[LLVMbugs] [Bug 372] NEW: [llvmgcc/libc] isnan(x) function should turn into llvm.isunordered(X, 0)

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.cs.uiuc.edu/bugs/show_bug.cgi?id=372

           Summary: [llvmgcc/libc] isnan(x) function should turn into
                    llvm.isunordered(X, 0)
           Product: tools
           Version: trunk
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: enhancement
          Priority: P2
         Component: llvm-gcc
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sabre at nondot.org


glibc is currently turning calls to isnan into calls to __isnan* functions
defined in glibc.  Since this is a single machine instruction and we can
generate it inline, it would be nice to be able to do so!

Unfortunately my fixincludes-fu is quite weak.  In particular, I am able to make
the fix, but I get all kinds of wierd errors from the resultant math.h
indicating that something bad happened.  The strange thing about this is that
when I diff the resultant math.h to the original one, only my fix has been
applied.  Because of this I really don't know what is going on, so I must punt
until someone with more fu can look at this.

Here's my fixincludes hack, which DOES work, but breaks things when math.h is fixed:
/* This transforms the glibc implementation of isnan(X) into
 * __builtin_isunordered(X,0) instead of calling __isnan*.
 */
fix = {
  hackname  = llvm_isnan_glibc;
  files     = math.h;
  select    = "^# ifdef __NO_LONG_DOUBLE_MATH\n(#[ \t]*define isnan)";
  c_fix     = format;
  c_fix_arg = "# ifdef __llvm__\n"
              "#   define isnan((X)) __builtin_isunordered(X, 0)\n"
              "# elif defined(__NO_LONG_DOUBLE_MATH)\n%1";
  test_text = 
  "# ifdef __NO_LONG_DOUBLE_MATH\n"
  "#  define isnan(x) \\\n";
};


This hacks on this macro in the glibc math.h file:

# ifdef __NO_LONG_DOUBLE_MATH
#  define isnan(x) \
     (sizeof (x) == sizeof (float) ? __isnanf (x) : __isnan (x))
# else
#  define isnan(x) \
     (sizeof (x) == sizeof (float)                                            \
      ? __isnanf (x)                                                          \
      : sizeof (x) == sizeof (double)                                         \
      ? __isnan (x) : __isnanl (x))
# endif

-Chris



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.