gwixp425bdt Intel Corporation, gwixp425bdt Datasheet - Page 30

• Manufacturer Part Number: gwixp425bdt
• Description: Intel Xp42x Product Line Of Network Processors And Ixc1100 Control Plane Processor
• Manufacturer: Intel Corporation

Intel
2.2.6
2.2.7
March 2005
30
®
IXP42X Product Line of Network Processors and IXC1100 Control Plane Processor
The I-cache can be enabled or disabled. Attribute bits within the descriptors — contained in the
ITLB of the IMMU — provide some control over an enabled I-cache.
When a needed line (eight 32-bit words) is not present in the I-cache, the line is fetched (critical
word first) from memory via a two-level, deep-fetch queue. The fetch queue allows the next
instruction to be accessed from the I-cache, but only when its data operands do not depend on the
execution results of the instruction being fetched via the queue.
Data Cache (D-Cache)
The D-cache can contain high-use data such as lookup tables and filter coefficients, allowing the
core access to data at core frequencies. This prevents core stalls caused by multi-cycle accesses to
external memory.
The 32-Kbyte D-cache is 32-set/32-way associative, where each set contains 32 ways and each
way contains a tag address, a cache line (32 bytes with one parity bit per byte) of data, two dirty
bits (one for each of two eight-byte groupings in a line), and one valid bit. For each of the 32 sets,
zero through 28 ways can be locked, unlocked, or used as local SRAM. Unlocked ways are
replaceable via a round-robin policy.
The D-cache (together with the mini-data cache) can be enabled or disabled. Attribute bits within
the descriptors, contained in the DTLB of the DMMU, provide significant control over an enabled
D-cache. These bits specify cache operating modes such as read and write allocate, write-back,
write-through, and D-cache versus mini-data cache targeting.
The D-cache (and mini-data cache) work with the load buffer and pend buffer to provide “hit-
under-miss” capability that allows the core to access other data in the cache after a “miss” is
encountered. The D-cache (and mini-data cache) works in conjunction with the write buffer for
data that is to be stored to memory.
Mini-Data Cache
The mini-data cache can contain frequently changing data streams such as MPEG video, allowing
the core access to data streams at core frequencies. This prevents core stalls caused by multi-cycle
accesses to external memory. The mini-data cache relieves the D-cache of data “thrashing” caused
by frequently changing data streams.
The 2-Kbyte, mini-data cache is 32-set/two-way associative, where each set contains two ways and
each way contains a tag address, a cache line (32 bytes with one parity bit per byte) of data, two
dirty bits (one for each of two eight-byte groupings in a line), and a valid bit. The mini-data cache
uses a round-robin replacement policy, and cannot be locked.
The mini-data cache (together with the D-cache) can be enabled or disabled. Attribute bits
contained within a coprocessor register specify operating modes write and/or read allocate, write-
back, and write-through.
The mini-data cache (and D-cache) work with the load buffer and pend buffer to provide “hit-
under-miss” capability that allows the core to access other data in the cache after a “miss” is
encountered. The mini-data cache (and D-cache) works in conjunction with the write buffer for
data that is to be stored to memory.
Document Number: 252479, Revision: 005
Datasheet