Is __attribute__ ((__mode__ (__TF__))) treated as x87 long double attribute?

[hjl-tools]

Clang parseModeAttrArg has

   case 'K': // KFmode - IEEE quad precision (__float128)
      ExplicitType = FloatModeKind::Float128;
      DestWidth = Str[1] == 'I' ? 0 : 128;
      break;
    case 'T':
      ExplicitType = FloatModeKind::LongDouble;
      DestWidth = 128;
      break;
    case 'I':
      ExplicitType = FloatModeKind::Ibm128;
      DestWidth = Str[1] == 'I' ? 0 : 128;
      break;
    }
    if (Str[1] == 'F') {
      IntegerMode = false;
    } else if (Str[1] == 'C') {
      IntegerMode = false;
      ComplexMode = true;
    } else if (Str[1] != 'I') {
      DestWidth = 0;
    }
    break;

On x86, TFmode in GCC is IEEE _Float128, not x87 long double. Is attribute ((mode (TF))) treated
as x87 long double attribute?

[llvmbot]

@llvm/issue-subscribers-clang-codegen

Author: None (hjl-tools)

Clang parseModeAttrArg has

   case 'K': // KFmode - IEEE quad precision (__float128)
      ExplicitType = FloatModeKind::Float128;
      DestWidth = Str[1] == 'I' ? 0 : 128;
      break;
    case 'T':
      ExplicitType = FloatModeKind::LongDouble;
      DestWidth = 128;
      break;
    case 'I':
      ExplicitType = FloatModeKind::Ibm128;
      DestWidth = Str[1] == 'I' ? 0 : 128;
      break;
    }
    if (Str[1] == 'F') {
      IntegerMode = false;
    } else if (Str[1] == 'C') {
      IntegerMode = false;
      ComplexMode = true;
    } else if (Str[1] != 'I') {
      DestWidth = 0;
    }
    break;

On x86, TFmode in GCC is IEEE _Float128, not x87 long double. Is attribute ((mode (TF))) treated
as x87 long double attribute?

[efriedma-quic]

TargetInfo::getRealTypeByWidth controls the lowering here. It should be long double on targets where long double is IEEE 128-bit, and __float128 on targets where it isn't.

(You can always write a test like typedef double x __attribute((mode(TF)));typedef long double x; to check what actually happens on a specific target.)