
6
PC7410
2141D–HIREL–02/04
–
–
–
–
–
–
–
–
–
Level 2 (L2) Cache Interface
–
Internal L2 cache controller and tags; external data SRAMs
–
512K, 1M and 2-Mbyte 2-way set associative L2 cache support
–
Copyback or write-through data cache (on a page basis or for all L2)
–
32-byte (512K), 64-byte (1M), or 128-byte (2M) sectored line size
–
Supports pipelined (register-register) synchronous burst SRAMs and
pipelined (register-register) late-write synchronous burst SRAMs
–
Supports direct mapped mode for 256K, 512K, 1M or 2 Mbytes of SRAM
(either all, half or none of L2 SRAM must be configured as direct mapped.
–
Core-to-L2 frequency divisors of ÷1, ÷1.5, ÷2, ÷2.5, ÷3, ÷3.5, and ÷4
supported
–
64-bit data bus which also support 32-bits bus mode
–
Selectable interface voltages of 1.8V and 2.5V
Memory Management Unit
–
128 entry, 2-way set associative instruction TLB
–
128 entry, 2-way set associative data TLB
–
Hardware reload for TLBs
–
Four instruction BATs and four data BATs
–
Virtual memory support for up to four petabytes (2
52
) of virtual memory
–
Real memory support for up to four gigabytes (2
32
) of physical memory
–
Snooped and invalidated for TLBI instructions
Efficient Data Flow
–
All data buses between VRF, load/store unit, dL1, iL1, L2 and the bus are
128 bits wide
–
dL1 is fully pipelined to provide 128 bits per cycle to/from the VRF
–
L2 is fully pipelined to provide 128 bits per L2 clock cycle to the L1s
–
Up to eight outstanding out-of-order cache misses between dL1 and L2/bus
–
Up to seven outstanding out-of-order transactions on the bus
–
Load folding to fold new dL1 misses into older outstanding load and store
misses to the same line
–
Store miss merging for multiple store misses to the same line. Only
coherency action taken (i.e., address only) for store misses merged to all 32
bytes of a cache line (no data tenure needed).
–
Two-entry finished store queue and four-entry completed store queue
between load/store unit and dL1
32K 32-byte line, 8-way set associative data cache (dL1)
Single-cycle cache access
Pseudo least-recently-used (LRU) replacement
Data cache supports AltiVec LRU and transient instructions algorithm
Copy-back or write-through data cache (on a page-per-page basis)
Supports all PowerPC memory coherency modes
Non-blocking instruction and data cache
Separate copy of data cache tags for efficient snooping
No snooping of instruction cache except for ICBI instruction