Fix CUDA build with clang++. #1997

Job	Run time
Test wasm (wasm32-unknown-unknown)	20s
Test build-std (aarch64-tvos)	1m 47s
Test build-std (aarch64-tvos-sim)	1m 53s
Test (stable)	1m 34s
Clippy	14s
Test build-std (x86_64-tvos)	2m 2s
Test (beta)	1m 22s
Test CUDA support	42s
Test (nightly)	1m 20s
Test (linux32)	1m 59s
MSRV (ubuntu-latest)	27s
Rustfmt	13s
Test (macos)	1m 53s
MSRV (windows-latest)	38s
semver-checks	1m 49s
Test (aarch64-macos)	1m 34s
MSRV (macos-latest)	31s
Test (aarch64-ios)	1m 29s
Test wasm32-wasip1-thread	31s
Test (win32)	1m 46s
Test (win64)	1m 44s
Test (mingw32)	1m 45s
Test (mingw64)	1m 45s
Test (windows-2019)	1m 47s
Test (windows-aarch64, windows-latest, stable, aarch64-pc-windows-msvc, --no-run)	1m 46s
Test (windows-clang, windows-2019, stable, x86_64-pc-windows-msvc, clang, clang++)	1m 44s
Test (windows-clang-cl, windows-2019, stable, x86_64-pc-windows-msvc, clang-cl, clang-cl)	1m 45s
	36m 20s

Provide feedback