SlideShare a Scribd company logo

Lec5 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Branch Predictor

Download as PPT, PDF
Hsien-Hsin Sean Lee, Ph.D., profile picture
Hsien-Hsin Sean Lee, Ph.D.

This document discusses branch prediction in computer architecture. It begins by explaining what information is predicted for branches - the direction and target. It then categorizes different types of branches and discusses the costs of branch misprediction. Various branch prediction techniques are presented, starting with simple 1-bit and 2-bit predictors, and progressing to more advanced correlating and global history predictors. The goal of branch prediction is to reduce penalties from mispredicted branches by speculatively executing the predicted path.

Devices & Hardware
Related topics:
Computer Architecture
1 of 45
Downloaded 69 times
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
ECE 4100/6100 Advanced Computer Architecture Lecture 5 Branch Prediction Prof. Hsien-Hsin Sean Lee School of Electrical and Computer Engineering Georgia Institute of Technology
2 Predict What? • Direction (1-bit) – Single direction for unconditional jumps and calls/returns – Binary for conditional branches • Target (32-bit or 64-bit addresses) – Some are easy • One: Uni-directional jumps • Two: Fall through (Not Taken) vs. Taken – Many: Function Pointer or Indirect Jump (e.g. jr r31)
3 Categorizing Branches 8% 10% 82% 19% 6% 75% 0% 20% 40% 60% 80% 100% Call/Return Jump Conditional Branch Frequency of branch instructions SPEC2000INT SPEC2000FP Source: H&P using Alpha
4 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue
5 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Mispredict
6 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue (flush entailed instructions and refetch) Mispredict
7 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Fetch the correct path
8 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Mispredict 8-issue Superscalar Processor (Worst case)
9 Why Branch is Predictable? for (i=0; i<100; i++) { …. } addi r10, r0, 100 add r1, r0, r0 L1: … … … … addi r1, r1, 1 bne r1, r10, L1 … … if (aa==2) aa = 0; if (bb==2) bb = 0; if (aa!=bb) …. addi r2, r0, 2 bne r10, r2, L_bb xor r10, r10, r10 L_bb: bne r11, r2, L_xx xor r11, r11, r11 L_xx: beq r10, r11, L_exit … Lexit:
10 Control Speculation • Execute instruction beyond a branch before the branch is resolved  Performance • Speculative execution • What if mis-speculated? need – Recovery mechanism – Squash instructions on the incorrect path • Branch prediction: Dynamic vs. Static • What to predict?
11 Static Branch Prediction • Uni-directional, always predict taken (or not taken) • Backward taken, Forward not taken – Need offset information (when?) • Compiler hints with branch annotation – When the info will be available? Post-decode?
12 Simplest Dynamic Branch Predictor • Prediction based on latest outcome • Index by some bits in the branch PC – Aliasing T NT T T NT NT . . . for (i=0; i<100; i++) { …. } addi r10, r0, 100 addi r1, r1, r0 L1: … … … … addi r1, r1, 1 bne r1, r10, L1 … … 0x40010100 0x40010104 0x40010108 … 0x40010A04 0x40010A08 How accurate? NT T 1-bit Branch History Table
13 Typical Table Organization Hash PC (32 bits) . . . . . 2N entries Prediction N bits FSM Update Logic table update Actual outcome
14 Simplest Dynamic Branch Predictor T NT T T NT NT . . . addi r10, r0, 100 addi r1, r1, r0 L1: add r21, r20, r1 lw r2, (r21) beq r2, r0, L2 … … j L3 L2: … … … L3: addi r1, r1, 1 bne r1, r10, L1 0x40010100 0x40010104 0x40010108 0x4001010c 0x40010110 0x40010210 0x40010B0c 0x40010B10 for (i=0; i<100; i++) { if (a[i] == 0) { … } … } NT T 1-bit Branch History Table
15 FSM of the Simplest Predictor • A 2-state machine • Change mind fast 00 11 If branch not taken If branch taken 00 11 Predict not taken Predict taken
16 Example using 1-bit branch history table for (i=0; i<44; i++) { …. } 00Pred Actual T T √ 11 11 √ T T √ 11 11 addi r10, r0, 4 addi r1, r1, r0 L1: … … addi r1, r1, 1 bne r1, r10, L1 NT 00   T 11  T √ 11 √ T T √ 11 11 NT 00  T 11  60% accuracy
17 2-bit Saturating Up/Down Counter Predictor Not Taken Taken Predict Not taken Predict taken ST: Strongly Taken WT: Weakly Taken WN: Weakly Not Taken SN: Strongly Not Taken 01/ WN 01/ WN 00/ SN 00/ SN 10/ WT 10/ WT 11/ ST 11/ ST MSB: Direction bit LSB: Hysteresis bit
18 2-bit Counter Predictor (Another Scheme) Not Taken Taken Predict Not taken Predict taken ST: Strongly Taken WT: Weakly Taken WN: Weakly Not Taken SN: Strongly Not Taken 01/ WN 01/ WN 00/ SN 00/ SN 11/ ST 11/ ST 10/ WT 10/ WT
19 Example using 2-bit up/down counter for (i=0; i<44; i++) { …. } 0101Pred Actual T T √ 1010 1111 √ T T √ 1111 1111 addi r10, r0, 4 addi r1, r1, r0 L1: … … addi r1, r1, 1 bne r1, r10, L1 NT 1010   T 1111 √ T √ 1111 √ T T √ 1111 1111 NT 1010  T 1111 √ 80% accuracy
20 Branch Correlation • Branch direction – Not independent – Correlated to the path taken • Example: Path 1-1 of b3 can be surely known beforehand • Track path using a 2-bit register if (aa==2) // b1 aa = 0; if (bb==2) // b2 bb = 0; if (aa!=bb) { // b3 ……. } b1 b2 b2 b3 b3 b3 1 (T) 1 1 0 (NT) 0 b3 0 Path: A:1-1 B:1-0 C:0-1 D:0-0 aa=0 bb=0 aa=0 bb≠2 aa≠2 bb=0 aa≠2 bb≠2 Code Snippet
21 Correlated Branch Predictor [PanSoRahmeh’92] • (M,N) correlation scheme – M: shift register size (# bits) – N: N-bit counter 2-bit counter hash . . . . X X Branch PC hash 2-bit counter . . . . 2-bit counter . . . . X X 2-bit counter . . . . 2-bit counter . . . . Prediction Prediction 2-bit shift register (global branch history) select Subsequent branch direction (2,2) Correlation Scheme 2-bit Sat. Counter Scheme 2w w Branch PC
22 Two-Level Branch Predictor [YehPatt91,92,93] • Generalized correlated branch predictor • 1st level keeps branch history in Branch History Register (BHR) • 2nd level segregates pattern history in Pattern History Table (PHT) 1 1 . . . . . 1 0 00…..00 00…..01 00…..10 11…..11 11…..10 Branch History Pattern Pattern History Table (PHT) Prediction Rc-k Rc-1 Rc: Actual Branch Outcome FSM Update Logic Branch History Register (BHR) (Shift left when update) N 2N entries Current StatePHT update
23 Branch History Register • An N-bit Shift Register = 2N patterns in PHT • Shift-in branch outcomes – 1 ⇒ taken – 0 ⇒ not taken • First-in First-Out • BHR can be – Global – Per-set – Local (Per-address)
24 Pattern History Table • 2N entries addressed by N-bit BHR • Each entry keeps a countercounter (2-bit or more) for prediction – Counter update: the same as 2-bit counter – Can be initialized in alternate patterns (01, 10, 01, 10, ..) • Alias (or interference) problem
25 Global History Schemes Global BHR Global PHT GAg Global BHR .. SetP(B) Per-set PHTs (SPHTs) GAs Global BHR .. Addr(B) Per-addr PHTs (PPHTs) GAp ** [PanSoRahmeh’92] similar to GAp . . . . . . . . . . . . . . . . . . . . . Set can be determined by branch opcode, compiler classification, or branch PC address.
26 GAs Two-Level Branch Prediction 01100110 BHR PC = 0x4001000C . . . PHT 00110110 . . 00110110 00110111 11111101 11111110 00000000 00000001 00000010 11111111 10 MSB = 1 Predict Taken The 2 LSBs are insignificant for 32-bit instruction
27 Predictor Update (Actually, Not Taken) 01100110 BHR PC = 0x4001000C . . . PHT 00110110 . . 00110110 00110111 11111101 11111110 00000000 00000001 00000010 11111111 1001 decremented 11001100 00111100 00111100 Wrong Predictio n • Update Predictor after branch is resolved
28 Per-Address History Schemes Global PHT PAg SetP(B) Per-set PHTs (SPHTs) PAs Addr(B) Per-addr PHTs (PPHTs) PAp . . . Addr(B) Per-addr BHT (PBHT) . . . Addr(B) Per-addr BHT (PBHT) . . . Addr(B) Per-set BHT (PBHT) •Ex: P6, Itanium •Ex: Alpha 21264’s local predictor . . . .. . . . . . . . . . .. . . . . . . . . .
29 PAs Two-Level Branch Predictor PC = 1110 0000 1001 1001 0010 1100 1110 1000 000 001 010 011 100 101 110 111 BHT 11010110 . . . PHT . . 11010101 11010110 11111101 11111110 00000000 00000001 00000010 11111111 MSB = 1 Predict Taken 11 110
30 Per-Set History Schemes Global PHT SAg SetP(B) Per-set PHTs (SPHTs) SAs Addr(B) Per-addr PHTs (PPHTs) SAp . . . Per-set BHT (SBHT) . . . SetH(B) Per-set BHT (SBHT) . . . SetH(B) Per-set BHT (SBHT) .. . . . . . . . . . . . . .. . . . . . . . . . SetH(B)
31 PHT Indexing Branch addr Global history Gselect 4/4 00000000 00000001 00000001 00000000 00000000 00000000 11111111 00000000 11110000 11111111 10000000 11110000 Insufficient History • Tradeoff between more history bits and address bits • Too many bits needed in Gselect ⇒ sparse table entries
32 Gshare Branch Predictor [McFarling93] • Tradeoff between more history bits and address bits • Too many bits needed in Gselect ⇒ sparse table entries • Gshare ⇒ Not to lose global history bits • Ex: AMD Athlon, MIPS R12000, Sun MAJC, Broadcom SiByte’s SB-1 Branch addr Global history Gselect 4/4 Gshare 8/8 00000000 00000001 00000001 00000001 00000000 00000000 00000000 00000000 11111111 00000000 11110000 11111111 11111111 10000000 11110000 01111111 GselectGselect 4/4: Index PHT by concatenateconcatenate low order 4 bits GshareGshare 8/8: Index PHT by {Branch address ⊕ Global history}
33 Gshare Branch Predictor . . . PHT . . 00 MSB = 0 Predict Not Taken 1 1 . . . . . 1 0 0 1 . . . . . 0 1 0 01. . . . .1 1 ⊕ PC Address Global BHR
34 Aliasing Example PHT BHR 1101 PC 0110 ---- XOR 1011 BHR 1001 PC 1010 ---- XOR 0011 1111 1110 1101 1100 1011 1010 1001 1000 0111 0110 0101 0100 0011 0010 0001 0000 PHT (indexed by 10) BHR 1101 PC 0110 ---- || 1001 BHR 1001 PC 1010 ---- || 1001 1111 1110 1101 1100 1011 1010 1001 1000 0111 0110 0101 0100 0011 0010 0001 0000 GApGAp GshareGshare
35 Hybrid Branch Predictor [McFarling93] • Some branches correlated to global history, some correlated to local history • Only update the meta-predictor when 2 predictors disagree P0P0 P1P1 . . . Choice (or Meta) Predictor Branch PC Final Prediction
36 Alpha 21264 (EV6) Hybrid Predictor Local History Table 1024 x 10 bits Single Local Predictor 1024 x 3 bits Global Predictor 4096 x 2 bits Choice Predictor 4096 x 2 bits Global history 12 Local prediction Global prediction Meta prediction Next Line/set Prediction L1 I-cache (64KB 2w) & TLB 4 instr./cycle4 instr./cycle Virtual address Final Branch Prediction PCPC 10 • A “tournament branch predictor” • Multi-predictor scheme w/ – Local predictorLocal predictor (~PAg) • Self-correlation – Global predictorGlobal predictor • Inter-correlation – Choice predictorChoice predictor as the decision maker: a 2-bit sat. counter to credit either local or global predictors. • Die size impact – History info tables ~2% – BTB ~ 2.7% (associated with I-$ on a per-line basis) • 2 cycle latency, we will discuss more later For Single-cycle Prediction
37 Alpha EV8 Branch Predictor Branch PC Global history F1 F2 F3 majority vote prediction G0 G1 Meta F4 Bimodal e-gskew predictor • Real silicon never sees the daylight • Use a 2Bc-gskew predictor (one form of enhanced gskew) – Bimodal predictor used as (1) static biased predictor and (2) part of e-gskew predictor – Global predictors G0 and G1 are part of e-gskew predictor – Table sizes: 352Kbits in total (208Kbits for prediction table; 144Kbits for hysteresis table.)
38 Branch Target Prediction • Try the easy ones first – Direct jumps – Call/Return – Conditional branch (bi-directional) • Branch Target Buffer (BTB) • Return Address Stack (RAS)
39 Branch Target Buffer (BTB) TargetTag TargetTag TargetTag… BTBBranch PC == == ==… ++ 4 Branch Target Predicted Branch Direction 0 1
40 Return Address Stack (RAS) • Different call sites make return address hard to predict – Printf() being called by many callers – The target of “return” instruction in printf() is a moving target • A hardware stack (LIFO) – Call will push return address on the stack – Return uses the prediction off of TOS
41 Return Address Stack • Does it always work? – Call depth – Setjmp/Longjmp – Speculative call? ++ 4 Call PC Push Return Address BTBBTB Return PC BTBBTB Return? • May not know it is a return instruction prior to decoding – Rely on BTB for speculation – Fix once recognize Return
42 Indirect Jump • Need Target Prediction – Many (potentially 230 for 32-bit machine) – In reality, not so many – Similar to predicting values • Tagless Target Prediction • Tagged Target Prediction
43 Tagless Target Prediction [ChangHaoPatt’97] 1 1 . . . . . 1 0 Branch History RegisterBranch History Register (BHR)(BHR) 00…..00 00…..01 00…..10 11…..11 11…..10 PC ⊗ BHR Pattern Target Cache (2N entries) Predicted Target Address Branch PCBranch PC HashHash • Modify the PHT to be a “Target Cache” – (indirect jump) ? (from target cache) : (from BTB) • Alias?
44 Tagged Target Prediction [ChangHaoPatt’97] • To reduce aliasing with set-associative target cache • Use branch PC and/or history for tags 1 1 . . . . . 1 0 BHR 00…..00 00…..01 00…..10 11…..11 11…..10 Target Cache (2n entries per way) Predicted Target Address Branch PC Hash n =? Tag Array
45 Multiple Branch Prediction • For a really wide machine – Across several basic blocks – Need to predict multiple branches per cycle • How to fetch non-contiguous instructions in one cycle? • Prediction accuracy extremely critical (will be reduced geometrically)

More Related Content

PDF
Vivado hls勉強会1（基礎編）
marsee101
 
PPTX
Introduction to DPDK
Kernel TLV
 
PDF
Intel dpdk Tutorial
Saifuddin Kaijar
 
PDF
Making Linux do Hard Real-time
National Cheng Kung University
 
PDF
LR parsing
ichikaz3
 
PDF
Smalltalkだめ自慢
Masashi Umezawa
 
PDF
Intel DPDK Step by Step instructions
Hisaki Ohara
 
PDF
CUDAのアセンブリ言語基礎のまとめ　PTXとSASSの概説
Takateru Yamagishi
 
Vivado hls勉強会1（基礎編）
marsee101
 
Introduction to DPDK
Kernel TLV
 
Intel dpdk Tutorial
Saifuddin Kaijar
 
Making Linux do Hard Real-time
National Cheng Kung University
 
LR parsing
ichikaz3
 
Smalltalkだめ自慢
Masashi Umezawa
 
Intel DPDK Step by Step instructions
Hisaki Ohara
 
CUDAのアセンブリ言語基礎のまとめ　PTXとSASSの概説
Takateru Yamagishi
 

What's hot (20)

PDF
CMake - Introduction and best practices
Daniel Pfeifer
 
PDF
深層学習向け計算機クラスター MN-3
Preferred Networks
 
PDF
RISC-V & SoC Architectural Exploration for AI and ML Accelerators
RISC-V International
 
PPT
Javaバイトコード入門
Kota Mizushima
 
PDF
２値化CNN on FPGAでGPUとガチンコバトル（公開版）
Hiroki Nakahara
 
PPTX
MySQLの運用でありがちなこと
Hiroaki Sano
 
PPTX
Understanding DPDK
Denys Haryachyy
 
PDF
DPDK In Depth
Kernel TLV
 
PDF
ELFの動的リンク
7shi
 
PDF
Deflate
7shi
 
PPTX
Deep Learningのための専用プロセッサ「MN-Core」の開発と活用（2022/10/19東大大学院「 融合情報学特別講義Ⅲ」）
Preferred Networks
 
PDF
OSC2011 Tokyo/Fall 濃いバナ(virtio)
Takeshi HASEGAWA
 
PDF
C++の話（本当にあった怖い話）
Yuki Tamura
 
PDF
スペクトラル・クラスタリング
Akira Miyazawa
 
PDF
「FPGA 開発入門：FPGA を用いたエッジ AI の高速化手法を学ぶ」
直久 住川
 
PDF
論文紹介：The wavelet matrix
Yuki Igarashi
 
PDF
条件分岐とcmovとmaxps
MITSUNARI Shigeo
 
PPTX
Gohan
Nachi Ueno
 
PPTX
LLVM Instruction Selection
Shiva Chen
 
PDF
MCC CTF講習会 pwn編
hama7230
 
CMake - Introduction and best practices
Daniel Pfeifer
 
深層学習向け計算機クラスター MN-3
Preferred Networks
 
RISC-V & SoC Architectural Exploration for AI and ML Accelerators
RISC-V International
 
Javaバイトコード入門
Kota Mizushima
 
２値化CNN on FPGAでGPUとガチンコバトル（公開版）
Hiroki Nakahara
 
MySQLの運用でありがちなこと
Hiroaki Sano
 
Understanding DPDK
Denys Haryachyy
 
DPDK In Depth
Kernel TLV
 
ELFの動的リンク
7shi
 
Deflate
7shi
 
Deep Learningのための専用プロセッサ「MN-Core」の開発と活用（2022/10/19東大大学院「 融合情報学特別講義Ⅲ」）
Preferred Networks
 
OSC2011 Tokyo/Fall 濃いバナ(virtio)
Takeshi HASEGAWA
 
C++の話（本当にあった怖い話）
Yuki Tamura
 
スペクトラル・クラスタリング
Akira Miyazawa
 
「FPGA 開発入門：FPGA を用いたエッジ AI の高速化手法を学ぶ」
直久 住川
 
論文紹介：The wavelet matrix
Yuki Igarashi
 
条件分岐とcmovとmaxps
MITSUNARI Shigeo
 
Gohan
Nachi Ueno
 
LLVM Instruction Selection
Shiva Chen
 
MCC CTF講習会 pwn編
hama7230
 
Ad

Viewers also liked (20)

PPT
Lec1 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Pipelining
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec6 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Instruction...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec4 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- ISA
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec7 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Dynamic Sch...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec20 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Da...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec6 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Can...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec14 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Se...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec1 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Intro
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec2 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Num...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec3 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Performance
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec10 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Mu...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec18 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- In...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec19 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Pr...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec8 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Qui...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec3 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- CMO...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec9 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Com...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec17 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Me...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec13 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Sh...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec16 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Fi...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec12 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Ad...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec1 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Pipelining
Hsien-Hsin Sean Lee, Ph.D.
 
Lec6 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Instruction...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec4 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- ISA
Hsien-Hsin Sean Lee, Ph.D.
 
Lec7 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Dynamic Sch...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec20 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Da...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec6 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Can...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec14 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Se...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec1 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Intro
Hsien-Hsin Sean Lee, Ph.D.
 
Lec2 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Num...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec3 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Performance
Hsien-Hsin Sean Lee, Ph.D.
 
Lec10 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Mu...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec18 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- In...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec19 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Pr...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec8 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Qui...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec3 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- CMO...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec9 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Com...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec17 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Me...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec13 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Sh...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec16 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Fi...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec12 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Ad...
Hsien-Hsin Sean Lee, Ph.D.
 
Ad

Similar to Lec5 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Branch Predictor (20)

PPT
ENG241-Week1-NumberSystemsaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
bao201875
 
PPT
Computer organiztion2
Umang Gupta
 
PDF
Datarepresentation2
gollasaidulu1
 
PPT
Data representation moris mano ch 03
thearticlenow
 
PPT
Lec8 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Dynamic Sch...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
LCDF3_Chap_10_P2.pptytttttyyyyyyyyyyyyyy
muradportal
 
PPT
Ch3
Shashank Mehrotra
 
PPT
DATA REPRESENTATION
Dr. Ajay Kumar Singh
 
PPT
Register transfer and microoperations
mahesh kumar prajapat
 
PPT
dataformats and data conversionscpu1.ppt
anilvarsha1
 
PPT
An introduction to data_representation.ppt
olisahchristopher
 
PPT
DATA REPRESENTATIONS and Data codes and formats.ppt
gnvivekananda4u
 
PPTX
PROCESSOR AND CONTROL UNIT
Amirthavalli Senthil
 
PPTX
Counter Register power point to learn good
6AEntertainment
 
PPT
microprocessors
Hossam Zein
 
PPTX
Chapter 1 digital design.pptx
AliaaTarek5
 
PPT
Micro operations
Ramakrishna Reddy Bijjam
 
PPT
Digital Logic & Design
Rokonuzzaman Rony
 
PPTX
PROCESSOR AND CONTROL UNIT - unit 3 Architecture
Gunasundari Selvaraj
 
PPT
Kaizen cso002 l1
asslang
 
ENG241-Week1-NumberSystemsaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
bao201875
 
Computer organiztion2
Umang Gupta
 
Datarepresentation2
gollasaidulu1
 
Data representation moris mano ch 03
thearticlenow
 
Lec8 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Dynamic Sch...
Hsien-Hsin Sean Lee, Ph.D.
 
LCDF3_Chap_10_P2.pptytttttyyyyyyyyyyyyyy
muradportal
 
Ch3
Shashank Mehrotra
 
DATA REPRESENTATION
Dr. Ajay Kumar Singh
 
Register transfer and microoperations
mahesh kumar prajapat
 
dataformats and data conversionscpu1.ppt
anilvarsha1
 
An introduction to data_representation.ppt
olisahchristopher
 
DATA REPRESENTATIONS and Data codes and formats.ppt
gnvivekananda4u
 
PROCESSOR AND CONTROL UNIT
Amirthavalli Senthil
 
Counter Register power point to learn good
6AEntertainment
 
microprocessors
Hossam Zein
 
Chapter 1 digital design.pptx
AliaaTarek5
 
Micro operations
Ramakrishna Reddy Bijjam
 
Digital Logic & Design
Rokonuzzaman Rony
 
PROCESSOR AND CONTROL UNIT - unit 3 Architecture
Gunasundari Selvaraj
 
Kaizen cso002 l1
asslang
 

More from Hsien-Hsin Sean Lee, Ph.D. (13)

PPT
Lec15 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Re...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec11 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- De...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec7 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Kar...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec5 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Boo...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec4 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- CMOS
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec15 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- EPIC VLIW
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec14 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech --- Coherence
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec13 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- SMP
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec13 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Multicore
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec12 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- P6, Netbur...
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec11 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part3
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec10 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part2
Hsien-Hsin Sean Lee, Ph.D.
 
PPT
Lec9 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part 1
Hsien-Hsin Sean Lee, Ph.D.
 
Lec15 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Re...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec11 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- De...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec7 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Kar...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec5 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- Boo...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec4 Intro to Computer Engineering by Hsien-Hsin Sean Lee Georgia Tech -- CMOS
Hsien-Hsin Sean Lee, Ph.D.
 
Lec15 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- EPIC VLIW
Hsien-Hsin Sean Lee, Ph.D.
 
Lec14 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech --- Coherence
Hsien-Hsin Sean Lee, Ph.D.
 
Lec13 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- SMP
Hsien-Hsin Sean Lee, Ph.D.
 
Lec13 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Multicore
Hsien-Hsin Sean Lee, Ph.D.
 
Lec12 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- P6, Netbur...
Hsien-Hsin Sean Lee, Ph.D.
 
Lec11 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part3
Hsien-Hsin Sean Lee, Ph.D.
 
Lec10 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part2
Hsien-Hsin Sean Lee, Ph.D.
 
Lec9 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Memory part 1
Hsien-Hsin Sean Lee, Ph.D.
 

Recently uploaded (20)

PDF
How NGOs Save Costs with Affordable IT Rentals
C Prompt Solutions Pvt Ltd
 
PPTX
KVL KCL ppt electrical electronics eee tiet
Gauri630804
 
PPTX
code of ethics.pptxdvhwbssssSAssscasascc
denielnaceno76
 
PPTX
STEEL- intro-1.pptxhejwjenwnwnenemwmwmwm
RahulGhanate1
 
PDF
Layer23-Switch.com The Cisco Catalyst 9300 Series is Cisco’s flagship stackab...
Layer23-switch.com
 
PDF
Smarter Security: How Door Access Control Works with Alarms & CCTV
Ighty Support
 
PPTX
Lecture-3-Computer-programming for BS InfoTech
UzumakiNamikaze
 
PDF
YKS Chrome Plated Brass Safety Valve Product Catalogue
SudeYksel
 
PPTX
Fundamentals of Computer.pptx Computer BSC
muhammadawais03085
 
PDF
Colorful Illustrative Digital Education For Children Presentation.pdf
mydidiisawitch6789
 
PPTX
Embeded System for Artificial intelligence 2.pptx
ifnadeksisa
 
PPTX
ATL_Arduino_Complete_Presentation_AI_Visuals.pptx
dushyantsharma1221
 
PPTX
executive branch_no record.pptxsvvsgsggs
alvendiajhorace
 
PPTX
了解新西兰毕业证（Wintec毕业证书）怀卡托理工学院毕业证存档可查的
unweq
 
PDF
Cableado de Controladores Logicos Programables
ssuserb7e1a9
 
PPT
FABRICATION OF MOS FET BJT DEVICES IN NANOMETER
Nikhil Raj
 
PPTX
PROGRAMMING-QUARTER-2-PYTHON.pptxnsnsndn
alexaqt551
 
PPTX
Nanokeyer nano keyekr kano ketkker nano keyer
Oscar Tebar Serrano
 
PPTX
quadraticequations-111211090004-phpapp02.pptx
marcusaviso1101
 
PPTX
1.pptxsadafqefeqfeqfeffeqfqeqfeqefqfeqfqeffqe
EarlVonneRoque
 
How NGOs Save Costs with Affordable IT Rentals
C Prompt Solutions Pvt Ltd
 
KVL KCL ppt electrical electronics eee tiet
Gauri630804
 
code of ethics.pptxdvhwbssssSAssscasascc
denielnaceno76
 
STEEL- intro-1.pptxhejwjenwnwnenemwmwmwm
RahulGhanate1
 
Layer23-Switch.com The Cisco Catalyst 9300 Series is Cisco’s flagship stackab...
Layer23-switch.com
 
Smarter Security: How Door Access Control Works with Alarms & CCTV
Ighty Support
 
Lecture-3-Computer-programming for BS InfoTech
UzumakiNamikaze
 
YKS Chrome Plated Brass Safety Valve Product Catalogue
SudeYksel
 
Fundamentals of Computer.pptx Computer BSC
muhammadawais03085
 
Colorful Illustrative Digital Education For Children Presentation.pdf
mydidiisawitch6789
 
Embeded System for Artificial intelligence 2.pptx
ifnadeksisa
 
ATL_Arduino_Complete_Presentation_AI_Visuals.pptx
dushyantsharma1221
 
executive branch_no record.pptxsvvsgsggs
alvendiajhorace
 
了解新西兰毕业证（Wintec毕业证书）怀卡托理工学院毕业证存档可查的
unweq
 
Cableado de Controladores Logicos Programables
ssuserb7e1a9
 
FABRICATION OF MOS FET BJT DEVICES IN NANOMETER
Nikhil Raj
 
PROGRAMMING-QUARTER-2-PYTHON.pptxnsnsndn
alexaqt551
 
Nanokeyer nano keyekr kano ketkker nano keyer
Oscar Tebar Serrano
 
quadraticequations-111211090004-phpapp02.pptx
marcusaviso1101
 
1.pptxsadafqefeqfeqfeffeqfqeqfeqefqfeqfqeffqe
EarlVonneRoque
 

Lec5 Computer Architecture by Hsien-Hsin Sean Lee Georgia Tech -- Branch Predictor

  • 1. ECE 4100/6100 Advanced Computer Architecture Lecture 5 Branch Prediction Prof. Hsien-Hsin Sean Lee School of Electrical and Computer Engineering Georgia Institute of Technology
  • 2. 2 Predict What? • Direction (1-bit) – Single direction for unconditional jumps and calls/returns – Binary for conditional branches • Target (32-bit or 64-bit addresses) – Some are easy • One: Uni-directional jumps • Two: Fall through (Not Taken) vs. Taken – Many: Function Pointer or Indirect Jump (e.g. jr r31)
  • 3. 3 Categorizing Branches 8% 10% 82% 19% 6% 75% 0% 20% 40% 60% 80% 100% Call/Return Jump Conditional Branch Frequency of branch instructions SPEC2000INT SPEC2000FP Source: H&P using Alpha
  • 4. 4 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue
  • 5. 5 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Mispredict
  • 6. 6 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue (flush entailed instructions and refetch) Mispredict
  • 7. 7 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Fetch the correct path
  • 8. 8 Branch Misprediction PC Next PC Fetch DriveAlloc Rename Queue Schedule Dispatch Reg File ExecFlags Br Resolve 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Single Issue Mispredict 8-issue Superscalar Processor (Worst case)
  • 9. 9 Why Branch is Predictable? for (i=0; i<100; i++) { …. } addi r10, r0, 100 add r1, r0, r0 L1: … … … … addi r1, r1, 1 bne r1, r10, L1 … … if (aa==2) aa = 0; if (bb==2) bb = 0; if (aa!=bb) …. addi r2, r0, 2 bne r10, r2, L_bb xor r10, r10, r10 L_bb: bne r11, r2, L_xx xor r11, r11, r11 L_xx: beq r10, r11, L_exit … Lexit:
  • 10. 10 Control Speculation • Execute instruction beyond a branch before the branch is resolved  Performance • Speculative execution • What if mis-speculated? need – Recovery mechanism – Squash instructions on the incorrect path • Branch prediction: Dynamic vs. Static • What to predict?
  • 11. 11 Static Branch Prediction • Uni-directional, always predict taken (or not taken) • Backward taken, Forward not taken – Need offset information (when?) • Compiler hints with branch annotation – When the info will be available? Post-decode?
  • 12. 12 Simplest Dynamic Branch Predictor • Prediction based on latest outcome • Index by some bits in the branch PC – Aliasing T NT T T NT NT . . . for (i=0; i<100; i++) { …. } addi r10, r0, 100 addi r1, r1, r0 L1: … … … … addi r1, r1, 1 bne r1, r10, L1 … … 0x40010100 0x40010104 0x40010108 … 0x40010A04 0x40010A08 How accurate? NT T 1-bit Branch History Table
  • 13. 13 Typical Table Organization Hash PC (32 bits) . . . . . 2N entries Prediction N bits FSM Update Logic table update Actual outcome
  • 14. 14 Simplest Dynamic Branch Predictor T NT T T NT NT . . . addi r10, r0, 100 addi r1, r1, r0 L1: add r21, r20, r1 lw r2, (r21) beq r2, r0, L2 … … j L3 L2: … … … L3: addi r1, r1, 1 bne r1, r10, L1 0x40010100 0x40010104 0x40010108 0x4001010c 0x40010110 0x40010210 0x40010B0c 0x40010B10 for (i=0; i<100; i++) { if (a[i] == 0) { … } … } NT T 1-bit Branch History Table
  • 15. 15 FSM of the Simplest Predictor • A 2-state machine • Change mind fast 00 11 If branch not taken If branch taken 00 11 Predict not taken Predict taken
  • 16. 16 Example using 1-bit branch history table for (i=0; i<44; i++) { …. } 00Pred Actual T T √ 11 11 √ T T √ 11 11 addi r10, r0, 4 addi r1, r1, r0 L1: … … addi r1, r1, 1 bne r1, r10, L1 NT 00   T 11  T √ 11 √ T T √ 11 11 NT 00  T 11  60% accuracy
  • 17. 17 2-bit Saturating Up/Down Counter Predictor Not Taken Taken Predict Not taken Predict taken ST: Strongly Taken WT: Weakly Taken WN: Weakly Not Taken SN: Strongly Not Taken 01/ WN 01/ WN 00/ SN 00/ SN 10/ WT 10/ WT 11/ ST 11/ ST MSB: Direction bit LSB: Hysteresis bit
  • 18. 18 2-bit Counter Predictor (Another Scheme) Not Taken Taken Predict Not taken Predict taken ST: Strongly Taken WT: Weakly Taken WN: Weakly Not Taken SN: Strongly Not Taken 01/ WN 01/ WN 00/ SN 00/ SN 11/ ST 11/ ST 10/ WT 10/ WT
  • 19. 19 Example using 2-bit up/down counter for (i=0; i<44; i++) { …. } 0101Pred Actual T T √ 1010 1111 √ T T √ 1111 1111 addi r10, r0, 4 addi r1, r1, r0 L1: … … addi r1, r1, 1 bne r1, r10, L1 NT 1010   T 1111 √ T √ 1111 √ T T √ 1111 1111 NT 1010  T 1111 √ 80% accuracy
  • 20. 20 Branch Correlation • Branch direction – Not independent – Correlated to the path taken • Example: Path 1-1 of b3 can be surely known beforehand • Track path using a 2-bit register if (aa==2) // b1 aa = 0; if (bb==2) // b2 bb = 0; if (aa!=bb) { // b3 ……. } b1 b2 b2 b3 b3 b3 1 (T) 1 1 0 (NT) 0 b3 0 Path: A:1-1 B:1-0 C:0-1 D:0-0 aa=0 bb=0 aa=0 bb≠2 aa≠2 bb=0 aa≠2 bb≠2 Code Snippet
  • 21. 21 Correlated Branch Predictor [PanSoRahmeh’92] • (M,N) correlation scheme – M: shift register size (# bits) – N: N-bit counter 2-bit counter hash . . . . X X Branch PC hash 2-bit counter . . . . 2-bit counter . . . . X X 2-bit counter . . . . 2-bit counter . . . . Prediction Prediction 2-bit shift register (global branch history) select Subsequent branch direction (2,2) Correlation Scheme 2-bit Sat. Counter Scheme 2w w Branch PC
  • 22. 22 Two-Level Branch Predictor [YehPatt91,92,93] • Generalized correlated branch predictor • 1st level keeps branch history in Branch History Register (BHR) • 2nd level segregates pattern history in Pattern History Table (PHT) 1 1 . . . . . 1 0 00…..00 00…..01 00…..10 11…..11 11…..10 Branch History Pattern Pattern History Table (PHT) Prediction Rc-k Rc-1 Rc: Actual Branch Outcome FSM Update Logic Branch History Register (BHR) (Shift left when update) N 2N entries Current StatePHT update
  • 23. 23 Branch History Register • An N-bit Shift Register = 2N patterns in PHT • Shift-in branch outcomes – 1 ⇒ taken – 0 ⇒ not taken • First-in First-Out • BHR can be – Global – Per-set – Local (Per-address)
  • 24. 24 Pattern History Table • 2N entries addressed by N-bit BHR • Each entry keeps a countercounter (2-bit or more) for prediction – Counter update: the same as 2-bit counter – Can be initialized in alternate patterns (01, 10, 01, 10, ..) • Alias (or interference) problem
  • 25. 25 Global History Schemes Global BHR Global PHT GAg Global BHR .. SetP(B) Per-set PHTs (SPHTs) GAs Global BHR .. Addr(B) Per-addr PHTs (PPHTs) GAp ** [PanSoRahmeh’92] similar to GAp . . . . . . . . . . . . . . . . . . . . . Set can be determined by branch opcode, compiler classification, or branch PC address.
  • 26. 26 GAs Two-Level Branch Prediction 01100110 BHR PC = 0x4001000C . . . PHT 00110110 . . 00110110 00110111 11111101 11111110 00000000 00000001 00000010 11111111 10 MSB = 1 Predict Taken The 2 LSBs are insignificant for 32-bit instruction
  • 27. 27 Predictor Update (Actually, Not Taken) 01100110 BHR PC = 0x4001000C . . . PHT 00110110 . . 00110110 00110111 11111101 11111110 00000000 00000001 00000010 11111111 1001 decremented 11001100 00111100 00111100 Wrong Predictio n • Update Predictor after branch is resolved
  • 28. 28 Per-Address History Schemes Global PHT PAg SetP(B) Per-set PHTs (SPHTs) PAs Addr(B) Per-addr PHTs (PPHTs) PAp . . . Addr(B) Per-addr BHT (PBHT) . . . Addr(B) Per-addr BHT (PBHT) . . . Addr(B) Per-set BHT (PBHT) •Ex: P6, Itanium •Ex: Alpha 21264’s local predictor . . . .. . . . . . . . . . .. . . . . . . . . .
  • 29. 29 PAs Two-Level Branch Predictor PC = 1110 0000 1001 1001 0010 1100 1110 1000 000 001 010 011 100 101 110 111 BHT 11010110 . . . PHT . . 11010101 11010110 11111101 11111110 00000000 00000001 00000010 11111111 MSB = 1 Predict Taken 11 110
  • 30. 30 Per-Set History Schemes Global PHT SAg SetP(B) Per-set PHTs (SPHTs) SAs Addr(B) Per-addr PHTs (PPHTs) SAp . . . Per-set BHT (SBHT) . . . SetH(B) Per-set BHT (SBHT) . . . SetH(B) Per-set BHT (SBHT) .. . . . . . . . . . . . . .. . . . . . . . . . SetH(B)
  • 31. 31 PHT Indexing Branch addr Global history Gselect 4/4 00000000 00000001 00000001 00000000 00000000 00000000 11111111 00000000 11110000 11111111 10000000 11110000 Insufficient History • Tradeoff between more history bits and address bits • Too many bits needed in Gselect ⇒ sparse table entries
  • 32. 32 Gshare Branch Predictor [McFarling93] • Tradeoff between more history bits and address bits • Too many bits needed in Gselect ⇒ sparse table entries • Gshare ⇒ Not to lose global history bits • Ex: AMD Athlon, MIPS R12000, Sun MAJC, Broadcom SiByte’s SB-1 Branch addr Global history Gselect 4/4 Gshare 8/8 00000000 00000001 00000001 00000001 00000000 00000000 00000000 00000000 11111111 00000000 11110000 11111111 11111111 10000000 11110000 01111111 GselectGselect 4/4: Index PHT by concatenateconcatenate low order 4 bits GshareGshare 8/8: Index PHT by {Branch address ⊕ Global history}
  • 33. 33 Gshare Branch Predictor . . . PHT . . 00 MSB = 0 Predict Not Taken 1 1 . . . . . 1 0 0 1 . . . . . 0 1 0 01. . . . .1 1 ⊕ PC Address Global BHR
  • 34. 34 Aliasing Example PHT BHR 1101 PC 0110 ---- XOR 1011 BHR 1001 PC 1010 ---- XOR 0011 1111 1110 1101 1100 1011 1010 1001 1000 0111 0110 0101 0100 0011 0010 0001 0000 PHT (indexed by 10) BHR 1101 PC 0110 ---- || 1001 BHR 1001 PC 1010 ---- || 1001 1111 1110 1101 1100 1011 1010 1001 1000 0111 0110 0101 0100 0011 0010 0001 0000 GApGAp GshareGshare
  • 35. 35 Hybrid Branch Predictor [McFarling93] • Some branches correlated to global history, some correlated to local history • Only update the meta-predictor when 2 predictors disagree P0P0 P1P1 . . . Choice (or Meta) Predictor Branch PC Final Prediction
  • 36. 36 Alpha 21264 (EV6) Hybrid Predictor Local History Table 1024 x 10 bits Single Local Predictor 1024 x 3 bits Global Predictor 4096 x 2 bits Choice Predictor 4096 x 2 bits Global history 12 Local prediction Global prediction Meta prediction Next Line/set Prediction L1 I-cache (64KB 2w) & TLB 4 instr./cycle4 instr./cycle Virtual address Final Branch Prediction PCPC 10 • A “tournament branch predictor” • Multi-predictor scheme w/ – Local predictorLocal predictor (~PAg) • Self-correlation – Global predictorGlobal predictor • Inter-correlation – Choice predictorChoice predictor as the decision maker: a 2-bit sat. counter to credit either local or global predictors. • Die size impact – History info tables ~2% – BTB ~ 2.7% (associated with I-$ on a per-line basis) • 2 cycle latency, we will discuss more later For Single-cycle Prediction
  • 37. 37 Alpha EV8 Branch Predictor Branch PC Global history F1 F2 F3 majority vote prediction G0 G1 Meta F4 Bimodal e-gskew predictor • Real silicon never sees the daylight • Use a 2Bc-gskew predictor (one form of enhanced gskew) – Bimodal predictor used as (1) static biased predictor and (2) part of e-gskew predictor – Global predictors G0 and G1 are part of e-gskew predictor – Table sizes: 352Kbits in total (208Kbits for prediction table; 144Kbits for hysteresis table.)
  • 38. 38 Branch Target Prediction • Try the easy ones first – Direct jumps – Call/Return – Conditional branch (bi-directional) • Branch Target Buffer (BTB) • Return Address Stack (RAS)
  • 39. 39 Branch Target Buffer (BTB) TargetTag TargetTag TargetTag… BTBBranch PC == == ==… ++ 4 Branch Target Predicted Branch Direction 0 1
  • 40. 40 Return Address Stack (RAS) • Different call sites make return address hard to predict – Printf() being called by many callers – The target of “return” instruction in printf() is a moving target • A hardware stack (LIFO) – Call will push return address on the stack – Return uses the prediction off of TOS
  • 41. 41 Return Address Stack • Does it always work? – Call depth – Setjmp/Longjmp – Speculative call? ++ 4 Call PC Push Return Address BTBBTB Return PC BTBBTB Return? • May not know it is a return instruction prior to decoding – Rely on BTB for speculation – Fix once recognize Return
  • 42. 42 Indirect Jump • Need Target Prediction – Many (potentially 230 for 32-bit machine) – In reality, not so many – Similar to predicting values • Tagless Target Prediction • Tagged Target Prediction
  • 43. 43 Tagless Target Prediction [ChangHaoPatt’97] 1 1 . . . . . 1 0 Branch History RegisterBranch History Register (BHR)(BHR) 00…..00 00…..01 00…..10 11…..11 11…..10 PC ⊗ BHR Pattern Target Cache (2N entries) Predicted Target Address Branch PCBranch PC HashHash • Modify the PHT to be a “Target Cache” – (indirect jump) ? (from target cache) : (from BTB) • Alias?
  • 44. 44 Tagged Target Prediction [ChangHaoPatt’97] • To reduce aliasing with set-associative target cache • Use branch PC and/or history for tags 1 1 . . . . . 1 0 BHR 00…..00 00…..01 00…..10 11…..11 11…..10 Target Cache (2n entries per way) Predicted Target Address Branch PC Hash n =? Tag Array
  • 45. 45 Multiple Branch Prediction • For a really wide machine – Across several basic blocks – Need to predict multiple branches per cycle • How to fetch non-contiguous instructions in one cycle? • Prediction accuracy extremely critical (will be reduced geometrically)