compare(IntPredicate::NE, step, step.get_type().const_zero(), "range_step_ne")
+        .unwrap();
+    ctx.make_assert(
+        generator,
+        not_zero,
+        "0:ValueError",
+        "step must not be zero",
+        [None, None, None],
+        ctx.current_loc,
+    );
+    ctx.builder
+        .build_call(len_func, &[start.into(), end.into(), step.into()], "calc_len")
+        .map(CallSiteValue::try_as_basic_value)
+        .map(|v| v.map_left(BasicValueEnum::into_int_value))
+        .map(Either::unwrap_left)
+        .unwrap()
+}

nac3core/src/codegen/numpy.rs

@@ -26,7 +26,7 @@ use super::{
 use crate::{
     symbol_resolver::ValueEnum,
     toplevel::{
-        helper::PrimDef,
+        helper::{arraylike_flatten_element_type, PrimDef},
         numpy::{make_ndarray_ty, unpack_ndarray_var_tys},
         DefinitionId,
     },
@@ -42,6 +42,7 @@ fn create_ndarray_uninitialized<'ctx, G: CodeGenerator + ?Sized>(
     ctx: &mut CodeGenContext<'ctx, '_>,
     elem_ty: Type,
 ) -> Result<NDArrayValue<'ctx>, String> {
+    let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
     let ndarray_ty = make_ndarray_ty(&mut ctx.unifier, &ctx.primitives, Some(elem_ty), None);
 
     let llvm_usize = generator.get_size_type(ctx.ctx);
@@ -54,7 +55,7 @@ fn create_ndarray_uninitialized<'ctx, G: CodeGenerator + ?Sized>(
 
     let ndarray = generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
 
-    Ok(NDArrayValue::from_pointer_value(ndarray, llvm_usize, None))
+    Ok(NDArrayValue::from_pointer_value(ndarray, llvm_elem_ty, llvm_usize, None))
 }
 
 /// Creates an `NDArray` instance from a dynamic shape.
@@ -127,7 +128,7 @@ where
     ndarray.store_ndims(ctx, generator, num_dims);
 
     let ndarray_num_dims = ndarray.load_ndims(ctx);
-    ndarray.create_dim_sizes(ctx, llvm_usize, ndarray_num_dims);
+    ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
 
     // Copy the dimension sizes from shape to ndarray.dims
     let shape_len = shape_len_fn(generator, ctx, shape)?;
@@ -143,7 +144,7 @@ where
             let shape_dim = ctx.builder.build_int_z_extend(shape_dim, llvm_usize, "").unwrap();
 
             let ndarray_pdim =
-                unsafe { ndarray.dim_sizes().ptr_offset_unchecked(ctx, generator, &i, None) };
+                unsafe { ndarray.shape().ptr_offset_unchecked(ctx, generator, &i, None) };
 
             ctx.builder.build_store(ndarray_pdim, shape_dim).unwrap();
 
@@ -194,12 +195,12 @@ pub fn create_ndarray_const_shape<'ctx, G: CodeGenerator + ?Sized>(
     ndarray.store_ndims(ctx, generator, num_dims);
 
     let ndarray_num_dims = ndarray.load_ndims(ctx);
-    ndarray.create_dim_sizes(ctx, llvm_usize, ndarray_num_dims);
+    ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
 
     for (i, &shape_dim) in shape.iter().enumerate() {
         let shape_dim = ctx.builder.build_int_z_extend(shape_dim, llvm_usize, "").unwrap();
         let ndarray_dim = unsafe {
-            ndarray.dim_sizes().ptr_offset_unchecked(
+            ndarray.shape().ptr_offset_unchecked(
                 ctx,
                 generator,
                 &llvm_usize.const_int(i as u64, true),
@@ -228,7 +229,7 @@ fn ndarray_init_data<'ctx, G: CodeGenerator + ?Sized>(
     let ndarray_num_elems = call_ndarray_calc_size(
         generator,
         ctx,
-        &ndarray.dim_sizes().as_slice_value(ctx, generator),
+        &ndarray.shape().as_slice_value(ctx, generator),
         (None, None),
     );
     ndarray.create_data(ctx, llvm_ndarray_data_t, ndarray_num_elems);
@@ -379,7 +380,7 @@ where
     let ndarray_num_elems = call_ndarray_calc_size(
         generator,
         ctx,
-        &ndarray.dim_sizes().as_slice_value(ctx, generator),
+        &ndarray.shape().as_slice_value(ctx, generator),
         (None, None),
     );
 
@@ -473,8 +474,8 @@ fn ndarray_broadcast_fill<'ctx, 'a, G, ValueFn>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, 'a>,
     res: NDArrayValue<'ctx>,
-    lhs: (BasicValueEnum<'ctx>, bool),
+    lhs: (Type, BasicValueEnum<'ctx>, bool),
-    rhs: (BasicValueEnum<'ctx>, bool),
+    rhs: (Type, BasicValueEnum<'ctx>, bool),
     value_fn: ValueFn,
 ) -> Result<NDArrayValue<'ctx>, String>
 where
@@ -487,8 +488,8 @@ where
 {
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
-    let (lhs_val, lhs_scalar) = lhs;
+    let (lhs_ty, lhs_val, lhs_scalar) = lhs;
-    let (rhs_val, rhs_scalar) = rhs;
+    let (rhs_ty, rhs_val, rhs_scalar) = rhs;
 
     assert!(
         !(lhs_scalar && rhs_scalar),
@@ -499,14 +500,26 @@ where
 
     // Assert that all ndarray operands are broadcastable to the target size
     if !lhs_scalar {
-        let lhs_val =
+        let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
-            NDArrayValue::from_pointer_value(lhs_val.into_pointer_value(), llvm_usize, None);
+        let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
+        let lhs_val = NDArrayValue::from_pointer_value(
+            lhs_val.into_pointer_value(),
+            llvm_lhs_elem_ty,
+            llvm_usize,
+            None,
+        );
         ndarray_assert_is_broadcastable(generator, ctx, res, lhs_val);
     }
 
     if !rhs_scalar {
-        let rhs_val =
+        let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
-            NDArrayValue::from_pointer_value(rhs_val.into_pointer_value(), llvm_usize, None);
+        let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
+        let rhs_val = NDArrayValue::from_pointer_value(
+            rhs_val.into_pointer_value(),
+            llvm_rhs_elem_ty,
+            llvm_usize,
+            None,
+        );
         ndarray_assert_is_broadcastable(generator, ctx, res, rhs_val);
     }
 
@@ -514,8 +527,14 @@ where
         let lhs_elem = if lhs_scalar {
             lhs_val
         } else {
-            let lhs =
+            let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
-                NDArrayValue::from_pointer_value(lhs_val.into_pointer_value(), llvm_usize, None);
+            let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
+            let lhs = NDArrayValue::from_pointer_value(
+                lhs_val.into_pointer_value(),
+                llvm_lhs_elem_ty,
+                llvm_usize,
+                None,
+            );
             let lhs_idx = call_ndarray_calc_broadcast_index(generator, ctx, lhs, idx);
 
             unsafe { lhs.data().get_unchecked(ctx, generator, &lhs_idx, None) }
@@ -524,8 +543,14 @@ where
         let rhs_elem = if rhs_scalar {
             rhs_val
         } else {
-            let rhs =
+            let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
-                NDArrayValue::from_pointer_value(rhs_val.into_pointer_value(), llvm_usize, None);
+            let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
+            let rhs = NDArrayValue::from_pointer_value(
+                rhs_val.into_pointer_value(),
+                llvm_rhs_elem_ty,
+                llvm_usize,
+                None,
+            );
             let rhs_idx = call_ndarray_calc_broadcast_index(generator, ctx, rhs, idx);
 
             unsafe { rhs.data().get_unchecked(ctx, generator, &rhs_idx, None) }
@@ -671,7 +696,7 @@ fn llvm_ndlist_get_ndims<'ctx, G: CodeGenerator + ?Sized>(
 fn llvm_arraylike_get_ndims<'ctx, G: CodeGenerator + ?Sized>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, '_>,
-    value: BasicValueEnum<'ctx>,
+    (ty, value): (Type, BasicValueEnum<'ctx>),
 ) -> IntValue<'ctx> {
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
@@ -679,7 +704,9 @@ fn llvm_arraylike_get_ndims<'ctx, G: CodeGenerator + ?Sized>(
         BasicValueEnum::PointerValue(v)
             if NDArrayValue::is_representable(v, llvm_usize).is_ok() =>
         {
-            NDArrayValue::from_pointer_value(v, llvm_usize, None).load_ndims(ctx)
+            let dtype = arraylike_flatten_element_type(&mut ctx.unifier, ty);
+            let llvm_elem_ty = ctx.get_llvm_type(generator, dtype);
+            NDArrayValue::from_pointer_value(v, llvm_elem_ty, llvm_usize, None).load_ndims(ctx)
         }
 
         BasicValueEnum::PointerValue(v) if ListValue::is_representable(v, llvm_usize).is_ok() => {
@@ -694,7 +721,6 @@ fn llvm_arraylike_get_ndims<'ctx, G: CodeGenerator + ?Sized>(
 fn ndarray_from_ndlist_impl<'ctx, G: CodeGenerator + ?Sized>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, '_>,
-    elem_ty: Type,
     (dst_arr, dst_slice_ptr): (NDArrayValue<'ctx>, PointerValue<'ctx>),
     src_lst: ListValue<'ctx>,
     dim: u64,
@@ -713,7 +739,7 @@ fn ndarray_from_ndlist_impl<'ctx, G: CodeGenerator + ?Sized>(
             let stride = call_ndarray_calc_size(
                 generator,
                 ctx,
-                &dst_arr.dim_sizes(),
+                &dst_arr.shape(),
                 (Some(llvm_usize.const_int(dim + 1, false)), None),
             );
 
@@ -727,6 +753,20 @@ fn ndarray_from_ndlist_impl<'ctx, G: CodeGenerator + ?Sized>(
                 |_, _| Ok(llvm_usize.const_int(1, false)),
                 |generator, ctx, _, i| {
                     let offset = ctx.builder.build_int_mul(stride, i, "").unwrap();
+                    let offset = ctx
+                        .builder
+                        .build_int_mul(
+                            offset,
+                            ctx.builder
+                                .build_int_truncate_or_bit_cast(
+                                    dst_arr.get_type().element_type().size_of().unwrap(),
+                                    offset.get_type(),
+                                    "",
+                                )
+                                .unwrap(),
+                            "",
+                        )
+                        .unwrap();
 
                     let dst_ptr =
                         unsafe { ctx.builder.build_gep(dst_slice_ptr, &[offset], "").unwrap() };
@@ -741,7 +781,6 @@ fn ndarray_from_ndlist_impl<'ctx, G: CodeGenerator + ?Sized>(
                     ndarray_from_ndlist_impl(
                         generator,
                         ctx,
-                        elem_ty,
                         (dst_arr, dst_ptr),
                         nested_lst_elem,
                         dim + 1,
@@ -760,7 +799,7 @@ fn ndarray_from_ndlist_impl<'ctx, G: CodeGenerator + ?Sized>(
 
         _ => {
             let lst_len = src_lst.load_size(ctx, None);
-            let sizeof_elem = ctx.get_llvm_type(generator, elem_ty).size_of().unwrap();
+            let sizeof_elem = dst_arr.get_type().element_type().size_of().unwrap();
             let sizeof_elem = ctx.builder.build_int_cast(sizeof_elem, llvm_usize, "").unwrap();
 
             let cpy_len = ctx
@@ -816,7 +855,8 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
 
     // object is an NDArray instance - copy object unless copy=0 && ndmin < object.ndims
     if NDArrayValue::is_representable(object, llvm_usize).is_ok() {
-        let object = NDArrayValue::from_pointer_value(object, llvm_usize, None);
+        let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
+        let object = NDArrayValue::from_pointer_value(object, llvm_elem_ty, llvm_usize, None);
 
         let ndarray = gen_if_else_expr_callback(
             generator,
@@ -878,7 +918,6 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
                 ndarray_sliced_copyto_impl(
                     generator,
                     ctx,
-                    elem_ty,
                     (ndarray, ndarray.data().base_ptr(ctx, generator)),
                     (object, object.data().base_ptr(ctx, generator)),
                     0,
@@ -892,6 +931,7 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
 
         return Ok(NDArrayValue::from_pointer_value(
             ndarray.map(BasicValueEnum::into_pointer_value).unwrap(),
+            llvm_elem_ty,
             llvm_usize,
             None,
         ));
@@ -1026,7 +1066,6 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
     ndarray_from_ndlist_impl(
         generator,
         ctx,
-        elem_ty,
         (ndarray, ndarray.data().base_ptr(ctx, generator)),
         object,
         0,
@@ -1099,7 +1138,6 @@ fn call_ndarray_eye_impl<'ctx, G: CodeGenerator + ?Sized>(
 fn ndarray_sliced_copyto_impl<'ctx, G: CodeGenerator + ?Sized>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, '_>,
-    elem_ty: Type,
     (dst_arr, dst_slice_ptr): (NDArrayValue<'ctx>, PointerValue<'ctx>),
     (src_arr, src_slice_ptr): (NDArrayValue<'ctx>, PointerValue<'ctx>),
     dim: u64,
@@ -1108,14 +1146,16 @@ fn ndarray_sliced_copyto_impl<'ctx, G: CodeGenerator + ?Sized>(
     let llvm_i1 = ctx.ctx.bool_type();
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
+    assert_eq!(dst_arr.get_type().element_type(), src_arr.get_type().element_type());
+
+    let sizeof_elem = dst_arr.get_type().element_type().size_of().unwrap();
+
     // If there are no (remaining) slice expressions, memcpy the entire dimension
     if slices.is_empty() {
-        let sizeof_elem = ctx.get_llvm_type(generator, elem_ty).size_of().unwrap();
-
         let stride = call_ndarray_calc_size(
             generator,
             ctx,
-            &src_arr.dim_sizes(),
+            &src_arr.shape(),
             (Some(llvm_usize.const_int(dim, false)), None),
         );
         let stride =
@@ -1133,13 +1173,13 @@ fn ndarray_sliced_copyto_impl<'ctx, G: CodeGenerator + ?Sized>(
     let src_stride = call_ndarray_calc_size(
         generator,
         ctx,
-        &src_arr.dim_sizes(),
+        &src_arr.shape(),
         (Some(llvm_usize.const_int(dim + 1, false)), None),
     );
     let dst_stride = call_ndarray_calc_size(
         generator,
         ctx,
-        &dst_arr.dim_sizes(),
+        &dst_arr.shape(),
         (Some(llvm_usize.const_int(dim + 1, false)), None),
     );
 
@@ -1162,9 +1202,29 @@ fn ndarray_sliced_copyto_impl<'ctx, G: CodeGenerator + ?Sized>(
         |generator, ctx, _, src_i| {
             // Calculate the offset of the active slice
             let src_data_offset = ctx.builder.build_int_mul(src_stride, src_i, "").unwrap();
+            let src_data_offset = ctx
+                .builder
+                .build_int_mul(
+                    src_data_offset,
+                    ctx.builder
+                        .build_int_cast(sizeof_elem, src_data_offset.get_type(), "")
+                        .unwrap(),
+                    "",
+                )
+                .unwrap();
             let dst_i =
                 ctx.builder.build_load(dst_i_addr, "").map(BasicValueEnum::into_int_value).unwrap();
             let dst_data_offset = ctx.builder.build_int_mul(dst_stride, dst_i, "").unwrap();
+            let dst_data_offset = ctx
+                .builder
+                .build_int_mul(
+                    dst_data_offset,
+                    ctx.builder
+                        .build_int_cast(sizeof_elem, dst_data_offset.get_type(), "")
+                        .unwrap(),
+                    "",
+                )
+                .unwrap();
 
             let (src_ptr, dst_ptr) = unsafe {
                 (
@@ -1176,7 +1236,6 @@ fn ndarray_sliced_copyto_impl<'ctx, G: CodeGenerator + ?Sized>(
             ndarray_sliced_copyto_impl(
                 generator,
                 ctx,
-                elem_ty,
                 (dst_arr, dst_ptr),
                 (src_arr, src_ptr),
                 dim + 1,
@@ -1219,7 +1278,7 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
             &this,
             |_, ctx, shape| Ok(shape.load_ndims(ctx)),
             |generator, ctx, shape, idx| unsafe {
-                Ok(shape.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None))
+                Ok(shape.shape().get_typed_unchecked(ctx, generator, &idx, None))
             },
         )?
     } else {
@@ -1227,7 +1286,7 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
         ndarray.store_ndims(ctx, generator, this.load_ndims(ctx));
 
         let ndims = this.load_ndims(ctx);
-        ndarray.create_dim_sizes(ctx, llvm_usize, ndims);
+        ndarray.create_shape(ctx, llvm_usize, ndims);
 
         // Populate the first slices.len() dimensions by computing the size of each dim slice
         for (i, (start, stop, step)) in slices.iter().enumerate() {
@@ -1259,7 +1318,7 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
                 ctx.builder.build_int_z_extend_or_bit_cast(slice_len, llvm_usize, "").unwrap();
 
             unsafe {
-                ndarray.dim_sizes().set_typed_unchecked(
+                ndarray.shape().set_typed_unchecked(
                     ctx,
                     generator,
                     &llvm_usize.const_int(i as u64, false),
@@ -1277,8 +1336,8 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
             (this.load_ndims(ctx), false),
             |generator, ctx, _, idx| {
                 unsafe {
-                    let dim_sz = this.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None);
+                    let dim_sz = this.shape().get_typed_unchecked(ctx, generator, &idx, None);
-                    ndarray.dim_sizes().set_typed_unchecked(ctx, generator, &idx, dim_sz);
+                    ndarray.shape().set_typed_unchecked(ctx, generator, &idx, dim_sz);
                 }
 
                 Ok(())
@@ -1293,7 +1352,6 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
     ndarray_sliced_copyto_impl(
         generator,
         ctx,
-        elem_ty,
         (ndarray, ndarray.data().base_ptr(ctx, generator)),
         (this, this.data().base_ptr(ctx, generator)),
         0,
@@ -1339,7 +1397,7 @@ where
             &operand,
             |_, ctx, v| Ok(v.load_ndims(ctx)),
             |generator, ctx, v, idx| unsafe {
-                Ok(v.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None))
+                Ok(v.shape().get_typed_unchecked(ctx, generator, &idx, None))
             },
         )
         .unwrap()
@@ -1376,8 +1434,8 @@ pub fn ndarray_elementwise_binop_impl<'ctx, 'a, G, ValueFn>(
     ctx: &mut CodeGenContext<'ctx, 'a>,
     elem_ty: Type,
     res: Option<NDArrayValue<'ctx>>,
-    lhs: (BasicValueEnum<'ctx>, bool),
+    lhs: (Type, BasicValueEnum<'ctx>, bool),
-    rhs: (BasicValueEnum<'ctx>, bool),
+    rhs: (Type, BasicValueEnum<'ctx>, bool),
     value_fn: ValueFn,
 ) -> Result<NDArrayValue<'ctx>, String>
 where
@@ -1390,8 +1448,8 @@ where
 {
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
-    let (lhs_val, lhs_scalar) = lhs;
+    let (lhs_ty, lhs_val, lhs_scalar) = lhs;
-    let (rhs_val, rhs_scalar) = rhs;
+    let (rhs_ty, rhs_val, rhs_scalar) = rhs;
 
     assert!(
         !(lhs_scalar && rhs_scalar),
@@ -1402,10 +1460,22 @@ where
 
     let ndarray = res.unwrap_or_else(|| {
         if lhs_scalar && rhs_scalar {
-            let lhs_val =
+            let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
-                NDArrayValue::from_pointer_value(lhs_val.into_pointer_value(), llvm_usize, None);
+            let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
-            let rhs_val =
+            let lhs_val = NDArrayValue::from_pointer_value(
-                NDArrayValue::from_pointer_value(rhs_val.into_pointer_value(), llvm_usize, None);
+                lhs_val.into_pointer_value(),
+                llvm_lhs_elem_ty,
+                llvm_usize,
+                None,
+            );
+            let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
+            let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
+            let rhs_val = NDArrayValue::from_pointer_value(
+                rhs_val.into_pointer_value(),
+                llvm_rhs_elem_ty,
+                llvm_usize,
+                None,
+            );
 
             let ndarray_dims = call_ndarray_calc_broadcast(generator, ctx, lhs_val, rhs_val);
 
@@ -1421,8 +1491,14 @@ where
             )
             .unwrap()
         } else {
+            let dtype = arraylike_flatten_element_type(
+                &mut ctx.unifier,
+                if lhs_scalar { rhs_ty } else { lhs_ty },
+            );
+            let llvm_elem_ty = ctx.get_llvm_type(generator, dtype);
             let ndarray = NDArrayValue::from_pointer_value(
                 if lhs_scalar { rhs_val } else { lhs_val }.into_pointer_value(),
+                llvm_elem_ty,
                 llvm_usize,
                 None,
             );
@@ -1434,7 +1510,7 @@ where
                 &ndarray,
                 |_, ctx, v| Ok(v.load_ndims(ctx)),
                 |generator, ctx, v, idx| unsafe {
-                    Ok(v.dim_sizes().get_typed_unchecked(ctx, generator, &idx, None))
+                    Ok(v.shape().get_typed_unchecked(ctx, generator, &idx, None))
                 },
             )
             .unwrap()
@@ -1495,10 +1571,10 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
         if let Some(res) = res {
             let res_ndims = res.load_ndims(ctx);
             let res_dim0 = unsafe {
-                res.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+                res.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
             };
             let res_dim1 = unsafe {
-                res.dim_sizes().get_typed_unchecked(
+                res.shape().get_typed_unchecked(
                     ctx,
                     generator,
                     &llvm_usize.const_int(1, false),
@@ -1506,10 +1582,10 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
                 )
             };
             let lhs_dim0 = unsafe {
-                lhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+                lhs.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
             };
             let rhs_dim1 = unsafe {
-                rhs.dim_sizes().get_typed_unchecked(
+                rhs.shape().get_typed_unchecked(
                     ctx,
                     generator,
                     &llvm_usize.const_int(1, false),
@@ -1558,15 +1634,10 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
 
     if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
         let lhs_dim1 = unsafe {
-            lhs.dim_sizes().get_typed_unchecked(
+            lhs.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_int(1, false), None)
-                ctx,
-                generator,
-                &llvm_usize.const_int(1, false),
-                None,
-            )
         };
         let rhs_dim0 = unsafe {
-            rhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+            rhs.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
         };
 
         // lhs.dims[1] == rhs.dims[0]
@@ -1605,7 +1676,7 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
                     },
                     |generator, ctx| {
                         Ok(Some(unsafe {
-                            lhs.dim_sizes().get_typed_unchecked(
+                            lhs.shape().get_typed_unchecked(
                                 ctx,
                                 generator,
                                 &llvm_usize.const_zero(),
@@ -1615,7 +1686,7 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
                     },
                     |generator, ctx| {
                         Ok(Some(unsafe {
-                            rhs.dim_sizes().get_typed_unchecked(
+                            rhs.shape().get_typed_unchecked(
                                 ctx,
                                 generator,
                                 &llvm_usize.const_int(1, false),
@@ -1642,7 +1713,7 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
 
         let common_dim = {
             let lhs_idx1 = unsafe {
-                lhs.dim_sizes().get_typed_unchecked(
+                lhs.shape().get_typed_unchecked(
                     ctx,
                     generator,
                     &llvm_usize.const_int(1, false),
@@ -1650,7 +1721,7 @@ pub fn ndarray_matmul_2d<'ctx, G: CodeGenerator>(
                 )
             };
             let rhs_idx0 = unsafe {
-                rhs.dim_sizes().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
+                rhs.shape().get_typed_unchecked(ctx, generator, &llvm_usize.const_zero(), None)
             };
 
             let idx = llvm_intrinsics::call_expect(ctx, rhs_idx0, lhs_idx1, None);
@@ -1981,11 +2052,18 @@ pub fn gen_ndarray_copy<'ctx>(
     let this_arg =
         obj.as_ref().unwrap().1.clone().to_basic_value_enum(context, generator, this_ty)?;
 
+    let llvm_elem_ty = context.get_llvm_type(generator, this_elem_ty);
+
     ndarray_copy_impl(
         generator,
         context,
         this_elem_ty,
-        NDArrayValue::from_pointer_value(this_arg.into_pointer_value(), llvm_usize, None),
+        NDArrayValue::from_pointer_value(
+            this_arg.into_pointer_value(),
+            llvm_elem_ty,
+            llvm_usize,
+            None,
+        ),
     )
     .map(NDArrayValue::into)
 }
@@ -2004,6 +2082,7 @@ pub fn gen_ndarray_fill<'ctx>(
     let llvm_usize = generator.get_size_type(context.ctx);
 
     let this_ty = obj.as_ref().unwrap().0;
+    let this_elem_ty = arraylike_flatten_element_type(&mut context.unifier, this_ty);
     let this_arg = obj
         .as_ref()
         .unwrap()
@@ -2014,10 +2093,12 @@ pub fn gen_ndarray_fill<'ctx>(
     let value_ty = fun.0.args[0].ty;
     let value_arg = args[0].1.clone().to_basic_value_enum(context, generator, value_ty)?;
 
+    let llvm_elem_ty = context.get_llvm_type(generator, this_elem_ty);
+
     ndarray_fill_flattened(
         generator,
         context,
-        NDArrayValue::from_pointer_value(this_arg, llvm_usize, None),
+        NDArrayValue::from_pointer_value(this_arg, llvm_elem_ty, llvm_usize, None),
         |generator, ctx, _| {
             let value = if value_arg.is_pointer_value() {
                 let llvm_i1 = ctx.ctx.bool_type();
@@ -2058,8 +2139,9 @@ pub fn ndarray_transpose<'ctx, G: CodeGenerator + ?Sized>(
 
     if let BasicValueEnum::PointerValue(n1) = x1 {
         let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, x1_ty);
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
-        let n_sz = call_ndarray_calc_size(generator, ctx, &n1.dim_sizes(), (None, None));
+        let n1 = NDArrayValue::from_pointer_value(n1, llvm_elem_ty, llvm_usize, None);
+        let n_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
 
         // Dimensions are reversed in the transposed array
         let out = create_ndarray_dyn_shape(
@@ -2074,7 +2156,7 @@ pub fn ndarray_transpose<'ctx, G: CodeGenerator + ?Sized>(
                     .builder
                     .build_int_sub(new_idx, new_idx.get_type().const_int(1, false), "")
                     .unwrap();
-                unsafe { Ok(n.dim_sizes().get_typed_unchecked(ctx, generator, &new_idx, None)) }
+                unsafe { Ok(n.shape().get_typed_unchecked(ctx, generator, &new_idx, None)) }
             },
         )
         .unwrap();
@@ -2111,7 +2193,7 @@ pub fn ndarray_transpose<'ctx, G: CodeGenerator + ?Sized>(
                             .build_int_sub(ndim_rev, llvm_usize.const_int(1, false), "")
                             .unwrap();
                         let dim = unsafe {
-                            n1.dim_sizes().get_typed_unchecked(ctx, generator, &ndim_rev, None)
+                            n1.shape().get_typed_unchecked(ctx, generator, &ndim_rev, None)
                         };
 
                         let rem_idx_val =
@@ -2177,8 +2259,9 @@ pub fn ndarray_reshape<'ctx, G: CodeGenerator + ?Sized>(
 
     if let BasicValueEnum::PointerValue(n1) = x1 {
         let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, x1_ty);
-        let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+        let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
-        let n_sz = call_ndarray_calc_size(generator, ctx, &n1.dim_sizes(), (None, None));
+        let n1 = NDArrayValue::from_pointer_value(n1, llvm_elem_ty, llvm_usize, None);
+        let n_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
 
         let acc = generator.gen_var_alloc(ctx, llvm_usize.into(), None)?;
         let num_neg = generator.gen_var_alloc(ctx, llvm_usize.into(), None)?;
@@ -2406,7 +2489,7 @@ pub fn ndarray_reshape<'ctx, G: CodeGenerator + ?Sized>(
         );
 
         // The new shape must be compatible with the old shape
-        let out_sz = call_ndarray_calc_size(generator, ctx, &out.dim_sizes(), (None, None));
+        let out_sz = call_ndarray_calc_size(generator, ctx, &out.shape(), (None, None));
         ctx.make_assert(
             generator,
             ctx.builder.build_int_compare(IntPredicate::EQ, out_sz, n_sz, "").unwrap(),
@@ -2454,17 +2537,22 @@ pub fn ndarray_dot<'ctx, G: CodeGenerator + ?Sized>(
 ) -> Result<BasicValueEnum<'ctx>, String> {
     const FN_NAME: &str = "ndarray_dot";
     let (x1_ty, x1) = x1;
-    let (_, x2) = x2;
+    let (x2_ty, x2) = x2;
 
     let llvm_usize = generator.get_size_type(ctx.ctx);
 
     match (x1, x2) {
         (BasicValueEnum::PointerValue(n1), BasicValueEnum::PointerValue(n2)) => {
-            let n1 = NDArrayValue::from_pointer_value(n1, llvm_usize, None);
+            let n1_dtype = arraylike_flatten_element_type(&mut ctx.unifier, x1_ty);
-            let n2 = NDArrayValue::from_pointer_value(n2, llvm_usize, None);
+            let n2_dtype = arraylike_flatten_element_type(&mut ctx.unifier, x2_ty);
+            let llvm_n1_data_ty = ctx.get_llvm_type(generator, n1_dtype);
+            let llvm_n2_data_ty = ctx.get_llvm_type(generator, n2_dtype);
 
-            let n1_sz = call_ndarray_calc_size(generator, ctx, &n1.dim_sizes(), (None, None));
+            let n1 = NDArrayValue::from_pointer_value(n1, llvm_n1_data_ty, llvm_usize, None);
-            let n2_sz = call_ndarray_calc_size(generator, ctx, &n1.dim_sizes(), (None, None));
+            let n2 = NDArrayValue::from_pointer_value(n2, llvm_n2_data_ty, llvm_usize, None);
+
+            let n1_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
+            let n2_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
 
             ctx.make_assert(
                 generator,
@@ -2501,7 +2589,7 @@ pub fn ndarray_dot<'ctx, G: CodeGenerator + ?Sized>(
                             .build_float_mul(e1, elem2.into_float_value(), "")
                             .unwrap()
                             .as_basic_value_enum(),
-                        _ => codegen_unreachable!(ctx),
+                        _ => codegen_unreachable!(ctx, "product: {}", elem1.get_type()),
                     };
                     let acc_val = ctx.builder.build_load(acc, "").unwrap();
                     let acc_val = match acc_val {
@@ -2515,7 +2603,7 @@ pub fn ndarray_dot<'ctx, G: CodeGenerator + ?Sized>(
                             .build_float_add(e1, product.into_float_value(), "")
                             .unwrap()
                             .as_basic_value_enum(),
-                        _ => codegen_unreachable!(ctx),
+                        _ => codegen_unreachable!(ctx, "acc_val: {}", acc_val.get_type()),
                     };
                     ctx.builder.build_store(acc, acc_val).unwrap();
 

nac3core/src/codegen/types/ndarray.rs

@@ -1,11 +1,17 @@
 use inkwell::{
     context::Context,
     types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
-    values::IntValue,
+    values::{IntValue, PointerValue},
     AddressSpace,
 };
+use itertools::Itertools;
 
-use super::ProxyType;
+use nac3core_derive::StructFields;
+
+use super::{
+    structure::{StructField, StructFields},
+    ProxyType,
+};
 use crate::codegen::{
     values::{ArraySliceValue, NDArrayValue, ProxyValue},
     {CodeGenContext, CodeGenerator},
@@ -19,6 +25,16 @@ pub struct NDArrayType<'ctx> {
     llvm_usize: IntType<'ctx>,
 }
 
+#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
+pub struct NDArrayStructFields<'ctx> {
+    #[value_type(usize)]
+    pub ndims: StructField<'ctx, IntValue<'ctx>>,
+    #[value_type(usize.ptr_type(AddressSpace::default()))]
+    pub shape: StructField<'ctx, PointerValue<'ctx>>,
+    #[value_type(i8_type().ptr_type(AddressSpace::default()))]
+    pub data: StructField<'ctx, PointerValue<'ctx>>,
+}
+
 impl<'ctx> NDArrayType<'ctx> {
     /// Checks whether `llvm_ty` represents a `ndarray` type, returning [Err] if it does not.
     pub fn is_representable(
@@ -67,35 +83,58 @@ impl<'ctx> NDArrayType<'ctx> {
         }
 
         let ndarray_data_ty = llvm_ndarray_ty.get_field_type_at_index(2).unwrap();
-        let Ok(_) = PointerType::try_from(ndarray_data_ty) else {
+        let Ok(ndarray_pdata) = PointerType::try_from(ndarray_data_ty) else {
             return Err(format!("Expected pointer type for `ndarray.2`, got {ndarray_data_ty}"));
         };
+        let ndarray_data = ndarray_pdata.get_element_type();
+        let Ok(ndarray_data) = IntType::try_from(ndarray_data) else {
+            return Err(format!(
+                "Expected pointer-to-int type for `ndarray.2`, got pointer-to-{ndarray_data}"
+            ));
+        };
+        if ndarray_data.get_bit_width() != 8 {
+            return Err(format!(
+                "Expected pointer-to-8-bit int type for `ndarray.1`, got pointer-to-{}-bit int",
+                ndarray_data.get_bit_width()
+            ));
+        }
 
         Ok(())
     }
 
+    // TODO: Move this into e.g. StructProxyType
+    #[must_use]
+    fn fields(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> NDArrayStructFields<'ctx> {
+        NDArrayStructFields::new(ctx, llvm_usize)
+    }
+
+    // TODO: Move this into e.g. StructProxyType
+    #[must_use]
+    pub fn get_fields(
+        &self,
+        ctx: &'ctx Context,
+        llvm_usize: IntType<'ctx>,
+    ) -> NDArrayStructFields<'ctx> {
+        Self::fields(ctx, llvm_usize)
+    }
+
     /// Creates an LLVM type corresponding to the expected structure of an `NDArray`.
     #[must_use]
-    fn llvm_type(
+    fn llvm_type(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> PointerType<'ctx> {
-        ctx: &'ctx Context,
+        // struct NDArray { num_dims: size_t, dims: size_t*, data: i8* }
-        dtype: BasicTypeEnum<'ctx>,
-        llvm_usize: IntType<'ctx>,
-    ) -> PointerType<'ctx> {
-        // struct NDArray { num_dims: size_t, dims: size_t*, data: T* }
         //
-        // * num_dims: Number of dimensions in the array
+        // * data    : Pointer to an array containing the array data
-        // * dims: Pointer to an array containing the size of each dimension
+        // * itemsize: The size of each NDArray elements in bytes
-        // * data: Pointer to an array containing the array data
+        // * ndims   : Number of dimensions in the array
-        let field_tys = [
+        // * shape   : Pointer to an array containing the shape of the NDArray
-            llvm_usize.into(),
+        // * strides : Pointer to an array indicating the number of bytes between each element at a dimension
-            llvm_usize.ptr_type(AddressSpace::default()).into(),
+        let field_tys =
-            dtype.ptr_type(AddressSpace::default()).into(),
+            Self::fields(ctx, llvm_usize).into_iter().map(|field| field.1).collect_vec();
-        ];
 
         ctx.struct_type(&field_tys, false).ptr_type(AddressSpace::default())
     }
 
-    /// Creates an instance of [`ListType`].
+    /// Creates an instance of [`NDArrayType`].
     #[must_use]
     pub fn new<G: CodeGenerator + ?Sized>(
         generator: &G,
@@ -103,24 +142,21 @@ impl<'ctx> NDArrayType<'ctx> {
         dtype: BasicTypeEnum<'ctx>,
     ) -> Self {
         let llvm_usize = generator.get_size_type(ctx);
-        let llvm_ndarray = Self::llvm_type(ctx, dtype, llvm_usize);
+        let llvm_ndarray = Self::llvm_type(ctx, llvm_usize);
 
-        NDArrayType::from_type(llvm_ndarray, llvm_usize)
+        NDArrayType { ty: llvm_ndarray, dtype, llvm_usize }
     }
 
-    /// Creates an [`NDArrayType`] from a [`PointerType`].
+    /// Creates an [`NDArrayType`] from a [`PointerType`] representing an `NDArray`.
     #[must_use]
-    pub fn from_type(ptr_ty: PointerType<'ctx>, llvm_usize: IntType<'ctx>) -> Self {
+    pub fn from_type(
+        ptr_ty: PointerType<'ctx>,
+        dtype: BasicTypeEnum<'ctx>,
+        llvm_usize: IntType<'ctx>,
+    ) -> Self {
         debug_assert!(Self::is_representable(ptr_ty, llvm_usize).is_ok());
 
-        NDArrayType {
+        NDArrayType { ty: ptr_ty, dtype, llvm_usize }
-            ty: ptr_ty,
-            dtype: ptr_ty
-                .get_element_type()
-                .try_into()
-                .expect("Expected BasicTypeEnum for dtype of NDArray"),
-            llvm_usize,
-        }
     }
 
     /// Returns the type of the `size` field of this `ndarray` type.
@@ -207,7 +243,7 @@ impl<'ctx> ProxyType<'ctx> for NDArrayType<'ctx> {
     ) -> Self::Value {
         debug_assert_eq!(value.get_type(), self.as_base_type());
 
-        NDArrayValue::from_pointer_value(value, self.llvm_usize, name)
+        NDArrayValue::from_pointer_value(value, self.dtype, self.llvm_usize, name)
     }
 
     fn as_base_type(&self) -> Self::Base {

nac3core/src/codegen/values/ndarray.rs

@@ -1,7 +1,7 @@
 use inkwell::{
-    types::{AnyTypeEnum, BasicTypeEnum, IntType},
+    types::{AnyType, AnyTypeEnum, BasicType, BasicTypeEnum, IntType},
     values::{BasicValueEnum, IntValue, PointerValue},
-    IntPredicate,
+    AddressSpace, IntPredicate,
 };
 
 use super::{
@@ -20,6 +20,7 @@ use crate::codegen::{
 #[derive(Copy, Clone)]
 pub struct NDArrayValue<'ctx> {
     value: PointerValue<'ctx>,
+    dtype: BasicTypeEnum<'ctx>,
     llvm_usize: IntType<'ctx>,
     name: Option<&'ctx str>,
 }
@@ -38,28 +39,21 @@ impl<'ctx> NDArrayValue<'ctx> {
     #[must_use]
     pub fn from_pointer_value(
         ptr: PointerValue<'ctx>,
+        dtype: BasicTypeEnum<'ctx>,
         llvm_usize: IntType<'ctx>,
         name: Option<&'ctx str>,
     ) -> Self {
         debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
 
-        NDArrayValue { value: ptr, llvm_usize, name }
+        NDArrayValue { value: ptr, dtype, llvm_usize, name }
     }
 
     /// Returns the pointer to the field storing the number of dimensions of this `NDArray`.
     fn ptr_to_ndims(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
-        let llvm_i32 = ctx.ctx.i32_type();
+        self.get_type()
-        let var_name = self.name.map(|v| format!("{v}.ndims.addr")).unwrap_or_default();
+            .get_fields(ctx.ctx, self.llvm_usize)
-
+            .ndims
-        unsafe {
+            .ptr_by_gep(ctx, self.value, self.name)
-            ctx.builder
-                .build_in_bounds_gep(
-                    self.as_base_value(),
-                    &[llvm_i32.const_zero(), llvm_i32.const_zero()],
-                    var_name.as_str(),
-                )
-                .unwrap()
-        }
     }
 
     /// Stores the number of dimensions `ndims` into this instance.
@@ -81,63 +75,51 @@ impl<'ctx> NDArrayValue<'ctx> {
         ctx.builder.build_load(pndims, "").map(BasicValueEnum::into_int_value).unwrap()
     }
 
-    /// Returns the double-indirection pointer to the `dims` array, as if by calling `getelementptr`
+    /// Returns the double-indirection pointer to the `shape` array, as if by calling
-    /// on the field.
+    /// `getelementptr` on the field.
-    fn ptr_to_dims(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
+    fn ptr_to_shape(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
-        let llvm_i32 = ctx.ctx.i32_type();
+        self.get_type()
-        let var_name = self.name.map(|v| format!("{v}.dims.addr")).unwrap_or_default();
+            .get_fields(ctx.ctx, self.llvm_usize)
-
+            .shape
-        unsafe {
+            .ptr_by_gep(ctx, self.value, self.name)
-            ctx.builder
-                .build_in_bounds_gep(
-                    self.as_base_value(),
-                    &[llvm_i32.const_zero(), llvm_i32.const_int(1, true)],
-                    var_name.as_str(),
-                )
-                .unwrap()
-        }
     }
 
     /// Stores the array of dimension sizes `dims` into this instance.
-    fn store_dim_sizes(&self, ctx: &CodeGenContext<'ctx, '_>, dims: PointerValue<'ctx>) {
+    fn store_shape(&self, ctx: &CodeGenContext<'ctx, '_>, dims: PointerValue<'ctx>) {
-        ctx.builder.build_store(self.ptr_to_dims(ctx), dims).unwrap();
+        ctx.builder.build_store(self.ptr_to_shape(ctx), dims).unwrap();
     }
 
     /// Convenience method for creating a new array storing dimension sizes with the given `size`.
-    pub fn create_dim_sizes(
+    pub fn create_shape(
         &self,
         ctx: &CodeGenContext<'ctx, '_>,
         llvm_usize: IntType<'ctx>,
         size: IntValue<'ctx>,
     ) {
-        self.store_dim_sizes(ctx, ctx.builder.build_array_alloca(llvm_usize, size, "").unwrap());
+        self.store_shape(ctx, ctx.builder.build_array_alloca(llvm_usize, size, "").unwrap());
     }
 
     /// Returns a proxy object to the field storing the size of each dimension of this `NDArray`.
     #[must_use]
-    pub fn dim_sizes(&self) -> NDArrayDimsProxy<'ctx, '_> {
+    pub fn shape(&self) -> NDArrayShapeProxy<'ctx, '_> {
-        NDArrayDimsProxy(self)
+        NDArrayShapeProxy(self)
     }
 
     /// Returns the double-indirection pointer to the `data` array, as if by calling `getelementptr`
     /// on the field.
     pub fn ptr_to_data(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
-        let llvm_i32 = ctx.ctx.i32_type();
+        self.get_type()
-        let var_name = self.name.map(|v| format!("{v}.data.addr")).unwrap_or_default();
+            .get_fields(ctx.ctx, self.llvm_usize)
-
+            .data
-        unsafe {
+            .ptr_by_gep(ctx, self.value, self.name)
-            ctx.builder
-                .build_in_bounds_gep(
-                    self.as_base_value(),
-                    &[llvm_i32.const_zero(), llvm_i32.const_int(2, true)],
-                    var_name.as_str(),
-                )
-                .unwrap()
-        }
     }
 
     /// Stores the array of data elements `data` into this instance.
     fn store_data(&self, ctx: &CodeGenContext<'ctx, '_>, data: PointerValue<'ctx>) {
+        let data = ctx
+            .builder
+            .build_bit_cast(data, ctx.ctx.i8_type().ptr_type(AddressSpace::default()), "")
+            .unwrap();
         ctx.builder.build_store(self.ptr_to_data(ctx), data).unwrap();
     }
 
@@ -149,7 +131,15 @@ impl<'ctx> NDArrayValue<'ctx> {
         elem_ty: BasicTypeEnum<'ctx>,
         size: IntValue<'ctx>,
     ) {
-        self.store_data(ctx, ctx.builder.build_array_alloca(elem_ty, size, "").unwrap());
+        let itemsize =
+            ctx.builder.build_int_cast(elem_ty.size_of().unwrap(), size.get_type(), "").unwrap();
+        let nbytes = ctx.builder.build_int_mul(size, itemsize, "").unwrap();
+
+        // TODO: What about alignment?
+        self.store_data(
+            ctx,
+            ctx.builder.build_array_alloca(ctx.ctx.i8_type(), nbytes, "").unwrap(),
+        );
     }
 
     /// Returns a proxy object to the field storing the data of this `NDArray`.
@@ -164,7 +154,7 @@ impl<'ctx> ProxyValue<'ctx> for NDArrayValue<'ctx> {
     type Type = NDArrayType<'ctx>;
 
     fn get_type(&self) -> Self::Type {
-        NDArrayType::from_type(self.as_base_value().get_type(), self.llvm_usize)
+        NDArrayType::from_type(self.as_base_value().get_type(), self.dtype, self.llvm_usize)
     }
 
     fn as_base_value(&self) -> Self::Base {
@@ -180,15 +170,15 @@ impl<'ctx> From<NDArrayValue<'ctx>> for PointerValue<'ctx> {
 
 /// Proxy type for accessing the `dims` array of an `NDArray` instance in LLVM.
 #[derive(Copy, Clone)]
-pub struct NDArrayDimsProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
+pub struct NDArrayShapeProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
 
-impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDimsProxy<'ctx, '_> {
+impl<'ctx> ArrayLikeValue<'ctx> for NDArrayShapeProxy<'ctx, '_> {
     fn element_type<G: CodeGenerator + ?Sized>(
         &self,
         ctx: &CodeGenContext<'ctx, '_>,
         generator: &G,
     ) -> AnyTypeEnum<'ctx> {
-        self.0.dim_sizes().base_ptr(ctx, generator).get_type().get_element_type()
+        self.0.shape().base_ptr(ctx, generator).get_type().get_element_type()
     }
 
     fn base_ptr<G: CodeGenerator + ?Sized>(
@@ -199,7 +189,7 @@ impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDimsProxy<'ctx, '_> {
         let var_name = self.0.name.map(|v| format!("{v}.data")).unwrap_or_default();
 
         ctx.builder
-            .build_load(self.0.ptr_to_dims(ctx), var_name.as_str())
+            .build_load(self.0.ptr_to_shape(ctx), var_name.as_str())
             .map(BasicValueEnum::into_pointer_value)
             .unwrap()
     }
@@ -213,7 +203,7 @@ impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDimsProxy<'ctx, '_> {
     }
 }
 
-impl<'ctx> ArrayLikeIndexer<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_> {
+impl<'ctx> ArrayLikeIndexer<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ctx, '_> {
     unsafe fn ptr_offset_unchecked<G: CodeGenerator + ?Sized>(
         &self,
         ctx: &mut CodeGenContext<'ctx, '_>,
@@ -252,10 +242,10 @@ impl<'ctx> ArrayLikeIndexer<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_>
     }
 }
 
-impl<'ctx> UntypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_> {}
+impl<'ctx> UntypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ctx, '_> {}
-impl<'ctx> UntypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_> {}
+impl<'ctx> UntypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ctx, '_> {}
 
-impl<'ctx> TypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_> {
+impl<'ctx> TypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ctx, '_> {
     fn downcast_to_type(
         &self,
         _: &mut CodeGenContext<'ctx, '_>,
@@ -265,7 +255,7 @@ impl<'ctx> TypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ct
     }
 }
 
-impl<'ctx> TypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayDimsProxy<'ctx, '_> {
+impl<'ctx> TypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ctx, '_> {
     fn upcast_from_type(
         &self,
         _: &mut CodeGenContext<'ctx, '_>,
@@ -282,10 +272,10 @@ pub struct NDArrayDataProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
 impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDataProxy<'ctx, '_> {
     fn element_type<G: CodeGenerator + ?Sized>(
         &self,
-        ctx: &CodeGenContext<'ctx, '_>,
+        _: &CodeGenContext<'ctx, '_>,
-        generator: &G,
+        _: &G,
     ) -> AnyTypeEnum<'ctx> {
-        self.0.data().base_ptr(ctx, generator).get_type().get_element_type()
+        self.0.dtype.as_any_type_enum()
     }
 
     fn base_ptr<G: CodeGenerator + ?Sized>(
@@ -318,15 +308,37 @@ impl<'ctx> ArrayLikeIndexer<'ctx> for NDArrayDataProxy<'ctx, '_> {
         idx: &IntValue<'ctx>,
         name: Option<&str>,
     ) -> PointerValue<'ctx> {
-        unsafe {
+        let sizeof_elem = ctx
+            .builder
+            .build_int_truncate_or_bit_cast(
+                self.element_type(ctx, generator).size_of().unwrap(),
+                idx.get_type(),
+                "",
+            )
+            .unwrap();
+        let idx = ctx.builder.build_int_mul(*idx, sizeof_elem, "").unwrap();
+        let ptr = unsafe {
             ctx.builder
                 .build_in_bounds_gep(
                     self.base_ptr(ctx, generator),
-                    &[*idx],
+                    &[idx],
                     name.unwrap_or_default(),
                 )
                 .unwrap()
-        }
+        };
+
+        // Current implementation is transparent - The returned pointer type is
+        // already cast into the expected type, allowing for immediately
+        // load/store.
+        ctx.builder
+            .build_pointer_cast(
+                ptr,
+                BasicTypeEnum::try_from(self.element_type(ctx, generator))
+                    .unwrap()
+                    .ptr_type(AddressSpace::default()),
+                "",
+            )
+            .unwrap()
     }
 
     fn ptr_offset<G: CodeGenerator + ?Sized>(
@@ -347,7 +359,20 @@ impl<'ctx> ArrayLikeIndexer<'ctx> for NDArrayDataProxy<'ctx, '_> {
             ctx.current_loc,
         );
 
-        unsafe { self.ptr_offset_unchecked(ctx, generator, idx, name) }
+        let ptr = unsafe { self.ptr_offset_unchecked(ctx, generator, idx, name) };
+
+        // Current implementation is transparent - The returned pointer type is
+        // already cast into the expected type, allowing for immediately
+        // load/store.
+        ctx.builder
+            .build_pointer_cast(
+                ptr,
+                BasicTypeEnum::try_from(self.element_type(ctx, generator))
+                    .unwrap()
+                    .ptr_type(AddressSpace::default()),
+                "",
+            )
+            .unwrap()
     }
 }
 
@@ -381,8 +406,17 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
         );
 
         let index = call_ndarray_flatten_index(generator, ctx, *self.0, indices);
+        let sizeof_elem = ctx
+            .builder
+            .build_int_truncate_or_bit_cast(
+                self.element_type(ctx, generator).size_of().unwrap(),
+                index.get_type(),
+                "",
+            )
+            .unwrap();
+        let index = ctx.builder.build_int_mul(index, sizeof_elem, "").unwrap();
 
-        unsafe {
+        let ptr = unsafe {
             ctx.builder
                 .build_in_bounds_gep(
                     self.base_ptr(ctx, generator),
@@ -390,7 +424,17 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
                     name.unwrap_or_default(),
                 )
                 .unwrap()
-        }
+        };
+        // TODO: Current implementation is transparent
+        ctx.builder
+            .build_pointer_cast(
+                ptr,
+                BasicTypeEnum::try_from(self.element_type(ctx, generator))
+                    .unwrap()
+                    .ptr_type(AddressSpace::default()),
+                "",
+            )
+            .unwrap()
     }
 
     fn ptr_offset<G: CodeGenerator + ?Sized>(
@@ -429,7 +473,7 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
                 let (dim_idx, dim_sz) = unsafe {
                     (
                         indices.get_unchecked(ctx, generator, &i, None).into_int_value(),
-                        self.0.dim_sizes().get_typed_unchecked(ctx, generator, &i, None),
+                        self.0.shape().get_typed_unchecked(ctx, generator, &i, None),
                     )
                 };
                 let dim_idx = ctx
@@ -455,7 +499,17 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
         )
         .unwrap();
 
-        unsafe { self.ptr_offset_unchecked(ctx, generator, indices, name) }
+        let ptr = unsafe { self.ptr_offset_unchecked(ctx, generator, indices, name) };
+        // TODO: Current implementation is transparent
+        ctx.builder
+            .build_pointer_cast(
+                ptr,
+                BasicTypeEnum::try_from(self.element_type(ctx, generator))
+                    .unwrap()
+                    .ptr_type(AddressSpace::default()),
+                "",
+            )
+            .unwrap()
     }
 }
 

nac3standalone/demo/src/ndarray.py

@@ -144,6 +144,7 @@ def test_ndarray_array():
 
     # Copy
     n2_cpy: ndarray[float, 2] = np_array(n2, copy=False)
+    output_ndarray_float_2(n2_cpy)
     n2_cpy.fill(0.0)
     output_ndarray_float_2(n2_cpy)