IBM z13s, Техническое руководство

Страница: 126 / 588

98
IBM z13s Technical Guide
under hardware control is possible because the R-unit has the full designed state in its buffer.
The principle is shown in Figure 3-12.
Figure 3-12 PU error detection and recovery
3.4.9 Branch prediction
Because of the ultra-high frequency of the PUs, the penalty for a wrongly predicted branch is
high. Therefore, a multi-pronged strategy for branch prediction, based on gathered branch
history that is combined with other prediction mechanisms, is implemented on each
microprocessor.
The BHT implementation on processors provides a large performance improvement.
Originally introduced on the IBM ES/9000 9021 in 1990, the BHT is continuously improved.
The BHT offers significant branch performance benefits. The BHT allows each PU to take
instruction branches based on a stored BHT, which improves processing times for calculation
routines. In addition to the BHT, z13s servers use various techniques to improve the
prediction of the correct branch to be run. The following techniques are used:
򐂰 Branch history table (BHT)
򐂰 Branch target buffer (BTB)
򐂰 Pattern history table (PHT)
򐂰 BTB data compression
The success rate of branch prediction contributes significantly to the superscalar aspects of
z13s servers because the architecture rules prescribe that, for successful parallel execution
of an instruction stream, the correctly predicted result of the branch is essential.
3.4.10 Wild branch
When a bad pointer is used or when code overlays a data area that contains a pointer to
code, a random branch is the result. This process causes a 0C1 or 0C4 abend. Random
branches are hard to diagnose because clues about how the system got there are not
evident.
With the wild branch hardware facility, the last address from which a successful branch
instruction was run is kept. z/OS uses this information with debugging aids, such as the
SLIP
command, to determine where a wild branch came from. It might also collect data from that
storage location. This approach decreases the number of debugging steps that are
necessary when you want to know from where the branch came from.
Soft Error Hard Error
No Error Fault
R-Unit
PU y PU x
Instruct ion State
Checkpoint