Fix complex types declared using mode TC

[eandrews]

Summary

This patch reverts incorrect IR - introduced in commit d11ec6f67e45 (https://reviews.llvm.org/rGd11ec6f67e45c630ab87bfb6010dcc93e89542fc) - for complex types declared using attribute((mode(TC))). TC mode corresponds to __float128 type.

[rjmccall]

For posterity in case someone tracks down this review: TC corresponds to an unspecified 128-bit format, which on some targets is a double-double format (like __ibm128_t) and on others is float128_t. The bug in the previous patch is that long double is only safe to use when it's known to be one of those formats.

Patch LGTM.

[qiucf]

Thanks. This issue looks a little bit like D109943.

[eandrews]

In D112975#3101720, @rjmccall wrote:

For posterity in case someone tracks down this review: TC corresponds to an unspecified 128-bit format, which on some targets is a double-double format (like __ibm128_t) and on others is float128_t. The bug in the previous patch is that long double is only safe to use when it's known to be one of those formats.

Patch LGTM.

Thanks for the note. I'll update the commit message when I check this in.

clang/lib/Basic/TargetInfo.cpp

case 128:
     // The caller explicitly asked for an IEEE compliant type but we still
     // have to check if the target supports it.
     if (ExplicitType == FloatModeKind::Float128)
       return hasFloat128Type() ? FloatModeKind::Float128
                                : FloatModeKind::NoFloat;
     if (ExplicitType == FloatModeKind::Ibm128)
       return hasIbm128Type() ? FloatModeKind::Ibm128
                              : FloatModeKind::NoFloat;
-    if (ExplicitType == FloatModeKind::LongDouble)
-      return ExplicitType;
+    if (&getLongDoubleFormat() == &llvm::APFloat::PPCDoubleDouble() ||
+        &getLongDoubleFormat() == &llvm::APFloat::IEEEquad())
+      return FloatModeKind::LongDouble;
+    if (hasFloat128Type())
+      return FloatModeKind::Float128;
     break;
   }
 
   return FloatModeKind::NoFloat;
 }
 
 /// getTypeAlign - Return the alignment (in bits) of the specified integer type
 /// enum. For example, SignedInt -> getIntAlign().

clang/test/CodeGenCXX/complex128.cpp

+// RUN: %clang_cc1 -emit-llvm -triple x86_64-unknown-linux-gnu %s -o - | FileCheck %s
+
+// Define __complex128 type corresponding to __float128 (as in GCC headers).
+typedef _Complex float __attribute__((mode(TC))) __complex128;
+
+void check() {
+  // CHECK: alloca { fp128, fp128 }
+  __complex128 tmp;
+}