Tagged with optix - NVIDIA Developer Forums

Solve linear system in OptiX / Dynamic memory

xindon — Wed, 16 May 2012 15:28:28 -0400

Hey OptiX experts :-),

in order to implement a moving least squares point set surface algorithm I need to solve a linear system for each ray. The matrix sizes are variable and can change from frame to frame.

1) As OptiX does not allow dynamic memory allocation (free/malloc or new/delete), how/where should the data of the linear system be stored best?

Each ray (or more precise: each instance of the intersection program) needs its own local memory for the linear system. I thought of allocating a huge global buffer, but I'm not sure how to do the indexing such that the different programs don't interfere.

2) Do I have to implement the solver manually or are the better/faster/easier implementations already available? (which maybe also solve the above local memory problem).

Thank you very much in advance!

Regards
Tim

Kernel launch fails if it contains rtTrace call

cneumann — Thu, 10 May 2012 12:21:35 -0400

It seems I can't get optix to actually trace any rays. As soon as I have a rtTrace call in my ray generation program the kernel launch fails (with exceptions enabled every launch index reports a RT_EXCEPTION_INTERNAL_ERROR). Not tracing rays and simply writing a color value to the output buffer works fine.
I can also successfully run the SDK examples (both precompiled and self built).

Since we've been struggling with this for a couple of days (at first suspecting a problem in my C++ wrapper code), we also wrote a logger library that intercepts all OptiX calls (it's LD_PRELOAD'ed) and shows arguments and return values, a log of a reproducer run is attached as well.
Comparing the log of our program with one from running an SDK example I can't find any substantial difference, but apparently there is something subtle going on ;)

Any suggestions on how to narrow down or solve the problem are very much appreciated. Thanks!

Platforms:
1) GTX 8800, Fedora 16 x86_64, driver 295.40, OptiX 2.5.0
2) GTX 560M, Win 7 64bit, driver 295.73, VS 2010, OptiX 2.5.0

OptiX 2.5.0 Ubuntu 10.10 64bit download is missing some examples

cneumann — Wed, 16 May 2012 11:23:15 -0400

The OptiX 2.5.0 download for Ubuntu 10.10 64bit (http://developer.nvidia.com/sites/default/files/akamai/tools/files/OptiX_2.5.0/NVIDIA-OptiX-SDK-2.5.0-ubuntu10.10-64.run) is missing the following directories/examples:

SDK/hybridShadows
SDK/isgShadows
SDK/isgReflections

as well as the precompiled versions of these.
If they were omitted intentionally SDK/CMakeLists.txt should be adjusted to not mention these directories in add_subdirectory() calls.

How to get the result in a data form?

milkiii — Mon, 14 May 2012 23:00:44 -0400

The demos feedback the result of ray tracing through a picture. If I now have a mesh which is consisted by several vertex, and I want to use a form to represent each vertex's light intensity, does any function or method to get this kind of result? Or, since PerRayData.result stores the result after ray tracing, can I get it in some way? It is so important to me, hoping I can get the answer. Thank u very much!

Fault with optix::AssertionFailure

milkiii — Wed, 09 May 2012 01:51:06 -0400

Now I am learning the tutorial in SDK and rewriting it by myself. When I ran the project to the function createProgramFromPTXFile, it checked error and exited, suggesting assertion failure. Thus I replaced the ptx file which generated by compiler with the one in demo, all things go well. I don't know how to solve this problem, whick part shall I modify.

AS Builder memory intensive! (v2.5.0)

Crog — Wed, 09 May 2012 07:24:14 -0400

In v2.5.0 the AS Acceleration build appears to be more memory intensive than previous versions. To my knowledge builds were previously not async which would explain this!

I wonder if it would be possible to tone down the builder or make it more intelligent in its execution when the amount of memory used by a process nears the 32bit limit. Its a problem as if lots of models need building the memory usage sky rockets. So we load 1gb of data, pass it to optix who makes a copy (2Gb) and then the build of 200 odd models takes it way way up! If we load the models in batches we haven't got the problem though and all the resoluting accelerations easily fit in memory if leaded from cache. If we could somehow tell optix to be a bit more careful in the memory used for building it could be useful or even make it single threaded as desired. Even better is if we could actually initiate the build process without having to call Launch which I would hope/expect will be a v3 feature...!?

Cheers,
Craig

Discuss ideas on possibly useful debugging features...

Crog — Wed, 09 May 2012 05:34:10 -0400

I have been making good use of the rtPrintf() in debugging recently and was wondering what other features OptiX could have in aiding debugging and diagnosis of issues.

One 'possibly silly' idea I decided to put out there is the inclusion of an rtGetKeyState( 'A' ) function! I know we could just set variables up ourselves but having this in-built is handy for switching lines of code on/off without having to get inside and recompile the core application. The inclusion of such a feature could be handy in cases for switching between test outputs and different modes for demonstrations so accelerating development. It is hopefully very little work for the Nvidia folks to implement...

Flucuating performance when increasing framesize in OptiX over Quadro Plex 7000

Mazenoz — Tue, 08 May 2012 15:11:32 -0400

Hello all,

I have developed new techniques implemented in OptiX to enable better performance of stereoscopic ray tracing (i.e. ray tracing two slightly offset images of the same scene). Strange is, when running the OptiX code on a Quadro Plex 7000, I get fluctuating performance when increasing the frame size. However I don't get such fluctuations when executing the code on a GeForce 545 M. See the graphs below:

Each step is averaged over 20 runs, where each run is executed for 20 seconds to compute the average fps.

I couldn't explain this behavior, could anyone please help me with this?

Real-time soft shadows in SceniX Optix using shadow maps issue?

jules123 — Wed, 02 May 2012 13:01:21 -0400

Hi,

I am implementing a hybrid solution of ray-tracing soft shadows that relies on shadows maps to speed things up. I have my shadow maps (depth) stored in sampler2DARRAY layers generated from CgFX shader code.

It works fine for primary rays as I can keep most of the visibility work in CgFX. The problem comes when rendering reflections and refractions as my renderer then crosses over to the RTFx shader code to do its work. I then would need access to the sampler2DARRAY shadow maps, from within RTFx.

I've asked a similar question before and the answer was to use demandUpload; meaning a GPU to CPU copy and then a CPU to GPU copy to give RTFx access to CGFx samplers. Because of the number of shadow maps this would quickly saturate, not a possible solution in this case.

Is there a GPU to GPU solution to enable TextureGL2DArray textures created in OpenGL that can be read directly by Optix in Scenix?

Thanks for any help,
Jules

Problem with Optix

Saouli — Tue, 01 May 2012 00:45:39 -0400

I am a biginner in Optix with Cuda, I tried to create a project with visual C++ 2008 for compile ambient occlusion in sdk optix

I have this error when I compile my solution

c:\program files\nvidia corporation\optix sdk 2.5.0\include\optixu\optixu_math_namespace.h(196) : fatal error C1189: #error : "optixu_math_namespace.h needs NOMINMAX defined on windows."

How i can compile this project using cmake or correct this error

Thank you so much.

Crash during buffer map with OptiX 2.5 after kernel launch and buffer resize

waltron — Thu, 12 Apr 2012 15:47:00 -0400

Edit: The problem still exists for me and so I am still interested in any answer. It would be realy nice to know, if anybody could reproduce the crash.

Hi there!

Some time ago, I tried to switch from 2.1 to 2.5 but had a problem because of program crashes. Unfortunately I could not provide you with a simple running/crashing example. Now I tried to switch again and did my homework.

The following example runs under Win7 on a GTS450 with Cuda 4.0 and driver version 296.10 while using the optix.1.dll of version 2.1.1. Just exchanging the dll with the 2.5 version crashes the program at the marked position.

I realy hope you can reproduce the crash and give me a hint about whats going wrong their. Of course I hope it's in my code so its easy to fix ;-)

Best regards
Andreas

PS: The code samples

The ray generation program


#include 
using namespace optix;

rtDeclareVariable(uint2, index, rtLaunchIndex, );
rtDeclareVariable(uint2, dim, rtLaunchDim, );

namespace FillBuffers
{
	rtDeclareVariable(float, FillValue, , );
	rtBuffer ResultBuffer;
}

using namespace FillBuffers;

RT_PROGRAM void optixKernel()
{
	ResultBuffer[index] = FillValue;
}

The steering code


#include "stdio.h";

#define NOMINMAX
#include "optix_world.h";

using namespace optix;

void main()
{
	std::cout << "Testing BufferResize" << std::endl;

	uint aWidth = 2;
	uint aHeight = 2;

	Context aCon = Context::create();
	Program aProg = aCon->createProgramFromPTXFile(
		"D://FillWithValue.x86.ptx", "optixKernel");

	aCon->setEntryPointCount(1);
	aCon->setRayGenerationProgram(0, aProg);

	Buffer aResultBuffer = aCon->createBuffer(RT_BUFFER_INPUT_OUTPUT, RT_FORMAT_FLOAT, aWidth, aHeight);
	aCon["FillBuffers::ResultBuffer"]->setBuffer(aResultBuffer);

	aProg["FillBuffers::FillValue"]->setFloat(1);

	aCon->validate();
	aCon->compile();
	aCon->launch(0, aWidth, aHeight);

	aWidth = aHeight = 3;
	aResultBuffer->setSize(aWidth, aHeight);

	// BOOM with OptiX 2.5 DLL, but just fine with OptiX 2.1
	aResultBuffer->map();

	std::cout << "The End";
	getchar();
}

Optix output buffer format

HolyCaesar — Tue, 01 May 2012 12:03:36 -0400

Hello,

I am new to cuda with optix. I got a problem with optix output buffer format. If I define the output buffer in .cu file as rtBuffer, then everything goes fine. However, If that output buffer defined as rtBuffer, then I just got black image and my program can pass the compiling. But the optix sample code is ok with rtBuffer.

One of my class is inherit simple scene class from sutil project, which comes with optix samples.

So, how can I get the image from this type of output buffer?

Thank you very much.

Does optiX always keep a copy of buffers on the host/CPU?

Crog — Tue, 01 May 2012 07:44:08 -0400

Hi,

This observation might be wrong but it appears OptiX will always keep a copy of buffers in host/CPU memory? This causes some issues when loading large scenes. Alsois the place where different buffer types are stored documented which may include possible single/multi-GPU differences?

Thanks for any help,
Craig

Optix with Geforce GTX 680 & 690

kaufmann — Mon, 30 Apr 2012 13:15:04 -0400

Has anyone tested Optix with the GTX 680 yet and can you report performance increases/decreases compared to a 580 / 590?

Will Optix 2.5 work with two GTX 690 right from the start next week or will we have to wait for an update / updated drivers?

thanks

k-nearest neighbor (kNN) search in OptiX?

jules123 — Mon, 23 Apr 2012 14:53:47 -0400

Hi,

A much used function for irradiance calculations in ray tracing is the k-nearest neighbor (kNN) search? In SceniX OptiX, I launch N randomly directed rays from a point and use the closest hit program to pick up the N intersection points.

I am wondering if NVIDIA can provide this as a highly optimized function call for a future release of OptiX?

Kind regards,
Jules

P.S. Where do we post to suggest new features?

Unknown error in rtProgramCreateFromPTXFile (Assertion 5308756)

xindon — Sat, 28 Apr 2012 10:57:18 -0400

Hi,

I'm really new to OptiX and I get the following error when trying to load a program:

Unknown error (Details: Function "_rtProgramCreateFromPTXFile" caught exception:
 Assertion failed: [5308756])

The line causing the error is simply:

optix::Program pinholeCamera = this->context->createProgramFromPTXFile("RayTracingKernels.ptx", "pinholeCamera");

According to the Visual Studio debugger the return code of rtProgramCreateFromPTXFile is -1.

The kernel file looks like this:


#include 
#include 

using namespace optix;

struct RayPayload
{
  float3 color;
  int    recursionDepth;
};

rtBuffer outputBuffer;

rtDeclareVariable(uint2, launchIndex, rtLaunchIndex, );
rtDeclareVariable(uint2, launchDimension, rtLaunchDim, );
rtDeclareVariable(optix::Ray, currentRay, rtCurrentRay,	);
rtDeclareVariable(RayPayload, currentRayPayload, rtPayload, );
rtDeclareVariable(float, tIntersection,	rtIntersectionDistance,	);

rtDeclareVariable(float3, eye, , );
rtDeclareVariable(float3, U, , );
rtDeclareVariable(float3, V, , );
rtDeclareVariable(float3, W, ,);

rtDeclareVariable(rtObject, topObject, , );
rtDeclareVariable(unsigned int,	defaultRayType,	, );
rtDeclareVariable(float, sceneEpsilon, , );
rtDeclareVariable(int, maxRecursionDepth, , );

rtDeclareVariable(float4, sphere, , );


/**
 * Pinhole camera ray generation program.
 */
RT_PROGRAM void pinholeCamera()
{
	float2 d = make_float2(launchIndex) / make_float2(launchDimension) * 2.f - 1.f;
	float3 rayOrigin = eye;
	float3 rayDirection = normalize(d.x*U + d.y*V + W);

	optix::Ray newRay = optix::make_Ray(rayOrigin, rayDirection, defaultRayType, sceneEpsilon, RT_DEFAULT_MAX);

	RayPayload payload;
	payload.color = make_float3(0.2f, 0.2f, 0.2f);
	payload.recursionDepth = 0;

	rtTrace(topObject, newRay, payload);   // <----- THIS LINE CAUSES THE ASSERTION ERROR

	outputBuffer[launchIndex] = make_float4(payload.color, 0.0f);
}

// .. several more RT_PROGRAMs..

The assertion error vanishes if I remove the call to "rtTrace(topObject, newRay, payload);" from the kernel.

Additional Information:
- OptiX 2.5.0
- CUDA 4.2
- Windows 7
- Visual Studio 2010 + Parallel NSight 2.2 RC2
- GeForce GTX 470
- Driver version: 301.32

- Compile parameters for kernel file:
- Code generation: compute_20,sm_20
- 64 bit

Changing code generation back to default settings (compute_10,sm_10) results in the same error, just with a different assertion code: 11403396

Using CUDA 4.0/4.1 or switching to 32 bit doesn't help either.

Is there some global size limit/recommendation for the OptiX kernel files? Currently I'm putting all RT_PROGRAMs and all variable declarations (see above) in a single large file.

Thank you very much!

Regards
Tim

SOLVED Problem in Optix and HD graphics ? No srgb support

Conkal — Sat, 28 Apr 2012 12:11:11 -0400

Hello ,

I am programming a Bidirectional Path tracer in Optix and I have encountered a problem , which is also the same using the "Path tracer" sample from the Nvidia Optix Samples.

(My computer is using intel i5-2500K (with iGpu intel HD-3000)and GTX560ti+ GTX460 , monitor is Samsung P2270, but the same happened in a couple of other computers. The drivers are the official ones with default parameters )

The problem is this :

1. When i connent my monitor to the GTX 560ti or GTX 460 graphics cards that i have, then the colors in the rendering are a lot brighter than when connecting the monitor to the iGPU of the processor.

2. The result of the "printscreen function"(pressing s in the rendering window) is the same with the result in the igpu .

What is happening ? Is it a bug from the sample code that copied by mistake or is it a problem with optix ? And more importantly WHICH RENDERING IS THE RIGHT ONE ?

I have uploaded the rendering of the default path tracer example and the same scene(almost) in mine.

Each one has three pictures , printscreen using the gtx , another one using the intel HD and the optix output from the "s" command.

.mtl and sample6 in Optix

fumble — Thu, 26 Apr 2012 08:28:14 -0400

Hi everyone,
I try to make a render with sample 6, but if i choose ambient occlusion shader or one bounce diffuse, the .mtl file is read but materials are not used.
If i want to have materials on my obj models i need to choose the default shader.

How can i had .mtl informations in anothers shader like ambient occlusion shader or one bounce diffuse?

Thank for your time!

Optix: Error Code 3

mcleary — Wed, 18 Apr 2012 23:27:18 -0400

I write an application that import collada files and display using OptiX. It works well for complex scenes etc. But, for some reason when trying to render a car model, it crashes with the following message

The NVIDIA OpenGL driver detected a problem with the display driver and is unable do continue. The application must close.

Error code: 3

And the OptiX generates an exception with the following message

Unknown error (Details: Function "_rtContextLaunch2D" caught exception: Encountered a CUDA error: driver().cuEventSynchronize(m_event) returned (999): Unknown, [6619195])

I have no idea what may be causing this error since my code works well for various scenes.

Any hints?

rtDeclareVariable vs rtBuffer performance in SceniX OptiX?

jules123 — Sun, 22 Apr 2012 16:32:29 -0400

Hi,

I have custom materials with up to 80 parameters (in each RTFx) declared using rtDeclareVariable at StateSet level. I am wondering, would performance be better to have one rtBuffer of material structures declared at Scene level, and then access them at StateSet level by setting an index for each StateSet that would point to the start of their given material structure in the rtBuffer?

How does SceniX Optix manage the rtDeclareVariable variables when switching between different StateSets for each launch_index/rtTrace?

Where are the rtDeclareVariables stored on the hardware? Are they stored in shared memory, registers, constant memory or accessed from GPU RAM (global memory) etc.?

Thanks for any help,
Jules

How to unbind a program from a context?

Tofic — Fri, 13 Apr 2012 14:15:59 -0400

Hello,

The scenario is like that:
1. I bind a miss program;
2. I change the computations and want to unbind this miss program.
If I call context->setMissProgram(0, NULL), I get an exception there.
I have to unbind it, as otherwise, I destroy this miss program and the context complains about it:
"OptiX Error: Invalid context (Details: Function "_rtContextLaunch2D" caught exception: Validation error: Uninitialized variable: $missProgram_2, [4915372])"
What's the solution in such case? Is it possible to disable/unbind this slot for miss program without destroying the context?

The same question relates to other program types (ray generation, exception, etc.)

Anton

NURBS intersection?

SBWillie — Mon, 09 Apr 2012 07:53:49 -0400

Hi, I'd like to implement NURBS ray tracing into my optix project. I'm looking for a little bit of guidance on how to implement the intersection and bounds programs. Thanks

Optix with Cuda, how to generate a ptx file

milkiii — Wed, 18 Apr 2012 01:28:57 -0400

I am a beginner in Optix with cuda. Doc in SDK says ptx file is compiled by nvcc, but when I compiled the project, it generated mang .gpu, .c,.i file. Though I choosed NVCC compilation type being -ptx and keep preprocessed files, I still find the ptx file. Is there any wrong with my object? Thank u very much,

Device texture limit of 128 exceeded

Lucia — Fri, 20 Apr 2012 16:52:20 -0400

Hi,

I'm trying to render an obj model, and I give this error:


OptiX Error: Invalid context (Details: Function "RTresult _rtContextValidate(RTcontext_api*)" caught exception: Validation error: Node validation failed for 'top_object':
Validation error: TextureSampler validation failed for 'diffuse_map':
Validation error: Device texture limit of 128 exceeded.  , [4915527], [4915337], [4915294])

I know it is due to the number of materials, but how I can fix this?

OptiX Transform Node Question

mcleary — Wed, 18 Apr 2012 14:02:25 -0400

If I have a moving object and a static object whose geometries does not change over time, just it's location, I can represent this object within an OptiX context like the following:


Group (NoAccel)
-- Transform
-- -- GeometryGroup (Sbvh) (Moving object)
-- -- -- GeometryInstance
-- Transform
-- -- GeometryGroup (Sbvh) (Static object)
-- -- -- GeometryInstance

I use the Sbvh builder since the underlying geometry is static for both objects. What I understand for the Programming Guide is that

OptiX does not rebuild any acceleration structures when the transform changes

When the transform node for the moving object changes, do I need to mark as dirty its underlying accelerations?

What is happening in my application is that the moving object disappears as if its bounding box stands still while the geometry moves.

Would it be possible to have tighter interop from OpenGL to OptiX in SceniX?

jules123 — Wed, 18 Apr 2012 08:09:57 -0400

Hi,

In particular eliminating texture to host to texture copies when using COLOR_ATTACHMENT output from an OpenGL fragment program as an OptiX rtTextureSampler input in RTFx. If it can all be kept on the GPU somehow that would be perfect.

There are new rasterizing and ray-tracing hybrid real-time rendering algorithms (meaning I'm coming up with some, and so are other people I presume) that really does need this feature. SceniX makes it easier to go from OptiX to OpenGL (a GPU to GPU memory copy I believe) but not the other way round.

Would this be possible?

Thanks for any help,
Jules

Is it possible to have SceniX OptiX launch dimension size different to its RTFx outputBuffer size?

jules123 — Tue, 17 Apr 2012 00:34:14 -0400

Hi,

I want not to launch rays through every pixel of the screen but only through screen coordinates read from a buffer/array.

So I want to be able to set up the launch dimension size to be the size of this buffer/array. Then in my RTFx code use the launch index to look up from the buffer/array the screen coordinate to trace the ray through. Hence the outputBuffer would need to be the size of the screen resolution and would be larger than the size of the buffer/array (launch dimension).

Is this possible?

Thanks for any help,
Jules

About .ptx file and Optix display

milkiii — Thu, 12 Apr 2012 08:47:18 -0400

It must be a simple question but it confuses me still. Now I am learning the demo, Sample 1, in OptiX Sdk,trying to write in a new project. I rerwote the .cpp file, and it compiles successfully.
However, it failed to run and said access violation. I am wondering the use of "char *ptx_filemane = ". Whether a .ptx file generated automatically，or I should do something about it. Moreover, in the sample, it used the head file "sutil.h" to help display the image, I must have some mistake when not using this file and doing nothing. Need your help, and thank you very much.

OpenGL color attachments to OptiX rtTextureSampler objects?

jules123 — Sun, 15 Apr 2012 21:11:06 -0400

Hi,

Damn, I think I just hit a roadblock :(

I need the color attachment texture object outputs from an OpenGL render as rtTextureSampler inputs to an OptiX renderer.

The two color attachments are defined as:
OpenGLtoRenderTargetGLFBO->setAttachment(nvgl::RenderTargetGLFBO::COLOR_ATTACHMENT0,
nvgl::TextureGLRectangle::create(GL_RGBA32F, GL_RGBA, GL_FLOAT))

OpenGLtoRenderTargetGLFBO->setAttachment(nvgl::RenderTargetGLFBO::COLOR_ATTACHMENT1,
nvgl::TextureGLRectangle::create(GL_RGBA32F, GL_RGBA, GL_FLOAT))

The RTFx program reads from two rtTextureSampler objects TEX0 and TEX1 defined:

rtTextureSampler TEX0;
rtTextureSampler TEX1;

How do I get the OpenGL COLOR_ATTACHMENT0 & COLOR_ATTACHMENT1 to be read by TEX0 & TEX1 in my OptiX RTFx program for best performance?

Thanks for any help,
Jules

Optix 2.5 (Feb 24 2012) crashes iMac with Snow Leopard

gerdmarmittsironacom — Mon, 16 Apr 2012 05:51:19 -0400

Optix 2.5 (Feb 24 2012) compiles well with installed XCode under Snow Leopard (all updates installed) and a GTS8800 GS (iMac 27"-Series). However, all SDK examples crashes when it comes to rtContextLaunch(). The weird thing: sample1 works fine (admittedly, I did only try ten examples or so).
I tryed an older release (Optix 2.1.1) ... all examples compile and run fine with the same configuration.

Monte Carlo Integration in optix sample

dfrontier — Sat, 10 Mar 2012 10:28:43 -0500

I need use Monte Carlo integration to optix sample "path_tracer". For example instead russian roulette which is there. Anybody know how start or make some simple monte carlo integration?

Thx.

add my own acceleration structure into Optix

xiaoxiao8612 — Fri, 06 Apr 2012 15:45:48 -0400

Does anyone know how to bring a new acceleration structure into Optix? I've developed a new structure based on kd-tree and I want to test the ray tracing performance with Opitx. But I really don't know where to start. It seems that I cannot access or change any code related to the acceleration object except using the existing builder and traverser.

Appreciate for any suggestion or discussion.

OptiX 2.5 not thread safe?

grangerfxgmailcom — Wed, 11 Apr 2012 16:30:49 -0400

I am attempting to run the OptiX traversal project from another thread rather than the main thread. When the function rtuTraversalPreprocess( t ) is called, I get the following exception:

First-chance exception at 0x000007fefdb3cacd in traversal.exe: Microsoft C++ exception: optix::GLContextCreateError at memory location 0x0339de60..

Repeated runs caused the graphics card driver to crash and recover on one occasion and to hard lock up the system on another. Most of the time, the program continued to run and successfully cast the rays, gathered the results and exited normally.

I suspect that this is a simple bug rather than some serious limitation. I hope that's true anyway because doing processing in the main thread is a bad design in an interactive program. I also tried disabling that function since the documentation said it is not needed and called automatically but I got the same exception later during the warm up rays. If there is something that this function does which must be called from the main thread (such as creating a GL context), there should be some way to do this from the main thread as part of initialization which only happens once early in the program.

BTW, we need an OptiX tag for posting these messages.

I attached two versions of the source. One is a .cpp file in which I was testing C++ compilation. The other is a .c file which is closer to the original sample code.

-Mark Granger

Array of rtTextureSampler

maxa — Thu, 12 Apr 2012 05:48:34 -0400

Array of rtTextureSampler in optix::Material is necessary to me.
It is possible?
Like - rtBuffer> Images;

OptiX v2.5.0 still not coping with many models well

Crog — Wed, 28 Mar 2012 11:04:51 -0400

Hi,

I have still got an issue with the v2.5.0 release in which thousands of models still produce a slow down. These are all static models so there is no rebuilding whatsoever and the strange behaviour is that if you move the camera away from the scene and look at nothing the render time never exceeds ~3Hz with 3,000 models that aren't even intersected.

I had observed this before but it appears there is a per instance overhead that is separate from the rendering times. Our ray-timing display shows a green image indicating that rays are taking no time whatsoever which shows the BVH is working in culling out any random/pointless intersections. The cause of this slowness within OptiX is a bit of an issue that is possibly in the validation code prior to the actual kernel launch. Or there is some data always transferred each launch that shouldn't be.

This issue can be easily reproduced using the Manta-Ray example and create a few thousand instances of the sphere and point the camera away from them.

Craig

Use Optix sampler without Cmake

fumble — Thu, 05 Apr 2012 06:05:10 -0400

hello,
i'm new in Optix world and i want to try to make new project in VS 2008.

My question: how can I use and modify the project "tutorial" without compiling all the examples included with SDK?

I try to create new project in VS2008 with "tutorial" files but don't work :/

Thanx for your help!

Proper way to render an animated scene with OptiX

mcleary — Tue, 10 Apr 2012 17:18:41 -0400

I'm writing a simple program which aims render of animated models using skinning algorithms. My question is, how is the proper way to implement this kind of algorithm in OptiX for triangle meshes?

My first guess is to transform each vertex of a triangle inside the intersection program in the current time and check the intersection with the "deformed" triangle. The main advantage is that I'll only need to map the "bone matrices" buffer between frames.

My second guess is to map the verticers buffer in CPU and do the math. When the intersection program runs again, all triangles will be "deformed". The main disadvantage is that the skinning math wil be handle by CPU.

I think my first guess would be the best way, but there is a more efficient way?

Is Optix 2.5 optimized for GTX680?

jogshy — Tue, 10 Apr 2012 13:50:00 -0400

I'm asking this because AFAIK CUDA4.1 is not prepared yet for the 680.
thx

Optix.NET: a .Net wrapper for OptiX

kch86 — Fri, 27 Jan 2012 19:06:31 -0500

I'm creating a .Net wrapper for optix and plan on releasing as an open source project under the MIT license. To that end I was wondering if it is possible to use some of the Optix SDK cuda shaders in sample programs. Any shaders I'd be using would of course retain their Nvidia license notice.

Thanks for any info :)

Edit: I sent this to optix-help@nvidia as well.
Edit 2: Heard back from David McAllister and got the go ahead.
Edit 3: changed the title to be a general thread for Optix.NET.

Compile C CUDA and Optix kernels in the same Visual Studio project

gemayelelias — Tue, 03 Apr 2012 11:25:47 -0400

Hello,

I am currently working on an Optix project where I need to render some data. However, I want to make this data change from frame to frame. I already have the functions that change the data written in C CUDA. I would like to integrate the two modules into the same project on visual studio.

p.s: to compile my Optix program I used the SDK's CMake files and modified them to include my project as part of the SDK's examples.

Any insight on how to make my project compile the .cu C CUDA files and the .cu Optix kernels together?

Thank you,

SM20 PTX: Unknown error, assertion #1049818. OptiX 2.5 only

Tofic — Tue, 20 Mar 2012 19:20:34 -0400

Hello,

It's started with OptiX 2.5, was working fine with OptiX 2.1.1 and CUDA 4.1. When I use any SM < 20, OptiX works fine. If I switch to SM>=20, I get this error:

OptiX Error: Unknown error (Details: Function "_rtContextCompile" caught excepti
on: Assertion failed: [1049818])

I use the same CUDA 4.1 compiler.

Anton

Bug report : Memory leak in Optix

Alex — Tue, 27 Mar 2012 20:41:39 -0400

Please find attached a source file that can show the memory leak.

"No binary for GPU" in Ptx assembly

bpeers — Mon, 26 Mar 2012 10:38:45 -0400

Hi, after upgrading a GTX480 to a GTX680 I get "No binary for GPU" from Optix 2.5, using 301.10 drivers (win7-64). Could it be that the internal cache holds on to compiled shaders that are 480 specific? How would I clear the optix cache? Recompiling source ptx shaders didn't help...

Thanks in advance!

Launch failed (Details: Function "_rtContextCompile" caught exception: : error: Encountered a CUDA error: createFromMemoryPtx returned (209): No binary for GPU, [7340218]
ptxas application ptx input, line 16; error : Module-scoped variables in .local state space are not allowed with ABI
ptxas fatal : Ptx assembly aborted due to errors [7340219])

How to start an Optix program?

milkiii — Wed, 28 Mar 2012 01:38:24 -0400

I'm now learning ray tracing and want to apply it to SAR simulation, find it impossible to find some tutorials or reference. For I've only used OGRE for a short time, do this need some CUDA skills? I feel so suffering as a newbie, hoping u can give me some advice, thank u very much!

Missing values in RTobjecttype enum?

cneumann — Tue, 27 Mar 2012 17:02:27 -0400

The enum RTobjecttype seems to miss the following entries:


RT_OBJECTTYPE_ACCELERATION
RT_OBJECTTYPE_GEOMETRY
RT_OBJECTTYPE_MATERIAL
RT_OBJECTTYPE_PROGRAM
RT_OBJECTTYPE_VARIABLE

Is this simply an oversight or is there a connection with rtVariable{Set,Get}Object() possibly not working with these types [1]? After all there is no defined return value for rtVariableGetType() when a variable is set to an object of these types - well there is RT_OBJECTTYPE_OBJECT, but that's not very helpful ;)

[1] I'm not saying it's not working (I haven't tried), just wondering ;)

Advanced materials possible in SceniX, OptiX?

jules123 — Mon, 26 Mar 2012 14:27:40 -0400

Hi,

I would like to implement some more advanced materials for the OptiX renderer in SceniX. I'm using the NVSGMaterial structure in my custom RTFx code for raytracing output. Is there a way to extend the structure to include custom properties? Currently from 'Using RTFx.pdf':

struct Material
{
float4 ambientColor;
float4 diffuseColor;
float4 emissiveColor;
float4 opacityColor;
float4 reflectivityColor;
float4 specularColor;
float reflectivity;
float specularExponent;
float ior;
float opacity;
};

I'd lack to add more properties. For example noise level, roughness, fog colour etc. Then I would handle each of them in my hit program.

To reiterate, can the SceniX scene graph StateSet handle custom material structures that can be used by OptiX program domains?

Thanks for any help,
Jules

OptiX v2.5.0 Assertion 2425493

Crog — Mon, 26 Mar 2012 05:42:14 -0400

Hi,

I have just got an assertion from optiX. I am not sure exactly what it is in relation yet. It is being caused from a section of code in which the order of things may have changed or something might be missing from a refactor. The number is below :

Unknown error (Details: Function "_rtContextLaunch2D" caught exception: Assertion failed: [2425493])

Cheers,
Craig

FXAA in SceniX?

jules123 — Fri, 23 Mar 2012 02:20:45 -0400

Hi,

For performance reasons, I wish to implement FXAA anti-aliasing in a CgFX shader to run in SceniX as a final pass.

I can get the outputBuffer with:

RTBufferSharedPtr buffer = renderTargetRT->getBuffer( "outputBuffer", renderTargetRT->getStereoTarget() );

RTBufferGLTextureWriteLock bufferLock( sharedPtr_cast(buffer) );
bufferLock->updateTexture();

Now I assume I've got to get the texture in the RTBuffer to become a COLOR_ATTACHMENT to a RendererGLFSQ, RendererGLFSQ->setCgFX to the FXAA shader, then set the texture to a sampler in the CgFX code.

Not sure if this is the correct approach? Not sure how to get the texture to become a COLOR_ATTACHMENT? And is this the fastest way?

Thanks for any help,
Jules

v2.5.0 LBvh not working properly multi GPU!?

Crog — Fri, 23 Mar 2012 11:59:06 -0400

Hi,

The new LBVh doesn't seem to work 'perfectly' multi gpu. I cannot see this mentioned in the documentation!? It runs amazingly fast but only one of the GPUs actually seems to get the updated acceleration structure. By the release notes it is obviously still early days but hopefully this can be rectified or mentioned in the docs.

Nvidia are probably working this but from what I can tell (from the bars in the render in which the acceleration is updated) that it would appear only one of the GPUs is ever doing LBvh calculations and the other is possibly idle. I would think using the second GPU for separate other models and splitting the load for generating the separate structures would make sense. That is unless the test case we have isn't big enough to trigger the need for two GPUs or the other GPU is actually doing some clever async processing of another kind on the side...

Weird slowdown with OptiX rtBuffer in SceniX

jules123 — Tue, 20 Mar 2012 05:22:04 -0400

Hi,

I want an array to hold data that is shared between my RayGenerationProgram and ClosestHitProgram

I use the code to set-up a buffer and to put it into the rtContext:

RTBufferSharedPtr rtBuffer;
nvrt::SmartRenderTargetRTBuiltin sRT = _OptixRenderer->getRenderTargetRT();
RTContextSharedPtr rtContext = sRT->getRTContext();
{
RTContextWriteLock rtctx(rtContext);
rtBuffer = rtctx->newBuffer(RT_BUFFER_INPUT_OUTPUT);
{
RTBufferWriteLock buf(rtBuffer);
buf->setFormat(RT_FORMAT_USER);
buf->setElementSize(sizeof(MyPointLight));
buf->setSize(16);
sRT->setVariable("MyBuffer", rtBuffer, false);
}
}

The trouble is the OptiX renderer slows to a crawl. I tested the RTFx code by removing everything to just

#include
#include

RT_PROGRAM void test2()
{
}

technique
(
entrypoint 0
{
RayGenerationProgram = test2;
}
)

Took all of 280 milliseconds to execute the RTFx code when I assumed it would take 0 millisecs. Without creating the rtBuffer everything returns back to normal speed.

I assume I'm doing something fundamentally wrong, like setting up the rtBuffer wrong? What could it be?

Thanks for any help,
Jules