Enable ORT TRT extension

Author: fpetrini15

CMakeLists.txt

@@ -1,4 +1,4 @@
-# Copyright (c) 2020-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2020-2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted provided that the following conditions
@@ -137,6 +137,15 @@ if(TRITON_ONNXRUNTIME_LIB_PATHS STREQUAL "")
   set(TRITON_ONNXRUNTIME_DOCKER_BUILD ON)
 endif()
 
+set(RHEL_BUILD OFF)
+if(LINUX)
+  file(STRINGS "/etc/os-release" DISTRO_ID_LIKE REGEX "ID_LIKE")
+  if(${DISTRO_ID_LIKE} MATCHES "rhel|centos")
+    set(RHEL_BUILD ON)
+  endif(${DISTRO_ID_LIKE} MATCHES "rhel|centos")
+endif(LINUX)
+set(TRITON_CORE_HEADERS_ONLY OFF)
+
 message(STATUS "Using Onnxruntime docker: ${TRITON_ONNXRUNTIME_DOCKER_BUILD}")
 
 if(NOT TRITON_ONNXRUNTIME_DOCKER_BUILD)
@@ -324,6 +333,9 @@ endif() # TRITON_ENABLE_ONNXRUNTIME_OPENVINO
 #
 if(TRITON_ONNXRUNTIME_DOCKER_BUILD)
   set(_GEN_FLAGS "")
+  if(${RHEL_BUILD} AND "${TRITON_BUILD_TENSORRT_HOME}" STREQUAL "")
+    set(TRITON_BUILD_TENSORRT_HOME "/usr/local/cuda/targets/${CMAKE_SYSTEM_PROCESSOR}-linux/")
+  endif(${RHEL_BUILD} AND "${TRITON_BUILD_TENSORRT_HOME}" STREQUAL "")
   if(NOT ${TRITON_BUILD_TARGET_PLATFORM} STREQUAL "")
     set(_GEN_FLAGS ${_GEN_FLAGS} "--target-platform=${TRITON_BUILD_TARGET_PLATFORM}")
   endif() # TRITON_BUILD_TARGET_PLATFORM

tools/gen_ort_dockerfile.py

@@ -105,7 +105,7 @@ def dockerfile_for_linux(output_file):
         df += """
 # The manylinux container defaults to Python 3.7, but some feature installation
 # requires a higher version.
-ARG PYVER=3.10
+ARG PYVER=3.12
 ENV PYTHONPATH=/opt/python/v
 RUN ln -sf /opt/python/cp${PYVER/./}* ${PYTHONPATH}
 
@@ -602,9 +602,8 @@ def preprocess_gpu_flags():
         "--target-platform",
         required=False,
         default=None,
-        help='Target for build, can be "linux", "windows" or "igpu". If not specified, build targets the current platform.',
+        help='Target for build, can be "linux", "windows", "rhel", or "igpu". If not specified, build targets the current platform.',
     )
-
     parser.add_argument(
         "--cuda-version", type=str, required=False, help="Version for CUDA."
     )
@@ -661,4 +660,4 @@ def preprocess_gpu_flags():
             FLAGS.ort_openvino = None
         dockerfile_for_windows(FLAGS.output)
     else:
-        dockerfile_for_linux(FLAGS.output)
+        dockerfile_for_linux(FLAGS.output)