Use libmkl_rt, or statically link against MKL

[ezyang]

Stack from ghstack:

• Don't use RTLD_GLOBAL to load _C. #31162 Don't use RTLD_GLOBAL to load _C.
• Pick parallel MKL implementation over sequential implementation. #31165 Pick parallel MKL implementation over sequential implementation.
• Use libmkl_rt, or statically link against MKL #31159 Use libmkl_rt, or statically link against MKL
• Uniformly apply Windows logic in cpp_extensions everywhere #31161 Actually get rid of Windows hacks in cpp_extensions.
• Stop using ctypes to interface with CUDA libraries. #31160 Stop using ctypes to interface with CUDA libraries.
• Move AutogradMeta and DeviceGuardImplInterface virtual methods out-of-line. #31176 Move AutogradMeta and DeviceGuardImplInterface virtual methods out-of-line.
• Add missing TORCH_CUDA_API annotation to throw_nccl_error #31157 Add missing TORCH_CUDA_API annotation to throw_nccl_error
• Remove dead CAFFE2_LIBS variable #31155 Remove dead CAFFE2_LIBS variable
• Remove LibIRC logic from cmake. #31152 Remove LibIRC logic from cmake.

Part of removing RTLD_GLOBAL for _C.

MKL doesn't like being in a local scope, unless you use the libmkl_rt.so
variant, so I adapted our CMake configuration to either accept that or
only allow static linking.

Signed-off-by: Edward Z. Yang ezyang@fb.com

[kostmo]

CircleCI build failures summary

As of commit ee279de:

• 2/2 failures introduced in this PR

Detailed failure analysis

One may explore the probable reasons each build failed interactively on the Dr. CI website.

1 new failure recognized by patterns

The following build failures don't appear to be due to upstream breakage:

pytorch_linux_xenial_py3_5_test (1/1)

Step: "Test" (details)

Dec 12 05:03:50 Ran 1471 tests in 462.501s
Dec 12 05:03:50 
Dec 12 05:03:50 FAILED (failures=1, skipped=681, expected failures=2)
Dec 12 05:03:50 
Dec 12 05:03:50 Generating XML reports...
Dec 12 05:03:50 Traceback (most recent call last):
Dec 12 05:03:50   File "test/run_test.py", line 456, in <module>
Dec 12 05:03:50     main()
Dec 12 05:03:50   File "test/run_test.py", line 449, in main
Dec 12 05:03:50     raise RuntimeError(message)
Dec 12 05:03:50 RuntimeError: test_nn failed!
Dec 12 05:03:50 + cleanup
Dec 12 05:03:50 + retcode=1
Dec 12 05:03:50 + set +x
Dec 12 05:03:50 =================== sccache compilation log ===================
Dec 12 05:03:50 =========== If your build fails, please take a look at the log above for possible reasons ===========
Dec 12 05:03:50 Compile requests                  7
Dec 12 05:03:50 Compile requests executed         6
Dec 12 05:03:50 Cache hits                        0
Dec 12 05:03:50 Cache misses                      6
Dec 12 05:03:50 Cache timeouts                    0

1 failure not recognized by patterns:

Job	Step
pytorch_macos_10_13_py3_test	Test

---

This comment was automatically generated by Dr. CI.
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 4 times.

[ezyang]

Hooo boy, this is the change that's causing the numeric errors.