Profile

Cover photo
Matt Pharr
Works at Google X Lab
Attended Stanford University
394 followers|45,245 views
AboutPostsPhotosVideos

Stream

Matt Pharr

Shared publicly  - 
 
 
CALL FOR SUBMISSION TOPICS.  Jaroslav Krivanek and I have been invited to organize a special computer-graphics-themed issue of TTSP, the journal on Transport Theory and Statistical Physics (which just changed its name to Journal of Computational and Theoretical Transport).  TTSP was founded with a charter of promoting cross-field awareness of methods for solving transport equations, both linear and non-linear, using exact or approximate methods, Monte Carlo or deterministic.  Boltzmann-like transport equations are at the foundation of many different disciplines in physics for describing the bulk behaviour of scattering neutrons, gas molecules, photons, waves, etc.  This invitation is a wonderful recognition by the transport theory community that we are solving interesting transport problems in computer graphics.  At this point we'd like to get a sense of the community's interest in participating.  Please contact me with specific topics if you have original research submissions in those topic that you think would be appropriate for this special issue.  Acceptable topics would be mostly contained in: Monte Carlo methods for global illumination or rendering participating media, variance reduction methods, analytic BSSRDF methods (basically, any problem where solving an integral equation is required to compute light transport).  Please contact us if you have any questions.
2
Robin Green's profile photo
 
I'm afraid my research on spherical basis functions is not analytical enough for the rest of the Physics world. There's a lot of handwaving going on in graphics, mainly because it's really hard to get actual physicists and mathematicians to join in and give a damn about computational efficiency.
Add a comment...

Matt Pharr

Shared publicly  - 
 
Aaron Lefohn originally shared:
 
Presentations from the Beyond Programmable Shading SIGGRAPH 2011 course are now posted: http://bps11.idav.ucdavis.edu/. There are a couple of talks missing, and those will get posted soon.
4
1
Add a comment...

Matt Pharr

Shared publicly  - 
 
Me with Charles Babbage's brain, at the Hunterian Archives at the Royal Academy of Surgeons.
2
Matt Pharr's profile photoPaul Lalonde's profile photo
3 comments
 
+Raul Aguaviva nice--looks like better light for photography as well. :-). I'll try to make it there!
Add a comment...

Matt Pharr

Shared publicly  - 
 
Is there a special loser of finnkampen prize I get if I attended but then am now back in my hotel room and about to go to sleep before midnight?
1
Add a comment...

Matt Pharr

Shared publicly  - 
 
First image from rudimentary REYES renderer. Next steps: fix image bugs, make implementation better (much room for improvement). Then onward to ispc-ization..
9
Matt Pharr's profile photoDaniel Wexler's profile photo
2 comments
 
Programming at HPG & SIGGRAPH dramatically increases bug generation.
Add a comment...

Matt Pharr

Shared publicly  - 
 
Dante's ~5th(?) circle: enough cut and paste to be painful, not enough to be worth it to refactor into a separate routine.
2
Add a comment...
Have him in circles
394 people
Johan Andersson's profile photo
Tom Duff's profile photo
Dan Goldman's profile photo
Matt Walsh's profile photo
Jaymin Kessler's profile photo
Chris Mills's profile photo

Matt Pharr

Shared publicly  - 
 
Great writeup on ray tracing vs. rasterization in c0de517e's blog. This very nicely addresses a lot of the myths about the purity/elegance/supposed greater efficiency of ray tracing over rasterization.

http://c0de517e.blogspot.com/2011/09/raytracing-myths.html
9
2
Eric Enderton's profile photoDavid Luebke's profile photoFredo Durand's profile photoBryan McNett's profile photo
5 comments
 
I think most of us understand that tracing vs. rasterization is a duality, like photons vs. waves. Some rendering is best described in terms of one, and other rendering in the other.

Contemporary game rasterization HW essentially traces eye ray packets in parallel into a triangle soup, and a game's scene graph is essentially a very shallow BVH (meshes bounded by spheres, containing vertexsets bounded by spheres) for testing against a single eye ray mega-packet (the view frustum.)

The contemporary model can be slightly generalized to support a deeper BVH and secondary rays.

Such generalization has its price, but so do the ubiquitous "X mapping" techniques that render a secondary scene into a texture, and then render a primary scene while reading from the texture.

Sometimes it is cheaper in HW, and often it is cleaner visually, to simply walk the secondary scene from within the main scene.
Add a comment...

Matt Pharr

Shared publicly  - 
 
With the other half of Babbage's brain. With a trip to Bletchley Park, I'll have the trifecta. 
1
Add a comment...

Matt Pharr

Shared publicly  - 
 
Inspired by Production Volume Rendering course by +Magnus Wrenninge and +Nafees Bin Zafar at siggraph, finished implementing a simple ray marching volume renderer in ispc at the airport. 2.3x speedup on the first go round (on one core). (Will probably include this as an example in the next release.)

Using a nice dataset from the Wavelet Turbulence for Fluid Sim paper by Theodore Kim et al from SIGGRAPH '08.
13
Robin Green's profile photoHector Yee's profile photoDoug Epps's profile photoJohn Bates's profile photo
5 comments
 
+Hector Yee tricubic interpolation is left as an exercise to the reader!

+John Bates just autovectorization. It's stepping packets of 4 rays through the volume together across the SIMD lanes. The main perf. issue (i.e. why "just" 2.3x) seems to be the fact that they follow paths through different sets of voxels, so the memory accesses to load voxel data end up being scalarized, which is slow...
Add a comment...

Matt Pharr

Shared publicly  - 
 
Rice Krispy-crusted bacon with a spicy maple syrup sauce. (@ "The Fish House" restaurant, Vancouver.)
4
1
Jeremy Shopf's profile photoEric Haines's profile photoFrancesco Banterle's profile photo
3 comments
 
Salty or sweet?
Add a comment...
People
Have him in circles
394 people
Johan Andersson's profile photo
Tom Duff's profile photo
Dan Goldman's profile photo
Matt Walsh's profile photo
Jaymin Kessler's profile photo
Chris Mills's profile photo
Work
Occupation
Software Engineer @ Google[x]
Skills
Programming
Employment
  • Google X Lab
    Software Engineer, 3 - present
  • Intel Corporation
    Principal Engineer
  • Neoptica
    Founder/CEO
  • NVIDIA
    MTS
  • Exluna
    Founder
  • Pixar
    Rendering R&D
Basic Information
Gender
Male
Story
Introduction
Rendering nerd turned compiler nerd.  See also my homepage.
Education
  • Stanford University
    Computer Science, M.S. and Ph.D., 1996 - 2001
  • Yale University
    Computer Science, B.S., 1989 - 1993
Links
YouTube
Other profiles