Changes to compile with Clang17

[czentgr]

The changes fix:

1. cast issue with the integer max to a floating point

    implicit conversion from 'long' to 'double' changes value from
       9223372036854775807 to 9223372036854775808
    implicit conversion from '__int128' to 'float' changes value from
       170141183460469231731687303715884105727
       to 1.70141183460469231731687E+38

2. template instantiation of simple vectors with complex types in tests

/root/oss/velox/./velox/vector/SimpleVector.h:370:17: error: invalid operands to binary expression ('const facebook::velox::Array<int>' and 'const facebook::velox::Array<int>')
  370 |     return left < right ? -1 : left == right ? 0 : 1;
      |            ~~~~ ^ ~~~~~
/root/oss/velox/./velox/vector/SimpleVector.h:163:19: note: in instantiation of member function 'facebook::velox::SimpleVector<facebook::velox::Array<int>>::comparePrimitiveAsc' requested here
  163 |     auto result = comparePrimitiveAsc(thisValue, otherValue);
      |        

3. Clang needs lib atomic symbols to be resolved for a number of executables.

/opt/rh/gcc-toolset-13/root/usr/lib/gcc/x86_64-redhat-linux/13/../../../../bin/ld: velox/exec/libvelox_exec.a(Task.cpp.o): in function `std::__atomic_base<unsigned int>::is_lock_free() const':
/opt/rh/gcc-toolset-13/root/usr/lib/gcc/x86_64-redhat-linux/13/../../../../include/c++/13/bits/atomic_base.h:460: undefined reference to `__atomic_is_lock_free'
/opt/rh/gcc-toolset-13/root/usr/lib/gcc/x86_64-redhat-linux/13/../../../../bin/ld: velox/exec/libvelox_exec.a(Task.cpp.o): in function `std::__atomic_base<long>::is_lock_free() const':
/opt/rh/gcc-toolset-13/root/usr/lib/gcc/x86_64-redhat-linux/13/../../../../include/c++/13/bits/atomic_base.h:460: undefined reference to `__atomic_is_lock_free'clang++: error: linker command failed with exit code 1 (use -v to see invocation)ninja: build stopped: subcommand failed.

[czentgr]

@majetideepak FYI

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	b30bcf8
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/668dbff5d838790008976e6d

[czentgr]

Remove old code.

[czentgr]

Need to work on this a bit more. Some of the deps built don't pick up the change. For example arrow:

16:48:46  #12 35.16 [203/860] Performing configure step for 'arrow_ep'
16:48:46  #12 35.16 -- Building using CMake version: 3.28.3
16:48:46  #12 35.16 -- The C compiler identification is GNU 12.2.1
16:48:46  #12 35.16 -- The CXX compiler identification is GNU 12.2.1

I guess the variables do not get inherited into bundled build with their own projects. Probably needs a few more changes.

Dependencies (from the setup scripts are also built with gcc).

[majetideepak]

Thanks, @czentgr . Changes look good to me.
We should add a CI test as well to ensure support is not broken.
@kgpai any thoughts on this?

[majetideepak]

I think we have to support building deps with clang17 as well.

[czentgr]

So recently Centos9 updated their clang from version 17 to 18 and it causes additional build breaks (around arrays that use a variable input size and possibly more). As a result I've installed clang15 (the only other easily available version from the repo) and built with that.
As a result of using clang15 I also found out how to make it available in ccache (not done by default after installing the package). Cmake doesn't seem to have a good way of changing the default compiler especially in external projects. Cmake picks up the CC and CXX environment variables - and this would apply to subprojects made when building.

Overall improvement of 99 TPCDS queries is about 8% compared to gcc12. Some queries are slightly slower and some are faster. Spread is 0.65 to 3.4 ratio. That is the worst query took almost twice as long (this one is a subsecond query) and best query improved by factor 3ish (19s down to 5.5s) .