Jump to content
  • Advertisement

michaelruecker

Member
  • Content Count

    33
  • Joined

  • Last visited

Everything posted by michaelruecker

  1. First question:   My particle systems currently supports a few blend modes. For example: Additive, multiplicative or alpha blending.   Everything works fine so far, yet I am struggling with the actual understanding.   Lets take additive blending as an example and assume we are drawing two particles. Either Particle 1 (P1) or Particle 2 (P2) is drawn first. Which leads to one of the images shown below:     On the right side is how it should look. The left side is how it shouldn't. The problem here is even though I use additive blending, which is for itself order independent, the output merger does some depth testing magic which results in the image shown on the left. The area T overlaps P2 and all fragments of P2 being removed (or never drawn).    My question is why or how the output merger exactly performs that depth test. And how that error on the left side of the image can occur. I mean it is clear that it is the drawing order that causes the problem but how the rasterizer does the depth test per fragment is a mystery.    I also know that if I disable depth write for my particles everything works fine since no fragments getting rejected. ----   Second question:   So far I have disabled depth writing for all blend modes but alpha blending. Since they are order independent blend modes. But now I am wondering if I even need depth writing for alpha blending. In fact I wonder if I ever need depth writing for any blend mode? Maybe it has some performance advantages?
  2. michaelruecker

    particles and depth testing

      Oh nice! Thank you that is what I was waiting for.
  3. michaelruecker

    particles and depth testing

    Thank you for your replies.     I am always drawing all particles per emitter in a single draw call. I never heard about using bounding boxes for particle sorting, but it sounds interesting.   Anyways, what I meant with performance is that, if I have the particles sorted anyways for an order dependent blending like alpha blending,I might as well activate depth writing. This would not produce any artifacts since the particles are drawn in correct order. Yet I still wonder if it has any advantage? Like maybe some fragments getting rejected in step 1 and don't have to be blended in step 2?
  4. michaelruecker

    Compute Shader: ThreadGroupSize and Performance

    Thank you for your answer!     This is where I got that quote: http://cvg.ethz.ch/teaching/2011spring/gpgpu/GPU-Optimization.pdf (Page 15). There isn't much more info though.   With my current implementation for my particle system, I have noticed that my performance drops, if I increase my thread group size from 64 to 128 on my nvidia card (had 1 million particles active -> 1 million threads). And I am not using shared memory. All I do is consume() a particle from one buffer, process the particle and append() it to the other buffer. These should be atomic operations. So there must be another reason why it might be bad to have bigger thread group sizes...   Also, I would like to write a few words about why it is critical to use the correct amounts of threads per thread group for my bachelor thesis. For that I need some reason why it might be bad to have too many threads per thread group. So any theoretical reason would help. (I could not find anything on the web so far)   While we are at it... there is this GTC presentation: http://www.nvidia.com/content/GTC/documents/1015_GTC09.pdf. On page 44 it says something about thread group size heristics.     I guess this is only true if you actually have enough work to do. So if you only need one thread group at a size of 512 you might want to lower it to 64 or even 32 and dispatch more groups. But it is not advised to start a few more thread groups if you only need 32 threads and your group size is already 32 just to have the other multiprocessor occupied. If i am correct? (Just asking because you have to be super precise when writing papers...)     Why is that? So that the multiprocessor can already load data for the next thread group to be processed?      This means having more warps in the queue and ready to be executed while the processor is still doing work on other warps, so that in case of latency it can switch out the warps and work on the warps in the queue instead?
  5. Hi Guys,   I have a few questions regarding ThreadGroupSize and performance.   1. No matter how many threads are in one thread group, it will always be executed by one SIMD / SMX (split in wavefronts / warps)? So lets say if I only need 1024 threads to process something and I start this in one group only, I have wasted performance since I could split it in smaller parts and have multiple SIMDs / SMXs working on it?   2. In case the above assumption is correct: If I dispatch only one thread group, will the other SIMD / SMX are blocked? Or do they work on other stuff like pixel processing, vector operations etc.? In other words... do all of them have to work on the same stuff or will the mix different things to keep them occupied?   3. Someone was writing this:      The first Statement makes perfectly sense. But the second... well I don't get it. Local memory is the main memory on the graphics card I guess not the thread shared memory? And could anyone explain what exactly happens that these races occur?   Thanks already!  
  6. RT @hearthstone_de: Folgt uns & RT diesen Tweet vor 16:00 Uhr am 15. November! Wir werden 50 zufällige Gewinner von Betazugängen auswählen.…
  7. I have particles which are processed to a billboard in the geometry shader.   These particles emit from a source point. This source point is attached to a placement. One can position, rotate and scale this placement.     What I would like to have is, that if I rotate the placement (which rotates the emitter) the emission direction is also rotated. So i.e. if my particles follow the y-axsis and I rotate it by 90° it would then follow the x-axis. Sounds more complicated than it is. Check these images to imagine it:     Basicly I am getting the WVP matrix from that placement(emitter). So all I need to do is project my particle with that matrix.   Here is what I tried:   float3 g_positions[4] =     {         float3( fHalfBillboardSizeX, fHalfBillboardSizeY, 0 ),         float3( -fHalfBillboardSizeX, fHalfBillboardSizeY, 0 ),         float3( fHalfBillboardSizeX, -fHalfBillboardSizeY, 0 ),         float3( -fHalfBillboardSizeX, -fHalfBillboardSizeY, 0 ),     }; // for all 4 billboard edge points (g_positions[4] ) do: float3 position =  g_positions[i]; float3 vInputPos = input[0].position; // that is the center point of the billboard. the actual particle position. output.position = float4(vInputPos, 1.0); output.position = mul(output.position, World); output.position = mul(float4(position, 1.0f), (float4x4)ViewInv ) + output.position; output.position = mul(output.position, View); output.position = mul(output.position, Projection); This does nearly work. It's just that my depth buffer seems to be broken with this. I see my particles through other objects for example. Any ideas why that is and how I could fix it?  
  8. michaelruecker

    Billboard WVP + rotation problem

    A friend of mine found the solution.   Apparently I corrupted the w value of the vector.   And the second problem is actually no problem.    Anyways thank you!
  9. michaelruecker

    Billboard WVP + rotation problem

    To visualize this I made a simple test scene and a video:     This was done with the settings mentioned above. As you can see there are actually two problems:   1. The depth buffer problem 2. The billboards aren't rotated accordingly. The arrow should point to the right when I rotate the placement.
  10. michaelruecker

    Random point on and in a conical frustum

    Thank you for your answer!   After taking your suggestion into consideration, I figured out that those are more operations that I am willing to sacrifice.    Actually it is not that bad if it is not evenly distributed. Even in Unity it isn't either. Guess they also figured, that it is too expensive.   Either way, I implemented a not evenly distributed version now which is sufficient for my needs.   Thank you in any way!
  11. Hey guys,   I am looking for two formulas to pick evenly distributed random points ON and IN a conical frustum. (I mean the first formula should find a random point ON the frustum and the second formula should give me random points IN that frustum.)     I had no success with Google so far. Does anyone have any useful links or knowledge about this?    
  12. I am currently stuck with a blending problem.   Basically I am having a lot of billboards each with an image and some black space around that image. All I want is to replace that black color with whatever is in the background so that it becomes invisible.   Original billboard image:   Background:   Expected result:   Yet I have tried everything I could imagine with my BlendStateDescription, DepthStencil and BlendFactor. Nothing bringt me to my expected result.   Without blending I get the following result: (As expected)     With : blendStateDescription.RenderTarget[0].SrcBlend = D3D11_BLEND_ONE; blendStateDescription.RenderTarget[0].DestBlend = D3D11_BLEND_ONE; blendStateDescription.RenderTarget[0].BlendOp = D3D11_BLEND_OP_ADD;   Or:   blendStateDescription.RenderTarget[0].SrcBlend = D3D11_BLEND_SRC_ALPHA; blendStateDescription.RenderTarget[0].DestBlend = D3D11_BLEND_ONE; blendStateDescription.RenderTarget[0].BlendOp = D3D11_BLEND_OP_ADD;   I get:     As stated earlier I have tried a hundred other combinations. Nothing successful yet.    Also I thought when I set my BlendFactor to {0.0f, 0.0f, 0.0f, 0.0f) it would handle the black color as my alpha channel. But it aparently does not. Actually nothing I do with SrcBlendAlpha, DestBlendAlpha and BlendOpAlpha has ANY effect.  
  13. michaelruecker

    Blending black area invisible

    Thank you both!   Apparently I had no alpha channel in my image, that's why it never worked.   Since I am not using additive blending now, my billboards flipping from back to front and vice versa. (In some frames billboard 1 is in front of 2 and then in the next frame it is the other way around)   I guess I have to sort the billboards by depth now am I correct about this or is there another way?
  14. I wonder if it is possible to update only a part of a buffer?   So lets say I need a huge buffer. But most of the time I will only use the first few bytes of that buffer. Only randomly I might need more.   Currently I am just having two buffers one is small and one big. The big buffer only gets updated if there is actually that much data needed.    But now I thought why one has to make it so complicated. Shouldn't there be a way to update only a part of a big buffer? (Of course it should not be slower than having two buffers) If so, any advice where i should look?     And if there is no way to update only a part of a structured buffer I have another question:   Can you have a pointer to a buffer inside HLSL? And if so how would be the syntax?   Like if I want to do something like this: StructuredBuffer<Structure> Small : register(t0); StructuredBuffer<Structure> Big: register(t1); Structure * MyBuffer; If (whatever) {   MyBuffer = &Small; } else {   MyBuffer = &Big; } RunSomeAlgorithmWithBuffer(MyBuffer);
  15. michaelruecker

    Update only a part of a buffer?

    Thank you so far. I have just tested it with:   D3D11_BOX destRegion; destRegion.left = 0; destRegion.right = iArrayEntries* sizeof(PARTICLE_SPAWN_DISTRIBUTION); destRegion.top = 0; destRegion.bottom = 1; destRegion.front = 0; destRegion.back = 1; m_pRenderAPI->m_pdevicecontext->UpdateSubresource(m_pbParticleDistribution, 0, &destRegion, aParticleDistribution, 0, 0); Seems to work. Yet no idea about the performance though. I just hope that it does what it says and only copies those few bytes over to the GPU.
  16. michaelruecker

    Update only a part of a buffer?

    Updating i.e. 16 Bytes of a huge buffer with this method would be the same as updating a 16 byte buffer completely? (If you have any experiences here)
  17. michaelruecker

    Many dispatch calls vs. higher ThreadGroupCount

    Thank you guys!     This is kind of what I am doing now.   I have a very small buffer and one overflow buffer. 99% of the time the small one will be used. Only if the system lags or if I have more data to process, the overflow buffer will be used as well. And in some very extrem situation or when doing a prewarm I run the overflow buffer in a loop.
  18. I wonder if it is performance wise much worse to do more dispatch calls instead of a higher ThreadGroupCount.     To explain this in a stupid example: ID3D11DeviceContext::Dispatch(10,1,1); vs for(int i = 0; i < 10; i++)   ID3D11DeviceContext::Dispatch(1,1,1); Has anyone tested this already or has some theoretical knowledge? 
  19. michaelruecker

    Many dispatch calls vs. higher ThreadGroupCount

    Ah thank you for that tip with GPUView. Before I try getting this to work maybe ATEFred knows the answer about the question if they run in parallel or series.   Because actually I would need them to run in series. I am doing something like this: for(int i = 0; i < x; i++) { // 1. update a small constant buffer which holds data for the upcoming particle spawns processed in the dispatch call   // 2. Dispatch(1,1,1); } So if this constant buffer gets overwritten by the second loop while the first dispatch is not done it would access wrong data. :/       I could also do it like this: // 1. bind 250kb data to hold the full spawn over time graph processed in this dispatch call // 2. Dispatch(x,1,1); The thing is, that the x is variable. Its random based on time passed. It can be 1 or it can be 30000. But mostly, like 99% of the time it will be 1.   Thats why I thought it might be better to just pass the data actually needed in small junks and process only that bit. And if it happens that I actually need more than 1 pass, I would just use a loop.   Maybe there is even another solution i haven't noticed yet.   If it was only possible to allocate memory dynamically... but no I have to allocate as much memory as is needed in the worst case.
  20. michaelruecker

    Many dispatch calls vs. higher ThreadGroupCount

    Thank you for the reply.   Yep I know about the thread group size. But for example if your shader is configured so that you have [numthreads(64, 1, 1)] as thread group size you could dispatch(1,1,1) (one) of that group and still have a full 64 threads running. As far as I know you should always run a multiple of 64.   So, if we take my first example again: ID3D11DeviceContext::Dispatch(10,1,1); // dispatches 10 x 64 threads -> 640 for(int i = 0; i < 10; i++)   ID3D11DeviceContext::Dispatch(1,1,1); // dispatches 10 x 64 threads -> 640 Lets say I am forced to use the second "solution" for some complicated reason. I wonder how much worse it will be. Or what other disadvantage I would have.
  21. Could anyone explain the differences between those two things please?   - DepthStencilView (http://msdn.microsoft.com/en-us/library/windows/desktop/ff476507(v=vs.85).aspx)  - DepthStencilState (http://msdn.microsoft.com/en-us/library/windows/desktop/ff476506(v=vs.85).aspx)  
  22. Ahh that cleared things up. Thank you very much!
  23. Alright, so lets say the engine I am working with, is not using a depth stencil state yet. But I need to set that state now for my stuff. Afterwards I would like to set it to what it was before I changed it.   To do this I would just use OMSetDepthStencilState(NULL, 0) and set the state to NULL again? Or do I actually have to create another state that with default parameters and set that active?   Also what do you exactly mean with:  
  24. So if one would not specify the state at all and just setting the view, depth testing would still work right?   Does it load some default settings then?
  25. Is there a way to to get the active element count of a AppendStructuredBuffer or ConsumeStructuredBuffer in HLSL?   I would like to access the element count on the shader side only.   (I guess I could use another buffer and fill it via CopyStructureCount. But was wondering if there is a direct method in HLSL)   All I found was a function called: GetDimensions (http://msdn.microsoft.com/en-us/library/windows/desktop/ff471449(v=vs.85).aspx).   But I think that function returns the maximal number of elements not the current element count.
  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

GameDev.net is your game development community. Create an account for your GameDev Portfolio and participate in the largest developer community in the games industry.

Sign me up!