|
| Some recent topics started on our forums |
|
Finding a cofounder...
|
My business partner and I currently find ourselves in a situation where we have a product that we're about ready to sell, but we have almost no business experience between us. We also feel that it would be silly to try and put on the business h... |
|
Posted February 14, 2009 5:25:51 AM
|
|
Bootstrap Engine
|
Hello all! This is a screenshot of a graphics engine that a friend and I have been working on for the last 2 and a half years. We're currently going into beta with it and are hoping to start licensing it shortly. If you'd like to find out mor... |
|
Posted April 3, 2007 1:22:27 AM
|
|
MS not supporting GL in longhorn?
|
According to www.OpenGL.org Microsoft is planning on dropping true OpenGL support in Windows Vista. According to the post on OpenGL.org, "
* OpenGL performance will be significantly reduced - perhaps as much as 50%
* OpenGL o... |
|
Posted August 5, 2005 5:01:29 PM
|
|
View All Topics Started By This User
|
|
| Some recent replies made on our forums |
|
Help with SSE/assembler
|
Jan,
The reason I suggested store forwarding rather than using a GPR is because the MOVSS instruction cannot load from a GPR (only m32 or xmm register, at least that I'm aware of). As for cache concerns, I'd say they're the single biggest concern... |
|
Posted November 7, 2010 8:53:36 PM
|
|
Help with SSE/assembler
|
Using a global variable is exactly what you want to avoid. This can cause cache misses, and in the event of a cache miss, this function will be *wayyyyy* slower than even the branching method. What you need to do is embed your constant into the ins... |
|
Posted November 6, 2010 10:43:58 PM
|
|
Help with SSE/assembler
|
Jan,
Ah, you're right about the latency. I was looking at the L2 cache latency, not the L1 cache latency. Sorry about the confusion! It always bums me out to see bad information on forums, and here I am spreading it =(.
I think you're r... |
|
Posted November 6, 2010 6:12:43 PM
|
|
Help with SSE/assembler
|
@Jan
If performance is a concern, then you definitely want to utilize the branch and not load a value from memory. Even in the event of an L1 cache hit, you'll still see a latency penalty that is roughly proportional to 60% of the cost of a branc... |
|
Posted November 6, 2010 3:45:27 PM
|
|
Help with SSE/assembler
|
I would really try to avoid NaNs. I should note that you could move your COMISS instruction up to test for zero prior to calculating the inverse square root. This would completely avoid NaNs, and you'd still end up with the correct result. In shor... |
|
Posted November 5, 2010 7:22:05 PM
|
|
Help with SSE/assembler
|
I wouldn't worry about the branch in this context. You're only going to skip a few instructions, so the cost of a branch miss is simply going to be a flush of the pipe (probably on the order of 16-20 clock cycles depending on the proc... |
|
Posted November 5, 2010 4:34:27 PM
|
|
Help with SSE/assembler
|
Look up the COMISS instruction. I believe that will help you achieve what you want. =)
Kevin B |
|
Posted November 5, 2010 4:31:51 PM
|
|
HL2 BSP Lightmap Coordinates
|
If you're using OpenGL, try this:
float lu = TextureInfo[Faces[faceId].texinfo].lightVecsU.dotProduct(Vertexes[vertexIndice])
+ TextureInfo[Faces[faceId].texinfo].lightOffsetU
- Faces[faceId].LightmapTextureMinsInLuxels[0]
+ 1.5;
... |
|
Posted September 17, 2010 12:46:25 PM
|
|
Aptitude in math?
|
Quote:
Not having one makes getting a job significantly harder. Entry level positions get hundreds of resumes. Not all of those are going to get phone screens, let alone an interview so you get to prove you know what you're doing. HR people can... |
|
Posted December 8, 2009 11:43:55 PM
|
|
Aptitude in math?
|
It makes me sad to see people go "oh, there is math there... I don't want to do that." It also makes me sad when I see things like, "Oh, you need college and a math degree!" It also makes me sad when someone looks to other people to tell them what... |
|
Posted December 8, 2009 8:06:51 PM
|
|
View All Replies Made By This User
|