fdlibm behaviour on non-HotSpot platforms

[gnu_andrew at member.fsf.org]

Hi,

Christian Thalinger (twisti) of the CACAO project has recently been looking at 
using the OpenJDK libraries with CACAO as the VM (via a suitable libjvm.so) 
to allow it to be built on non-HotSpot supported platforms (e.g. ppc, ppc64, 
alpha).  I've just been testing this on ppc64 this morning.

The trigonometric functions (sin, cos, tan) seem to run into difficulties on 
these new platforms.  On x86, x86_64 and sparc, sin(0.7) returns the correct 
answer of 0.644217687237691.  On alpha, the same test 
gives -0.9589410493575494 while it returns the input value of 0.7 on both ppc 
and ppc64.  Ironically, gij on ppc64 gives the correct answer.

This was spotted when running SPECs check:

5$ cacao -cp ./check.jar spec/jbb/validity/PepTest
LOG: [0x30021cd0] JVM_GetStackAccessControlContext: IMPLEMENT ME!
LOG: [0x30021cd0] JVM_GetStackAccessControlContext: IMPLEMENT ME!
LOG: [0x30021cd0] JVM_FindSignal: name=HUP
LOG: [0x30021cd0] JVM_RegisterSignal: sig=0, handler=0x2, IMPLEMENT ME!
LOG: [0x30021cd0] JVM_FindSignal: name=INT
LOG: [0x30021cd0] JVM_RegisterSignal: sig=0, handler=0x2, IMPLEMENT ME!
LOG: [0x30021cd0] JVM_FindSignal: name=TERM
LOG: [0x30021cd0] JVM_RegisterSignal: sig=0, handler=0x2, IMPLEMENT ME!
LOG: [0x30021cd0] JVM_GetStackAccessControlContext: IMPLEMENT ME!
LOG: [0x30021cd0] JVM_GetStackAccessControlContext: IMPLEMENT ME!
LOG: [0x30021cd0] JVM_GetSystemPackage: IMPLEMENT ME!
LOG: [0x30021cd0] JVM_GetSystemPackage: IMPLEMENT ME!
testIf:     OK
testArray:  OK
testBitOps: OK
testFor:    OK
testDiv:    OK
testTableSwitch:  OK
testLookupSwitch: OK
testHiddenField:  OK
checkRemainders:  long double OK
checkMathFcts: sin(0.7) evaluated to: 0.7, expected: 0.644218
cos(0.7) evaluated to: 1.0, expected: 0.764842
acos(0.7) evaluated to: 1.5707963267948966, expected: 0.795399
atan(0.7) evaluated to: 0.7, expected: 0.610726
 
******************************************
Some math function failed
******************************************
Primes less than 50: 2 3 5 7 11 13 17 19 23 29 31 37 41 43 47 
testExc1(simple throw/catch):  OK
testExc2(skip catch clauses):  OK
testExc3(catch in inner):      OK
testExc4(catch in outer):      OK
testExc5(rethrow):             OK
testExc6(throw accross call):  OK
testExc7(throw accr. 2 calls): OK
loopExitContinueInExceptionHandler: OK
testStringHash:  OK
testClone:       OK
testObjectArray: OK
checkInstanceOf: OK
checkInterfaceInstanceOf: OK
testWaitNull: OK
testVarAndMethodNameClash: OK
testMisk: OK
testGC: OK
testRandom : OK
PepTest: error

Any idea what the problem may be and when this may be rectified?

Thanks,
-- 
Andrew :)