confusing "cproj" errors for test-float on tilegx

confusing "cproj" errors for test-float on tilegx

[Chris Metcalf]

I'm not sure how to address these errors; can anyone suggest what might
be going wrong here?  test-float passed in the glibc 2.21 testing, so
maybe this is due to Joseph's change to test all rounding modes??
(Since tile doesn't have rounding modes, that doesn't seem that likely.)
But the last time cproj tests appear to have been touched was in
c980f2f4fe0f5 ("Fix cproj handling of (finite, NaN) arguments (bug 15531).")
but that was back in August 2013, certainly prior to the last time I
ran the test suite for glibc 2.21.

The double and idouble tests pass with the freshly-updated ULPs, but
the float and ifloat tests both fail with the appended output: cproj()
of various complex infinities should be yielding infinity, but instead
yields a very tiny positive number.

testing float (without inline functions)
Failure: Test: Real part of: cproj (inf + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf + +0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf - 0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf + +0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf - 0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (+0 + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (+0 - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-0 + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-0 - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf + 1.0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf - 1.0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf + 1.0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf - 1.0 i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (1.0 + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (1.0 - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-1.0 + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-1.0 - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf + qNaN i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (inf - qNaN i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf + qNaN i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-inf - qNaN i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (qNaN + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (qNaN - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-qNaN + inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf
Failure: Test: Real part of: cproj (-qNaN - inf i)
Result:
  is:          3.57331108402828353086e-43   0x1.fe000000000000000000p-142
  should be:   inf   inf

Test suite completed:
   9998 test cases plus 9169 tests for exception flags and
     9169 tests for errno executed.
   28 errors occurred.

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com

Re: confusing "cproj" errors for test-float on tilegx

[Joseph Myers]

This sounds like a compiler bug to me (whether in the compilation of 
cprojf or something else involved in the test).

-- 
Joseph S. Myers
joseph@codesourcery.com