[0/2] unnormal test case fixes

Message ID 20210104142443.3034204-1-siddhesh@sourceware.org


Siddhesh Poyarekar Jan. 4, 2021, 2:24 p.m. UTC
  The following two patches fix issues that Joseph pointed out in the
unnormal tests.

The first patch fixes the breakage caused by moving nan-pseudo-number.h
and in the process also fixes the macro condition used to limit
conditions under which the test is built.

The second patch limits the macro condition further so that the unnormal
tests build only for long double and Float64x on intel96.  Further, it
cleans up the test macro names and casts test arguments to the type
being tested.

Verified with build-many-glibcs.py for ia64, m68k, i686 and aarch64 and
with a native build and test for x86_64.  Florian is helping me with a
full build-many-glibcs.py run since I don't have compilers for all
architectures yet.

Siddhesh Poyarekar (2):
  Drop nan-pseudo-number.h usage from tests
  Limit unnormal tests to supported long double formats

 math/libm-test-driver.c                     | 17 +++++++++--------
 math/libm-test-fpclassify.inc               | 18 +++++++++---------
 math/libm-test-isinf.inc                    | 18 +++++++++---------
 math/libm-test-isnan.inc                    | 18 +++++++++---------
 math/libm-test-issignaling.inc              | 18 +++++++++---------
 math/libm-test-support.h                    |  2 ++
 math/test-float64x.h                        |  1 +
 math/test-ldouble.h                         |  1 +
 sysdeps/ieee754/ldbl-96/nan-pseudo-number.h |  4 +---
 sysdeps/x86/fpu/nan-pseudo-number.h         |  1 -
 10 files changed, 50 insertions(+), 48 deletions(-)