Search found 138 matches

by shiqiu1105
Sat Feb 16, 2013 4:48 am
Forum: General Development
Topic: The path tracer in SmallVCM, and more questions on BDPT
Replies: 6
Views: 4137

The path tracer in SmallVCM, and more questions on BDPT

I have been reading the source code of SmallVCM. The path tracer in it is a little weird to me. It has the next event estimation. However, whenever the path hits an area light source, it also adds the contribution of directly hitting the light source. In pbrt, only the previous vertex is specular wo...
by shiqiu1105
Fri Feb 15, 2013 10:47 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

SOA is good for SIMD processing, because it allows you to read 4 coordinates at once in an 128bit register. wakeup call ! we have already 512bit in place. I know that, but the principle is the same. :) The man wanted to know what SOA is. Okay, but why it is good for SIMD? For example, a Vec4 would ...
by shiqiu1105
Fri Feb 15, 2013 6:56 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

mpeterson wrote:
jbikker wrote:SOA is good for SIMD processing, because it allows you to read 4 coordinates at once in an 128bit register.
wakeup call ! we have already 512bit in place.
What do you mean?
by shiqiu1105
Fri Feb 15, 2013 3:19 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

The best approach developed by Timo does not use persistent thread ! Check the paper ;-) Also, it depends of what you want to do... if you want to improve this code to support instances etc... by example... and maybe depend of the video card and the scene ! It is up to you to test... or check the r...
by shiqiu1105
Fri Feb 15, 2013 2:18 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

Thanks for you guys answer! They helped a lot :D

I do want to know, if persistent threads are not efficient nowadays, what's the efficient approach now then?
by shiqiu1105
Fri Feb 15, 2013 2:33 am
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

I have also downloaded his code. He uses a lot of SOA pointers which makes it quite hard to read..
by shiqiu1105
Fri Feb 15, 2013 2:27 am
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

Probably you should re-read the paper a few times. Packet traversal is simply tracing a group of rays together at once. Even sans wide hardware ALU benefits, it should improve memory access, since the groups of rays traced together will be coherent. Traditionally, it's been tuned to CPU SIMD, but n...
by shiqiu1105
Thu Feb 14, 2013 10:12 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Re: Can't understand Timo Aila's GPU traversal paper

What's packet trevsal? I thought that's on the CPU using SIMD
by shiqiu1105
Thu Feb 14, 2013 9:28 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9434

Can't understand Timo Aila's GPU traversal paper

I have been reading Alia's paper on the GPU ray travsal.

But don't quite understand it.

So basically, to do fast GPU ray tracing, we need to launch enough threads to fill the gpu,and use while-while loop?

Anything like QBVH he's using?

Can anyone who understand the paper give a short review :) ?
by shiqiu1105
Wed Feb 13, 2013 7:40 pm
Forum: Links & Papers
Topic: Ph.D. thesis
Replies: 21
Views: 27544

Re: Ph.D. thesis

tomasdavid wrote:Because your Ph.D. thesis is about what you have done, not about what someone else did. :-)
Oh? I thought path regeneration and stream path tracing are also someone else's works.