• Advertisement
Sign in to follow this  

DX11 [Compute Shader] Groupshared memory as slow as VRAM

This topic is 1956 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

As far as I know, I have one of the earlier and cheaper mobile graphics cards that supported DX11. It's a AMD Radeon HD 5730M.

I've started optimizing graphics algorithms by porting them to compute shader and improving them by sharing memory and synchronizing the threads. This way I could improve the runtime of my Bloom from [eqn]O(n)[/eqn] to [eqn]O(log(n))[/eqn] per pixel.

But that was only the theoretical runtime. In reality, the algorithm performed so much worse than the original linear algorithm. I'm pretty sure I know the reason. Instead of let's say 32 read operations and 1 write operation, the algorithm now needs 1 read operation from VRAM, 5 read operations from groupshared memory, 5 write operations to groupshared memory and 1 write operation to VRAM.

Overall groupshared memory being L1 Cache should be way faster than 32 read operations from VRAM and it's even way less operations because of the algorithm having logarithmic runtime, but it's way slower (8ms instead of 0.5ms). The slowdown could be because of memory bank conflicts. But could they really cause such an enormous slowdown?

To me it looks like my graphics card might not even have an actual L1 cache residing on the Wavefront as groupshared memory at all. It performs just as bad as a UAV residing in VRAM would. So maybe they simply wrote a driver that uses 32kb of reserved memory in the VRAM as groupshared memory. Could that be the case or is it the bank conflicts?

I wish there were tools that could shine more light on such problems. Graphics cards and the tools should be more transparent in what's actually going on, so that the developers could improve the algorithms even further.

Update: After reading through NVidias CUDA documentation my shaders don't even cause any bank conflicts at all. Each half warp (16 threads) always accesses 16 different memory banks. Just a whole block (1024 threads) accesses them multiple times, which is normal and has nothing to do with bank conflicts. Edited by CryZe

Share this post


Link to post
Share on other sites
Advertisement
I suppose it's possible that your hardware doesn't actually have on-chip shared memory and just uses global memory instead, but I've not heard of that ever being the case. Although mobile hardware isn't usually well-documented, so who knows. You could try using GPU PerfStudio or AMD's APP profiling suite, but I'm not sure if either those will give you enough information to narrow down the problem. Perhaps you might want to try running some samples that make use of shared memory to see if they also perform poorly on your hardware.

Also just so you know, shared memory isn't L1. On AMD and Nvidia hardware It's its own special type of on-chip memory, and it's separate from the caches.

Share this post


Link to post
Share on other sites
[quote name='MJP' timestamp='1347309681' post='4978690']
Also just so you know, shared memory isn't L1. On AMD and Nvidia hardware It's its own special type of on-chip memory, and it's separate from the caches.
[/quote]
"As mentioned in Section F.4.1, for devices of compute capability 2.x and higher, the same on-chip memory is used for both L1 and shared memory, and how much of it is dedicated to L1 versus shared memory is configurable for each kernel call." Source: [url="http://developer.download.nvidia.com/compute/DevZone/docs/html/C/doc/CUDA_C_Programming_Guide.pdf"]CUDA Programming Guide[/url]

I've tried both the PerfStudio and AMD's APP Profiler. But they didn't work at all. PerfStudio wasn't able to catch a frame (endlessly trying to connect, even though it was already connected) and the APP Profiler showed me an error message in both of its modes. I'll probably try it again tomorrow.

[quote name='MJP' timestamp='1347309681' post='4978690']
Perhaps you might want to try running some samples that make use of shared memory to see if they also perform poorly on your hardware.
[/quote]
Oh, that's a good idea. I remember that the OIT11 Sample from the DirectX Sample Browser performs incredibly bad on my hardware (9FPS at 320x240). I don't know if it performs bad in comparison to the other samples on other hardware as well, though. I'll take a look into it's source to check out why it might perform that bad.

I'll also try to implement a bandwidth heavy compute shader that either performs an enormous amount of write operations to shared memory or to shared memory while causing as many bank conflicts as possible or to global memory. If the performance is the same the chances that my graphics card uses on chip memory as shared memory are pretty much zero. Edited by CryZe

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
  • Advertisement
  • Popular Now

  • Advertisement
  • Similar Content

    • By AxeGuywithanAxe
      I wanted to see how others are currently handling descriptor heap updates and management.
      I've read a few articles and there tends to be three major strategies :
      1 ) You split up descriptor heaps per shader stage ( i.e one for vertex shader , pixel , hull, etc)
      2) You have one descriptor heap for an entire pipeline
      3) You split up descriptor heaps for update each update frequency (i.e EResourceSet_PerInstance , EResourceSet_PerPass , EResourceSet_PerMaterial, etc)
      The benefits of the first two approaches is that it makes it easier to port current code, and descriptor / resource descriptor management and updating tends to be easier to manage, but it seems to be not as efficient.
      The benefits of the third approach seems to be that it's the most efficient because you only manage and update objects when they change.
    • By evelyn4you
      hi,
      until now i use typical vertexshader approach for skinning with a Constantbuffer containing the transform matrix for the bones and an the vertexbuffer containing bone index and bone weight.
      Now i have implemented realtime environment  probe cubemaping so i have to render my scene from many point of views and the time for skinning takes too long because it is recalculated for every side of the cubemap.
      For Info i am working on Win7 an therefore use one Shadermodel 5.0 not 5.x that have more options, or is there a way to use 5.x in Win 7
      My Graphic Card is Directx 12 compatible NVidia GTX 960
      the member turanszkij has posted a good for me understandable compute shader. ( for Info: in his engine he uses an optimized version of it )
      https://turanszkij.wordpress.com/2017/09/09/skinning-in-compute-shader/
      Now my questions
       is it possible to feed the compute shader with my orignial vertexbuffer or do i have to copy it in several ByteAdressBuffers as implemented in the following code ?
        the same question is about the constant buffer of the matrixes
       my more urgent question is how do i feed my normal pipeline with the result of the compute Shader which are 2 RWByteAddressBuffers that contain position an normal
      for example i could use 2 vertexbuffer bindings
      1 containing only the uv coordinates
      2.containing position and normal
      How do i copy from the RWByteAddressBuffers to the vertexbuffer ?
       
      (Code from turanszkij )
      Here is my shader implementation for skinning a mesh in a compute shader:
      1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 struct Bone { float4x4 pose; }; StructuredBuffer<Bone> boneBuffer;   ByteAddressBuffer vertexBuffer_POS; // T-Pose pos ByteAddressBuffer vertexBuffer_NOR; // T-Pose normal ByteAddressBuffer vertexBuffer_WEI; // bone weights ByteAddressBuffer vertexBuffer_BON; // bone indices   RWByteAddressBuffer streamoutBuffer_POS; // skinned pos RWByteAddressBuffer streamoutBuffer_NOR; // skinned normal RWByteAddressBuffer streamoutBuffer_PRE; // previous frame skinned pos   inline void Skinning(inout float4 pos, inout float4 nor, in float4 inBon, in float4 inWei) {  float4 p = 0, pp = 0;  float3 n = 0;  float4x4 m;  float3x3 m3;  float weisum = 0;   // force loop to reduce register pressure  // though this way we can not interleave TEX - ALU operations  [loop]  for (uint i = 0; ((i &lt; 4) &amp;&amp; (weisum&lt;1.0f)); ++i)  {  m = boneBuffer[(uint)inBon].pose;  m3 = (float3x3)m;   p += mul(float4(pos.xyz, 1), m)*inWei;  n += mul(nor.xyz, m3)*inWei;   weisum += inWei;  }   bool w = any(inWei);  pos.xyz = w ? p.xyz : pos.xyz;  nor.xyz = w ? n : nor.xyz; }   [numthreads(1024, 1, 1)] void main( uint3 DTid : SV_DispatchThreadID ) {  const uint fetchAddress = DTid.x * 16; // stride is 16 bytes for each vertex buffer now...   uint4 pos_u = vertexBuffer_POS.Load4(fetchAddress);  uint4 nor_u = vertexBuffer_NOR.Load4(fetchAddress);  uint4 wei_u = vertexBuffer_WEI.Load4(fetchAddress);  uint4 bon_u = vertexBuffer_BON.Load4(fetchAddress);   float4 pos = asfloat(pos_u);  float4 nor = asfloat(nor_u);  float4 wei = asfloat(wei_u);  float4 bon = asfloat(bon_u);   Skinning(pos, nor, bon, wei);   pos_u = asuint(pos);  nor_u = asuint(nor);   // copy prev frame current pos to current frame prev pos streamoutBuffer_PRE.Store4(fetchAddress, streamoutBuffer_POS.Load4(fetchAddress)); // write out skinned props:  streamoutBuffer_POS.Store4(fetchAddress, pos_u);  streamoutBuffer_NOR.Store4(fetchAddress, nor_u); }  
    • By mister345
      Hi, can someone please explain why this is giving an assertion EyePosition!=0 exception?
       
      _lightBufferVS->viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&_lightBufferVS->position), XMLoadFloat3(&_lookAt), XMLoadFloat3(&up));
      It looks like DirectX doesnt want the 2nd parameter to be a zero vector in the assertion, but I passed in a zero vector with this exact same code in another program and it ran just fine. (Here is the version of the code that worked - note XMLoadFloat3(&m_lookAt) parameter value is (0,0,0) at runtime - I debugged it - but it throws no exceptions.
          m_viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&m_position), XMLoadFloat3(&m_lookAt), XMLoadFloat3(&up)); Here is the repo for the broken code (See LightClass) https://github.com/mister51213/DirectX11Engine/blob/master/DirectX11Engine/LightClass.cpp
      and here is the repo with the alternative version of the code that is working with a value of (0,0,0) for the second parameter.
      https://github.com/mister51213/DX11Port_SoftShadows/blob/master/Engine/lightclass.cpp
    • By mister345
      Hi, can somebody please tell me in clear simple steps how to debug and step through an hlsl shader file?
      I already did Debug > Start Graphics Debugging > then captured some frames from Visual Studio and
      double clicked on the frame to open it, but no idea where to go from there.
       
      I've been searching for hours and there's no information on this, not even on the Microsoft Website!
      They say "open the  Graphics Pixel History window" but there is no such window!
      Then they say, in the "Pipeline Stages choose Start Debugging"  but the Start Debugging option is nowhere to be found in the whole interface.
      Also, how do I even open the hlsl file that I want to set a break point in from inside the Graphics Debugger?
       
      All I want to do is set a break point in a specific hlsl file, step thru it, and see the data, but this is so unbelievably complicated
      and Microsoft's instructions are horrible! Somebody please, please help.
       
       
       

    • By mister345
      I finally ported Rastertek's tutorial # 42 on soft shadows and blur shading. This tutorial has a ton of really useful effects and there's no working version anywhere online.
      Unfortunately it just draws a black screen. Not sure what's causing it. I'm guessing the camera or ortho matrix transforms are wrong, light directions, or maybe texture resources not being properly initialized.  I didnt change any of the variables though, only upgraded all types and functions DirectX3DVector3 to XMFLOAT3, and used DirectXTK for texture loading. If anyone is willing to take a look at what might be causing the black screen, maybe something pops out to you, let me know, thanks.
      https://github.com/mister51213/DX11Port_SoftShadows
       
      Also, for reference, here's tutorial #40 which has normal shadows but no blur, which I also ported, and it works perfectly.
      https://github.com/mister51213/DX11Port_ShadowMapping
       
  • Advertisement