floating point function approximators
play

floating-point function approximators in FPGAs David B. Thomas - PowerPoint PPT Presentation

Oct 08, 2022 •197 likes •726 views

FloatApprox : faithfully rounded floating-point function approximators in FPGAs David B. Thomas Imperial College London 1 David Thomas, Imperial College, dt10@ic.ac.uk FloPoCo : Parameterised primitives 2 David Thomas, Imperial College,

  1. FloatApprox : faithfully rounded floating-point function approximators in FPGAs David B. Thomas Imperial College London 1 David Thomas, Imperial College, dt10@ic.ac.uk

  2. FloPoCo : Parameterised primitives 2 David Thomas, Imperial College, dt10@ic.ac.uk

  3. FloPoCo : Parameterised primitives 3 David Thomas, Imperial College, dt10@ic.ac.uk

  4. FloPoCo : Parameterised primitives 4 David Thomas, Imperial College, dt10@ic.ac.uk

  5. FloatApprox : Parameterised anything Approximation Input format interval Output format 5 David Thomas, Imperial College, dt10@ic.ac.uk

  6. FloatApprox : Parameterised anything 6 David Thomas, Imperial College, dt10@ic.ac.uk

  7. FloatApprox • Architecture for FPGA function approximation – Deeply pipelined – Floating-point in and out – Faithfully rounded • Method and tool for approximating functions – Handles any most twice-differentiable functions – Completely automated: expression -> VHDL – Designed for reliability rather than optimality 7 David Thomas, Imperial College, dt10@ic.ac.uk

  8. 1. Motivation 2. The FloatApprox approach 1. Range reduction and approximation method 2. Evaluation architecture 3. Evaluation in hardware 8 David Thomas, Imperial College, dt10@ic.ac.uk

  9. Floating-point IP: Requirements • Faithfully rounded – Make every bit count – Tractable error analysis • Pipelined for 250MHz+ clock rate – Must be pipelined: RAM and DSPs are multi-cycle – HLS tools have retiming built-in • Working RTL (circuit) implementation – A paper can’t be synthesised 10 David Thomas, Imperial College, dt10@ic.ac.uk

  10. What floating-point IP is available? Source Pipelined Faithful RTL add, mul, div FloPoCo Yes Yes Yes log, exp FloPoCo Yes Yes Yes sin, cos FPLibrary No Yes Yes Altera Yes Yes Altera flow only Xilinx Yes ? Vivado HLS only log1p Altera Yes Yes Altera flow only expm1 Altera Yes No OpenCL only erf Altera Yes No OpenCL only 11 David Thomas, Imperial College, dt10@ic.ac.uk

  11. What floating-point IP is available? Source Pipelined Faithful RTL add, mul, div FloPoCo Yes Yes Yes log, exp FloPoCo Yes Yes Yes sin, cos FPLibrary No Yes Yes Altera Yes Yes Altera flow only Xilinx Yes ? Vivado HLS only log1p Altera Yes Yes Altera flow only expm1 Altera Yes No OpenCL only erf Altera Yes No OpenCL only 12 David Thomas, Imperial College, dt10@ic.ac.uk

  12. What floating-point IP is available? Source Pipelined Faithful RTL add, mul, div FloPoCo Yes Yes Yes log, exp FloPoCo Yes Yes Yes sin, cos FPLibrary No Yes Yes Altera Yes Yes Altera flow only Xilinx Yes ? Vivado HLS only log1p Altera Yes Yes Altera flow only expm1 Altera Yes No OpenCL only erf Altera Yes No OpenCL only 13 David Thomas, Imperial College, dt10@ic.ac.uk

  13. Motivation for FloatApprox • We currently have : + , - , * , / , log , exp – Use existing IP: FloPoCo, Xilinx, Altera, ... • We should have : log1p, expm1, erf, sin, acos, ... – What FloatApprox does badly... ... but better than anything else available • What I want : sqrt(-2 log(x)), 1/(1+exp(-x)) – What FloatApprox does well 14 David Thomas, Imperial College, dt10@ic.ac.uk

  14. Goals of FloatApprox • Approximation : FloatApprox as a tool – Convert any function f(x) to RTL – Able to handle most smooth functions – Suitable for automated use • Input : data-types, range, function • Output : faithfully rounded circuit • Architecture : FloatApprox as generated IP – Pipelined – Faithfully rounded – Working RTL 15 David Thomas, Imperial College, dt10@ic.ac.uk

  15. FloatApprox : Approximation • Given a function f t how do we create f a ? • Segment the function so that each segment is: 1. Contained in one input binade 2. Monotonically increasing or decreasing in range 3. Contained in one output binade 4. FaithfulReal: can approx. with real degree d poly 5. FaithfulFixed: can faithfully approximate with fixed-point polynomial of degree d 16 David Thomas, Imperial College, dt10@ic.ac.uk

  16. FloatApprox : Approximation • Given a function f t how do we create f a ? • Segment the function so that each segment is: 1. Contained in one input binade 2. Monotonically increasing or decreasing in range 3. Contained in one output binade 4. FaithfulReal: can approx. with real degree d poly 5. FaithfulFixed: can faithfully approximate with fixed-point polynomial of degree d 17 David Thomas, Imperial College, dt10@ic.ac.uk

  17. Example: Input function over reals 4.0 3.5  0 . 95 sin( 0 . 1 ) x   0 . 06 y x 3.0 x   0 16 x 2.5 y 1.5 0.5 0 2 4 6 8 10 12 14 16 x 18 David Thomas, Imperial College, dt10@ic.ac.uk

  18. Move to float representation 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 19 David Thomas, Imperial College, dt10@ic.ac.uk

  19. 1 : Segment using input binades 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 20 David Thomas, Imperial College, dt10@ic.ac.uk

  20. 2 : Make segments monotonic 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 21 David Thomas, Imperial College, dt10@ic.ac.uk

  21. 3 : Segment using output binades 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 22 David Thomas, Imperial College, dt10@ic.ac.uk

  22. 3 : Segment using output binades 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 23 David Thomas, Imperial College, dt10@ic.ac.uk

  23. 3 : Segment using output binades 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 24 David Thomas, Imperial College, dt10@ic.ac.uk

  24. 3 : Segment using output binades 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 25 David Thomas, Imperial College, dt10@ic.ac.uk

  25. 4 : Split to degree d polynomials 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 26 David Thomas, Imperial College, dt10@ic.ac.uk

  26. 4 : Split to degree d polynomials 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 27 David Thomas, Imperial College, dt10@ic.ac.uk

  27. 4 : Split to degree d polynomials 2 4 2 3 2 2 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4 28 David Thomas, Imperial College, dt10@ic.ac.uk

  28. • Segments form a partition on input domain • Segment domains and ranges cover one binade • All segments can be faithfully calculated as degree d fixed-point polynomial 2 1 2 0 2 -1 2 -2 2 -3 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 2 4

  29. Real-world issues • Lots of corner cases to worry about – Crossing from negative to positive to NaN is fun – Method should be faithful by construction • Calculations performed using mpfr and sollya – Mostly interval arithmetic via sollya – Occasionally bisection search in mpfr • Speed of approximation is an issue – Single precision usually takes minutes – Double precision often takes hours

  30. FloatApprox : Architecture flags s expnt significand x Segmentation ë S i û ≤ i ≤ é S i ù Table-Lookup flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 31 David Thomas, Imperial College, dt10@ic.ac.uk

  31. Compile-time configuration flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 32 David Thomas, Imperial College, dt10@ic.ac.uk

  32. Evaluation: Input flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 33 David Thomas, Imperial College, dt10@ic.ac.uk

  33. Evaluation: Segmentation flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 34 David Thomas, Imperial College, dt10@ic.ac.uk

  34. Evaluation: Segmentation flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 35 David Thomas, Imperial College, dt10@ic.ac.uk

  35. Evaluation: Segmentation flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 36 David Thomas, Imperial College, dt10@ic.ac.uk

  36. Evaluation: Segmentation flags s expnt significand 2 1 2 0 x 2 -1 Segmentation ë S i û ≤ i ≤ é S i ù 2 -2 2 -3 Table-Lookup 2 -4 2 -4 2 -3 2 -2 2 -1 2 0 2 1 2 2 2 3 flags s expnt c 0 c 1 ... c d c 0 c 1 c d x Fixed-Point Polynomial y = ∑ c i x i flags s expnt significand 37 David Thomas, Imperial College, dt10@ic.ac.uk

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Lecture 3 Floating Point Representations 1 Floating-point arithmetic We often incur

Lecture 3 Floating Point Representations 1 Floating-point arithmetic We often incur

ECE 0142 Computer Organization Lecture 3 Floating Point Representations 1 Floating-point arithmetic We often incur floating-point programming. Floating point greatly simplifies working with large (e.g., 2 70 ) and small (e.g., 2 -17 )

638 views • 30 slides
Floating-point numbers Fractional binary numbers IEEE floating-point standard Floating-point

Floating-point numbers Fractional binary numbers IEEE floating-point standard Floating-point

Floating-point numbers Fractional binary numbers IEEE floating-point standard Floating-point operations and rounding Lessons for programmers Many more details we will skip (its a 58-page standard) See CSAPP 2.4 for more detail. 1

381 views • 15 slides
15-213 The course that gives CMU its Zip! Floating Point Sept 6, 2006 Topics Topics

15-213 The course that gives CMU its Zip! Floating Point Sept 6, 2006 Topics Topics

15-213 The course that gives CMU its Zip! Floating Point Sept 6, 2006 Topics Topics IEEE Floating Point Standard Rounding Floating Point Operations Mathematical properties class03.ppt 15-213, F06 Floating Point

1.03k views • 34 slides
Floating point Today ! IEEE Floating Point Standard ! Rounding ! Floating Point Operations !

Floating point Today ! IEEE Floating Point Standard ! Rounding ! Floating Point Operations !

Floating point Today ! IEEE Floating Point Standard ! Rounding ! Floating Point Operations ! Mathematical properties Next time ! The machine model Chris Riesbeck, Fall 2011 Monday, October 3, 2011 Checkpoint Monday, October 3, 2011 IEEE

1.07k views • 28 slides
2/10/2020 Today: Floating Point Background: Fractional binary numbers IEEE floating point

2/10/2020 Today: Floating Point Background: Fractional binary numbers IEEE floating point

2/10/2020 Today: Floating Point Background: Fractional binary numbers IEEE floating point standard: Definition Floating Point Example and properties Rounding, addition, multiplication CSci 2021: Machine Architecture and

411 views • 7 slides
9/20/2018 Today: Floating Point Background: Fractional binary numbers IEEE floating point

9/20/2018 Today: Floating Point Background: Fractional binary numbers IEEE floating point

9/20/2018 Today: Floating Point Background: Fractional binary numbers IEEE floating point standard: Definition Floating Point Example and properties Rounding, addition, multiplication CSci 2021: Machine Architecture and

339 views • 7 slides
CS 356 Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent

CS 356 Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent

3.1 CS 356 Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent very small numbers (fractions) and very large numbers Avogadros Number: +6.0247 * 10 23 Plancks Constant: +6.6254 * 10 -27

789 views • 33 slides
Machine numbers: how floating point numbers are stored? Floating-point number representation

Machine numbers: how floating point numbers are stored? Floating-point number representation

Machine numbers: how floating point numbers are stored? Floating-point number representation What do we need to store when representing floating point numbers in a computer? ! = 1. & 2 ! Initially, different floating-point

379 views • 18 slides
Floating point How arithmetic operations mathematics involving floating point numbers

Floating point How arithmetic operations mathematics involving floating point numbers

Numerical and Scientific Computing with Applications David F . Gleich CS 314, Purdue September 7, 2016 In this class: Floating point How arithmetic operations mathematics involving floating point numbers work. Next class IEEE

812 views • 11 slides
Floating Point Data CSCI 125 & 161 / ENGR 144 Floating point numbers are not exact

Floating Point Data CSCI 125 & 161 / ENGR 144 Floating point numbers are not exact

Floating Point Data CSCI 125 & 161 / ENGR 144 Floating point numbers are not exact Value 0.1 is very close to 1/10, but not Lecture 13 precisely equal to it 0.1 10 = 0.000110011001100110011... 2 Cannot rely on accuracy of

313 views • 3 slides
Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent very

Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent very

3.1 Unit 3 IEEE 754 Floating Point Representation 3.2 Floating Point Used to represent very small numbers (fractions) and very large numbers Avogadros Number: +6.022 10 23 Boltzmanns Constant: +1.38 10 -23 32 or

1.13k views • 38 slides
Formal verification of floating-point algorithms John Harrison Intel Corporation Floating

Formal verification of floating-point algorithms John Harrison Intel Corporation Floating

Formal verification of floating-point algorithms 1 Formal verification of floating-point algorithms John Harrison Intel Corporation Floating point algorithm verification HOL Light Floating point numbers and formats HOL floating

636 views • 28 slides
ACCURATE FLOATING-POINT SUMMATION IN CUB URI VERNER Summer intern OUTLINE Who needs accurate

ACCURATE FLOATING-POINT SUMMATION IN CUB URI VERNER Summer intern OUTLINE Who needs accurate

ACCURATE FLOATING-POINT SUMMATION IN CUB URI VERNER Summer intern OUTLINE Who needs accurate floating-point summation? ! Round-off error: source and recovery A new method for accurate FP summation on a GPU Added as a function to the open-source

619 views • 28 slides
Debugging Floating-Point Debugging Floating-Point Debugging Floating-Point Math in Racket Math

Debugging Floating-Point Debugging Floating-Point Debugging Floating-Point Math in Racket Math

Debugging Floating-Point Debugging Floating-Point Debugging Floating-Point Math in Racket Math in Racket Math in Racket Neil Toronto RacketCon 2013 Racket Floating-Point Support Racket Floating-Point Support Racket Floating-Point Support

914 views • 88 slides
DLX Floating Point Extend MIPS Pipeline to Floating Point Operations Functional units

DLX Floating Point Extend MIPS Pipeline to Floating Point Operations Functional units

DLX Floating Point Extend MIPS Pipeline to Floating Point Operations Functional units more complex than simple integer ALU Require several clock cycles for a FP arithmetic operation Add: 4 cycles, Multiply: 7 cycles, Divide: 25

377 views • 22 slides
Universal Value Function Approximators Tom Schaul, Dan Horgan, Karol Gregor, Dave Silver

Universal Value Function Approximators Tom Schaul, Dan Horgan, Karol Gregor, Dave Silver

Universal Value Function Approximators Tom Schaul, Dan Horgan, Karol Gregor, Dave Silver Motivation Forecasts about the environment = temporally abstract predictions (questions) not necessarily related to reward (unsupervised)

812 views • 23 slides
Detailed Routing Find actual geometric layout of each Global Routing G oba out g net within

Detailed Routing Find actual geometric layout of each Global Routing G oba out g net within

11/2/2018 DET A IL ED RO UT ING PRO F. INDRA NIL SENG UPT A DEPA RT DEPA RT MENT MENT O F C O MPUT O F C O MPUT ER SC IENC E A ND ENG INEERING ER SC IENC E A ND ENG INEERING Detailed Routing Find actual geometric layout of each

759 views • 33 slides
part 4 May, 2012 Dr. Belal Gharaibeh 1 The Back Low back disorders (LBD) have been

part 4 May, 2012 Dr. Belal Gharaibeh 1 The Back Low back disorders (LBD) have been

Biomechanics analysis part 4 May, 2012 Dr. Belal Gharaibeh 1 The Back Low back disorders (LBD) have been identified as one of the most common and significant musculoskeletal problems that results in substantial amounts of morbidity,

510 views • 14 slides
OSCARonDebian: Contributions from the Google Summer of Code 2005 program Ram Kumar DANGETTI

OSCARonDebian: Contributions from the Google Summer of Code 2005 program Ram Kumar DANGETTI

OSCARonDebian: Contributions from the Google Summer of Code 2005 program Ram Kumar DANGETTI Institut National des Telecommunications Geoffroy VALLEE Oak Ridge National Laboratory Introduction SoC 2005 allows us to have a pretty good OoD

705 views • 13 slides
Tropical Volume by Tropical Ehrhart Polynomials Georg Loho London School of Economics Matthias

Tropical Volume by Tropical Ehrhart Polynomials Georg Loho London School of Economics Matthias

Tropical Volume by Tropical Ehrhart Polynomials Georg Loho London School of Economics Matthias Schymura Ecole Polytechnique F ed erale de Lausanne September 24, 2019 Discrete Geometry with a View on Symplectic and Tropical Geometry

823 views • 56 slides
Realtime Search with Lucene Michael Busch @michibusch michael@twitter.com buschmi@apache.org

Realtime Search with Lucene Michael Busch @michibusch michael@twitter.com buschmi@apache.org

Realtime Search with Lucene Michael Busch @michibusch michael@twitter.com buschmi@apache.org Monday, June 7, 2010 1 Realtime Search with Lucene Agenda Introduction - Near-realtime Search (NRT) - Searching DocumentsWriters RAM buffer

1.06k views • 87 slides
Urban Building Usage Labeling by Geometric and Context Analyses of the Footprint Data Hai Huang,

Urban Building Usage Labeling by Geometric and Context Analyses of the Footprint Data Hai Huang,

Institute of Cartography and Geoinformatics | Leibniz Universitt Hannover Urban Building Usage Labeling by Geometric and Context Analyses of the Footprint Data Hai Huang, Birgit Kieler and Monika Sester Introduction The Usage (use and

915 views • 22 slides
Fiscal 2018 Fourth Quarter Earnings September 26, 2018 Safe Harbor Statements in this

Fiscal 2018 Fourth Quarter Earnings September 26, 2018 Safe Harbor Statements in this

Fiscal 2018 Fourth Quarter Earnings September 26, 2018 Safe Harbor Statements in this presentation that are not historical are considered forward-looking statements and are subject to change based on various factors and uncertainties

556 views • 26 slides
Section 6: Kinematics Section 6: Kinematics 6-1 Biomechanics - angular kinematics Same as

Section 6: Kinematics Section 6: Kinematics 6-1 Biomechanics - angular kinematics Same as

Section 6: Kinematics Section 6: Kinematics 6-1 Biomechanics - angular kinematics Same as linear kinematics, but There is one vector along th the moment arm. t There is one vector F RMA perpendicular to the perpendicular

319 views • 21 slides

More recommend