Instruction Set Architecture
Instruction Set Architecture (ISA)
* Instruction set architecture of a machine
fills the semantic gap between the user
and the machine.
* ISA serves as the starting point for the
design of a new machine or modification
of an existing one.
* We will look at the main features of an
instruction set architecture and learn how
to design one.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Topics for Discussions
* Classifications of ISA
* Memory Addressing
* Operations in the Instruction Set
* Type and Size of Operands
* Encoding an Instruction Set
* Role of Compilers
* An Example ISA -- MIPS
(in particular, MIPS64)
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Assumed Prior Knowledge
* Instruction: Label Operator Operands
* Example: LOOP: MOVZ R1, R2
* Interpretation:
Move the contents of register R1
(source operand) to R2 (destination
operand) when the test flag is 0.
* Operands can be registers, memory
address, or absolute values (immediate).
* Instruction cycle:
Fetch, Decode, Execute, Store
* Basic memory and register addressing
* Registers Program Counter (PC) and
Instruction Register (IR).
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Classifying ISA
Stack Architecture:
Operands are implicit. They are on the top
of the stack. For example, a binary
operation pops the top two elements of the
stack, applies the operation and pushes the
result back on the stack.
Accumulator Architecture:
One of the operands is implicitly the
accumulator. Usually one of the source
operand and the destination operand is
assumed to be the accumulator.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Classifying ISA (contd.)
General Purpose Register Architecture:
Operands are explicit: either memory
operands or register operands. Any
instruction can access memory. ISA
design is orthogonal.
Load/Store Architecture:
Only load/store instructions can access
memory. All other instructions use
registers. Also referred to as registerregister architecture.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Classifying ISA (contd.) - Sample Code
Stack Accumulator GP Register Load/Store
Push A
Load A
Load R1, A
Load R1, A
Push B
Add B
Add R1, B
Load R2, B
Add
Store C
Store C, R1
Add R3,R1,R2
Pop C
Huang - Spring 2008 CSE 340
Store C, R3
Instruction Set Architecture
Memory Addressing
* ISA design must define how memory
addresses are interpreted and specified in
the instructions.
* Most machines (ISAs) are byte addressed.
That is, a byte (8 bits) is the smallest
memory size that has a unique address.
Other sizes are: half word, word, double
word ...
* There are two different conventions for
ordering bytes within a word:
Little Endian and Big Endian.
* Little Endian: The byte with address XXX...00
is the least significant byte in the word.
Big Endian: The byte with address XXX...00 is
the most significant byte in the word.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Addressing Modes
* Addressing modes was one of the ways designers
attempted to close the gap between HLL and the
machine language. (By providing modes that a
compiler can directly use to map many of the HLL
features such as array indexing, base register
addressing, position independent coding (PCrelative.)
* Examples: Indexed mode array access,
auto increment and auto decrement to allow
processing of array elements in sequence
and so on.
* Some popular addressing modes: Register,
Immediate, Displacement, register-indirect,
indexed, absolute, memory-indirect,
auto increment and auto decrement.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Choice of Addressing Modes
* Study the measurements related to the usage of
the various addressing modes.
* For example, if the displacement mode is used,
you may further study the distribution of how far
the displacement is.
* Then, select the modes depending on their usage,
and the specific attributes of the mode
depending on experimental results with a select
set of benchmarks.
Huang - Spring 2008 CSE 340
Instruction Set Architecture
Operations in the Instruction Set
* Data transfer instructions.
* Arithmetic and logic instructions.
* Instructions for control flow: conditional and
unconditional branches, jumps, procedure calls
and procedure returns.
* System calls.
* Floating point instructions.
* Decimal instructions.
* String instructions.
* Graphics instructions.
Huang - Spring 2008 CSE 340
10
Instruction Set Architecture
Behind the Scenes Operations
* Condition codes and condition register: Should
this be set after every instruction or only when
specified?
* For procedure calls, does the caller or called save
the state of computation?
* Position independence: Most common way to
specify branch destination is to supply a
displacement that is added to the Program
Counter (PC). Using PC-relative addressing
permits the code to run independently of where
it is loaded.
Huang - Spring 2008 CSE 340
11
Instruction Set Architecture
Type and Size of Operands
* Type: Integer, floating point, single precision
float, double precision float, character.
* Floating point: IEEE 754 format.
* Support for decimal data: Direct decimal
operations, packed decimal.
* byte, word, double word, quad word, etc.
* With SPEC92 suite of benchmarks, on 32-bit
address machines, word is the size commonly
used for integer computation and 64-bit (double)
is common for floating point computation.
Huang - Spring 2008 CSE 340
12
Instruction Set Architecture
Encoding an Instruction Set
* Operator of the instruction is encoded into
opcode.
* When there are large number of addressing modes
and more than one memory operands are allowed
in an instruction, each operand must have an
address specifier.
* Otherwise if it is like the load/store machine, the
operands can be encoded along with the opcode
itself.
* The number of addressing modes and number
of registers have a significant effect on instruction
length.
Huang - Spring 2008 CSE 340
13
Instruction Set Architecture
Variations in Instruction Encoding
* Variable:
Can support any number of operands.
Number of operands is a part of the opcode.
Each operand has an address specifier.
(e.g. VAX, 1-53 bytes)
* Fixed:
Fixed format always has the same number of
operands. How they are decoded may differ
with opcode. (e.g. SPARC (4 bytes), MIPS,
PowerPC)
* Hybrid:
Multiple formats of fixed length. Format
specified as a part of the opcode. (e.g. IBM 370,
Intel 80X86 series (7 different formats of 1-5
bytes)
Huang - Spring 2008 CSE 340
14
Instruction Set Architecture
The Role of Compilers
Front-End
High-Level
Optimization
Transform HLL to
common intermediate
form
Procedure inlining &
loop transformations
Global and local opt.
Global Optimizer & register allocation
Code Generator
Huang - Spring 2008 CSE 340
Machine dependent
opt.
15
Instruction Set Architecture
The Role of Compilers (contd.)
* Front-end:
One per language. Once the source code is
transformed into a common intermediate
code, the rest of the phases are the same for
any HLL.
* High-level optimization:
Involves working on the source code.
Example: online expansion of functions.
* Local optimization:
Within a sequence of code.
The straight line section of the code.
Huang - Spring 2008 CSE 340
16
Instruction Set Architecture
The Role of Compilers (contd.)
* Global optimization:
External local optimization across branches
and introduce a set of transformations aimed
at optimizing loops (Loop Optimization).
* Register allocation is a very important function.
Recent register allocation techniques are based
on graph coloring.
* Machine-dependent optimizations exploit some
architectural features.
Huang - Spring 2008 CSE 340
17
Instruction Set Architecture
Compiler Technology and ISA
* Todays compiler technology (in the form of
optimization and others) helps in making such
architectures as load/store a viable solution.
* As an ISA designer, one may help the compiler
writer generate efficient code. What are the
features in ISA that can help compilation?
Huang - Spring 2008 CSE 340
18
Instruction Set Architecture
ISA Features to Support Efficient Compilation
* Regularity:
Three components of instruction set: operation,
data types and addressing modes should be
orthogonal. (independent)
e.g. VAXs arithmetic operations are equally
applicable to register or memory operands.
* Provide primitives and not solutions:
Features that match a particular HLL feature are
often not usable by other HLL or other statements.
e.g. Index register to support arrays. Instead a
general purpose register would serve the purpose
and other situations well.
Huang - Spring 2008 CSE 340
19
Instruction Set Architecture
ISA Features ... (contd.)
* Simplify trade-off among alternatives:
An HLL statement may be translated into several
different equivalent sets of instructions.
-- Which is better? The longer or shorter
sequence? With cache and pipelining this
tradeoff becomes complex.
-- Example:
In register-memory architecture, how
many times a memory variable has been
referenced before it can be moved into a
register?
Huang - Spring 2008 CSE 340
20
Instruction Set Architecture
ISA Features ... (contd.)
* Provide instructions that bind quantities known
at compile time as constants -if something is known at compile time,
resolve it at compile time.
* Counter Example -- CALLS of VAX ISA
CALLS is a procedure call in VAXs ISA.
Even though the list of registers to be saved on
transfer control to the procedure is known at
compile-time, the register saving process is
included in the semantics of CALLS! So
cannot be substituted with instructions at
compile time.
Huang - Spring 2008 CSE 340
21
Instruction Set Architecture
Summarizing the ISA Features
* How to decide the general class of ISA?
* How is the memory going to be addressed?
* What are the addressing modes?
* What are the allowed operations?
* The types of operands?
* How are instructions encoded?
* How to add features that help compiler generate
efficient code?
Huang - Spring 2008 CSE 340
22
