Make offload-compress the default (#1704)

* Make offload-compress the default
* Add guard for --offload-compress since it was introduced in ROCm 6.2
* Address some of Nilesh's feedback.
* Reorganize for code cleanliness
* Improve comment
* Compress gpu code at link and compile time

CMakeLists.txt

@@ -819,6 +819,15 @@ endif()
 if (HAVE_PARALLEL_JOBS)
   target_compile_options(rccl PRIVATE -parallel-jobs=12)
 endif()
+
+if (ROCM_VERSION VERSION_GREATER_EQUAL "60200") 
+  target_compile_options(rccl PRIVATE --offload-compress)    # Compress GPU code at compile time.
+  target_link_libraries(rccl PRIVATE --offload-compress)     # Compress GPU code at link time.
+  message(STATUS "--offload-compress enabled - ROCm version >= 6.2.0")
+else()
+  message(STATUS "--offload-compress disabled - ROCm version < 6.2.0")
+endif()
+
 target_compile_options(rccl PRIVATE -Werror=uninitialized)
 target_compile_options(rccl PRIVATE -Werror=sometimes-uninitialized)
 target_compile_options(rccl PRIVATE -Wall)