Installing ITensor with g++ and Accelerate on macos fails

[mtfishman]

Installing ITensor with g++ and Accelerate on macos fails. This was pointed out by Daniel, and I just reproduced it. Normally, g++ is just an alias clang++, but if I explicitly install g++ (for example with Homebrew), and then try to compile ITensor with that g++ compiler and Accelerate, then the ITensor build process fails at lapack_wrap.cc with errors like:

Compiling itensor/tensor/lapack_wrap.cc with optimizations
In file included from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Headers/../Frameworks/vecLib.framework/Headers/vecLib.h:25,
                 from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Headers/Accelerate.h:20,
                 from /Users/mfishman/software/itensor/itensor/tensor/lapack_wrap.h:84,
                 from tensor/lapack_wrap.cc:16:
/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Frameworks/vecLib.framework/Headers/vBasicOps.h: In function 'vUInt16 vU16HalfMultiply(vUInt16, vUInt16)':
/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Frameworks/vecLib.framework/Headers/vBasicOps.h:379:84: note: use '-flax-vector-conversions' to permit conversions between vectors with differing element types or numbers of subparts
  379 |   vUInt16   __vbasicops_vB) { return _mm_mullo_epi16(__vbasicops_vA, __vbasicops_vB); }
      |                                                                                    ^
/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Frameworks/vecLib.framework/Headers/vBasicOps.h:379:54: error: cannot convert 'vUInt16' {aka '__vector(8) short unsigned int'} to '__m128i' {aka '__vector(2) long long int'}
  379 |   vUInt16   __vbasicops_vB) { return _mm_mullo_epi16(__vbasicops_vA, __vbasicops_vB); }
      |                                                      ^~~~~~~~~~~~~~
      |                                                      |
      |                                                      vUInt16 {aka __vector(8) short unsigned int}
In file included from /usr/local/Cellar/gcc/9.2.0_3/lib/gcc/9/gcc/x86_64-apple-darwin18/9.2.0/include/xmmintrin.h:1255,
                 from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Frameworks/vecLib.framework/Headers/vecLibTypes.h:92,
                 from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Headers/../Frameworks/vecLib.framework/Headers/vecLib.h:21,
                 from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Accelerate.framework/Headers/Accelerate.h:20,
                 from /Users/mfishman/software/itensor/itensor/tensor/lapack_wrap.h:84,
                 from tensor/lapack_wrap.cc:16:
...

Maybe Accelerate uses features that are not available in g++? It may be a more fundamental problem, but if so we may want to point this out in the installation instructions to warn people about trying this combination.

[mtfishman]

Just to confirm, compiling ITensor with g++ and openblas on macos works for me, so this is an Accelerate specific problem.

[XiaoTailong]

g++ with openblas and clang++ with (macos) Accelerate framework. Match is necessary.