12/9/2015

ORGANISASI & SEMESTER 3

PERTEMUAN KE-10
ARSITEKTUR GEMBONG EDHI SETYAWAN

KOMPUTER

1 s1 / TI / semest er 3 / 3 sks / reguler

 2 Gembong Edhi Setyawan

RISC
TIK:

s1 / TI / semester 3 / 3 sks / reguler

 3 Gembong Edhi Setyawan

Major Advances in
Computers(1)
 The family concept
 IBM System/360 1964
 DEC PDP-8
 Separates architecture from implementation

 Microporgrammed control unit

 Idea by Wilkes 1951
 Produced by IBM S/360 1964

 Cache memory
 IBM S/360 model 85 1969

s1 / TI / semester 3 / 3 sks / reguler

 4 Gembong Edhi Setyawan

Major Advances in
Computers(2)
 Solid State RAM
 (See memory notes)
 Microprocessors
 Intel 4004 1971
 Pipelining
 Introduces parallelism into fetch execute cycle
 Multiple processors

s1 / TI / semester 3 / 3 sks / reguler

 5 Gembong Edhi Setyawan

The Next Step - RISC

 Reduced Instruction Set Computer

 Key features
 Large number of general purpose registers
 or use of compiler technology to optimize register use
 Limited and simple instruction set
 Emphasis on optimising the instruction pipeline

s1 / TI / semester 3 / 3 sks / reguler

 6 Gembong Edhi Setyawan

Comparison of processors
CISC RISC Superscalar
IBM DEC VAX Intel Motorola MIPS IBM Intel
370/168 11/780 486 88000 R4000 RS/6000 80960
1973 1978 1989 1988 1991 1990 1989

No. of instruction
208 303 235 51 94 184 62

Instruction size (octets)

2-6 2-57 1-11 4 32 4 4 or 8

Addressing modes
4 22 11 3 1 2 11

GP Registers
16 16 8 32 32 32 23-256

Control memory (k bytes) (microprogramming)

420 480 246 0 0 0 0

s1 / TI / semester 3 / 3 sks / reguler

 7 Gembong Edhi Setyawan

Driving force for CISC

 Software costs far exceed hardware costs
 Increasingly complex high level languages
 Semantic gap
 Leads to:
 Large instruction sets
 More addressing modes
 Hardware implementations of HLL statements
 e.g. CASE (switch) on VAX

s1 / TI / semester 3 / 3 sks / reguler

 8 Gembong Edhi Setyawan

Intention of CISC
 Ease compiler writing
 Improve execution efficiency
 Complex operations in microcode
 Support more complex HLLs

s1 / TI / semester 3 / 3 sks / reguler

 9 Gembong Edhi Setyawan

Execution Characteristics
 Operations performed
 Operands used
 Execution sequencing
 Studies have been done based on programs written
in HLLs
 Dynamic studies are measured during the execution
of the program

s1 / TI / semester 3 / 3 sks / reguler

 10 Gembong Edhi Setyawan

Operations
 Assignments
 Movement of data
 Conditional statements (IF, LOOP)
 Sequence control
 Procedure call-return is very time consuming
 Some HLL instruction lead to many machine code
operations

s1 / TI / semester 3 / 3 sks / reguler

 11 Gembong Edhi Setyawan

Relative Dynamic Frequency

Dynamic Machine Instruction Memory Reference
Occurrence (Weighted) (Weighted)
Pascal C Pascal C Pascal C
Assign 45 38 13 13 14 15
Loop 5 3 42 32 33 26
Call 15 12 31 33 44 45
If 29 43 11 21 7 13
GoTo - 3 - - - -
Other 6 1 3 1 2 1

s1 / TI / semester 3 / 3 sks / reguler

 12 Gembong Edhi Setyawan

Operands
 Mainly local scalar variables
 Optimisation should concentrate on accessing local
variables

Pascal C Average
Integer constant 16 23 20
Scalar variable 58 53 55
Array/structure 26 24 25

s1 / TI / semester 3 / 3 sks / reguler

 13 Gembong Edhi Setyawan

Procedure Calls
 Very time consuming
 Depends on number of parameters passed
 Depends on level of nesting
 Most programs do not do a lot of calls followed by
lots of returns
 Most variables are local
 (c.f. locality of reference)

s1 / TI / semester 3 / 3 sks / reguler

 14 Gembong Edhi Setyawan

Implications
 Best support is given by optimising most used and
most time consuming features
 Large number of registers
 Operand referencing
 Careful design of pipelines
 Branch prediction etc.
 Simplified (reduced) instruction set

s1 / TI / semester 3 / 3 sks / reguler

 15 Gembong Edhi Setyawan

Large Register File

 Software solution
 Require compiler to allocate registers
 Allocate based on most used variables in a given time
 Requires sophisticated program analysis

 Hardware solution
 Have more registers
 Thus more variables will be in registers

s1 / TI / semester 3 / 3 sks / reguler

 16 Gembong Edhi Setyawan

Registers for Local Variables

 Store local scalar variables in registers
 Reduces memory access
 Every procedure (function) call changes locality
 Parameters must be passed
 Results must be returned
 Variables from calling programs must be restored

s1 / TI / semester 3 / 3 sks / reguler

 17 Gembong Edhi Setyawan

Register Windows
 Only few parameters
 Limited range of depth of call
 Use multiple small sets of registers
 Calls switch to a different set of registers
 Returns switch back to a previously used set of
registers

s1 / TI / semester 3 / 3 sks / reguler

 18 Gembong Edhi Setyawan

Register Windows cont.

 Three areas within a register set
 Parameter registers
 Local registers
 Temporary registers
 Temporary registers from one set overlap parameter
registers from the next
 This allows parameter passing without moving data

s1 / TI / semester 3 / 3 sks / reguler

 19 Gembong Edhi Setyawan

Overlapping Register Windows

s1 / TI / semester 3 / 3 sks / reguler

 20 Gembong Edhi Setyawan

Circular Buffer diagram

s1 / TI / semester 3 / 3 sks / reguler

 21 Gembong Edhi Setyawan

Operation of Circular Buffer

 When a call is made, a current window pointer is
moved to show the currently active register window
 If all windows are in use, an interrupt is generated
and the oldest window (the one furthest back in the
call nesting) is saved to memory
 A saved window pointer indicates where the next
saved windows should restore to

s1 / TI / semester 3 / 3 sks / reguler

 22 Gembong Edhi Setyawan

Global Variables
 Allocated by the compiler to memory
 Inefficient for frequently accessed variables
 Have a set of registers for global variables

s1 / TI / semester 3 / 3 sks / reguler

 23 Gembong Edhi Setyawan

Registers v Cache
 Large Register File Cache

 All local scalars Recently used local scalars

 Individual variables Blocks of memory
 Compiler assigned global variables Recently used global variables
 Save/restore based on procedure Save/restore based on nesting
caching algorithm
 Register addressing Memory addressing

s1 / TI / semester 3 / 3 sks / reguler

 24 Gembong Edhi Setyawan

Referencing a Scalar -
Window Based Register File

s1 / TI / semester 3 / 3 sks / reguler

 25 Gembong Edhi Setyawan

Referencing a Scalar - Cache

s1 / TI / semester 3 / 3 sks / reguler

 26 Gembong Edhi Setyawan

Compiler Based Register

Optimization
 Assume small number of registers (16-32)
 Optimizing use is up to compiler
 HLL programs have no explicit references to registers
 usually - think about C - register int
 Assign symbolic or virtual register to each candidate variable
 Map (unlimited) symbolic registers to real registers
 Symbolic registers that do not overlap can share real registers
 If you run out of real registers some variables use memory

s1 / TI / semester 3 / 3 sks / reguler

 27 Gembong Edhi Setyawan

Graph Coloring
 Given a graph of nodes and edges
 Assign a color to each node
 Adjacent nodes have different colors
 Use minimum number of colors
 Nodes are symbolic registers
 Two registers that are live in the same program fragment are
joined by an edge
 Try to color the graph with n colors, where n is the number of
real registers
 Nodes that can not be colored are placed in memory

s1 / TI / semester 3 / 3 sks / reguler

 28 Gembong Edhi Setyawan

Graph Coloring Approach

s1 / TI / semester 3 / 3 sks / reguler

 29 Gembong Edhi Setyawan

Why CISC (1)?

 Compiler simplification?
 Disputed…
 Complex machine instructions harder to exploit
 Optimization more difficult

 Smaller programs?
 Program takes up less memory but…
 Memory is now cheap
 May not occupy less bits, just look shorter in symbolic
form
 More instructions require longer op-codes
 Register references require fewer bits
s1 / TI / semester 3 / 3 sks / reguler
 30 Gembong Edhi Setyawan

Why CISC (2)?

 Faster programs?
 Bias towards use of simpler instructions
 More complex control unit
 Microprogram control store larger
 thus simple instructions take longer to execute

 It is far from clear that CISC is the appropriate

solution

s1 / TI / semester 3 / 3 sks / reguler

 31 Gembong Edhi Setyawan

RISC Characteristics
 One instruction per cycle
 Register to register operations
 Few, simple addressing modes
 Few, simple instruction formats
 Hardwired design (no microcode)
 Fixed instruction format
 More compile time/effort

s1 / TI / semester 3 / 3 sks / reguler

 32 Gembong Edhi Setyawan

RISC v CISC
 Not clear cut
 Many designs borrow from both philosophies
 e.g. PowerPC and Pentium II

s1 / TI / semester 3 / 3 sks / reguler

 33 Gembong Edhi Setyawan

RISC Pipelining
 Most instructions are register to register
 Two phases of execution
 I: Instruction fetch
 E: Execute
 ALU operation with register input and output
 For load and store
 I: Instruction fetch
 E: Execute
 Calculate memory address
 D: Memory
 Register to memory or memory to register s1
operation
/ TI / semester 3 / 3 sks / reguler
 34 Gembong Edhi Setyawan

Effects of Pipelining

s1 / TI / semester 3 / 3 sks / reguler

 35 Gembong Edhi Setyawan

Optimization of Pipelining
 Delayed branch
 Does not take effect until after execution of following
instruction
 This following instruction is the delay slot

s1 / TI / semester 3 / 3 sks / reguler

 36 Gembong Edhi Setyawan

Normal and Delayed Branch

Address NormalDelayed Optimized
100 LOAD X,A LOAD X,A LOAD X,A
101 ADD 1,A ADD 1,A JUMP 105
102 JUMP 105 JUMP 105 ADD 1,A
103 ADD A,B NOOP ADD A,B
104 SUB C,B ADD A,B SUB C,B
105 STORE A,Z SUB C,B STORE A,Z
106 STORE A,Z

s1 / TI / semester 3 / 3 sks / reguler

 37 Gembong Edhi Setyawan

Use of Delayed Branch

s1 / TI / semester 3 / 3 sks / reguler

 38 Gembong Edhi Setyawan

Controversy
 Quantitative
 compare program sizes and execution speeds
 Qualitative
 examine issues of high level language support and use of
VLSI real estate
 Problems
 No pair of RISC and CISC that are directly comparable
 No definitive set of test programs
 Difficult to separate hardware effects from complier effects
 Most comparisons done on “toy” rather than production
machines
 Most commercial devices are a mixture
s1 / TI / semester 3 / 3 sks / reguler
 39 Gembong Edhi Setyawan

Required Reading
 Stallings chapter 12
 Manufacturer web sites

s1 / TI / semester 3 / 3 sks / reguler

40 Gembong Edhi Setyawan

s1 / TI / semester 3 / 3 sks / reguler