refactor: consume InfiniRT runtime headers

.github/ci_config.yml

@@ -15,7 +15,7 @@ platforms:
         BASE_IMAGE: nvcr.io/nvidia/pytorch:24.10-py3
         SKIP_APT: "1"
         PIP_INDEX_URL: https://pypi.tuna.tsinghua.edu.cn/simple
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_NVIDIA=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -51,7 +51,7 @@ platforms:
       - /lib/firmware:/lib/firmware
       - /usr/src:/usr/src
       - /lib/modules:/lib/modules
-    setup: python -m pip install packaging exceptiongroup typing-extensions pygments pybind11 libclang && python -m pip install . --no-build-isolation --no-deps --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: python -m pip install scikit-build-core packaging exceptiongroup typing-extensions pygments pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_ILUVATAR=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix -DCMAKE_CUDA_ARCHITECTURES=OFF -DCMAKE_CUDA_COMPILER=/usr/local/corex-4.3.0.20250624/bin/clang++ -DCUDAToolkit_ROOT=/usr/local/corex-4.3.0.20250624 && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && python -m pip install . --no-build-isolation --no-deps --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -81,7 +81,7 @@ platforms:
       - "--privileged"
       - "--ulimit=memlock=-1"
       - "--ulimit=stack=67108864"
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_METAX=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -109,7 +109,7 @@ platforms:
         PIP_INDEX_URL: https://pypi.org/simple
     docker_args:
       - "--privileged"
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_MOORE=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -136,7 +136,7 @@ platforms:
         PIP_INDEX_URL: https://pypi.org/simple
     docker_args:
       - "--privileged"
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_CAMBRICON=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -172,7 +172,7 @@ platforms:
       - "--group-add=video"
     volumes:
       - /opt/hyhal:/opt/hyhal:ro
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_HYGON=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       gpu:
         type: unittest
@@ -209,7 +209,7 @@ platforms:
       - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi:ro
     env:
       ASCEND_HOME_PATH: /usr/local/Ascend/ascend-toolkit/latest
-    setup: pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
+    setup: pip install scikit-build-core pybind11 libclang pyyaml && rm -rf /tmp/infinirt-src /tmp/infinirt-build /tmp/infinirt-prefix && git clone --depth 1 ${INFINI_RT_GIT_URL:-https://gh-proxy.com/https://github.com/InfiniTensor/InfiniRT.git} /tmp/infinirt-src && cmake -S /tmp/infinirt-src -B /tmp/infinirt-build -DWITH_CPU=ON -DWITH_ASCEND=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/infinirt-prefix && cmake --build /tmp/infinirt-build -j$(nproc) && cmake --install /tmp/infinirt-build && pip install .[dev] --no-build-isolation --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON --config-settings=cmake.define.INFINI_RT_ROOT=/tmp/infinirt-prefix
     jobs:
       npu:
         type: unittest

CMakeLists.txt

@@ -294,8 +294,6 @@ if(WITH_TORCH)
     message(STATUS "Found PyTorch: ${TORCH_INCLUDE_DIRS}")
 endif()
 
-include_directories(${CMAKE_CURRENT_SOURCE_DIR}/src)
-
 # Only one CUDA-like GPU backend can be enabled at a time.
 set(_gpu_backend_count 0)
 foreach(_gpu_backend WITH_NVIDIA WITH_ILUVATAR WITH_HYGON WITH_METAX WITH_MOORE WITH_ASCEND)
@@ -502,6 +500,90 @@ if(WITH_HYGON AND NOT EXISTS "${DTK_ROOT}/llvm/lib/LLVMgold.so")
     set(PYBIND11_ENABLE_EXTRAS OFF)
 endif()
 
+set(INFINI_RT_ROOT "$ENV{INFINI_RT_ROOT}" CACHE PATH "Installed InfiniRT prefix")
+set(INFINI_RT_SOURCE_DIR "" CACHE PATH "InfiniRT source directory for development builds")
+set(INFINI_RT_INCLUDE_DIRS "" CACHE STRING "InfiniRT include directories")
+set(INFINI_RT_LIBRARY "" CACHE FILEPATH "InfiniRT library")
+
+function(_infiniops_configure_infinirt)
+    if(TARGET infinirt)
+        get_target_property(_infinirt_include_dirs infinirt INTERFACE_INCLUDE_DIRECTORIES)
+        if(_infinirt_include_dirs)
+            set(INFINI_RT_INCLUDE_DIRS "${_infinirt_include_dirs}" CACHE STRING "InfiniRT include directories" FORCE)
+        endif()
+        return()
+    endif()
+
+    if(INFINI_RT_SOURCE_DIR)
+        if(NOT EXISTS "${INFINI_RT_SOURCE_DIR}/CMakeLists.txt")
+            message(FATAL_ERROR
+                "InfiniRT source tree not found at `${INFINI_RT_SOURCE_DIR}`. "
+                "Unset `INFINI_RT_SOURCE_DIR` to use an installed InfiniRT package.")
+        endif()
+
+        add_subdirectory("${INFINI_RT_SOURCE_DIR}" "${CMAKE_BINARY_DIR}/InfiniRT")
+        set(INFINI_RT_INCLUDE_DIRS
+            "${INFINI_RT_SOURCE_DIR}/include;${INFINI_RT_SOURCE_DIR}/generated/include"
+            CACHE STRING "InfiniRT include directories" FORCE)
+        return()
+    endif()
+
+    set(_infinirt_include_hints ${INFINI_RT_INCLUDE_DIRS})
+    if(INFINI_RT_ROOT)
+        list(APPEND _infinirt_include_hints "${INFINI_RT_ROOT}/include")
+    endif()
+
+    find_path(_INFINI_RT_INCLUDE_DIR
+        NAMES infini/rt.h
+        HINTS ${_infinirt_include_hints}
+        PATH_SUFFIXES include
+    )
+
+    if(INFINI_RT_LIBRARY)
+        set(_INFINI_RT_LIBRARY "${INFINI_RT_LIBRARY}")
+    else()
+        set(_infinirt_library_hints "")
+        if(INFINI_RT_ROOT)
+            list(APPEND _infinirt_library_hints
+                "${INFINI_RT_ROOT}/lib"
+                "${INFINI_RT_ROOT}/lib64")
+        endif()
+        find_library(_INFINI_RT_LIBRARY
+            NAMES infinirt
+            HINTS ${_infinirt_library_hints}
+        )
+    endif()
+
+    if(NOT _INFINI_RT_INCLUDE_DIR OR NOT _INFINI_RT_LIBRARY)
+        message(FATAL_ERROR
+            "InfiniRT installed headers/library were not found. "
+            "Set `INFINI_RT_ROOT` to an installed InfiniRT prefix, or set "
+            "`INFINI_RT_INCLUDE_DIRS` and `INFINI_RT_LIBRARY` explicitly. "
+            "For development builds from source, set `INFINI_RT_SOURCE_DIR`.")
+    endif()
+
+    set(_infinirt_resolved_include_dirs "")
+    foreach(_infinirt_include_dir IN LISTS _infinirt_include_hints)
+        if(_infinirt_include_dir AND EXISTS "${_infinirt_include_dir}")
+            list(APPEND _infinirt_resolved_include_dirs "${_infinirt_include_dir}")
+        endif()
+    endforeach()
+    list(APPEND _infinirt_resolved_include_dirs "${_INFINI_RT_INCLUDE_DIR}")
+    list(REMOVE_DUPLICATES _infinirt_resolved_include_dirs)
+
+    add_library(infinirt SHARED IMPORTED GLOBAL)
+    set_target_properties(infinirt PROPERTIES
+        IMPORTED_LOCATION "${_INFINI_RT_LIBRARY}"
+        INTERFACE_INCLUDE_DIRECTORIES "${_infinirt_resolved_include_dirs}"
+    )
+
+    set(INFINI_RT_INCLUDE_DIRS "${_infinirt_resolved_include_dirs}" CACHE STRING "InfiniRT include directories" FORCE)
+    set(INFINI_RT_LIBRARY "${_INFINI_RT_LIBRARY}" CACHE FILEPATH "InfiniRT library" FORCE)
+    message(STATUS "Using installed InfiniRT: ${_INFINI_RT_LIBRARY}")
+endfunction()
+
+_infiniops_configure_infinirt()
+
 add_subdirectory(src)
 
 if(NOT GENERATE_PYTHON_BINDINGS)

CONTRIBUTING.md

@@ -69,13 +69,21 @@ Branch names use the format `<type>/xxx-yyyy-zzzz`, where `<type>` matches the P
 Using Nvidia as an example:
 
 ```bash
-pip install .[dev] -C cmake.define.WITH_CPU=ON -C cmake.define.WITH_NVIDIA=ON
+pip install .[dev] \
+  -C cmake.define.INFINI_RT_ROOT=/path/to/infinirt-prefix \
+  -C cmake.define.WITH_CPU=ON \
+  -C cmake.define.WITH_NVIDIA=ON
 ```
 
-Auto-detection is supported for some platforms, so you can also simply run:
+`/path/to/infinirt-prefix` is the InfiniRT install prefix, typically the value
+used for InfiniRT's `CMAKE_INSTALL_PREFIX`. It should contain
+`include/infini/rt.h` and `lib/libinfinirt.so`.
+
+Auto-detection is supported for some platforms, so you can also let InfiniOps
+detect the device backends while still pointing it at the installed InfiniRT:
 
 ```bash
-pip install .[dev]
+pip install .[dev] -C cmake.define.INFINI_RT_ROOT=/path/to/infinirt-prefix
 ```
 
 > `[dev]` installs optional development dependencies (e.g. `pytest`) that are not needed for production but required for development and testing. After the first install, subsequent installs only need `pip install .`.
@@ -94,6 +102,7 @@ For routine development and pull requests, start with a smoke build plus the smo
 
 ```bash
 python -m pip install .[dev] --no-build-isolation --no-deps \
+  --config-settings=cmake.define.INFINI_RT_ROOT=/path/to/infinirt-prefix \
   --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON
 python -m pytest tests -m smoke -q
 ```
@@ -251,5 +260,7 @@ using T = TypeMapType<Device::Type::kCpu, ListGet<0>(list_tag)>;
 2. **Segmentation fault during tests**: Run `pytest -n 1`.
 3. **`Unknown CMake command "pybind11_add_module"`**: Install pybind11 with `pip install pybind11[global]`. See the [pybind11 installation guide](https://pybind11.readthedocs.io/en/stable/installing.html).
 4. **Auto-detection (`AUTO_DETECT_DEVICES`) fails**: Some machines may not expose devices in expected paths (e.g. `/dev/nvidia*`). Use explicit CMake defines instead (e.g. `-C cmake.define.WITH_NVIDIA=ON`).
-5. **`bash: pytest: command not found`**: Use `python -m pytest`.
-6. **`CUBLAS_STATUS_INVALID_VALUE` in `cublasSgemmStridedBatched`**: PyTorch version issue. Downgrade to `torch<=2.9.1`.
+5. **InfiniRT headers or library are not found**: Set `INFINI_RT_ROOT` to the InfiniRT install prefix, or set `INFINI_RT_INCLUDE_DIRS` and `INFINI_RT_LIBRARY` explicitly.
+6. **`ImportError: libinfinirt.so: cannot open shared object file`**: Reinstall InfiniOps from a build that was configured with the InfiniRT install prefix so the Python wheel can bundle the runtime library.
+7. **`bash: pytest: command not found`**: Use `python -m pytest`.
+8. **`CUBLAS_STATUS_INVALID_VALUE` in `cublasSgemmStridedBatched`**: PyTorch version issue. Downgrade to `torch<=2.9.1`.

README.md

@@ -8,21 +8,33 @@ InfiniOps is a high-performance, cross-platform operator library supporting mult
 - CMake 3.18+
 - Python 3.10+
 - Hardware-specific SDKs (e.g. CUDA Toolkit, MUSA Toolkit)
+- Installed InfiniRT headers and library
 
 ## Installation
 
-Install with pip (recommended):
+Install InfiniRT first, then build InfiniOps with the InfiniRT install prefix:
 
 ```bash
-pip install .
+pip install . -C cmake.define.INFINI_RT_ROOT=/path/to/infinirt-prefix
 ```
 
-This auto-detects available platforms on supported backends. To specify platforms explicitly:
+`/path/to/infinirt-prefix` is the directory passed to InfiniRT as
+`CMAKE_INSTALL_PREFIX`; it should contain `include/infini/rt.h` and
+`lib/libinfinirt.so`.
+
+InfiniOps auto-detects available platforms on supported backends. To specify
+platforms explicitly:
 
 ```bash
-pip install . -C cmake.define.WITH_CPU=ON -C cmake.define.WITH_NVIDIA=ON
+pip install . \
+  -C cmake.define.INFINI_RT_ROOT=/path/to/infinirt-prefix \
+  -C cmake.define.WITH_CPU=ON \
+  -C cmake.define.WITH_NVIDIA=ON
 ```
 
+The Python wheel installs the required InfiniRT shared library next to the
+InfiniOps extension so `import infini.ops` can load its runtime dependency.
+
 ### CMake Options
 
 | Option | Description | Default |
@@ -36,6 +48,10 @@ pip install . -C cmake.define.WITH_CPU=ON -C cmake.define.WITH_NVIDIA=ON
 | `-DWITH_CAMBRICON=[ON\|OFF]` | Compile the Cambricon implementation | OFF |
 | `-DWITH_ASCEND=[ON\|OFF]` | Compile the Ascend implementation | OFF |
 | `-DAUTO_DETECT_DEVICES=[ON\|OFF]` | Auto-detect available platforms | ON |
+| `-DINFINI_RT_ROOT=<path>` | InfiniRT install prefix containing `include/` and `lib/` | `$INFINI_RT_ROOT` |
+| `-DINFINI_RT_INCLUDE_DIRS=<paths>` | Explicit InfiniRT include directories | empty |
+| `-DINFINI_RT_LIBRARY=<path>` | Explicit path to `libinfinirt.so` | empty |
+| `-DINFINI_RT_SOURCE_DIR=<path>` | InfiniRT source tree for development builds | empty |
 
 If no accelerator options are provided and auto-detection finds nothing, `WITH_CPU` is enabled by default.
 

examples/runtime_api.h

@@ -2,6 +2,7 @@
 #define INFINI_OPS_EXAMPLES_RUNTIME_API_H_
 
 #include "device.h"
+#include "runtime.h"
 
 #ifdef WITH_NVIDIA
 #include "native/cuda/nvidia/ops/gemm/cublas.h"
@@ -23,8 +24,9 @@
 #include "native/ascend/ops/gemm/kernel.h"
 #include "native/ascend/runtime_.h"
 #elif WITH_CPU
+#include <infini/rt/cpu/runtime_.h>
+
 #include "native/cpu/ops/gemm/gemm.h"
-#include "native/cpu/runtime_.h"
 #else
 #error "One `WITH_*` backend must be enabled for the examples."
 #endif