Yes, the kHoleNanInt64 is quiet Nan on MIPS, but everything is working well
regarding hole nan checks because kHoleNanInt64 has distinctive bit pattern
0xfff7ffff ffffffff, the FPU on MIPS HW always generate 7ff7ffff fffffffff
for
qNaN.
Hole nan checks compares exact bit-pattern of upper 32 bit, so even hole
nan is
qnan on MIPS
it is correctly recognized. We only had problems with verification of fixed
double arrays
in debug mode on HW which this CL addresses.
The best part is that if kHoleNanInt64 is passed through the MIPS fpu it
gets
converted to
7ff7ffff .... so we can do similar fast canonicalization optimization in
upcoming CL
as it is done for ia32 or arm. All that thanks to the chosen bit patter for
kHoleNanInt64.
https://codereview.chromium.org/912833003/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
