Some new stuff..

[nAo]

SONY:
Sony Shading engine:
Surface computer and computing method using the same


XENON (ehhee ):
System and method for parallel execution of data generation tasks

CELL:
Lowered PU power usage method and apparatus
External message passing method and apparatus
SUBARRAY CONTROL AND SUBARRAY CELL ACCESS IN A MEMORY MODULE
Latch type sense amplifier method and apparatus
Apparatus and method of wordline/bitline redundancy control using shift registers in an SRAM
Low skew, power efficient local clock signal generation system
UNIFIED LOCAL CLOCK BUFFER STRUCTURES
Power saving in FPU with gated power based on opcodes and data
Power saving in a floating point unit using a multiplier and aligner bypass

Enjoy!
Marco

 

[Guden Oden]

Can some mad person read these and post up summaries in a language fit for human consumption?

 

[hovz]

lol

 

[Brimstone]

I thought Michael Abrash didn't work for Microsoft anymore? Doesn't he remain a good friend of John Carmack and the last time around when Microsoft was creating the X-Box, John Carmack and Tim Sweeney were both giving opinions on what they would like to see. So once again both Carmack and Sweeney are influencing the Microsoft console architechture?

 

[one]

Xenon patent is interesting.
It mainly talks about a Gekko-like optimized communication method between CPU and GPU, but other than that, it talks about roles of cores in a CPU module. A core is likely to have SMT. The Xenon CPU has 2 or more cores, 1 for Host CPU, 1 or more for Geometry CPU. (When the CPU has only one core, which is unlikely though, threads can do those roles.) So it's more like Emotion Engine, though it seems that it's not a combination of different types of cores, but using the same type cores for different CPU tasks.

 

[one]

SONY:
Sony Shading engine:
Surface computer and computing method using the same

This patent is quite old judging from Foreign Application Data (Nov 1, 1999 JP)

 

[nAo]

Can some mad person read these and post up summaries in a language fit for human consumption?

The Xenon patent is the most interesting, imho.
It shows a multicore cpu being able to lock a portion of its (shared between cores) L2 cache.
One CPU can fill this cache portion with data to be consumed by the GPU.
AFter a fill the CPU kick a DMA transaction to the GPU and the GPU, at the same time, tells the CPU how many data are left to fetch.
The CPU(s) have also a modified instruction set with special instructions to perform dot products and compressione/decompression of geometry data.
I believe these compression/decompression instructions are very simple and fast inctructions to convert 1/2/3/4 components vectors from fp to various integer representations (signed and unsigned bytes, shorts and longs) and back to fp.
Obviously all this machinery is best suited when the CPU generates procedural geometry..like subdivision surfaces or any other kind of high order surface that has to be tesselated (dynamic lodded characters? ), there are infinite possibilites anyway.
A very good sign of the things that will come..

ciao,
Marco

 

[Fafalada]

Yep Xenon patent is neat, and it confirms a few things I already suggested before, like the L2 cache locking - it's a feature IBM has been using in their CPUs for quite awhile already, (including in Gekko, although in that case it was L1 cache that was lockeable in that manner).

I believe these compression/decompression instructions are very simple and fast inctructions to convert 1/2/3/4 components vectors from fp to various integer representations (signed and unsigned bytes, shorts and longs) and back to fp.

If we look back at Gekko again, it uses composite load/store+conversion instructions for this, it's a neat way to make compression/decompression free, unlike most other instruction sets which require you to execute an extra op or two to convert.
I guess chances are good that Xenon ISA adopts these also.

like subdivision surfaces or any other kind of high order surface that has to be tesselated (dynamic lodded characters?

C'mon Marco, no need to hide it, we all know you're thinking about multiresolution representations

 

[nAo]

If we look back at Gekko again, it uses composite load/store+conversion instructions for this, it's a neat way to make compression/decompression free, unlike most other instruction sets which require you to execute an extra op or two to convert.

Emh..PS2?
Thanks for the info..I never messed with Gekko..

C'mon Marco, no need to hide it, we all know you're thinking about multiresolution representations

You got me

 

[pc999]

C'mon Marco, no need to hide it, we all know you're thinking about multiresolution representations

So no more blocky persons/monsters etc...

 

[arhra]

Interesting. Doing a patent search for the names of some of the guys listed as Inventors finds patents assigned to WebTV and Cagent (ie, the guys who designed the 3DO M2). I guess that acquisition didn't go totally to waste in the end after all

 

[Farid]

C'mon Marco, no need to hide it, we all know you're thinking about multiresolution representations

So you expect MRM to be possible, at little cost, on Xenon? Neat feature, to say the least.

 

[AzBat]

Look for this patent application as well...

Hardware-accelerated anti-aliased graphics

Abstract
In order to render a primitive, the primitive is subdivided into trapezoids and triangles. The subdivision occurs using scanline-aligned lines. These simple scanline-aligned regions are further subdivided so that the primitive is divided into simple scanline-boundaried trapezoids and other complex scan shapes. The simple scanline-boundaried trapezoids are rasterized. One rasterization method uses a texture map containing slope-based coverage information to edge areas. Gouraud shading may be used to provide the anti-aliasing effects on the scanline-boundaried trapezoids. The simple scanline-boundaried trapezoids may also be rasterized using a software rasterizer. Complex scans are rasterized using a software rasterizer. As data is already rasterized, it is thereby efficiently transferred to the GPU.

Seems to be filed the same day as the other one. Possible Xbox2 feature?

Tommy McClain

 

[j^aws]

...
A very good sign of the things that will come..

ciao,
Marco

<twiddles thumbs and waits>

Btw, I'm expecting an end of year summay on all the patents posted this year at B3D!

 

[Deepak]

How do I make sense of these patents. I am interested in 2D/3D stuff and all those tech stuff goes over my head (that why I never post in these thread but diligently read them ). My knowledge about graphics is almost nil. What books should I start with?

 

[Megadrive1988]

Interesting. Doing a patent search for the names of some of the guys listed as Inventors finds patents assigned to WebTV and Cagent (ie, the guys who designed the 3DO M2). I guess that acquisition didn't go totally to waste in the end after all

this is good to hear. I've been hoping that the acquisition of Cagent and their M2 & MX technologies would somehow be put to use in Xenon.

 

[Inane_Dork]

Look for this patent application as well...

Seems to be filed the same day as the other one. Possible Xbox2 feature?

Tommy McClain

I've read a good part of it, but I am not convinced it is a Xenon feature. But I'm not convinced the Xenon will be without it, either.

The argument against basically is that the system described is in no way slanted towards consoles. The exemplary system is a PC, and the process is said to work in offline rendering as well as real-time.

The argument for is that this method is supposed to reduce memory demands, which seems to be a big deal for MS with the X2. Also, recently some ATi employee posted here about a totally different method for doing AA. This may have been it.



EDIT: The patent seems to be mostly about hardware anti-aliasing for vector graphics, not polygons. It may help with HUDs and text rendering on Xenon games, but that's about the only relation I see at the moment.

 

[AzBat]

I've read a good part of it, but I am not convinced it is a Xenon feature. But I'm not convinced the Xenon will be without it, either.

The argument against basically is that the system described is in no way slanted towards consoles. The exemplary system is a PC, and the process is said to work in offline rendering as well as real-time.

The argument for is that this method is supposed to reduce memory demands, which seems to be a big deal for MS with the X2. Also, recently some ATi employee posted here about a totally different method for doing AA. This may have been it.



EDIT: The patent seems to be mostly about hardware anti-aliasing for vector graphics, not polygons. It may help with HUDs and text rendering on Xenon games, but that's about the only relation I see at the moment.

Thanks for the info. Is it possible this could help with the Xbox2's possible use of MRM technology?

Tommy McClain

 

[Inane_Dork]

MRM? I Googled and found multi-resolution mesh, is that what you mean?

If so, I don't think this has any relation to that. This is more like micropolygons applied to vector graphics. I think it was implicit in the patent that there was no depth buffer, thus everything was drawn from back to front and AA was handled through modified transparency. That would not work for 3D graphics, I don't think.

 

[AzBat]

MRM? I Googled and found multi-resolution mesh, is that what you mean?

Yes.

If so, I don't think this has any relation to that. This is more like micropolygons applied to vector graphics. I think it was implicit in the patent that there was no depth buffer, thus everything was drawn from back to front and AA was handled through modified transparency. That would not work for 3D graphics, I don't think.

Okie dokie.

Tommy McClain