Reduce the size of the cugraph libraries (#1503)

By explicitly telling nvcc's fatbin pass to always compress device code
we can ensure that our binaries are the smallest possible size.

See rapidsai/cudf#7583 for additional context.

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - Rick Ratzel (https://github.com/rlratzel)
  - Brad Rees (https://github.com/BradReesWork)

URL: #1503

cpp/CMakeLists.txt

@@ -116,6 +116,7 @@ set(FAISS_GPU_ARCHS "${FAISS_GPU_ARCHS} -gencode arch=compute_${ptx},code=comput
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-extended-lambda --expt-relaxed-constexpr")
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Werror=cross-execution-space-call -Wno-deprecated-declarations -Xptxas --disable-warnings")
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler -Wall,-Wno-error=sign-compare,-Wno-error=unused-but-set-variable")
+set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xfatbin=-compress-all")
 
 # Option to enable line info in CUDA device compilation to allow introspection when profiling /
 # memchecking