Bug 6244933 was caused by assuming that on x86 machines with SSE2 that
since c2 would use the xmm registers it wouldn't pollute the fpu stack.
The test case for that bug (Java2D) showed that this assumption was
incorrect. It is believed to by the arg setup for the d2i wrapper.
Because we always have to assume the c2 dirties the stack on SSE2 machines
the jni native wrapper code suffers about a 10% hit compared to c1 on
BobV's jni microbenchmark.

###@###.### 2005-04-06 18:35:04 GMT