Intel

1
Intels P6Processor FamilyArchitecture
?a?a??t??? Ge?????? ?? 438 ?etapt???a??
?????f?????? 3? ?ate????s? ?????µ??e?
????te?t?????? ?p?????st?? ?e?µe???? ???µ??? 2001
2
???????e?e? ?pe?e??ast?? Intel IA-32

• P5
• Pentium
• Pentium MMX

• P6
• Pentium Pro
• Pentium II
• Celeron
• Pentium III

• Netburst
• Xeon
• Pentium IV

3
?a µ??? t?? ???????e?a? P6

• Pentium PRO
• 133-200MHz
• 66MHz bus
• 0.5µ
• 256, 512 1?? L2 cache
• 88KB L1 cache

• Pentium II
• ? ?ata?a??t??? ??d?s? t?? PPRO
• ???te???? L2 Cache 512KB Slot 1 vs Socket
• 233-300MHz(Klamath)-450MHz (Deschutes)
• 66-100MHz bus
• 0.25µ
• ?a??te??? 16bit code
• 16KB L1 cache

• Celeron
• ??????µ??? e?d?s? t?? PII
• 66-100MHz bus
• 128-256 L2 Cache
• 266MHz 1.2GHz
• 0.25µ-0.18-0.13µ
• SIMD
• 32KB L1 Cache

• Pentium III
• 450550MHz(Katmai)1.33GHz (Coppermine)
• 133MHz bus
• 0.25µ 0.18µ-013µ
• SIMD epe?t?se??
• 256KB on chip cache
• Single Edge Contact Cartridge v2

4
??ta????sµ??(?at? t?? eµf???s? t?? ?e???? P6)

• CPUs
• AMDs K5
• NexGens Nx586
• Cyrixs M1

• ??e??e?t?µata
• ?a??te?? pipelining
• Chip te???????a? 0.5µ a????? ?a? 0.35µ
• 133MHz CPU Clock
• On chip L2 cache 256KB

• St???e?a p?? eµfa?????ta? ap? a?ta????st??
• Micro-ops
• Dynamic Execution (out of order)
• Branch Prediction

5
P6 S???pt???
6
? d?af??? t?? a???te?t?????? P6

• S?µf??es? ??a???? / ???µ??
• ?p?t?????e? ße?t??s? t?? ap?d?s?? µe ?p?????se?
  f????? te???????e? µ??µ??.
• ??a??st?p???s? t?? ????s?? st? bus t??
  s?st?µat??. ? P6 ???s?µ?p??e? ????te?? t?? 25
  t?? bandwidth a?t?? gt ?e??ss?te?e? CPUs ?/?a?
  IO.
• S?µf??es? Se???a??? ??t??es?? ??t????
• Dynamic Execution Out Of Order, Data flow
  Analysis, Branch Prediction, Speculative
  Execution.
• Deeper Pipelining
• ?e?t??s? ?atas?e???
• ?a?µ?? Sµ?????s??.
• ????s? s?????t?ta? ?e?t?????a? a????? e??te????
  ?a? ?pe?ta ?a? es?te???? .

7
Level 3 Super scalar engine.

• ????a??e? t? µ??e??? t?? ???e f?s?? t?? pipeline
  ?a? a????e? ? a???µ?? t?? f?se??
• 3 ??????? µ???de? e?t??es??
• Fetch/Decode Unit
• Geµ??e? t? Instruction pool µe e?t???? µe ß?s?
  t?? instruction pointer. Se pe??pt?s? branch
  p??spa?e? ?a p??ß???e? st???.
• Dispatch/execute Unit
• ?p? t? instruction pool e?te?e? ?p??a e?t???
  ??e? d?a??s?µ??? t??? te?est?? t??. Dataflow
  analysis ??a ?a e?t?p?ste? ? se??? µe t?? ?p??a
  ?a e?te?este? ta??te?a t? p????aµµa
• ?a ap?te??sµata de? ???st???p?????ta?
  (speculative execution)
• Retirement Unit
• ??et??e? t? Instruction Pool ??a e?t???? ??
  ?p??e? p??pe? ?a ???st???p??????? ?a? st???e? t??
  a??a??? st? µ??µ?.

• ?p?d?se??
• Fetch/Decode µ???? 3 e?t???? a?? ?????
• Dispatch/execute µ???? 5 e?t???? a?? ????? (3
  t?p???)
• Retirement µ???? 3 e?t???? a?? ?????

8
Non Blocking Caches

• Dual Ported Caches
• L1 Performance
• ??a LOAD ?a? µ?a STORE ta?t?????a se ???e ?????.
• Non-Blocking - 4 Stage Pipelined L2 Cache
• ? ap?t???a e??es?? µ?a? ?ata????s?? de? p???a?e?
  pa?s? ?e?t?????a? t?? CPU.
• 4 a?t?se?? e??p??et???ta? ?a? ep?p???? 12
  µp????? ?a ß??s???ta? se a?aµ???.
• Transaction Bus
• ?p?st????? d?s???????. 4 a?t?se?? µp????? ?a
  e???eµ??? ??a ??a? p6 e?? 8 ??a t? s????? t??
  d?a????.
• ?e ?????ta? ?????? se a?aµ????.
• Out Of Order Bus
• ?? p??t?????? d?a?e???s?? t?? d?a???? µe
  µ??a??sµ??? Retry Defer, se ??a MP pe??ß?????
  ap?????eta? e?t?? se???? (Out Of Order).
• 64GB Cacheable memory

9
MP Ready

• ?p????????a µe L2 Cache se ?d??t??? d?a???
• ?pe?e?????s? ??a???? ?p???ste?? Hardware
  ?p?st??????
• MESI
• Snooping memory
• G?a t?? e?????t?ta t?? ?at?stas?? t?? cache.
• 4 CPUs ??a p???? e?µet???e?s? d?a????
• ??s?µ?t?s? APIC
• ?p???ste?? Hardware ?p?st??????

10
P6 ???f??a
Dual Independent Bus Architecture (DIB) ??a????
es?te????? ?e?t?????a? a?e???t?t?? a?t?? t??
ep????????a? µe t? s?st?µa, ?? ?p???? ?e?t???????
se e?te??? d?af??et???? s????t?te?. L1 Code
Cache 4 way set associative, ???a??µ??? se
??aµµ?? t?? 32 bytes, pa?a??????e? t?? a?t?se??
Next_IP L1 Data Cache 2 way set associative
???a??µ??? se ??aµµ?? t?? 32 bytes, pa?a??????e?
t?? ?e???t??e? d?e????se?? t?? µ???d?? e?t??es??
e?t????. Registers 40 f?s???? ?ata????t?? t?? 64
bits ??a ???s? se renaming. 8 data registers
???????, ??at?? st?? p????aµµat?st?. 8 FP
Registers ?a? 8 MMX Registers ?? ?p????
ta?t????ta? µe t??? FP. 6 16 bit segment
registers ?a? ??a? flag register. Performance
Counters ?s?te????? ?ata????t?? p?? t?????
stat?st??? st???e?a.
11
P6 Functional Units Diagram

• Key points
• In order Section
• Instruction Cache
• Instruction Fetch / Decode Units
• Data Cache
• Out of Order Section
• Execution Engine
• Micro-ops (load / store)
• Reservation Stations

12
Pipeline Stages
??e???? branch, a??a?? Next ID
Commit
??t??es? ??t????
??t??s? pe??st??f? s?µe??s? 16 bytes
S?µe??s? ?a? ap?st??? se Instruction Pool
???µ???s? IP
?p???d???p???s?
?et???µas?a ?ata????t??
?p?fas? ?a? ???t?s? ap? IP e?t???? p??? e?t??es?
13
Instruction Fetching

• Instruction Prefetcher
• Next IP ?????e? t?? ep?µe?? ??s? µ??µ?? ap? t??
  ?p??a ??te?ta? e?t???.
• Instruction Cache ?pa?t? µe 16 st????sµ??a
  bytes st?? a?t?s? t?? Next IP
• 3 instruction Decoders pa?a?aµß????? ta 16 bytes
  af?? pe??st?af??? ?a? s?µe?????? ?? e?t????
  (start/end)
• 128bit bus

14
Instruction Decoders

• Decorders
• ??a? general decoder µ?a macro op a?? ?????.
• ??? simple decoders µ?a micro op a?? ?????. ??
  s??a?t?s??? macro op gt stall ??a ??a ????? ?a?
  µetaf??? t?? st?? general
• Micro ops Macro ops
• Simple instructions of the register-register form
  are only one micro-op.
• Load instructions are only one micro-op.
• Store instructions have two micro-ops.
• Simple read-modify instructions are two
  micro-ops.
• Simple instructions of the register-memory form
  have two to three micro-ops.
• Simple read-modify write instructions are four
  micro-ops.
• Complex instructions generally have more than
  four micro-ops, therefore they take multiple
  cycles to decode.

15
Instruction Decoding (µ-ops)

• ? Decoder µetaf???e? ???e IA32 e?t??? se ??a
  t??ad??? µ-op
• 2-logical sources, 1 logical destination
• S?????? 1 IA32 instruction -gt 1 µ-op
• ??? sp???a 1 IA32 instruction -gt 4 µ-op
• ???? sp???a Microcode Instruction Sequencer.
• Microcode µ?a se??? ap? p??e???aµ??e? a???????e?
  ap??? µ-ops.

16
Instruction Decoding (RAT)

• ?? µ-ops ap?st????ta? st?? RAT (Register Alias
  Table) e? se???
• ???p??e?ta? register renaming ???????
  ?ata????t?? IA32 se f?s????? P6
• ?? e?t???? µ-ops pe??st???????ta? µe bits
  ?at?stas?? (allocator stage)
• ?? ap?t??esµa p????e?ta? st? ???? a?aµ????
  e?t???? (instruction pool)
• Instruction pool ReOrder Buffer (ROB)
  ???p??e?ta? sa? content addressable memory t?? 40
  e???af??

17
Instruction Dispatch/Execution

• ?p??????ta? e?t???? ap? t? Instruction Pool µe
  ß?s? t? status t???.
• ?????eta? a? e??a? d?a??s?µ?? ?? te?est??
• ?????eta? a? e??a? d?a??s?µ? ? apa?t??µe?? µ???da
  e?t??es??.
• ? RS ap?µa????e? t?? e?t??? ap? t? ROB ?a? t??
  p????e? st? µ???da. ?e? ??e? s?µas?a ? se??? st?
  ROB.
• 5 ported RS (?p?? st? s??µa)
• max 5 µ-ops per cycle, 3 sustained
• Data flow analysis (?e?d? FIFO)

18
Instruction Dispatch/Execution (Branches)

• ??p??e? ap? t?? µ-ops e??a? branches.
• S?µe?????ta? (tagged) ?a??? e?s????ta? e?t??
  se???? µe t? d?e????s? µetap?d?s?? ?a? t?
  d?e????s? ap???????.
• ?ta? t? branch p?a?µat??? e?te?este? s????????ta?
  ta st???e?a µe e?e??a p?? ?p?t????a?.
• Se ep?t???a t? branch ep??????eta? (retirement)
  ?a??? ?a? ??e? ?? e?t???? µeta?? a?t?? ?a? t??
  ep?µe??? branch (speculative execution).
• ?? BTB p??ß??pe? t?? pe??ss?te?e? a??? ??? ??e?.
  ?e????e? 512 e???af?? p??????µe??? branches ?a?
  st????.
• Se ap?t???a ? Jump Execution Unit (JEU)
  ap?µa????e? ??e? t?? e?t???? ap? t? ROB ?a? t?
  pipeline s??e???e? ap? t? ??a s?st? d?e????s?.
• ??st??
• Not taken on hit ?aµ?a ep?pt?s?
• Taken on hit ?a??st???s? e??? ?????? (p??s?
  fetch ?a? issue)
• Mis-predicted e????st? ??st?? 9 ?????? (t?
  µ???? t?? in order issue pipeline, ??a ????? IF
  ?a? t? ????? t?? ep?????s?? t?? s?st?? branch).
  ??p??? 10, µ???st? 26 ??????.
• Static predictions, conditional ?a? unconditional
  on-hit 5-6 ??????.

19
Branch Prediction (Dynamic)

• 4 branch predictions/??aµµ? 128bits.
• 2 level adaptive(Yeh method).
• 4 bits p????f???a? a?? branch. ???ß??pe?
  a???????e? branches.
• Return Stack Buffer (RET instructions)
• ??e???? BTB ??a p??????µe?? e?t??es? t?? branch.
• ?? de? ?p???e? BTB entry ???eta? static
  prediction gt e??µ???s? BTB.
• ???s???? e?t???? CMOV.

20
Branch Prediction (Static)
21
Retire Unit

• ??et??e? t?? ?at?stas? t?? µ-ops st? ROB.
• ?se? ????? ?????????e? ?a p??pe? ?a ap?µa????????
  ap? t? ROB.
• ???pe? ?a f???t?se? ??a t?? ???? t???s? t??
  se???? t??? sa? IA32 instructions ?a? µ???sta e?
  µeta?? interrupts, traps, faults, breakpoints ?a?
  mis-predictions.
• ?a??? ap?fas??e? t?? ep?µe?e? e?t???? p???
  ep?????s?, a?t?? t?? p????e?, in-order st??
  Retirement Register file.

22
Bus Interface Unit

• ??? t?p?? e?t???? loads (1 µ-op, address, width,
  register) ?a? stores (2 µ-ops, µ?a ?e??? t?
  d?e????s? ?a? µ?a ta ded?µ??a)
• ??t? ta stores de? e?te????ta? speculatively
  ??at? de? ?p???e? d??at?t?ta undo.
• ??t? ta stores de? a?as??t?ss??ta? µeta?? t???.
• ??a store e?te?e?ta? µ??? ?ta? ?a? ta d?? µ-ops
  t?? e??a? ?t??µa ?a? de? e???eµ??? p??????µe?a
  stores
• ?? MOB e??a? ??t? sa? RS ?a? reorder buffer ??a
  loads ?a? stores, t? ?p??? ep?t??pe? loads ?a
  pe????? loads ?a? stores ?a? ta epa?e????e? ?ta?
  e?de??µe?e? s?????e? µp???a??sµat?? a????ta?
  (dependency of resources)

23
St???e?a ??t??es??

• ??a load ?a? ??a store st?? ?d?a d?e????s?
  µp????? ?a e?te?est??? st?? ?d?? ?????.
• ? ?a??st???s? t?? stores de? ??e? µe???? s?µas?a
  st?? ap?d?s? (3-5 ?at? t?? Intel).
• Register Renaming ???e read e??? ???????
  register a?af??eta? st?? ?d?? f?s???. ???e
  ep?µe?? write a?af??eta? se ??? f?s??? register.
• ?e? e??a? d??at?? ?a e?te?este? µ?a FMUL se ?????
  aµ?s?? ep?µe?? t?? p??????µe??? t??.
• FPU stages ?p?? P5, d??
• ?etat??p? te?est?? se es?te???? format
• ??t??es? ?e?t?????a? se ?????te?? ep?ped?
  a???ße?a?
• St???????p???s? ?a? µetat??p? te?est?? se
  standard format
• ??af??? sf??µat??

24
Execution Modes

• Protected mode. The native state of the
  processor. All instructions and architectural
  features are available, providing the highest
  performance and capability. Recommended mode for
  all new applications and operating systems. Also
  offers the ability to directly execute
  real-address mode 8086 software in a protected,
  multi-tasking environment. This (Virtual-8086)
  mode is not actually a processor modebut a
  protected mode attribute that can be enabled for
  any task.
• Real-address mode. Provides the programming
  environment of the Intel 8086 processor, with a
  few extensions (such as the ability to switch to
  protected or system management mode). The
  processor is placed in real-address mode
  following power-up or a reset. From real-address
  mode, only a single instruction is required to
  switch to protected mode.
• System management mode. A standard architectural
  feature unique to all Intel processors, beginning
  with the Intel386 SL processor. Provides an
  operating system or executive with a transparent
  mechanism for implementing platform-specific
  functions such as power management. The processor
  enters SMM the external SMM interrupt pin (SMI)
  is activated or an SMI is received from the
  advanced programmable interrupt controller
  (APIC). In SMM, the processor switches to a
  separate address space while saving the entire
  context of the currently running program or task.
  SMM-specific code may then be executed
  transparently. Upon returning from SMM, the
  processor is placed back into its state prior to
  the system management interrupt.

25
Addressing Modes

• Flat memory model memory appears to a program
  as a single, continuous address space, called a
  linear address space. Code (a programs
  instructions), data, and the procedure stack are
  all contained in this address space. The linear
  address space is byte addressable.
• Segmented memory model memory appears to a
  program as a group of independent address spaces
  called segments. Code, data, and stacks are
  typically contained in separate segments. To
  address a byte in a segment, a program must issue
  a logical address (far pointer), which consists
  of a segment selector and an offset. The segment
  selector identifies the segment to be accessed
  and the offset identifies a byte in the address
  space of the segment. Up to 16,383 segments of
  different sizes and types. The processor
  translates each logical address into a linear
  address to access a memory location,
  transparently to the application program.
  Increases the reliability of programs and
  systems.
• Real-address model same as the Intel 8086
  processor, for backward compatibility. Uses a
  specific implementation of segmented memory in
  which the linear address space for the program
  and the operating system/executive consists of an
  array of equally sized segments.
• In Protected Mode all modes can be used
• In Real and SM mode only the Real-address mode is
  available

26
???f??a T?µata

• Data flow analysis
• ?? pe??ss?te?e? t?? µ?a? e?t???? e??a? ?t??µe?
  st? ROB p??? ap?st??? se µ?a µ???da e?t??es??
  t?te ???eta? ep????? ß?se? ?e?d? FIFO a??????µ??
  ? ?at??????.
• ECC
• Se ep??tas? t?? e?????? ?s?t?µ?a? sta s?µata t??
  address bus ?a? t?? e?????? d?s???????,
  ?p?st????e? ??e??? ?a? d?????s? sfa??µt?? sta
  data signals ?a? ??a t? d?a??? t?? L2 Cache a???
  ?a? t? d?a??? s?st?µat??. ?ts? p??stat????ta?
  ???s?µa ded?µ??a, af?? d????????ta? single bit
  errors ?a? e?t?p????ta? ta double bit errors. ?a
  sf??µata µp????? ?a ?ata???f??ta? ?a? ?ts? st?
  s????e?a ?a e?t?p????ta? ?? ast???e? t??
  s?st?µat??.
• MMX
• Math Matrix Extensions. ??p?? SIMD.
• ?p??tas? t?? ßas???? ?epe?t????? e?t???? t?? x86
  ??a t? d?a?e???s? e????a? video ?a? ????, t?
  ?p??? eµfa???eta? st??? P5.
• Streaming SIMD extensions
• 70 ??e? e?t???? p?? ep?t??p??? d?a?e???s?
  e????a?, 3D ape?????s??, ????, video ?a?
  a?a?????s?? f????.
• Intel Processor Serial Number (PII ?a? PIII)
• ??s????a ??t??es?? Unaligned e?t????
• ??t???? p?? p????pt??? ap? 16bit ??d??a t??
  pa?e????t??. ?????ase st? ??????? eµf???s? t??
  PII
• PGE (page global enable)
• ?p?t??pe? t? s?µe??s? se??d?? sa? Global (p?
  kernel pages) ?ste ?a µ?? e??a?a?????ta? ??
  e???af?? t?? TLB ?at? t? context switching.

27
MESI MODEL
St???? ? pa?a???????s? t?? ?at?stas?? t?? ??aµµ??
t?? cache ????? t?? ?d??? ap?????? t???.

• Modified
• ?a ded?µ??a ????? a??a?te? ???? write hit. Ta
  p??pe? ?a ??aft??? p?s? st?? ????a µ??µ?.
• Exclusive
• ?a ded?µ??a p??a??tata pe??????ta? µ??? st?
  t????? cache
• Shared
• ?a ded?µ??a ß??s???ta? p??a??tata ?a? se ???a
  caches a??? de? ????? ???e? a??a???
• Invalid
• ?????? ?at?stas? ?p?? ? ??aµµ? t?? cache e??a?
  ?????

28
Associative Caches / CAM

• Associative
• ??a block a?t?st?????eta? se ?p??ad?p?te ??aµµ?
  t?? cache
• ? d?e????s? µetaf???eta? se tag ?a? word
• ?? tag µ??ad??? p??sd?????e? ??a block
• ?? tag ???e ??aµµ?? e????eta? ??a s?µpt?s? gt
  ?a??st???s? a?a??t?s??
• Set Associative
• ?????eta? se sets
• ???e set pe???aµß??e? ??p??e? ??aµµ?? (lines)
• ??a ded?µ??? block a?t?st?????eta? se µ?a ap? t??
  ??aµµ?? e??? ?a? µ??? set (n-way ?p?? n a???µ??
  ??aµµ??) ?? 2-way set associative pe????e? sets
  t?? d?? ??aµµ??. ????ap?? blocks a?t?st?????? st?
  ?d?? set.
• ?e???eta? d?aµµat??? ? ?????? a?a??t?s??
• CAM
• ?a ?d?a ded?µ??a pe?????f??? t? s?µe??
  ap????e?s??, pe????????ta? t?? a????e? a?a??t?s??
• ??p??a ded?µ??a ????? ??a ?a? µ??? ??a s?µe??
  ap????e?s?? t? ?p??? e??a? p???a????sµ???
• ?a pa??µ??a ded?µ??a ap????e???ta? se ?e?t??????
  ??se??
• ? ?a??? pa???e? ta ded?µ??a ?a? pa???e? t?
  d?e????s? µ?sa se ??a µ??? ?????
• ???sfata ???p??????a? hardware components CAM.

29
?p?te??sµata

• 90 ???? p??ß?e?? branches a??µa ?a? ?ta? a?t?
  ß??s???ta? se µe???? ß????
• 25 ???s? t?? bandwidth d?a???? s?st?µat??
• ?e?t??s? ep?d?se?? e?t???? a???a???

30
S?????t???
31
Rerefences

• Intels P6 Uses Decoupled Superscalar Design by
  Linley Gwennap (Microdesign Resources vol 9 No 2
  , 16 February 1995)
• The P6 Architecture Background Information for
  Developers by Intel Corporation
  (1995-p6arc.pdf)
• IA-32 Intel Architecture Software
  DevelopersManual Volume 2 Instruction Set
  Reference by Intel Corporation (2001
  24547104.pdf)
• Pentium Pro Family Developers Manual
  Vol2Programmers Reference Manual , by Intel
  Corporation (Dec 1995 - 24269101.pdf)
• The Pentium Pro At 150, 166 , 180 and 200MHz,
  by Intel Corporation (Jun 1997 - 24276905.pdf)
• The Intel Architecture Optimization Manual, by
  Intel Corporation (1997 -24281601.pdf)
• Pentium II Developers Manual, by Intel
  Corporation (Oct 1999 - 24350201.pdf)
• The Intel Celeron Processor up to 1.1GHz
  Datasheet, by Intel Corporation (Aug 2001 -
  24365819.pdf)
• The P6 Family of Processors Hardware
  Developers Manual, by Intel Corporation (Sep
  1998 - 24400101.pdf)
• Pentium III Processor at 450 MHz to 1.13 GHz
  Datasheet, by Intel Corporation (Jul 2000 -
  24445208.pdf)

32
Functional Units (e?a??a?t??? ???)