Intel Itanium Architecture
Datapath
SLItaniumCPU -> instruction fetch -> bundle/template decoder -> predicate/register files -> execution clusters -> load/store -> register-stack/speculation machinery -> MMU/TLB -> cache/interconnect -> retirement/architectural state
Bundle execution
A bundle supplies three instruction slots. The template describes instruction-slot categories and stop information.
SLeeLa represents each bundle as a scheduling object rather than pretending the three slots are ordinary x86-style variable-length instructions.
Predicate execution
Each instruction has predicate context. A false predicate suppresses the architectural operation while preserving the instruction's place in the scheduled stream.
Register rotation
Procedure calls and software-pipelined loops can rotate portions of the register file. Rotation is represented explicitly in the register-window/rename layer.
EPIC scheduling
The generic model exposes issue groups, stop boundaries, execution resources, dependencies, and latency. Actual Itanium processor issue width and cluster topology remain implementation-specific.
Compatibility
IA-32 compatibility is kept as a separate execution subsystem/profile and is not merged into the IA-64 decoder.