tag manually checked hot functions with @[direct_array_access]

datasets/mnist_test.v

@@ -2,10 +2,14 @@
 import vtl.datasets
 
 fn test_mnist() {
+	unbuffer_stdout()
+	println('start')
 	mnist := datasets.load_mnist()!
+	println('mnist dataset loaded')
 
 	assert mnist.train_features.shape == [60000, 28, 28]
 	assert mnist.test_features.shape == [10000, 28, 28]
 	assert mnist.train_labels.shape == [60000]
 	assert mnist.test_labels.shape == [10000]
+	println('done')
 }

src/creation.v

@@ -93,6 +93,7 @@ pub fn from_1d[T](arr []T, params TensorData) !&Tensor[T] {
 
 // from_2d takes a two dimensional array of floating point values
 // and returns a two-dimensional Tensor if possible
+@[direct_array_access]
 pub fn from_2d[T](a [][]T, params TensorData) !&Tensor[T] {
 	mut arr := []T{cap: a.len * a[0].len}
 	for i in 0 .. a.len {

src/fun.v

@@ -135,6 +135,7 @@ pub fn (t &Tensor[T]) as_strided[T](shape []int, strides []int) !&Tensor[T] {
 
 // transpose permutes the axes of an tensor in a specified
 // order and returns a view of the data
+@[direct_array_access]
 pub fn (t &Tensor[T]) transpose[T](order []int) !&Tensor[T] {
 	mut ret := t.view()
 	n := order.len
@@ -187,6 +188,7 @@ fn fabs(x f64) f64 {
 }
 
 // slice returns a tensor from a variadic list of indexing operations
+@[direct_array_access]
 pub fn (t &Tensor[T]) slice[T](idx ...[]int) !&Tensor[T] {
 	mut newshape := t.shape.clone()
 	mut newstrides := t.strides.clone()
@@ -266,6 +268,7 @@ pub fn (t &Tensor[T]) slice[T](idx ...[]int) !&Tensor[T] {
 
 // slice_hilo returns a view of an array from a list of starting
 // indices and a list of closing indices.
+@[direct_array_access]
 pub fn (t &Tensor[T]) slice_hilo[T](idx1 []int, idx2 []int) !&Tensor[T] {
 	mut newshape := t.shape.clone()
 	mut newstrides := t.strides.clone()

src/fun_logical.v

@@ -105,7 +105,7 @@ pub fn (t &Tensor[T]) array_equiv[T](other &Tensor[T]) bool {
 	return true
 }
 
-@[inline]
+@[direct_array_access; inline]
 fn handle_equal[T](vals []T, _ []int) bool {
 	mut equal := true
 	for v in vals {

src/iter.v

@@ -129,6 +129,7 @@ fn handle_flatten_iteration[T](mut s TensorIterator[T]) T {
 	return val
 }
 
+@[direct_array_access]
 fn tensor_backstrides[T](t &Tensor[T]) []int {
 	rank := t.rank()
 	shape := t.shape

src/split.v

@@ -6,6 +6,7 @@ module vtl
 // integer that does not equally divide the axis. For an array of length
 // l that should be split into n sections, it returns l % n sub-arrays of
 // size l//n + 1 and the rest of size l//n.
+@[direct_array_access]
 pub fn (t &Tensor[T]) array_split[T](ind int, axis int) ![]&Tensor[T] {
 	ntotal := t.shape[axis]
 	neach := ntotal / ind
@@ -125,7 +126,11 @@ pub fn (t &Tensor[T]) dsplit_expl[T](ind []int) ![]&Tensor[T] {
 
 // splitter implements a generic splitting function that contains the underlying functionality
 // for all split operations
+@[direct_array_access]
 fn (t &Tensor[T]) splitter[T](axis int, n int, div_points []int) ![]&Tensor[T] {
+	if n > 0 && div_points.len <= n {
+		return error('splitter error, div_points.len <= n')
+	}
 	mut subary := []&Tensor[T]{}
 	sary := t.swapaxes(axis, 0)!
 	for i in 0 .. n {

src/util.v

@@ -60,6 +60,7 @@ pub fn (mut t Tensor[T]) ensure_memory[T]() {
 
 // assert_shape_off_axis ensures that the shapes of Tensors match
 // for concatenation, except along the axis being joined
+@[direct_array_access]
 fn assert_shape_off_axis[T](ts []&Tensor[T], axis int, shape []int) ![]int {
 	mut retshape := shape.clone()
 	for t in ts {
@@ -91,6 +92,7 @@ fn assert_shape[T](shape []int, ts []&Tensor[T]) ! {
 
 // is_col_major_contiguous checks if an array is contiguous with a col-major
 // memory layout
+@[direct_array_access]
 fn is_col_major_contiguous(shape []int, strides []int, ndims int) bool {
 	if ndims == 0 {
 		return true
@@ -114,6 +116,7 @@ fn is_col_major_contiguous(shape []int, strides []int, ndims int) bool {
 
 // is_row_major_contiguous checks if an array is contiguous with a row-major
 // memory layout
+@[direct_array_access]
 fn is_row_major_contiguous(shape []int, strides []int, ndims int) bool {
 	if ndims == 0 {
 		return true
@@ -150,6 +153,7 @@ fn clip_axis(axis int, size int) !int {
 }
 
 // strides_from_shape returns the strides from a shape and memory format
+@[direct_array_access]
 fn strides_from_shape(shape []int, memory MemoryFormat) []int {
 	mut accum := 1
 	mut result := []int{len: shape.len}
@@ -207,6 +211,7 @@ fn shape_with_autosize(shape []int, size int) !([]int, int) {
 
 // filter_shape_not_strides removes 0 size dimensions from the shape
 // and strides of an array
+@[direct_array_access]
 fn filter_shape_not_strides(shape []int, strides []int) !([]int, []int) {
 	mut newshape := []int{}
 	mut newstrides := []int{}