ndstrides: [3] Implement indexing/slicing #513

lyken · 2024-08-28T13:48:48+08:00

lyken commented

2024-08-28 13:48:48 +08:00

This PR implements indexing/slicing with ndarray strides.

Slicing an ndarray no longer copies ndarray->data and only creates a new ndarray with different strides but points to the same data ndarray->data, which is very cheap. This is only possible with the addition of strides to NAC3's ndarray definition. This is how NumPy implements ndarrays too.
... and np.newaxis have been implemented using IRRT, but they are only there to help with implementing other aspects of ndarray with strides (e.g., NDArrayObject::atleast_nd() in a later commit, it uses np.newaxis internally to prepend 1s to the dimension; or in matmul, np.newaxis and ... are used to deal with 1D matrices). You cannot write my_array[..., none] in NAC3 Python as the type inferencer/code generator cannot understand it. I have encountered some problems when trying to implement this into NAC3:
1. ... does not have a concrete type. #486. Probably requires a hack on the typechecker.
2. Consider my_array[..., none] again; none has type <class 'Option'>, but NumPy wants a NoneType. This causes issues in @portable.

Extra note: After this PR, there is finally a proper way to start developing ndarray with strides with nac3standalone, since now you can print the contents of an ndarray and debug:

@extern
def output_float64(x: float):
    ...

# This requires ndarray indexing to be implemented.
def output_ndarray_float_2(n: ndarray[float, Literal[2]]):
    for r in range(len(n)):
        for c in range(len(n[r])):
            output_float64(n[r][c])

def run() -> int32:
    xs = np_full([3, 4], 9999.9999)
    output_ndarray_float_2(xs)
    return 0

This PR implements indexing/slicing with ndarray strides. - Slicing an ndarray no longer copies `ndarray->data` and only creates a new ndarray with different strides but points to the same data `ndarray->data`, which is very cheap. This is only possible with the addition of strides to NAC3's ndarray definition. This is how NumPy implements ndarrays too. - `...` and `np.newaxis` have been implemented using IRRT, but they are only there to help with implementing other aspects of ndarray with strides (e.g., `NDArrayObject::atleast_nd()` in a later commit, it uses `np.newaxis` internally to prepend 1s to the dimension; or in `matmul`, `np.newaxis` and `...` are used to deal with 1D matrices). You cannot write `my_array[..., none]` in NAC3 Python as the type inferencer/code generator cannot understand it. I have encountered some problems when trying to implement this into NAC3: 1. `...` does not have a concrete type. https://git.m-labs.hk/M-Labs/nac3/issues/486. Probably requires a hack on the typechecker. 2. Consider `my_array[..., none]` again; `none` has type `<class 'Option'>`, but NumPy wants a `NoneType`. This causes issues in `@portable`. Extra note: After this PR, there is finally a proper way to start developing ndarray with strides with nac3standalone, since now you can print the contents of an ndarray and debug: ```python @extern def output_float64(x: float): ... # This requires ndarray indexing to be implemented. def output_ndarray_float_2(n: ndarray[float, Literal[2]]): for r in range(len(n)): for c in range(len(n[r])): output_float64(n[r][c]) def run() -> int32: xs = np_full([3, 4], 9999.9999) output_ndarray_float_2(xs) return 0 ```

lyken added 3 commits 2024-08-28 13:48:48 +08:00

5537645395

core/irrt: add Slice and Range

Needed for implementing general ndarray indexing.

Currently the IRRT slice and range have nothing to do with NAC3's slice
and range.

fda7f8d827

core/irrt: rename NDIndex to NDIndexInt

The name `NDIndex` is used in later commits.

510dbfc70e

core/ndstrides: implement ndarray indexing

The functionality for `...` and `np.newaxis` is there in IRRT, but there
is no implementation of them for @kernel Python expressions because of
#486.

lyken added a new dependency 2024-08-28 13:49:00 +08:00

#512 ndstrides: [2] `{Any,Tuple,List}Object` + Implement `np_{zeros,ones,full,empty}` and `len()`

lyken added a new dependency 2024-08-28 13:52:09 +08:00

#514 ndstrides: [4] Implement `np_array()`

lyken added 1 commit 2024-08-28 16:44:13 +08:00

40cf1c747f

core/ndstrides: remove duplicated gen_slice()

lyken force-pushed ndstrides-3-indexing from 40cf1c747f to fdd5d37490

2024-08-30 14:11:18 +08:00

Compare

lyken commented

2024-08-30 14:11:23 +08:00

Rebased.

lyken force-pushed ndstrides-3-indexing from fdd5d37490 to a40cdde8d2

2024-08-30 14:13:05 +08:00

Compare

lyken commented

2024-08-30 14:13:24 +08:00

Rebased to fix some formatting issues.

derppening force-pushed ndstrides-3-indexing from a40cdde8d2 to 0206c3989a

2024-10-17 13:57:21 +08:00

Compare

derppening force-pushed ndstrides-3-indexing from 0206c3989a to c71bd81278

2024-10-17 16:08:55 +08:00

Compare

derppening force-pushed ndstrides-3-indexing from c71bd81278 to 3d8e66a54e

2024-10-17 16:30:23 +08:00

Compare

derppening force-pushed ndstrides-3-indexing from 3d8e66a54e to 8f9d2d82dd

2024-10-18 14:23:16 +08:00

Compare

This pull request can be merged automatically.

You are not authorized to merge this pull request.

You can also view command line instructions.

Step 1:

From your project repository, check out a new branch and test the changes.

git checkout -b ndstrides-3-indexing ndstrides-2-basic

git pull origin ndstrides-3-indexing

Step 2:

Merge the changes and update on Gitea.

git checkout ndstrides-2-basic

git merge --no-ff ndstrides-3-indexing

git push origin ndstrides-2-basic

Sign in to join this conversation.

No reviewers

No Label

No Milestone

No Assignees

2 Participants

Notifications

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Blocks

#514 ndstrides: [4] Implement `np_array()`

M-Labs/nac3

Depends on

#512 ndstrides: [2] `{Any,Tuple,List}Object` + Implement `np_{zeros,ones,full,empty}` and `len()`

M-Labs/nac3

Reference: M-Labs/nac3#513