Intel Processor Computer Hardware User Manual


 
A-2 March, 2003 Developers Manual
Intel
®
80200 Processor based on Intel
®
XScale
Microarchitecture
Compatibility: Intel
®
80200 Processor vs. SA-110
Feature / Parameter Brief Description or Note SA-110
Intel
®
80200
Processor
Main Execution Pipeline Scalar, in-order execution, single issue ••
RISC Superpipeline
Pipeline with more than usual number of pipe
stages. Allows greater operating frequency.
(5 stage)
(7 stage)
Out-of order completion Instructions may finish out of program order.
Concurrent execution in 3
pipes
Instructions may occupy the MAC, ALU, and
data-cache pipes concurrently.
ALU
Number of cycles ALU takes to complete
instruction.
11
a
a. A 32-bit shift and ALU operation takes 1.5 cycles.
Register Scoreboarding
Allows instructions in different pipelines to
execute as long as there are no data hazards.
Dynamic branch prediction
128 entry branch table address cache holds
history of branches taken and not taken.
Branch misprediction
penalty
(in cycles) 1 4
MAC Pipeline Handles all multiply operations
Dedicated 40-bit
accumulator
Allows 256 accumulates before scaling of data is
necessary.
1 cycle 16x32 MAC
sustained
Early terminate
MAC may finish instruction and return results in
any of the MAC pipestages.
n/a
Instruction Cache ••
Geometry / Replacement
policy
Both the SA-110 and the Intel
®
80200 processor
are 32-way set associative with round-robin
replacement. They differ in size.
16K 32K
Lockable by line
Instructions may be locked into the instruction
cache with line granularity, preventing future
eviction.
Fetch buffers Buffer incoming external memory operations.
Data Cache / Mini Data
Cache
Replacement Policy round-robin round-robin
Primary Data Cache
Geometry
Both SA-110 and
the Intel
®
80200 processor are
32-way set associative with round-robin
replacement. They differ in size.
16K,
32 ways
b
b. SA1100
32K,
32 way
Mini Data Cache
Geometry
Set associative
512 bytes,
2 ways
2K,
2 ways
Data RAM
Software can re-map portions of the data cache
into data-RAM on a line granularity.
Hit under miss
Allow accesses to the data cache while there are
outstanding miss requests to external memory
Write-back
Store operation that hits cache is not written to
external memory.
••
Write-through
Store operations are written to external memory
even if cache is hit.
Fill Buffer Buffer incoming external memory operations. ••
Pending Buffer
Collects memory requests that hit an
outstanding load
Write Buffer 8 entry write buffer ••
Write Buffer coalescing
Number of entries that a new store request can
coalesce to in the write buffer
last entry All entries