Is `CUDA_ENFORCE_LOWER` still needed?

[jakirkham]

Previously CUDA_ENFORCE_LOWER was used with CUDA_VER_10_1_UP, which ensured the CUDA version was at least 10.1. Here was the explanation on how to use CUDA_ENFORCE_LOWER:

raft/cpp/include/raft/core/cusparse_macros.hpp

Lines 15 to 25 in e74a688

	// Notes:
	//(1.) CUDA_VER_10_1_UP aggregates all the CUDA version selection logic;
	//(2.) to enforce a lower version,
	//
	//`#define CUDA_ENFORCE_LOWER
	// #include <raft/sparse/detail/cusparse_wrappers.h>`
	//
	// (i.e., before including this header)
	//
	#define CUDA_VER_10_1_UP (CUDART_VERSION >= 10010)
	#define CUDA_VER_12_4_UP (CUDART_VERSION >= 12040)

Also here was a usage example:

raft/cpp/include/raft/spectral/detail/matrix_wrappers.hpp

Lines 216 to 218 in e74a688

	#if not defined CUDA_ENFORCE_LOWER and CUDA_VER_10_1_UP
	auto size_x = transpose ? nrows_ : ncols_;
	auto size_y = transpose ? ncols_ : nrows_;

However support for CUDA 9, 10 (and even CUDA 11) have been dropped. Meaning CUDA_VER_10_1_UP is always true and thus is no longer used. As a result, recent code looks like this:

raft/cpp/include/raft/spectral/detail/matrix_wrappers.hpp

Lines 217 to 219 in 6725e0f

	#if not defined CUDA_ENFORCE_LOWER
	auto size_x = transpose ? nrows_ : ncols_;
	auto size_y = transpose ? ncols_ : nrows_;

---

Given this, is CUDA_ENFORCE_LOWER still needed?

• If so, what does it mean for CUDA 12+ code?
• If not, what should happen to the code it guards?