[PATCH v2 0/5] Improve test when building with clang

[Adhemerval Zanella]

The math change fixes most regressions when building with clang. On
x86_64, there are only three regressions (math/test-ldouble-scalb,
math/test-tgmath, and math/test-tgmath2), while on aarch64, there are
two (math/test-tgmath and math/test-tgmath2).

Changes from v1:
- Add regression tests for fpclassify and isinf fixes.
- Use indirection on tst-vfprintf-user-type.
- Add clang bug report for the ldlb-128 powl change.
- Enable tst-atomic on clang.

Adhemerval Zanella (5):
  x86: Do not use __builtin_fpclassify for _Float64x/long double
  x86: Do not use __builtin_isinf_sign for _Float64x/long double
  stdio: Fix tst-vfprintf-user-type on clang
  math: Use math_opt_barrier on ldbl-128 powl underflow/overflow
    handling
  misc: Enable tst-atomic for clang

 bits/fp-builtin-denormal.h              | 32 +++++++++
 math/Makefile                           |  1 +
 math/math.h                             | 94 ++++++++++++++++++++++++-
 misc/tst-atomic.c                       |  9 ---
 stdio-common/tst-vfprintf-user-type.c   |  4 +-
 sysdeps/ieee754/ldbl-128/e_powl.c       |  6 +-
 sysdeps/x86/bits/fp-builtin-denormal.h  | 29 ++++++++
 sysdeps/x86/fpu/Makefile                |  1 +
 sysdeps/x86/fpu/test-builtin-denormal.c | 57 +++++++++++++++
 9 files changed, 218 insertions(+), 15 deletions(-)
 create mode 100644 bits/fp-builtin-denormal.h
 create mode 100644 sysdeps/x86/bits/fp-builtin-denormal.h
 create mode 100644 sysdeps/x86/fpu/test-builtin-denormal.c

-- 
2.43.0