On 02/19/2013 12:40 PM, Torbjorn Granlund wrote: > The apparent symptom is that an > > adox (%rdi,%rcx,8), %r9 > > where the 8 bytes at the memory location contains just zeros, > and r9 is some random value, results in OF being set.
Interesting. > I only have Intel's 327831.pdf document which does not accurately > document adcx or adox. Their detailed manuals are aparently silent > about these insns. Try http://software.intel.com/sites/default/files/319433-014.pdf which is the most recent copy of the extensions document. > > Does qemu ever fold foldable constructs like > > move $0, %rax > add %rax, %rax Yes it does. That can be partially counteracted by using -singlestep, as the optimizer will no longer see the extended blocks. But on the other hand that will also change how the translator optimizes the condition codes, so it can once again hide the bug. r~
