RFR 8225337 [lworld] TestNewAcmp.java fails with C1 when java.lang.ClassValue::* are compiled

Tobias Hartmann tobias.hartmann at oracle.com
Tue Jun 11 10:18:12 UTC 2019


Hi Ioi,

looks good to me.

Best regards,
Tobias

On 10.06.19 07:02, Ioi Lam wrote:
> https://bugs.openjdk.java.net/browse/JDK-8225337
> http://cr.openjdk.java.net/~iklam/valhalla/8225337-acmp-wrong-call-to-ValueBootstrapMethods-isSubstitutable.v01/
> 
> 
> In C1, when an acmp operation cannot be handled in-line, we make a call
> to ValueBootstrapMethods::isSubstitutable. We the call is made, registers are
> spilled onto the stack. However, the call is made only conditionally, but
> C1's register allocator can't handle this -- it think the registers are spilled
> for all possible paths taken by the acmp operation.
> 
> The fix is to move the call into a new SubstitutabilityCheckStub. That's the
> standard way of how C1 handles a slow path.
> 
> Thanks
> - Ioi
> 
> 
> 
> 
> 
> 


More information about the valhalla-dev mailing list