Expose enablement of TensorExpr fuser as env variable

[bwasti]

This commit allows one to use an environment variable to enable the fuser in torch/csrc/jit/tensorexpr/

PYTORCH_TENSOREXPR=1 python benchmark.py

This commit also changes the registration to happen by default, removing the requirement for the python exposed "_jit_register_tensorexpr_fuser"

[dr-ci]

💊 CircleCI build failures summary and remediations

As of commit 600f887 (more details on the Dr. CI page):

---

✅ None of the build failures appear to be your fault 💚

---

• 2/2 broken upstream at merge base 925cdd5 on Mar 24 from 3:57pm to 5:19pm (3 commits; 50eb1a3 - 36e3c00)

  Please rebase on the viable/strict branch (expand for instructions)

  If your commit is newer than viable/strict, you can try basing on an older, stable commit:

  git fetch https://github.com/pytorch/pytorch viable/strict
  git rebase --onto FETCH_HEAD $(git merge-base origin/master HEAD)
  

  If your commit is older than viable/strict:

  git fetch https://github.com/pytorch/pytorch viable/strict
  git rebase FETCH_HEAD
  

  Check out the recency history of this "viable master" tracking branch.

---

1 job timed out:

• pytorch_macos_10_13_py3_test

---

🚧 1 upstream failure:

These were probably caused by upstream breakages:

• pytorch_linux_xenial_cuda10_2_cudnn7_py3_gcc7_test on Mar 24 from 3:57pm to 5:19pm (3 commits; 50eb1a3 - 36e3c00)

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker.

This comment has been revised 3 times.

[ZolotukhinM]

If we start to always register our pass, we should probably just add it to the graph executor pipeline directly. The existing machinery is only used because it is optional.

[ZolotukhinM]

If PYTORCH_TENSOREXPR env variable is set then torch._C._jit_set_texpr_fuser_enabled(False/True) calls will have no effect. Is it a desired behavior?

[bwasti]

for me, yea. One can override the program with PYTORCH_TENSOREXPR=0

Right now, for benchmarks, this is really useful because I don't need to inspect internals of various scripts. In the future, I see the python interface as eventually becoming deprecated anyway (once it is always on), and this functioning the same way regardless

[ZolotukhinM]

Nit: why underscore at the end?

[bwasti]

previous code used texpr_fuser_enabled directly, whereas code should now use the function tensorExprFuserEnabled. the underscore is my attempt at "privatizing" the variable so folks don't use it directly

[facebook-github-bot]

@bwasti has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@bwasti merged this pull request in a3e10d2.