Upload
others
View
16
Download
0
Embed Size (px)
Citation preview
Processor Architecture
“God created the integers, all else is the work of man” Leopold
Kronecker
(He believed in the reduction of all mathematics to arguments involving
only the integers and a finite number of steps) (link)
Moving forward
� We are studying architecture—not assembly code (ASM)
� But ASM is the way we manipulate and study the system
on the architecture level
� So we are learning ASM—to understand architecture
� We are also not “married” to the Intel x86 (and
derivatives)
� But it still rules the desktop/laptop (deslaptop?) world
� And other architectures are similar
� So we will use a fake ASM similar to but different from
(and simpler than) the X86 ���� called Y86
� And study the hardware design of the architecture more
� And build one
C Declaration Resolution
� See http://www.unixwiz.net/techtips/reading-cdecl.html
Chapter 4: Processor Architecture� How does the hardware execute the instructions?
� We’ll see by studying an example system
� Based on simple instruction set devised for this purpose
� 8086 evolved and is complex, non-intuitive and tangled
� Important to ASM programmers and compiler writers, not
useful for architecture studies
� Y86, inspired by x86
� Fewer data types, instructions, addressing modes
� Simpler encodings
� Reasonably complete for integer programs
� We’ll design hardware to implement Y86 ISA
� Basic building blocks
� Sequential implementation
� Pipelined implementation
Instruction Set Architecture� Defines interface between hardware
and software
� Software spec is assembly language
� State: registers, memory
� Instructions, encodings
� Hardware must execute instructions
correctly
� May use variety of transparent tricks
to make execution fast.
� Results must match sequential
execution.
� ISA is a layer of abstraction
� Above: how to program machine
� Below: what needs to be built
ISA
Compiler OS
CPUDesign
CircuitDesign
ChipLayout
ApplicationProgram
Where Are We Now?
CS142 & 124
IT344
Y86 Processor and System State
� Program Registers
� Same 8 as with IA32. Each 32 bits
� Condition Codes
� Single-bit flags as in x86: OF (Overflow), ZF (Zero), SF (Negative)
� Program Counter
� Indicates address of instruction
� Memory
� Byte-addressable storage, words in little-endian byte order
� Stat
� Indicates exceptional outcomes (bad opcode, bad address, halt)
%eax
%ecx
%edx
%ebx
%esi
%edi
%esp
%ebp
RF: Program registers
ZF SF OF
CC: Condition
codes
PC
DMEM: Memory
Stat: Program Status
Y86 Instructions
� Format
� 1 to 6 bytes of information read from memory
� Can determine instruction length from first byte
� Not as many instruction types, and simpler encoding than IA32
� Each accesses and modifies some portion of the CPU and system
state
� Program registers
� Condition codes
� Program counter
� Memory contents
Encoding Registers� Each register has 4-bit ID
� Similar encoding used in IA32
� But we never deciphered encoding to notice!
� Register ID 0xF indicates “no register”
� Will use this in our hardware design in multiple places
� Could otherwise encode register # in 3 bits
� Simplifies decoding of instructions
%eax
%ecx
%edx
%ebx
%esi
%edi
%esp
%ebp
0
1
2
3
6
7
4
5
Instruction Example� Addition instruction
� Add value in register rA to that in register rB
� Store result in register rB
� Y86 allows addition to be applied to register data only
� Set condition codes based on result
� Two-byte encoding
� First byte indicates instruction type
� Second gives source and destination registers
� e.g., addl %eax,%esi has encoding 60 06
addl rA, rB 6 0 rA rB
Encoded RepresentationGeneric Form
Arithmetic and Logical Operations
� Refer to generically as “OPl”
� Encodings differ only by
“function code”
� Low-order 4 bits in first
instruction word
� All set condition codes as side
effect
addl rA, rB 6 0 rA rB
subl rA, rB 6 1 rA rB
andl rA, rB 6 2 rA rB
xorl rA, rB 6 3 rA rB
Add
Subtract (rA from rB)
And
Exclusive-Or
Instruction Code Function Code
Move Operations
� Similar to the IA32 movl instruction
� Simpler format for memory addresses
� Separated into different instructions to simplify hardware
implementation
rrmovl rA, rB 2 0 rA rB Register --> Register
Immediate --> Registerirmovl V, rB 3 0 F rB V
Register --> Memoryrmmovl rA, D(rB) 4 0 rA rB D
Memory --> Registermrmovl D(rB), rA 5 0 rA rB D
Move Instruction Examples
irmovl $0xabcd, %edx movl $0xabcd, %edx 30 82 cd ab 00 00
IA32 Y86 Encoding
rrmovl %esp, %ebx movl %esp, %ebx 20 43
mrmovl -12(%ebp),%ecxmovl -12(%ebp),%ecx 50 15 f4 ff ff ff
rmmovl %esi,0x41c(%esp)movl %esi,0x41c(%esp)
—movl $0xabcd, (%eax)
—movl %eax, 12(%eax,%edx)
—movl (%ebp,%eax,4),%ecx
40 64 1c 04 00 00
Jump Instructions
� Refer to generically as “jXX”
� Encodings differ only by
“function code”
� Based on values of condition
codes
� Same as IA32 counterparts
� Encode full destination address
� Unlike PC-relative
addressing in IA32
jmp Dest 7 0
Jump Unconditionally
Dest
jle Dest 7 1
Jump When Less or Equal
Dest
jl Dest 7 2
Jump When Less
Dest
je Dest 7 3
Jump When Equal
Dest
jne Dest 7 4
Jump When Not Equal
Dest
jge Dest 7 5
Jump When Greater or Equal
Dest
jg Dest 7 6
Jump When Greater
Dest
Stack Operations
� Decrement %esp by 4
� Store word from rA to memory at %esp
� Like IA32
� Read word from memory at %esp
� Save in rA
� Increment %esp by 4
� Like IA32
pushl rA a 0 rA 8
popl rA b 0 rA 8
Same stack conventions as IA32
Subroutine Call and Return
� Push address of next instruction onto stack
� Start executing instructions at Dest
� Like IA32
� Pop value from stack
� Use as address for next instruction
� Like IA32
call Dest 8 0 Dest
ret 9 0
Miscellaneous Instructions
� Don’t do anything
� Stop executing instructions
� IA32 has comparable instruction, but it can’t be executed in user
mode
� We will use this instruction to stop the simulator
nop 0 0
halt 1 0
Other Useful instructions
� DWIM Do What I Mean
� FLI Flash Lights Impressively
� HCF Halt and Catch Fire
� BBW Branch Both Ways
� JTZ Jump to Twilight Zone
� LAP Laugh At Programmer
� WSWW Work in Strange and Wondrous Ways
� KPE Kill Programmer on Error
� More here
Y86 Instruction Set (complete set)
Byte 0 1 2 3 4 5
pushl rA A 0 rA F
jXX Dest 7 fn Dest
popl rA B 0 rA F
call Dest 8 0 Dest
rrmovl rA, rB 2 0 rA rB
irmovl V, rB 3 0 F rB V
rmmovl rA, D(rB) 4 0 rA rB D
mrmovl D(rB), rA 5 0 rA rB D
OPl rA, rB 6 fn rA rB
ret 9 0
nop 0 0
halt 1 0
addl 6 0
subl 6 1
andl 6 2
xorl 6 3
jmp 7 0
jle 7 1
jl 7 2
je 7 3
jne 7 4
jge 7 5
jg 7 6
Writing Y86 Code
� Best to use C compiler as much as possible
� Write code in C
� Compile for IA32 with gcc -S
� Hand translate into Y86
� Coding example
� Find number of elements in null-terminated list
int len1(int a[]);
5043
6125
7395
0
a
⇒⇒⇒⇒ 3
Y86 Code Generation Example
� First try
� Write typical array code
� Compile with gcc -O2 -S
� Problem
� Hard to do array indexing on
Y86: no scaled addressing
modes/* Find number of elements in
null-terminated list */
int len1(int a[])
{
int len;
for (len = 0; a[len]; len++)
;
return len;
}
L18:
incl %eax
cmpl $0,(%edx,%eax,4)
jne L18
x86 code
Y86 Code Generation Example #2
� Second try
� Revise C source to use pointers
� Compile with gcc -O2 -S
� Result
� Doesn’t use indexed addressing
/* Find number of elements in
null-terminated list */
int len2(int a[])
{
int len = 0;
while (*a++)
len++;
return len;
}
L5:
movl (%edx),%eax
incl %ecx
addl $4,%edx
testl %eax,%eax
jne L5
x86 code
Y86 Code Generation Example #3
� IA32 code
� Setup
� Y86 code
� Setup
len2:
pushl %ebp
xorl %ecx,%ecx
movl %esp,%ebp
movl 8(%ebp),%edx
movl (%edx),%eax
je L7
len2:
pushl %ebp # Save %ebp
xorl %ecx,%ecx # len = 0
rrmovl %esp,%ebp # Set frame
mrmovl 8(%ebp),%edx # Get a // ptr
mrmovl (%edx),%eax # Get *a
je L7 # Goto exit
Hand translation
Y86 Code Generation Example #4
� IA32 code
� Loop + Finish
� Y86 code
� Loop + Finish
L5:
movl (%edx),%eax
incl %ecx
addl $4,%edx
testl %eax,%eax
jne L5
movl %ebp,%esp
movl %ecx,%eax
popl %ebp
ret
L5:
mrmovl (%edx),%eax # Get *a
irmovl $1,%esi
addl %esi,%ecx # len++
irmovl $4,%esi
addl %esi,%edx # a++
andl %eax,%eax # *a == 0?
jne L5 # No--Loop
rrmovl %ebp,%esp # Pop
rrmovl %ecx,%eax # Rtn len
popl %ebp
ret
Hand translation
Y86 Program Structure
� Programmer must do
more work; no
compiler, linker, run-
time system
� Make program
placement explicit
� Stack initialization
must be explicit (addr.
0x100)
� Must ensure code
is not overwritten!
� Must initialize data
� Can use symbolic
names
irmovl Stack,%esp # Set up stack
rrmovl %esp,%ebp # Set up frame
irmovl List,%edx
pushl %edx # Push argument
call len2 # Call Function
halt # Halt
.align 4
List: # List of elements
.long 5043
.long 6125
.long 7395
.long 0
# Function
len2:
. . .
# Allocate space for stack
.pos 0x100
Stack:
Assembling Y86 Program
� Generates “object code” file eg.yo
� Actually looks like disassembler output
� ASCII file to make it easy for you to read
unix> yas eg.ys
0x000: 308400010000 | irmovl Stack,%esp # Set up stack
0x006: 2045 | rrmovl %esp,%ebp # Set up frame
0x008: 308218000000 | irmovl List,%edx
0x00e: a028 | pushl %edx # Push argument
0x010: 8028000000 | call len2 # Call Function
0x015: 10 | halt # Halt
0x018: | .align 4
0x018: | List: # List of elements
0x018: b3130000 | .long 5043
0x01c: ed170000 | .long 6125
0x020: e31c0000 | .long 7395
0x024: 00000000 | .long 0
Simulating Y86 Program
� Instruction set simulator
� Computes effect of each instruction on processor state
� Prints changes in state from original
unix> yis eg.yo
Stopped in 41 steps at PC = 0x16. Exception 'HLT', CC Z=1 S=0 O=0
Changes to registers:
%eax: 0x00000000 0x00000003
%ecx: 0x00000000 0x00000003
%edx: 0x00000000 0x00000028
%esp: 0x00000000 0x000000fc
%ebp: 0x00000000 0x00000100
%esi: 0x00000000 0x00000004
Changes to memory:
0x00f4: 0x00000000 0x00000100
0x00f8: 0x00000000 0x00000015
0x00fc: 0x00000000 0x00000018
CISC Instruction Sets
� CISC: Complex Instruction Set Computer
� Dominant style of machines designed prior to ~1980
� Stack-oriented instruction set
� Use stack to pass arguments, save program counter
� Explicit push and pop instructions
� Arithmetic instructions can access memory
� addl %eax, 12(%ebx,%ecx,4)
� Requires memory read and write + complex address calculation
� Condition codes
� Set as side effect of arithmetic and logical instructions
� Philosophy
� Add instructions to perform “typical” programming tasks
RISC Instruction Sets
� Reduced Instruction Set Computer
� Early projects at IBM, Stanford (Hennessy), and Berkeley (Patterson)
� Fewer, simpler instructions in ISA (initially)
� Takes more to perform same operations (relative to CISC)
� But an instruction can execute faster on simpler hardware
� Register-oriented instruction set
� Many more (typically ≥ 32) registers
� Used for arguments, return value and address, temporaries
� Only load and store instructions can access memory
� Similar to Y86 mrmovl and rmmovl
� No condition codes
� Test instructions return 0/1 in general purpose register
Example: MIPS Registers
Example: MIPS Instructions
Op Ra Rb Offset
Op Ra Rb Rd Fn00000
R-R
Op Ra Rb Immediate
R-I
Load/Store
addu $3,$2,$1 # Register add: $3 = $2+$1
addu $3,$2,3145 # Immediate add: $3 = $2+3145
sll $3,$2,2 # Shift left: $3 = $2 << 2
lw $3,16($2) # Load Word: $3 = M[$2+16]
sw $3,16($2) # Store Word: M[$2+16] = $3
Op Ra Rb Offset
Branch
beq $3,$2,dest # Branch when $3 = $2
CISC vs. RISC Debate
� Strong opinions at the time!
� CISC arguments
� Easy for compiler (bridge semantic gap)
� Concise object code (memory was expensive)
� RISC arguments
� Simple is better for optimizing compilers
� A simple CPU can be made to run very fast
� Current status
� For desktop processors, choice of ISA not a technical issue
� With enough hardware, anything can be made to run fast
� Code compatibility more important
� For embedded processors, RISC makes sense
� Smaller, cheaper, less power
4.1 Summary
� Y86 instruction set architecture
� Similar state and instructions as IA32
� Simpler encodings
� Small instruction set
� Y86 somewhere between CISC and RISC
� Changes from x86 consistent with RISC principles
4.2: Logic Design: A Brief Review
� Fundamental hardware requirements
� Communication
� How to get values from one place to another
� Computation
� Storage
� All are simplified by restricting to 0s and 1s
� Communication
� Low or high voltage on wire
� Computation
� Compute Boolean functions
� Storage
� Store bits of information
How many different components
needed to make a computer?� Answer: a zillion NAND (or NOR) gates is all it takes.
� Can create NOT, AND, OR, NOR
� NAND w. inputs tied together is NOT
� NAND followed by NOT = AND
� NAND w. NOTs on inputs = OR
� OR followed by NOT = NOR
� (Exercise: Create an XOR)
� With AND, NAND, OR, NOR, NOT you can build
� A 1-bit adder � cascade to n-bit adder
� A one bit register � Cascade to n-bit register
� A clock
� Use registers, adders, clocks and glue logic and you have a
computer
Communication: Digital Signals
� Use voltage thresholds to extract discrete values from continuous
signal
� Simplest version: 1-bit signal
� Either high range (1) or low range (0)
� With guard range between them
� Not strongly affected by noise or low quality circuit elements
� Can make circuits simple, small, and fast
Voltage
Time
0 1 0
Computation: Logic Gates
� Outputs are Boolean functions of inputs
� Respond continuously to changes in inputs
� After some small delay
Voltage
Time
a
ba && b
Rising Delay Falling Delay
Combinational Circuits
� Acyclic network of logic gates
� Continuously responds to changes on primary inputs
� Primary outputs become (after some delay) Boolean functions of
primary inputs
Acyclic Network
PrimaryInputs
PrimaryOutputs
Bit Equality
� Generate 1 if a and b are equal
� Hardware control language (HCL)
� Very simple hardware description language
� Boolean operations have syntax similar to C logical operations
� We’ll use it to describe control logic for processors
� Much more convenient than drawing gates
� Assumes compiler exists to turn HCL into gate equivalent
Bit equala
b
eq
bool eq = (a&&b)||(!a&&!b)
HCL Expression
Word Equality
� 32-bit word size
� HCL representation
� Equality operation
� Generates Boolean value
b31
Bit equal
a31
eq31
b30
Bit equal
a30
eq30
b1
Bit equal
a1
eq1
b0
Bit equal
a0
eq0
Eq
=B
A
Eq
Word-Level Representation
bool Eq = (A == B)
HCL Representation
Bit-Level Multiplexer
� Control signal s
� Data signals a and b
� Output a when s=1, b when s=0
Bit MUX
b
s
a
out
bool out = (s&&a)||(!s&&b)
HCL Expression
Word Multiplexer
� Select input word A or B depending
on control signal s
� HCL representation
� Case expression
� Series of test : value pairs
� Result value determined by first
successful test
Word-Level Representation
HCL Representation
b31
s
a31
out31
b30
a30
out30
b0
a0
out0
int Out = [
s : A;
1 : B;
];
s
B
A
OutMUX
OFZFCF
OFZFCF
OFZFCF
OFZFCF
Arithmetic Logic Unit
� Combinational logic
� Continuously responding to inputs
� Control signal selects function computed
� Corresponding to 4 arithmetic/logical operations in Y86
� Also computes values for condition codes
A
L
U
Y
X
X + Y
0
A
L
U
Y
X
X - Y
1
A
L
U
Y
X
X & Y
2
A
L
U
Y
X
X ^ Y
3
A
B
A
B
A
B
A
B
Edge-Triggered Latch (Flip Flop)
� Only in latching mode for
brief period
� On rising clock edge
� Value latched depends on
data as clock rises
� Output remains stable at all
other times
Q+
Q–
R
S
D
C
Data
ClockTTrigger
C
D
Q+
Time
T
Storage: Registers
� Each stores word of data (one byte in above register)
� Different from program registers (e.g., %eax)
� Collection of edge-triggered latches
� Loads input on rising edge of clock
I O
Clock
D
CQ+
D
CQ+
D
CQ+
D
CQ+
D
CQ+
D
CQ+
D
CQ+
D
CQ+
i7
i6
i5
i4
i3
i2
i1
i0
o7
o6
o5
o4
o3
o2
o1
o0
Clock
Structure
Register Operation
� Stores data bits
� For most of time acts as barrier between input and output
� As clock rises, loads input
State = x
Rising
clock�Output = xInput = y
x�
State = y
Output = y
y
State Machine Example
� Accumulator circuit
� Load or accumulate
on each cycle
Comb. Logic
A
L
U
0
OutMUX
0
1
Clock
In
Load
x0 x1 x2 x3 x4 x5
x0 x0+x1 x0+x1+x2 x3 x3+x4 x3+x4+x5
Clock
Load
In
Out
Storage: Random-Access Memory
� Stores multiple words of memory
� Address input specifies which word to read or write
� Register file
� Holds values of program registers
– %eax, %esp, etc.
� Register identifier serves as address
– ID 0xF implies no read or write performed
� Multiple Ports
� Can read and/or write multiple words simultaneously
– Each has separate address and data input/output
Registerfile
A
B
WdstW
srcA
valA
srcB
valB
valW
Read ports Write port
Clock
Register File Timing
� Reading
� Like combinational logic
� Output data generated based on input
address
� After some delay
� Writing
� Like register (a few slides ago)
� Update only as clock rises
Registerfile
A
B
srcA
valA
srcB
valB
y
2Register
fileW
dstW
valW
Clock
x2
Rising
clock� �Register
fileW
dstW
valW
Clock
y2
x2
2
x
4.2 Summary
� Computation
� Performed by combinational logic
� Computes Boolean functions
� Continuously reacts to input changes
� Storage
� Registers
� Hold single words
� Loaded as clock rises
� Random-access memories
� Hold multiple words
� Multiple read and write ports possible
� Read word anytime address input changes
� Write word only on rising clock edge