master_june24: __CUDACC__ macros prevent HIP support

[valassi]

Another issue introduced in #830 and being reviewed in #882.

Macro CUDACC should not be used when the code is meant to support also HIP

https://github.com/valassi/madgraph4gpu/blob/8e312bc02d9d072615fcb1052b5db54754498517/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc#L345

#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
      if( channelIds != 0 )
      {
        channelids_sv = CID_ACCESS::kernelAccessConst( channelIds );
#if defined __CUDACC__ or !defined MGONGPU_CPPSIMD
        if( channelids_sv == 1 ) numerators_sv += cxabs2( amp_sv[0] );
        denominators_sv += cxabs2( amp_sv[0] );
#else
        for( int i = 0; i < neppV; ++i )
        {
          if( channelids_sv[i] == 1 ) numerators_sv += cxabs2( amp_sv[0] );
          denominators_sv += cxabs2( amp_sv[0] );
        }
#endif
      }
#endif

[roiser]

HI @valassi , thank you for spotting this omission, shall I fix it or do you want to do it?

[valassi]

I'll do it, it's faster

[valassi]

This is fixed by this commit eeb42ee in PR #882

Closing

[valassi]

channelids_sv = CID_ACCESS::kernelAccessConst( channelIds );
#if defined CUDACC or !defined MGONGPU_CPPSIMD
if( channelids_sv == 1 ) numerators_sv += cxabs2( amp_sv[0] );

PS These specific lines have disappeared completely (because the SIMD implementation was wrong #894
But there were other places where 830 introduced CUDACC instead of MGONGPUCPP_GPUIMPL