The Microarchitecture Level
Chapter 4
The Data Path (1)
Figure 4-1. The data path of the example
microarchitecture used in this chapter.
The Data Path (2)
Figure 4-2. Useful combinations of ALU signals
and the function performed.
Data Path Timing (1)
Figure 4-3. Timing diagram of one data path cycle.
Data Path Timing (2)
Activities of subcycles with subcycle length:
Memory Operation
Figure 4-4. Mapping of the bits in MAR to the address bus.
Microinstructions (1)
Functional Signal Groups:
9 Signals to control writing data from C bus into registers.
9 Signals to enable registers onto B bus for ALU input.
8 Signals to control ALU and shifter functions.
2 Signals to indicate memory read/write via MAR/MDR.
1 Signal to indicate memory fetch via PC/MBR.
Microinstructions (2)
Figure 4-5. The microinstruction format for the Mic-1.
Microinstructions (3)
Groups of signals:
Addr – Contains address of potential next microinstruction.
JAM – Determines how te next microinstruction selected.
ALU – ALU and shifter functions.
C – Selects which registers written from C bus.
Mem – Memory functions.
B – Selects B bus source; encoded as shown.
Microinstruction Control: The Mic-1 (1)
The sequencer must produce two kinds of information each cycle:
• The state of every control signal in the system
• The address of the microinstruction that is to be executed next
Microinstruction Control: The Mic-1 (2)
Figure 4-6. The complete block diagram of our example
microarchitecture, the Mic-1.
Microinstruction Control: The Mic-1 (3)
Figure 4-6. The complete block diagram of our example
microarchitecture, the Mic-1.
Microinstruction Control: The Mic-1 (4)
In all cases, MPC can take on only one of two possible values:
• The NEXT ADDRESS
• The NEXT ADDRESS with the high-order bit ORed with 1
Microinstruction Control: The Mic-1 (5)
Figure 4-7. A microinstruction with JAMZ set to 1
has two potential successors.
Stacks (1)
Figure 4-8. Use of a stack for storing local variables. (a) While A is
active. (b) After A calls B. (c) After B calls C. (d) After C and B
return and A calls D.
Stacks (2)
Figure 4-9. Use of an operand stack for doing
an arithmetic computation
The IJVM Memory Model (1)
Defined areas of memory
•
•
•
•
The constant pool
The Local variable frame
The operand stack
The method area
The IJVM Memory Model (2)
Figure 4-10. The various parts of the IJVM memory.
The IJVM Instruction Set (1)
Figure 4-11. The IJVM instruction set. The operands
byte, const, and varnum are 1 byte. The operands
disp, index, and offset are 2 bytes.
The IJVM Instruction Set (2)
Figure 4-11. The IJVM instruction set. The operands
byte, const, and varnum are 1 byte. The operands
disp, index, and offset are 2 bytes.
The IJVM Instruction Set (3)
Figure 4-12. (a) Memory before executing INVOKEVIRTUAL.
(b) After executing it.
The IJVM Instruction Set (4)
Figure 4-13. (a) Memory before executing IRETURN.
(b) After executing it.
Compiling Java to IJVM (1)
Figure 4-14. (a) A Java fragment. (b) The corresponding Java
assembly language. (c) The IJVM program in hexadecimal.
Compiling Java to IJVM (2)
Figure 4-15. The stack after each instruction of Fig. 4-14(b).
Microinstructions and Notation
Figure 4-16. All permitted operations. Any of above operations
may be extended by adding ‘‘<< 8’’ to them to shift result left by
1 byte. Example: common operation H = MBR << 8.
Implementation of IJVM Using the Mic-1
Fig. 4-17. The microprogram for the Mic-1 (1 of 5)
Implementation of IJVM Using the Mic-1
Fig. 4-17. The microprogram for the Mic-1 (2 of 5)
Implementation of IJVM Using the Mic-1
Fig. 4-17. The microprogram for the Mic-1 (3 of 5)
Implementation of IJVM Using the Mic-1
Fig. 4-17. The microprogram for the Mic-1 (4 of 5)
Implementation of IJVM Using the Mic-1
Fig. 4-17. The microprogram for the Mic-1 (5 of 5)
Implementation of IJVM Using the Mic-1 (2)
Figure 4-18. The BIPUSH instruction format
Implementation of IJVM Using the Mic-1 (3)
Figure 4-19. (a) ILOAD with a 1-byte index.
(b) WIDE ILOAD with a 2-byte index.
Implementation of IJVM Using the Mic-1 (4)
Figure 4-20. The initial microinstruction sequence for ILOAD and
WIDE ILOAD. The addresses are examples.
Implementation of IJVM Using the Mic-1 (5)
Figure 4-21. The IINC instruction has two different operand fields
Implementation of IJVM Using the Mic-1 (6)
Figure 4-22. The situation at the start of various microinstructions.
(a) Main1. (b) goto1. (c) goto2. (d) goto3. (e) goto4.
Speed versus Cost
Basic approaches for increasing the speed of execution:
• Reduce # of clock cycles needed to execute an instruction
• Simplify organization so that clock cycle can be shorter
• Overlap execution of instructions
Merging Interpreter Loop with Microcode (1)
Figure 4-23. Original microprogram sequence for executing POP.
Merging Interpreter Loop with Microcode (2)
Figure 4-24. Enhanced microprogram sequence
for executing POP
Three-Bus Architecture (1)
Figure 4-25. Mic-1 code for executing ILOAD
Three-Bus Architecture (2)
Figure 4-26. Three-bus code for executing ILOAD.
Instruction Fetch Unit (1)
For every instruction the following operations may occur:
•
•
•
•
•
PC passed through ALU and incremented.
PC used to fetch next byte in instruction stream.
Operands read from memory.
Operands written to memory.
The ALU does computation and results stored back.
Instruction Fetch Unit (2)
Figure 4-27. A fetch unit for the Mic-1.
Instruction Fetch Unit (3)
Figure 4-28. A finite-state machine for implementing the IFU.
Design with Prefetching: The Mic-2 (1)
Figure 4-29. The data path for Mic-2.
Design with Prefetching: The Mic-2 (2)
Figure 4-29. The data path for Mic-2.
Pipelined Design: The Mic-3 (1)
Major components to the actual data path cycle:
• The time to drive the selected registers onto the A and B buses
• The time for the ALU and shifter to do their work
• The time for the results to get back to the registers to be stored
Pipelined Design: The Mic-3 (2)
Figure 4-30. The microprogram for the Mic-2 (part 1 of 3).
Pipelined Design: The Mic-3 (2)
Figure 4-30. The microprogram for the Mic-2 (part 2 of 3).
Pipelined Design: The Mic-3 (2)
Figure 4-30. The microprogram for the Mic-2 (part 3 of 3).
Pipelined Design: The Mic-3 (3)
Figure 4-31. The three-bus data path used in the Mic-3.
Pipelined Design: The Mic-3 (3)
Figure 4-31. The three-bus data path used in the Mic-3.
Pipelined Design: The Mic-3 (4)
Figure 4-32. The Mic-2 code for SWAP.
Pipelined Design: The Mic-3 (5)
Figure 4-33. The implementation of SWAP on the Mic-3.
Pipelined Design: The Mic-3 (6)
Figure 4-34. Graphical illustration of how a pipeline works.
Pipelined Design: The Mic-3 (6)
Figure 4-34. Graphical illustration of how a pipeline works.
Seven-Stage Pipeline: The Mic-4 (1)
Figure 4-35. The main components of the Mic-4.
Seven-Stage Pipeline: The Mic-4 (2)
Figure 4-36. The Mic-4 pipeline.
Cache Memory
Figure 4-37. A system with three levels of cache.
Direct-Mapped Caches (1)
Each cache entry consists of three parts:
• Valid bit indicates whether there is any valid data in this entry
• Tag with unique, 16-bit value identifying corresponding line of
memory from which data came
• Data field contains copy of data in memory.
Holds one cache line of 32 bytes.
Direct-Mapped Caches (2)
Figure 4-38. (a) A direct-mapped cache.
(b) A 32-bit virtual address.
Direct-Mapped Caches (2)
Any given place in memory maps to
exactly 1 cache location
How many words map to each cache
line?
232/212 = 220, or 1,048,576
A direct mapped cache with a 20 bit tag and 212 locations.
Direct Mapped Cache [contd…]
• What is the size of cache ?
4K
• If I read
0000 0000 0000 0000 0000 0000 1000 0001
• What is the index number checked ? 64
Direct-Mapped Caches (2)
A
B
C
D
E
F
G
H
I
J
H
K
L
M
…
Betty
Bob
Jim
It’s like a parking lot where you must park in the (only) slot marked
with the first letter of your last name.
Direct-Mapped Caches (3)
TAG field corresponds to Tag bits stored in cache entry.
LINE field indicates which cache entry holds corresponding
data, if present.
WORD field tells which word within a line is referenced.
BYTE field usually not used, but if only single byte is
requested, tells which byte within word is needed.
Direct Mapped
Set-Associative Caches
Figure 4-39. A four-way set-associative cache.
Associative Caches
• Block 12 placed in 8 block cache:
– Fully associative, direct mapped, 2-way set associative
– S.A. Mapping = Block Number Modulo Number Sets
Fully associative:
block 12 can go
anywhere
Block
no.
01234567
Direct mapped:
block 12 can go
only into block 4
(12 mod 8)
Block
no.
01234567
Set associative:
block 12 can go
anywhere in set 0
(12 mod 4)
Block
no.
Block-frame address
Block
no.
1111111111222222222233
01234567890123456789012345678901
01234567
Set Set Set Set
0 1 2 3
Set Associative Cache
• N-way set associative: N entries for each Cache Index
– N direct mapped caches operates in parallel
• Example: Two-way set associative cache
– Cache Index selects a “set” from the cache
– The two tags in the set are compared to the input in parallel
– Data is selected based on the tag result
Valid
Cache Tag
:
:
Adr Tag
Compare
Cache Index
Cache Data
Cache Data
Cache Block 0
Cache Block 0
:
:
Sel1 1
Mux
0 Sel0
OR
Hit
Cache Block
Cache Tag
:
Compare
Valid
:
Example: 4-way set associative
Cache
What is the cache size in this case ?
•
Disadvantages of Set Associative
Cache
N-way Set Associative Cache versus Direct Mapped Cache:
– N comparators vs. 1
– Extra MUX delay for the data
– Data comes AFTER Hit/Miss decision and set selection
• In a direct mapped cache, Cache Block is available BEFORE
Hit/Miss:
Valid
Cache Tag
:
:
Adr Tag
Compare
Cache Index
Cache Data
Cache Data
Cache Block 0
Cache Block 0
:
:
Sel1 1
Mux
0 Sel0
OR
Hit
Cache Block
Cache Tag
:
Compare
Valid
:
Fully Associative Cache
• Fully Associative Cache
– Forget about the Cache Index
– Compare the Cache Tags of all cache entries in parallel
– Example: Block Size = 32 B blocks, we need N 27-bit
comparators
• By
31 definition: Conflict Miss = 0 for a fully associative cache
4
Cache Tag (27 bits long)
0
Byte Select
Ex: 0x01
=
Byte 31
=
Byte 63
:
Valid Bit Cache Data
Byte 1 Byte 0
:
Cache Tag
Byte 33 Byte 32
=
=
:
=
:
:
Cache Misses
• Compulsory (cold start or process migration, first reference):
first access to a block
– “Cold” fact of life: not a whole lot you can do about it
– Note: If you are going to run “billions” of instruction,
Compulsory Misses are insignificant
• Capacity:
– Cache cannot contain all blocks access by the program
– Solution: increase cache size
• Conflict (collision):
– Multiple memory locations mapped
to the same cache location
– Solution 1: increase cache size
– Solution 2: increase associativity
• Coherence (Invalidation): other process (e.g., I/O) updates
memory
Branch Prediction
Figure 4-40. (a) A program fragment. (b) Its translation to a
generic assembly language.
Dynamic Branch Prediction (1)
Figure 4-41. (a) 1-bit branch history. (b) 2-bit branch history.
(c) Mapping between branch instruction address, target address.
Dynamic Branch Prediction (2)
Figure 4-42. A 2-bit finite-state machine for branch prediction.
Out-of-Order Execution, Register Renaming (1)
Figure 4-43. A superscalar CPU with in-order
issue and in-order completion.
Out-of-Order Execution, Register Renaming (2)
Figure 4-43. A superscalar CPU with in-order
issue and in-order completion.
Out-of-Order Execution, Register Renaming (3)
Figure 4-44. Operation of a superscalar CPU with out-of-order
issue and out-of order completion.
Out-of-Order Execution, Register Renaming (4)
Figure 4-44. Operation of a superscalar CPU with out-of-order
issue and out-of order completion.
Speculative Execution
Figure 4-45. (a) A program fragment.
(b) The corresponding basic block graph.
Core i7’s Sandy Bridge Microarchitecture
Figure 4-46. The block diagram of the Core i7’s Sandy Bridge
microarchitecture.
Core i7’s Sandy Bridge Pipeline (1)
Figure 4-47. A simplified view of the Core i7 data path.
Core i7’s Sandy Bridge Pipeline (2)
Scheduler queues send micro-ops into the 6 functional units:
•
•
•
•
•
•
ALU 1 and the floating-point multiply unit
ALU 2 and the floating-point add/subtract unit
ALU 3 and branch processing and floating-point compare unit
Store instructions
Load instructions 1
Load instructions 2
OMAP4430’s Cortex A9 Microarchitecture
Figure 4-48. The block diagram of the OMAP4430’s
Cortex A9 microarchitecture.
OMAP4430’s Cortex A9 Pipeline (1)
Figure 4-49. A simplified representation of the
OMAP4430’s Cortex A9 pipeline.
OMAP4430’s Cortex A9 Pipeline (2)
Figure 4-49. A simplified representation of the
OMAP4430’s Cortex A9 pipeline.
Microarchitecture of the ATmega168
Microcontroller
Figure 4-50. The microarchitecture of the ATmega168.
End
Chapter 4