[Bug math/16576] New: PowerPC: glibc attempts to use fsqrt instruction on processors that do not implement it

[Bug math/16576] New: PowerPC: glibc attempts to use fsqrt instruction on processors that do not implement it

[adrian at humboldt dot co.uk]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

            Bug ID: 16576
           Summary: PowerPC: glibc attempts to use fsqrt instruction on
                    processors that do not implement it
           Product: glibc
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: adrian at humboldt dot co.uk

In sysdeps/powerpc/fpu/math_private.h, __CPU_HAS_FSQRT is set true on all 64
bit PowerPC processors. Recent 64-bit PowerPC processors (e5500 core) from
Freescale do not implement this instruction, causing any application using
sqrt() to fail.

  28 # if __WORDSIZE == 64 || defined _ARCH_PWR4
  29 #  define __CPU_HAS_FSQRT 1
  30 # else
  31 #  define __CPU_HAS_FSQRT ((GLRO(dl_hwcap) & PPC_FEATURE_64) != 0)
  32 # endif

There is a patch for this in OpenEmbedded, but it provides a compile time fix
which is less useful for Linux distributions. A runtime probe for the existence
of the fsqrt instruction would be of wider use.

http://lists.openembedded.org/pipermail/openembedded-core/2012-September/068696.html

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

[Bug math/16576] [powerpc] glibc attempts to use fsqrt instruction on processors that do not implement it

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

Joseph Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
               Host|                            |powerpc*-*-*
            Summary|PowerPC: glibc attempts to  |[powerpc] glibc attempts to
                   |use fsqrt instruction on    |use fsqrt instruction on
                   |processors that do not      |processors that do not
                   |implement it                |implement it

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

[Bug math/16576] PowerPC: glibc attempts to use fsqrt instruction on processors that do not implement it

[joseph at codesourcery dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=16576

--- Comment #1 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
The correct compile-time test is _ARCH_PPCSQ (that's based on the same 
condition GCC uses to decide whether to generate fsqrt instructions 
itself).  Not __WORDSIZE == 64, not _ARCH_PWR4.

Has Freescale (or anyone else) added an HWCAP / HWCAP2 bit to the kernel 
to indicate whether fsqrt is available at runtime?  That's what's needed 
for any correct runtime probe (as opposed to the present incorrect probe 
based on PPC_FEATURE_64).

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

[Bug math/16576] [powerpc] glibc attempts to use fsqrt instruction on processors that do not implement it

[adrian at humboldt dot co.uk]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

--- Comment #2 from Adrian Cox <adrian at humboldt dot co.uk> ---
Unfortunately there is no HWCAP bit for the presence (or absence) of fsqrt.

The best I can find in the kernel source is that fsqrt is available if
PPC_FEATURE_64 is set and PPC_FEATURE_BOOKE is not set.

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

[Bug math/16576] [powerpc] glibc attempts to use fsqrt instruction on processors that do not implement it

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |security-

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

[Bug math/16576] [powerpc] glibc attempts to use fsqrt instruction on processors that do not implement it

[azanella at linux dot vnet.ibm.com]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

Adhemerval Zanella Netto <azanella at linux dot vnet.ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |azanella at linux dot vnet.ibm.com

--- Comment #3 from Adhemerval Zanella Netto <azanella at linux dot vnet.ibm.com> ---
Fixed upstream by 08cee2a464f614a6d4275b5af6c52481f1aa16e6.

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

[Bug math/16576] [powerpc] glibc attempts to use fsqrt instruction on processors that do not implement it

[azanella at linux dot vnet.ibm.com]

https://sourceware.org/bugzilla/show_bug.cgi?id=16576

Adhemerval Zanella Netto <azanella at linux dot vnet.ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #4 from Adhemerval Zanella Netto <azanella at linux dot vnet.ibm.com> ---
Fixed upstream by 08cee2a464f614a6d4275b5af6c52481f1aa16e6.

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