Вы находитесь на странице: 1из 14

Internal

Architecture of
ADSP 21xx Family
BY
SWARNASHREE S (ENC-33)
SMRUTI JADHAV (ENC-9)
Introduction
 ADSP-21xx family DSP's are used in
high speed numeric processing
applications.
 ADSP-21xx architecture consists
of Five Internal Buses to ensure
data transfer:
1. Program memory address
2. Data memory address
3. Program memory data
4. Data memory data
5. Result
Computational Unit

 The processor contains three -independent computational units,


1. ALU
2. MAC (Multiplier-accumulator)
3. and the barrel shifter.
 The computational units process 16-bit data directly.
 ALU is 16 bits wide with two 16 bit input ports and one output port.
 The ALU provides a standard set of arithmetic and logic functions.
Block Diagram
Buses

1. PMA and DMA buses.


 They are used internally for addresses associated with Program and
data memory.
 The PMD and DMD are used for data associated with memory spaces.
 Off Chip, the buses are multiplexed into a single external address bus
and a single external data bus.
 The address spaces are selected by the appropriate control signal.
2. Result bus.
The result (R) bus transfers the intermediate results directly between various
computational units.

3. PMA bus is 14-bits wide allowing direct access of up to 16k words of code and
data. PMD bus is 24 bits wide to accommodate the 24 bit instruction width.

4. The DMA bus 14 bits wide allowing direct access of up to 16k words of data. The
DMD bus is 16 bit wide.
5. The DMD bus provides a path for the contents of any register in the processor to be
transferred to any other register or to any external data memory location in a single
cycle.
The PMD bus can also be used to transfer data to and from the
computational units thro direct path or via PMD-DMD bus exchange unit.
Computational Unit

ALU Features

1. Add, subtract, Negate, increment, decrement, Absolute value, AND, OR, EX-
OR, Not etc.

2. Bitwise operators, Constant operators

3. Multi-precision Math Capability

4. Divide Primitives and overflow support.


MAC:

 The MAC performs high speed single-cycle multiply/add and


multiply/subtract operations. MAC has two 16 bit input ports and
one 32 bit product output port.
 32 bit product is passed to a 40 bit adder/subtractor which adds or
subtracts the new product from the content of the multiplier result
(MR).
 It also contains a 40 bit accumulator which provides 8 bit overflow in
successive additions to ensure that no loss of data occurs.
SHIFTER:

 The shifter performs a complete set of shifting functions like logical


and arithmetic shifts, normalization, denormalization etc
 ALU, MAC and shifter are connected to DMD bus on one side and
to R bus on other side. All three sections contains input and output
registers which are accessible from the internal DMD bus.
 Computational operations generally take the operands from input
registers and load the result into an output register.
DATA ADDRESS GENERATORS:

 The two DAG's provides memory addresses when memory data is


transferred to or from the input or output registers.
 Each DAG keeps track of up to four address pointers. Hence jumps,
branching types of instructions are implemented within one cycle.
 With two independent DAG's, the processor can generate two
address simultaneously for dual operand fetches.
 DAG1 can supply addresses to data memory only.
 DAG2 can supply addresses to either data memory or program
memory.
 When the appropriate mode bit is set in mode status register
(MSTAT), the output address of DAG1 is bit-reversed before being
driven onto the address bus. This feature facilitates addressing in
radix-2 FFT algorithm.
PROGRAM SEQUENCER:

 The program sequencer exchanges data with DMD bus.


 It supplies instruction address to program memory.
 The sequencer is driven by the instruction register which holds the
currently executing instruction. The instruction register introduces a
single level of pipelining into the program flow.
 Instructions are fetched and loaded into the instruction register
during one processor cycle, and executed during the following
cycle while the next instruction is pre-fetched.
 The cache memory stores up to 16 previously executed instructions.
Thus data memory on PMD bus is more efficient.
Thank you