And after some debugging, I found the following symptom: the binary content of the double-precision float "3. 1415926535" is correct, which is 0x400921fb54411744
and it's displayed correct on Android 2.2, while on Android 4.0 it displayed as d=<36.<2=2=3<<==<22=323 So, anything related with java println needs to be modified when porting to new chip? Thanks in advance for any clue --- Regards, WANG.Jiong On 02/02/2012 06:04 PM, WANG.Jiong wrote: > Hi all, > > Currently, I am porting Android ICS to a new chip, now I am blocked > by one Dalvik bug. > > when I run dalvik unit test under dalvik/tests, several testcases > failed, like 017-float. > > After exploration, I found all of these failure are caused by the > same reason: double-precision float are handled incorrectly. > > For 017-float, the test result is: > > --- expected.txt 2011-12-31 19:14:46.000000000 +0800 > +++ output.txt 2012-02-02 17:57:07.881173911 +0800 > @@ -1,3 +1,3 @@ > -base values: d=3.1415926535 f=3.1415927 > -base values: d=3.1415926535 f=3.1415927 > -base values: f=3.1415927 d=3.1415926535 > +base values: d=<36.<2=2=3<<==<22=323 f=003.1415927 > +base values: d=<36.<2=2=3<<==<22=323 f=003.1415927 > +base values: f=003.1415927 d=<36.<2=2=3<<==<22=323 > > the value of d are incorrect, and the value of f are prefixed by two > extra '0'. > > while I print the value of 'f' only, it will get correct value > "3.1415927" > > currently, I only enabled c stub based interpreter > > My question is: > > when porting Dalvik to new chip, is there anything special I need > to do to support double-precision? like I need to define any special > macro, so something else > > Thanks in advance > > --- > Regards, > WANG.Jiong > -- unsubscribe: android-porting+unsubscr...@googlegroups.com website: http://groups.google.com/group/android-porting
