[Issue]: release/internal/3.3.x branch has merge failure in third_party/amd/language/hip/libdevice.py

[lamikr]

Problem Description

TheRock rocm/release/2.7x pytorch builds use the Triton from branch release/internal/3.3.x
When testing it, I get following error because there is a merge conflict committed in to file

third_party/amd/language/hip/libdevice.py

Error can be detected by cloning the triton and checking with editor with following commands:

git clone https://github.com/ROCm/triton.git
cd triton
git checkout release/internal/3.3.x
vi ./third_party/amd/language/hip/libdevice.py

Then there is following merge error starting from the line 76

@core.extern
<<<<<<< HEAD
def fast_dividef(arg0, arg1, _builder=None):
=======
def fast_tanhf(arg0, _semantic=None):
    return core.extern_elementwise("", "", [arg0], {
        (core.dtype("fp32"), ): ("__triton_hip_fast_tanhf", core.dtype("fp32")),
    }, is_pure=True, _semantic=_semantic)


@core.extern
def fast_dividef(arg0, arg1, _semantic=None):
>>>>>>> 353540798c ([AMD] Add fast_tanhf to libdevice (#7780))
    return core.extern_elementwise("", "", [arg0, arg1], {
        (core.dtype("fp32"), core.dtype("fp32")): ("__triton_hip_fast_fdividef", core.dtype("fp32")),
    }, is_pure=True, _builder=_builder)

Operating System

Ubuntu 22.04

CPU

AMD EPYC 9654 96-Core Processor

GPU

MI300x

ROCm Version

therock 7.0 pytorch build

ROCm Component

No response

Steps to Reproduce

It can be tested with

python triton/python/tutorials/01-vector-add.py
which shows error:

Traceback (most recent call last):
  File "therock_sdk_pytorch_rel/external-builds/pytorch/triton/python/tutorials/01-vector-add.py", line 23, in <module>
    import triton
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/__init__.py", line 21, in <module>
    from .compiler import compile, CompilationError
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/compiler/__init__.py", line 1, in <module>
    from .compiler import CompiledKernel, ASTSource, IRSource, compile, make_backend, LazyDict
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/compiler/compiler.py", line 13, in <module>
    from .code_generator import ast_to_ttir
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/compiler/code_generator.py", line 11, in <module>
    from .. import language
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/language/__init__.py", line 5, in <module>
    from . import extra
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/language/extra/__init__.py", line 19, in <module>
    spec.loader.exec_module(module)
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/language/extra/hip/__init__.py", line 1, in <module>
    from . import libdevice
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/language/extra/hip/__init__.py", line 1, in <module>
    from . import libdevice
  File "therock_sdk_pytorch_rel/.venv/lib/python3.12/site-packages/triton/language/extra/hip/libdevice.py", line 88
    >>>>>>> 353540798c ([AMD] Add fast_tanhf to libdevice (#7780))

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response