On 10/21/2014 07:50 PM, Peter Maydell wrote: > On 9 October 2014 15:00, Chen Gang <gang.chen.5...@gmail.com> wrote: [] >> --- a/disas/libvixl/a64/instructions-a64.h >> +++ b/disas/libvixl/a64/instructions-a64.h >> @@ -95,30 +95,6 @@ const unsigned kDoubleExponentBits = 11; >> const unsigned kFloatMantissaBits = 23; >> const unsigned kFloatExponentBits = 8; >> >> -const float kFP32PositiveInfinity = rawbits_to_float(0x7f800000); >> -const float kFP32NegativeInfinity = rawbits_to_float(0xff800000); ...
> Upstream's plan for fixing this is to turn these into > 'extern const double foo' in the header with the definition > in an appropriate .cc file. Given that, I think I'm happy for > us to take this patch in QEMU for the moment, with the expectation > that the next libvixl drop will just obsolete it. If upstream already have a patch for this, why not take the upstream's solution now, so we wont need to revert our solution before applying next drop from upstream? BTW, how the "upstream drop" is done, anyway? If it is something like git merge, we may need to care, but if it is just cp -a upstream/* . when there's no reason to... ;) Thanks, /mjt
