diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S b/sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S index 4f1c17680d..3648e4a69f 100644 --- a/sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S +++ b/sysdeps/powerpc/powerpc32/power4/fpu/w_sqrt.S @@ -25,20 +25,20 @@ sets the appropriate floating point exceptions. Extended checking is only needed to set errno (via __kernel_standard) if the input value is negative. - + The fsqrt will set FPCC and FU (Floating Point Unordered or NaN to indicated that the input value was negative or NaN. Use Move to Condition Register from FPSCR to copy the FPCC field to cr1. The branch on summary overflow transfers control to w_sqrt to process any error conditions. Otherwise we can return the result directly. - + This part of the function is a leaf routine, so no need to stack a frame or execute prologue/epilogue code. This means it is safe to transfer directly to w_sqrt as long as the input value (f1) is preserved. Putting the sqrt result into f2 (double parameter 2) allows passing both the input value and sqrt result into the extended wrapper so there is no need to recompute. - + This tactic avoids the overhead of stacking a frame for the normal (non-error) case. Until gcc supports prologue shrink-wrapping this is the best we can do. */ |