Add oneDNN support for "where" operator

[bgawrych]

Description

Implementation of where operator utilizing few oneDNN primitives.
Implementation creates two masks:

• condition is met
• condition not met
  In the next step these mask are applied for left operand (condition met) and for right operand (condition not met). In the end there is sum of both masked input giving proper result.

Performance:

Script:

import mxnet as mx
import time

dims = [128, 512, 1024, 4096]
print("shape;time")
for dim1 in dims:
    for dim2 in dims:
        shape = (dim1, dim2)
        a = mx.np.random.uniform(-1.0, 1.0, shape).astype(mx.np.float32)
        b = mx.np.random.uniform(-1.0, 1.0, shape).astype(mx.np.float32)
        cond = mx.np.random.uniform(0,1, size=shape)
        cond = mx.np.round(cond)
        mx.nd.waitall()
        tic = time.time()
        for i in range(10000):
            out = mx.np.where(cond, a, b)
            out.wait_to_read()
        toc = time.time()
        print(f"{shape};{toc-tic}")

command:
OMP_NUM_THREADS=32 GOMP_CPU_AFFINITY="0-31:1" python script.py

Benchmarked on AWS c6i.16xlarge

[mxnet-bot]

Hey @bgawrych , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [miscellaneous, clang, windows-gpu, centos-gpu, unix-gpu, sanity, windows-cpu, centos-cpu, edge, website, unix-cpu]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[bgawrych]

@mxnet-bot run ci [windows-gpu, centos-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [windows-gpu, centos-gpu]

[bgawrych]

@mxnet-bot run ci [unix-cpu, centos-gpu]

[bgawrych]

@mxnet-bot run ci [unix-cpu, centos-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-cpu, centos-gpu]