[Bug math/19375] New: powerpc: incorrect results for POWER7 logb with negative subnormals

["adhemerval.zanella at linaro dot org" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=19375

            Bug ID: 19375
           Summary: powerpc: incorrect results for POWER7 logb with
                    negative subnormals
           Product: glibc
           Version: 2.23
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: adhemerval.zanella at linaro dot org
  Target Milestone: ---

The 'math/test-double' and 'math/test-double-finite' are both failing on POWER7
or newer machine with:

testing double (finite-math-only)
Failure: Test: logb (-min_subnorm_value)
Result:
 is:         -1.0110000000000000e+03  -0x1.f980000000000p+9
 should be:  -1.0740000000000000e+03  -0x1.0c80000000000p+10
 difference:  6.3000000000000000e+01   0x1.f800000000000p+5
 ulp       :  277076930199552.0000
 max.ulp   :  0.0000
Failure: Test: logb_downward (-min_subnorm_value)
Result:
 is:         -1.0110000000000000e+03  -0x1.f980000000000p+9
 should be:  -1.0740000000000000e+03  -0x1.0c80000000000p+10
 difference:  6.3000000000000000e+01   0x1.f800000000000p+5
 ulp       :  277076930199552.0000
 max.ulp   :  0.0000
Failure: Test: logb_towardzero (-min_subnorm_value)
Result:
 is:         -1.0110000000000000e+03  -0x1.f980000000000p+9
 should be:  -1.0740000000000000e+03  -0x1.0c80000000000p+10
 difference:  6.3000000000000000e+01   0x1.f800000000000p+5
 ulp       :  277076930199552.0000
 max.ulp   :  0.0000
Failure: Test: logb_upward (-min_subnorm_value)
Result:
 is:         -1.0110000000000000e+03  -0x1.f980000000000p+9
 should be:  -1.0740000000000000e+03  -0x1.0c80000000000p+10
 difference:  6.3000000000000000e+01   0x1.f800000000000p+5
 ulp       :  277076930199552.0000
 max.ulp   :  0.0000

This is due a issue on 'sysdeps/powerpc/power7/fpu/s_logb.c' for handling
subnormal negative numbers (it does not clear the sign bit before issue the
leading 0-bits builtin).

-- 
You are receiving this mail because:
You are on the CC list for the bug.