[Offload] Rework handling for loading vendor runtimes

[jhuber6]

Summary:
We previously had multiple options for this, this patch replaces them
with LIBOMPTARGET_DLOPEN_PLUGINS= to be a list of plugins to
dynamically use. It defaults to everything right now. This ignores the
host plugin because the libffi dependency is going to be removed
soon hopefully in #91264.

[llvmbot]

@llvm/pr-subscribers-offload

@llvm/pr-subscribers-backend-amdgpu

Author: Joseph Huber (jhuber6)

Changes

Summary:
We previously had multiple options for this, this patch replaces them
with LIBOMPTARGET_DLOPEN_PLUGINS= to be a list of plugins to
dynamically use. It defaults to everything right now. This ignores the
host plugin because the libffi dependency is going to be removed
soon hopefully in #91264.

---

Full diff: https://github.com/llvm/llvm-project/pull/93073.diff

5 Files Affected:

• (modified) offload/CMakeLists.txt (+3)
• (modified) offload/cmake/Modules/LibomptargetGetDependencies.cmake (-8)
• (modified) offload/plugins-nextgen/amdgpu/CMakeLists.txt (+1-2)
• (modified) offload/plugins-nextgen/cuda/CMakeLists.txt (+2-2)
• (modified) openmp/docs/SupportAndFAQ.rst (+9)

diff --git a/offload/CMakeLists.txt b/offload/CMakeLists.txt
index c3dcebfb7301b..8bbdb94c0f654 100644
--- a/offload/CMakeLists.txt
+++ b/offload/CMakeLists.txt
@@ -146,6 +146,9 @@ endif()
 message(STATUS "Building the offload library with support for "
                "the \"${LIBOMPTARGET_PLUGINS_TO_BUILD}\" plugins")
 
+set(LIBOMPTARGET_DLOPEN_PLUGINS "${LIBOMPTARGET_PLUGINS_TO_BUILD}" CACHE STRING
+    "Semicolon-separated list of plugins to use 'dlopen' for runtime linking")
+
 set(LIBOMPTARGET_ENUM_PLUGIN_TARGETS "")
 foreach(plugin IN LISTS LIBOMPTARGET_PLUGINS_TO_BUILD)
   set(LIBOMPTARGET_ENUM_PLUGIN_TARGETS
diff --git a/offload/cmake/Modules/LibomptargetGetDependencies.cmake b/offload/cmake/Modules/LibomptargetGetDependencies.cmake
index e37b86b2a81fa..c296f7ea3863e 100644
--- a/offload/cmake/Modules/LibomptargetGetDependencies.cmake
+++ b/offload/cmake/Modules/LibomptargetGetDependencies.cmake
@@ -3,7 +3,6 @@
 #
 # libffi : required to launch target kernels given function and argument
 #          pointers.
-# CUDA : required to control offloading to NVIDIA GPUs.
 
 include (FindPackageHandleStandardArgs)
 
@@ -43,13 +42,6 @@ endif()
 find_package(FFI QUIET)
 set(LIBOMPTARGET_DEP_LIBFFI_FOUND ${FFI_FOUND})
 
-################################################################################
-# Looking for CUDA...
-################################################################################
-
-find_package(CUDAToolkit QUIET)
-set(LIBOMPTARGET_DEP_CUDA_FOUND ${CUDAToolkit_FOUND})
-
 ################################################################################
 # Looking for NVIDIA GPUs...
 ################################################################################
diff --git a/offload/plugins-nextgen/amdgpu/CMakeLists.txt b/offload/plugins-nextgen/amdgpu/CMakeLists.txt
index 7630e3788dae0..cb9ab8c240848 100644
--- a/offload/plugins-nextgen/amdgpu/CMakeLists.txt
+++ b/offload/plugins-nextgen/amdgpu/CMakeLists.txt
@@ -13,8 +13,7 @@ target_sources(omptarget.rtl.amdgpu PRIVATE src/rtl.cpp)
 target_include_directories(omptarget.rtl.amdgpu PRIVATE
                            ${CMAKE_CURRENT_SOURCE_DIR}/utils)
 
-option(LIBOMPTARGET_FORCE_DLOPEN_LIBHSA "Build with dlopened libhsa" ON)
-if(hsa-runtime64_FOUND AND NOT LIBOMPTARGET_FORCE_DLOPEN_LIBHSA)
+if(hsa-runtime64_FOUND AND NOT "amdgpu" IN_LIST LIBOMPTARGET_DLOPEN_PLUGINS)
   message(STATUS "Building AMDGPU plugin linked against libhsa")
   target_link_libraries(omptarget.rtl.amdgpu PRIVATE hsa-runtime64::hsa-runtime64)
 else()
diff --git a/offload/plugins-nextgen/cuda/CMakeLists.txt b/offload/plugins-nextgen/cuda/CMakeLists.txt
index fa5559c5e7dcb..3a3ed68670490 100644
--- a/offload/plugins-nextgen/cuda/CMakeLists.txt
+++ b/offload/plugins-nextgen/cuda/CMakeLists.txt
@@ -10,8 +10,8 @@ add_target_library(omptarget.rtl.cuda CUDA)
 
 target_sources(omptarget.rtl.cuda PRIVATE src/rtl.cpp)
 
-option(LIBOMPTARGET_FORCE_DLOPEN_LIBCUDA "Build with dlopened libcuda" ON)
-if(LIBOMPTARGET_DEP_CUDA_FOUND AND NOT LIBOMPTARGET_FORCE_DLOPEN_LIBCUDA)
+find_package(CUDAToolkit QUIET)
+if(CUDAToolkit_FOUND AND NOT "cuda" IN_LIST LIBOMPTARGET_DLOPEN_PLUGINS)
   message(STATUS "Building CUDA plugin linked against libcuda")
   target_link_libraries(omptarget.rtl.cuda PRIVATE CUDA::cuda_driver)
 else()
diff --git a/openmp/docs/SupportAndFAQ.rst b/openmp/docs/SupportAndFAQ.rst
index 9e6974dfbb13d..a158422befd07 100644
--- a/openmp/docs/SupportAndFAQ.rst
+++ b/openmp/docs/SupportAndFAQ.rst
@@ -454,6 +454,15 @@ Q: What command line options can I use for OpenMP?
 We recommend taking a look at the OpenMP 
 :doc:`command line argument reference <CommandLineArgumentReference>` page.
 
+Q: Can I build the offloading runtimes without CUDA or HSA?
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+By default, the offloading runtime will load the associated vendor runtime 
+during initialization rather than directly linking against them. This allows the 
+program to be built and run on many machine. If you wish to directly link 
+against these libraries, use the ``LIBOMPTARGET_DLOPEN_PLUGINS=""`` option to 
+suppress it for each plugin. The default value is every plugin enabled with 
+``LIBOMPTARGET_PLUGINS_TO_BUILD``.
+
 Q: Why is my build taking a long time?
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 When installing OpenMP and other LLVM components, the build time on multicore