Shorten the torch_cpu_library comments and give aten_bridge its real reason

The same three-line comment sat at five link sites. It explained why the
line does not link torch, but not why it does not link plain torch_cpu,
and "in a CUDA build" read as our own CUDA option rather than how
PyTorch was built.

Replace it with one line at each site that names both reasons: torch
adds libtorch_cuda with a CUDA build of PyTorch, and torch_cpu adds MKL
to the link. At aten_bridge the old wording was also misleading. That
target is a static library, so it is never loaded on its own. The link
matters because a static library passes it on to _training_lib, so the
comment now says that.

Author: shoumikhin

CMakeLists.txt

@@ -1450,9 +1450,8 @@ if(EXECUTORCH_BUILD_PYBIND)
       aten_bridge PUBLIC C10_USING_CUSTOM_GENERATED_MACROS
     )
     target_compile_options(aten_bridge PUBLIC ${_pybind_compile_options})
-    # torch_cpu_library is the CPU half of torch. Linking torch would also force
-    # libtorch_cuda into DT_NEEDED in a CUDA build, and a CPU-only torch could
-    # then not load this library.
+    # aten_bridge is static, so this link reaches _training_lib, where torch
+    # adds libtorch_cuda with a CUDA build of PyTorch and torch_cpu adds MKL.
     if(EXECUTORCH_BUILD_SHARED)
       target_link_libraries(
         aten_bridge PRIVATE torch_cpu_library c10 executorch_shared

extension/llm/custom_ops/CMakeLists.txt

@@ -230,9 +230,7 @@ if(EXECUTORCH_BUILD_KERNELS_LLM_AOT)
   endif()
   executorch_target_link_shared_runtime(custom_ops_aot_lib)
 
-  # torch_cpu_library is the CPU half of torch. Linking torch would also force
-  # libtorch_cuda into DT_NEEDED in a CUDA build, and a CPU-only torch could
-  # then not load this library.
+  # torch adds libtorch_cuda with a CUDA build of PyTorch; torch_cpu adds MKL.
   target_link_libraries(
     custom_ops_aot_lib PUBLIC cpublas torch_cpu_library extension_tensor
                               extension_threadpool

extension/llm/runner/CMakeLists.txt

@@ -122,9 +122,7 @@ if(EXECUTORCH_BUILD_PYBIND)
   )
   # Link with the extension_llm_runner library and its dependencies.
   #
-  # torch_cpu_library is the CPU half of torch. Linking torch would also force
-  # libtorch_cuda into DT_NEEDED in a CUDA build, and a CPU-only torch could
-  # then not load this library.
+  # torch adds libtorch_cuda with a CUDA build of PyTorch; torch_cpu adds MKL.
   target_link_libraries(
     _llm_runner PRIVATE extension_llm_runner tokenizers::tokenizers
                         portable_lib ${TORCH_PYTHON_LIBRARY} torch_cpu_library

extension/training/CMakeLists.txt

@@ -53,9 +53,7 @@ if(EXECUTORCH_BUILD_PYBIND)
   # is resolved from it rather than from the whole-archive-forcing static
   # `executorch` target, so this module shares the one backend registry.
   #
-  # torch_cpu_library is the CPU half of torch. Linking torch would also force
-  # libtorch_cuda into DT_NEEDED in a CUDA build, and a CPU-only torch could
-  # then not load this library.
+  # torch adds libtorch_cuda with a CUDA build of PyTorch; torch_cpu adds MKL.
   if(EXECUTORCH_BUILD_SHARED)
     set(_pybind_training_dep_libs ${TORCH_PYTHON_LIBRARY} etdump aten_bridge
                                   torch_cpu_library extension_training

tools/cmake/Codegen.cmake

@@ -334,9 +334,7 @@ function(gen_custom_ops_aot_lib)
   set_target_properties(${GEN_LIB_NAME} PROPERTIES CXX_STANDARD 20)
   target_compile_definitions(${GEN_LIB_NAME} PRIVATE USE_ATEN_LIB=1)
   include_directories(${TORCH_INCLUDE_DIRS})
-  # torch_cpu_library is the CPU half of torch. Linking torch would also force
-  # libtorch_cuda into DT_NEEDED in a CUDA build, and a CPU-only torch could
-  # then not load this library.
+  # torch adds libtorch_cuda with a CUDA build of PyTorch; torch_cpu adds MKL.
   target_link_libraries(${GEN_LIB_NAME} PRIVATE torch_cpu_library)
 
   executorch_target_link_options_shared_lib(${GEN_LIB_NAME})