-
NdArray
-
Index tensor
-
Permute tensor
-
Broadcast tensors
-
Basic tensor element ops
-
Slice tensor
-
Display and Debug tensor
-
Modify strides
- Flip
-
Arange
- Try using iters
-
Tensor to contiguous
- Reshape
-
[?] Modify linspace
-
Flip all dims as contiguous
- Slice after flip all dims
-
Dimension-wise ops
-
[>] Matmul
- 2D (naive)
- ND (naive)
- [?] Strassen
-
Conv & cross-correlation
- All types naive 1d and 2d
- [?] Generalize for higher dims?
- Replace padding with slicing
- Handle cases where kernel > input
- [?] FFT
- Perform using strides?
-
[>] Padding tensor
- [?] Better method?
-
Kernel / Op Fusion
-
Other ndarray functions
-
-
Autograd
- Forw
- Rev
-
[>] Optimisations
-
Better error handling
-
SIMD
-
GPU