• Name: Elie TOURNIER
• Email: tournier.elie@gmail.com

---

"Soft" double precision floating point support.

by Elie (Hopetech) TOURNIER, for GSoC 2016

---

1. Organization

This project will take place with X.org foundation.

The X.Org project "provides a free and open source implementation of the X Window System".

Mesa3D, a graphics library, is part of the X.org project.

"Mesa is an open-source implementation of the OpenGL specification - a system for rendering interactive 3D graphics - which provide OpenGL support to users of X on Linux, FreeBSD and other operating systems."

During the "Google Summer of Code", I will contribute to Mesa by implementing a double precision floating point library.

2. Aims

GPUs natively support single precision, but only OpenGL 4.0 class GPUs have hardware support for double precision. The goal of this project is to implement a library of double precision operations in pure GLSL 1.30 for this GPU using bit twiddling operations and integer math. There are many library of software double precision floating point for devices that lack floating-point hardware. I should therefore translate one of this library (generally write in C) to pure GLSL 1.30.

3. Constraints

• Main Mesa code is under MIT license. It will be important throughout the project period to only use compatible license files. You can find a list of licenses here.
• The code must comply with the Mesa coding style.
• The code must be commented by respecting the Mesa standard.
• Each double precision value would be stored in a uvec2.

4. Deliverables

• A library with the following functions:

• Add.

• Negate.

• Absolute value.

• Multiply.

• Reciprocal.

• Convert to single precision.

• Convert from single precision.

• Comparison.

• Nice to have:

• Rsqrt.

• Log.

• Exp.

These last functions are not necessary for the implementation of GL_ARB_gpu_shader_fp64 (see What's next section) but may be useful for other uses.

5. Schedule

For each implemented function, I shall realize tests, documentation and then submit my code to the community for reviewing.

• Before coding period:

  • During this period I will remain in constant touch with my mentor and the all the Mesa community.
  • Study current CPU "soft" double precision floating point libraries to familiarize myself with bit twiddling operations.
  • Prepare tests architecture.

• May 23rd - May 29th:

  • Implement absolute value function.

uvec2 abs_fp64(uvec2 a);

• May 30th - June 5th:
  • Implement add function.

uvec2 add_fp64(uvec2 a, uvec2 b);

• June 6th - June 12th:
  • Implement negate function.

uvec2 neg_fp64(uvec2 a);

• June 13th - June 19th:
  • Implement multiply function.

uvec2 mul_fp64(uvec2 a, uvec2 b);

• June 20th - June 26th:

  • Mid-term evaluation.

• June 27th - July 3rd:

  • Implement conversion to single precision.

uint fp64-to-fp32(uvec2 a);

• July 4th - July 10th:
  • Implement conversion from single precision.

uvec2 fp32-to-fp64(uint a);

• July 11th - July 17th:
  • Implement comparison.

/**
 * Returns `true` if the double-precision floating-point value `a' is equal to the
 * corresponding value `b', and `false` otherwise.
 */
bool eq_fp64(uvec2 a, uvec2 b);
/**
 * Returns true if the double-precision floating-point value `a' is less than or
 * equal to the corresponding value `b', and false otherwise.
 */
bool le_fp64(uvec2 a, uvec2 b);
/**
 * Returns true if the double-precision floating-point value `a' is less than
 * the corresponding value `b', and false otherwise.
 */
bool lt_fp64(uvec2 a, uvec2 b);

• July 18th - July 24th:
  • Port reciprocal and rsqrt from NIR to GLSL 1.30.

uvec2 recip_fp64(uvec2 a);
uvec2 rsqrt_fp64(uvec2 a);

• July 25th - July 31st:
  • Port floor, ceil and truncate from NIR to GLSL 1.30.

uvec2 floor_fp64(uvec2 a);
uvec2 ceil_fp64(uvec2 a);
uvec2 trunc_fp64(uvec2 a);

• August 1st - August 7th:

  • Time for unpredictable delay.

• August 8th - August 14th:

  • Time for unpredictable delay.

• August 15th - August 22nd:

  • Final evaluation.

6. What's next?

• Implement log and exponential functions. This can be done in August if I have enough time.
• A stretch goal would be to use this library of functions to implement GL_ARB_gpu_shader_fp64 on all GPUs for which Mesa supports GLSL 1.30.

7. Student Information

7.1 Personal Information

• Name: Elie TOURNIER
• Email: tournier.elie@gmail.com
• IRC: Elie_
• GitHub: GitHub
• Linkedin: Linkedin
• Resume: Resume
• Country: France
• Time zone: GMT+1

7.2 Education

• Engineering school: Telecom Physique Strasbourg, France
• Subject/Major: Medical Information and Communication Technology (Sandwich Course)
• Year: Last year

7.3 Work Experience

I am currently working on the development of an image quality measurement software for a French medical X-ray sensors company named Thales Electron Devices. My code is written in C++ with BOOST library. To build the software, I use CMake.

7.4 Summer Availability

• Exams End Date: 05/26/2016
• Time Commitments: I'm available all the summer. However, as I will be graduate in September, I will find work this summer which will take me some of my time. But the GSoC remains my main priority.

8. Why did I choose this subject?

I chose this subject for two reasons. First because I want to contribute to something I daily use. Secondly because GPU are the most beautiful component of the computer.