[core] Account for both ellipsisMask and begin / end masks in strided slice.

[annxingyuan]

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

---

This change is 

[pyu10055]

Reviewable status: 0 of 1 approvals obtained (waiting on @annxingyuan and @pyu10055)

---

tfjs-core/src/ops/slice_util.ts, line 118 at r1 (raw file):

        newIndices[ellipsisInsertionIndex] = 0;
      } else {
        newIndices.splice(

Since you are iterating from low to high, I am curious why you need to insert and pop from back, can you replace the value at the index with 0?

---

tfjs-core/src/ops/strided_slice.ts, line 96 at r1 (raw file):

  // Normalize the start, end and strides.
  if (ellipsisAxes.length && numInterpolatedAxes > 0) {

when ellipsis mask is not zero, will the condition be not true?
meaning if the startForAxis and stopForAxis method need the ellipsisMask as parameter?

[pyu10055]

Thank you for fixing this.

Reviewable status: 0 of 1 approvals obtained (waiting on @annxingyuan and @pyu10055)

[annxingyuan]

Reviewable status: 0 of 1 approvals obtained (waiting on @pyu10055)

---

tfjs-core/src/ops/slice_util.ts, line 118 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

Since you are iterating from low to high, I am curious why you need to insert and pop from back, can you replace the value at the index with 0?

Done

Thanks for catching this - this logic was leftover from the previous implementation, when newIndices were initialized to begin / end.

---

tfjs-core/src/ops/strided_slice.ts, line 96 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

when ellipsis mask is not zero, will the condition be not true?
meaning if the startForAxis and stopForAxis method need the ellipsisMask as parameter?

Sometimes ellipsisAxis.length > 0 but numInterpolatedAxes = 0, this is the case when the ellipsis only applies to one axis because begin / end have the same dimensionality as the input shape.

[pyu10055]

THanks!