Generated Backend

[micmelesse]

This pr does the following

• updates the old backend to a new generated backend
  • this will make maintance much easier
• adds a tool (scripts/amd/gen_hip_backend.py) used to generate the backend
  • the tool is based on the parsing lib tree-sitter
  • it parses the c++, python, cmake and tablgen files in this fork and modifies them to be used for the backend
  • the modifications involve altering Macros (defines, includes), imports, types, functions, namespaces, strings.
• adds a github actions tests for both upstream testing and fork testing
  • in the future this will allows us to move away from jenkins
  • fork test
    • is the same as the jenkins ci
  • upstream test
    • clones the upstream
    • checkouts it out to the last shared commit with upstream
    • applies a patch
    • checks out this branch as a backend
    • builds
    • runs unit tests
  • when reviewing you can ignore python/third_party/hip/gen. It is the generated files from the rest of the fork. If the upstream test works It should be fine.

[scxiao]

https://github.com/ROCmSoftwarePlatform/triton/blob/3538df7d4cddbc0770c52bffef17e25ad68792e4/python/triton/third_party/hip/hip_backend.py#L580 It cannot hard coded to 64 since Navi31 has wave size 32. The PR for dynamic warp size is at:

https://github.com/ROCmSoftwarePlatform/triton/pull/244

[micmelesse]

moveing the work on the ci to #455. Generating the backend might not be needed.