Transcript instruction set

Instruction Set Architectures
1
Outline
 Overview of the Instruction Set Level
 Addressing Modes
 Stack-Based Instruction Sets
 Compilation Process
2
Instruction Set Architecture
 An instruction set architecture (ISA) is the interface
between the software and the processor.
 Sometimes referred to as simply the architecture of the
processor.
 The ISA consists of the elements that are visible and relevant
to building software that runs on that processor:
 Instruction set
 Memory model
 Registers (including special purpose registers)
3
ISAs and Microarchitecture
 The internal design of the processor is the microarchitecture
(next unit in the course).
 Not necessary to understand the internal design to write
functional programs.
 For optimal performance, it is also helpful to understand the
microarchitecture too.
 The separation of ISA and microarchitecture allows different
microarchitectures to implement the same ISA.
4
Instruction Set
 What are key design issues when designing an instruction set?
 Key questions:
 Is it compatible with its predecessor?
 Can I run my old OS on it?
 Will it run all my existing application programs unmodified?
5
Memory Model
Key decisions for memory:
 Byte-addressable vs. word-addressable
 How big is a word?
 Memory alignment restrictions
6
Registers
 All processors have some registers that are visible at the ISA level.
 How many general-purpose registers should a machine have?
7
Special Registers
 A processor will have some special registers that may or may
not be directly specified within an assembly language
instruction.
 Most are indirectly accessed – read and/or written without
being specified by the instruction.
 Examples:
 Program counter: Pointer to current instruction.
 Stack pointer: Pointer to the top of the program stack.
 Status flags register: Contains flags for overflow, zero, etc.
 It is called the Program Status Word (PSW) in the text.
8
Example: Intel Core i7 Registers (32bits)
*EAX: Accumulator
*EBX: Pointers
*ECX: Length of string
*EDX: Used along with EAX to store 64-bit
products and dividends
*ESI: Source string
*EDI: Destination string
EBP: Frame pointer
ESP: Stack pointer
CS-GS, SS: Segment registers (archaic)
EIP: Instruction pointer (program counter)
EFLAGS: Status register
*Can be used as a general-purpose register
9
EFLAGS Register
10
FP and MMX Registers
In addition to these registers, the Core i7 also includes:
 Eight 80-bit floating point registers
 Eight 64-bit MMX registers
 Sixteen 128-bit XMM registers
11
Packed Registers
The MMX and XMM stored pack multiple pieces of data
into a single register:
Source:
Intel
12
Vector Operations
Special instruction operate on these values as such. This is
useful for vector calculations:
13
Variable-Length vs. Fixed-Length
Instruction Sets
 Variable-length instruction set: The length of an individual
instruction can vary.
 Advantages:
 Does not waste space (instructions with few operands can be small)
 Can add new instructions to the instruction set.
 Example: Intel
 Fixed-length instruction set: Each instruction is the same size.
 Advantages:
 Easy to decode (significantly less hardware required)
 Can determine where the instruction boundaries are
 Example: ANNA, MIPS
WINNER?
 Fixed-length instruction sets
14
Load-Store Architecture
 Load-Store Architecture: Only load and store instructions can
access memory. Other instructions must use registers to access data.
 Advantages:
 Handling memory accesses separately from other computations simplifies and
improves the performance of the underlying microarchitecture.
 Example: ANNA, MIPS
 Non Load-Store Architecture: Any instruction can access memory
to get its operands and/or store its result.
 Advantages:
 More flexibility for the assembly language programmer.
 Fewer instructions are needed.
 Example: Intel
WINNER?
 Load-Store Architecture
15
Instruction Types
What kinds of instructions should an instruction set have?
16
Number of Operands
 Another key decision: “How many operands in arithmetic and
logical operations?”
 ANNA instructions have three operands: one destination and
two source operands.
 It is also possible to have fewer operands.
 Fewer operands  smaller instruction sizes
 Fewer operands  less flexibility
17
Number of Operands
 Two operands: One destination also serves as the source.
 One operand: Accumulator-based instruction sets. An
accumulator serves as an implicit source and destination.
 Zero operands: Stack-based instruction sets
18
Outline
 Overview of the Instruction Set Level
 Addressing Modes
 Stack-Based Instruction Sets
 Compilation Process
19
Addressing Modes
 An addressing mode is a way of determining where the
operand for a particular instruction is located.
 Different types:
 Immediate addressing
 Register addressing
 Memory addressing
 direct, register indirect, indexed, stack
 PC-relative addressing
20
Immediate Addressing
 Immediate addressing: Operand comes from immediate
in instruction
 Example from ANNA:
21
Register Addressing
 Register addressing: Operand comes from a register
 Example from ANNA:
22
Direct Addressing
 Direct Addressing: Operand comes from memory. The
memory address is given in the instruction as an immediate.
 Example:
23
Direct Addressing Issues
 Memory location is hard-coded (OK for global variables,
doesn't work for local variables)
 Requires an instruction format that allows for the entire
address to be included in the immediate.
 ANNA does not support direct addressing for this reason.
24
Register Indirect Addressing
 Register Indirect Addressing: Operand comes from
memory. The memory address is stored in a register. The
register acts as a pointer in this addressing mode.
 How is it beneficial over direct addressing?
 Example from ANNA:
25
Indexed Addressing
 Indexed Addressing: Operand comes from memory.
The memory address is computed by adding a base
address that comes from a register and an offset (or
index) that comes an immediate.
 Also called: base + offset addressing
 Example from ANNA:
26
Indexed Addressing
Indexed addressing is useful for:
27
Stack Addressing
 Stack Addressing: Operand comes from memory,
specifically the top of the stack. The address comes from a
special stack pointer register.
 Example from IJVM:
 IADD: pop two words from stack; push their sum
28
PC-Relative Addressing
 PC-relative addressing: A variant on indexing addressing
that uses the PC as the base register.
 Only useful for branch / jump instructions.
 Example from ANNA:
29
PC-Relative Addressing Advantages
 Do not need to know absolute address of target ahead of
time, only how far away the target is.
 Allows code to be relocated to a different part of the address
space without modification.
30
Outline
 Overview of the Instruction Set Level
 Addressing Modes
 Stack-Based Instruction Sets
 Compilation Process
31
Stack-Based Instruction Sets
 A stack-based instruction set uses a stack to store local
variables and temporary values.
 This stack is the same program stack accessed by variables.
 No registers except for special registers.
 The most popular example of a stack-based instruction set is
the JavaVirtual Machine (JVM) instruction set.
 Also known as Java bytecodes.
 The book and these notes briefly describe IJVM or Integer JVM
– a subset of integer JVM instructions.
32
Stack-Based Instruction Set Operations
Arithmetic operations:
1. Pop the top two values on the stack.
2. Perform the operation.
3. Push the result onto the top of the stack.
Memory operations:
 Load: push value from memory onto the stack.
 Store: pop value from stack into memory.
33
IJVM Memory Model
 Recall that programs consists of four memory sections:
 Code: Machine code instructions
 Data: Global variables / constants used by the program
 Stack: Stores parameters and local variables
 Heap: Memory dynamically allocated by the program
 In “register-based” instruction sets, registers are used to store
frequently used local variables and temporary results.
 The stack is used if there are enough registers.
 In stack-based instructions sets, all local variables and
temporary results are stored on the stack.
34
IJVM Memory Model
CPP: Constant Pool Pointer
SP: Stack Pointer
LV: Local Variable Frame Pointer
PC: Program Counter
35
IJVM Instructions
36
IJVM Example
F
Java code
37
IJVM code
machine code
IJVM Example: Stack During Execution
38
Outline
 Overview of the Instruction Set Level
 Addressing Modes
 Stack-Based Instruction Sets
 Compilation Process
39
Source Code to Execution
Source
Assembly
Compiler
Assembler
Library
Library
Library
Loader
40
DLL
DLL
DLL
Object File
Linker
Executable
Compiler
 The compiler converts C++ (or any language) into assembly
code.
 In g++, run compiler only using –S:
g++ -S prog.cpp
(produces prog.s)
 Two major parts: front-end and back-end
 Front-end
 Parses high level language
 Checks for syntax errors
 Back-end
 Optimizes code
 Allocates registers
 Produces assembly code
41
Assembler
 The assembler converts assembly file (.s file) to object file (.o
file).
 In g++, use –c to stop at the object code level:
g++ -c prog.s
g++ -c prof.cpp
(both produce prog.o)
 Generally a simple translation: Assembly instructions map to a
machine instruction.
 Assembler provides directives, pseudo-instructions, and labels to
make programming easier.
 The assembler typically makes two passes.
 During the first pass, it determines the addresses of the labels.
 The actual conversion to machine code occurs during the second
pass.
42
Multiple File Compilation
 Individual source code files are compiled and assembled
separately.
 The linker is responsible for linking the individual source
code files into a single executable.
 Even if all the code were contained in a single code, the
object file is still not executable: the format of an object
file and executable are different.
• A library is a package of (related) object files.
43
Producing Machine Code
 The assembler does not have enough information to produce
executable machine code.
 References to labels in another file:
 May need to call a function in another file.
 May need to access a variable in another file.
 Cannot fill in the immediate values.
 Assembler assumes file starts at address 0 (or another fixed
address).
 Not every file can start at 0.
 Need to relocate the file and change the direct addresses.
44
Object Files
Object file contain more than machine code. The format
of an object file is as follows:
Identification
Entry Point Table
External Reference Table
Code
Data / Constants
Relocation Dictionary
Debug Info
End of Module
45
Object Files
 Identification: Contains the starting points and sizes of the
remaining sections and other information needed by the linker.
 Entry Point Table: A table of <label, address> pairs that are
visible to other object files. The labels correspond to functions
and global variables that other object files might want to access.
 Example: If object file A has defined function foo, an entry to foo
with its starting address appears in the entry point table.
 External Reference Table: A list of labels used in this file but are
not defined and expected to be another file. Contains a list of
functions called and global variables used that are defined in a
different file.
 Example: If object file B calls foo but foo is defined in a different file,
the label foo will appear in the external reference table.
46
Object Files
 Code: Contains the machine code.
 Data / Constants: Contains initialized global variables and
constants.
 Relocation Dictionary: Contains a list of instructions or data
entries that refer to direct addresses. If the linker relocates the file
to a different place in memory, these instructions will need to be
updated with their new addresses.
 Debug Info: Additional information regarding variable names and
source code line numbers that is used by debuggers.
 This section is only present if compiled with debugging turned on (-g in g++).
 End of module: Marks the end of the object file, possibly with a
checksum to check for corrupt files and additional information.
47
Linker
 The linker stitches independently created object files into a single
executable file. In particular:
1.
Gather the addresses for all labels in the entry point tables.
• An error occurs if the same label appears twice.
2.
Replace all external references with the appropriate addresses.
• An error occurs if an external reference is not found.
3.
Place all the code together into one code section.
• The code in each file is guaranteed to stay together.
4.
5.
Place all the data together into one data section.
Relocate all the addresses reference in the program to reflect the new
addresses.
• Not needed for PC-relative targets.
6.
48
Produce the executable file.
Linking Example
Executable
File
A
File
B
Text
Text
Data
Text
Text
Text
Data
Data
Data
File
C
49
Text
Data
Data
Loader
 The loader copies the executable file from disk into
memory.
 Asks the operating system to schedule the executable as a
new process.
 Used to be a straight forward process.
 Now portions of the compiling / assembly / linking process are
deferred to load time (or even run time).
50
Dynamically Linked Libraries (DLLs)
 Dynamically linked libraries are not linked nor loaded
until run-time.
 Many system libraries are linked dynamically.
 Advantages:
 Only load portions of the program that are actually executed.
 Multiple processes can share the same code in memory.
 Easier to update libraries (don’t need to recreate new executables).
 Disadvantages
 Slower – overhead due to linking while the program is running.
 Complexity – OS support is needed.
51
Thank You!
52