Improve ROCm support. (#780)

- updates to support ROCm 5.2
- workarounds in tests where NV tools were used unconditionally
- implemented `get_num_blocks()` and `add_memfence()` for AMD GPU
- backported from history some atomics
- added bf16 support
- minor warnings cleanup
- added dockerfile to run on a ROCm enabled machine

Co-authored-by: B1tway <andrew.shukshov@gmail.com>
Co-authored-by: Andrey Shukshov <36711069+B1tway@users.noreply.github.com>

triton_rocm_20-52.Dockerfile

@@ -0,0 +1,14 @@
+FROM rocm/pytorch:rocm5.2.3_ubuntu20.04_py3.7_pytorch_1.12.1
+
+# build triton
+RUN export TRITON_USE_ROCM=ON MI_GPU_ARCH=gfx90a
+
+# Unit Tests 
+# to run unit tests
+# 1. build this Dockerfile
+#    docker build --build-arg -f triton_rocm_20-52.Dockerfile -t triton_rocm52 .
+# 2. run docker container
+#    docker run -it --rm --network=host --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --name triton --ipc=host --shm-size 16G --device=/dev/kfd --device=/dev/dri triton_rocm52:latest
+# 3. run core unit tests on a rocm machine
+#    cd ~/triton/python
+#    pytest --verbose test/unit/language/test_core.py | tee test_core.log