To replicate your environment, I followed these steps:
I have CMake 3.20 and Ninja 1.10.2 installed system-wide and in the PATH.
Then, I created D:\clang-mingw.cmake with the following contents:
set(CMAKE_C_COMPILER "C:/Program Files/LLVM/bin/clang.exe")
set(CMAKE_CXX_COMPILER "C:/Program Files/LLVM/bin/clang++.exe")
set(CMAKE_C_COMPILER_TARGET x86_64-windows-gnu)
set(CMAKE_CXX_COMPILER_TARGET x86_64-windows-gnu)
set(CMAKE_SYSROOT "C:/msys64/mingw64")
set(CMAKE_EXE_LINKER_FLAGS_INIT "-fuse-ld=lld")
set(CMAKE_MODULE_LINKER_FLAGS_INIT "-fuse-ld=lld")
set(CMAKE_SHARED_LINKER_FLAGS_INIT "-fuse-ld=lld")
This is called a toolchain file. These are used to inform CMake about non-standard compiler and library setups like this one.
Setting the CMAKE_<LANG>_COMPILER variables (naturally) points CMake to the compilers you want to use. In this case, that's Clang.
Setting CMAKE_<LANG>_COMPILER_TARGET configures Clang to use the x86_64-windows-gnu target when compiling. This is important for CMake's compiler detection and sanity checking steps.
Setting CMAKE_SYSROOT informs Clang where to find all of the standard libraries and headers (ie. those from mingw-w64).
Finally, we ensure LLD is used by adding -fuse-ld=lld to the linker flags used when linking executables, loadable modules, and shared libraries (but not static libraries, since no linker is needed, just the archiver). The CMAKE_{EXE,MODULE,SHARED}_LINKER_FLAGS_INIT variables are meant to be set from the toolchain for this purpose.
Then from a normal command prompt (i.e. not a Visual Studio command prompt or MSYS2 bash), I ran:
D:\SDL2-2.0.14>cmake -G Ninja -S . -B build -DCMAKE_TOOLCHAIN_FILE=D:/clang-mingw.cmake -DCMAKE_BUILD_TYPE=Debug -DDIRECTX=OFF -DSDL_TEST=OFF -DSDL_SHARED=OFF -DSDL_STATIC=ON
-- The C compiler identification is Clang 12.0.0
-- The CXX compiler identification is Clang 12.0.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files/LLVM/bin/clang.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files/LLVM/bin/clang++.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Could NOT find PkgConfig (missing: PKG_CONFIG_EXECUTABLE)
-- Performing Test HAVE_GCC_WALL
-- Performing Test HAVE_GCC_WALL - Success
[... output clipped ...]
The configure command I used does little more than replicate the SDL-specific settings you used in your question. It just selects Ninja as the build system generator (you could of course replace this with a different one if you so desired) and sets the toolchain file to the one above.
It then builds without errors.
D:\SDL2-2.0.14>cmake --build build
[... output clipped ...]
[163/164] C:\PROGRA~1\LLVM\bin\clang.exe --target=x86_64-windows-gnu --sysroot=C:/msys64/mingw64 -DUSING_GENERATED_CONFIG_H -Iinclude -I../include -idirafter "D:/SDL2-2.0.14/src/video/khronos" "-ID:/SDL2-2.0.14/src/hidapi/hidapi" -msse3 -msse2 -msse -m3dnow -mmmx -Wshadow -fvisibility=hidden -Wdeclaration-after-statement -Werror=declaration-after-statement -fno-strict-aliasing -Wall -g -MD -MT CMakeFiles/SDL2-static.dir/src/video/yuv2rgb/yuv_rgb.c.obj -MF CMakeFiles\SDL2-static.dir\src\video\yuv2rgb\yuv_rgb.c.obj.d -o CMakeFiles/SDL2-static.dir/src/video/yuv2rgb/yuv_rgb.c.obj -c ../src/video/yuv2rgb/yuv_rgb.c
[164/164] cmd.exe /C "cd . && "C:\Program Files\CMake\bin\cmake.exe" -E rm -f libSDL2d.a && C:\PROGRA~1\LLVM\bin\llvm-ar.exe qc libSDL2d.a @CMakeFiles\SDL2-static.rsp && cd ."
