Hip kernel fix

[reyna-abhyankar]

Description of changes:

Fix hip kernels for recently merged operators

Related Issues:

Linked Issues:

• Issue #

Issues closed by this PR:

• Closes #

---

This change is 

[lockshaw]

Reviewed 6 of 6 files at r1, all commit messages.
Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @lambda7xx and @reyna-abhyankar)

---

lib/kernels/src/hip/combine_kernels.cpp line 58 at r1 (raw file):

                    GenericTensorAccessorR const &input,
                    GenericTensorAccessorW const &output,
                    DataType data_type) {

Is this necessary, or can you get the data type from GenericTensorAccessor?

---

lib/kernels/src/hip/concat_kernels.cpp line 42 at r1 (raw file):

                    GenericTensorAccessorW const &output,
                    std::vector<GenericTensorAccessorR> const &inputs,
                    int num_inputs,

This can just be derived from inputs.size()

---

lib/kernels/src/hip/concat_kernels.cpp line 43 at r1 (raw file):

                    std::vector<GenericTensorAccessorR> const &inputs,
                    int num_inputs,
                    ff_dim_t legion_axis) {

Uh, if this is an ff_dim_t why is it called legion_axis? Shouldn't it probably just be axis (no ff needed as it's in the type)

---

lib/kernels/src/hip/concat_kernels.cpp line 77 at r1 (raw file):

  assert(num_inputs <= MAX_NUM_INPUTS);
  ArrayShape shape = output_grad.shape;
  calc_blk_size(num_blocks, output_blk_size, shape, legion_axis);

Suggestion:

  calc_blk_size(num_blocks, output_blk_size, output_grad.shape, legion_axis);

[lockshaw]

Reviewed 8 of 8 files at r2, all commit messages.
Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @lambda7xx and @reyna-abhyankar)