I understand that in 2d, images are cached in x and y directions.
But in 1d, why would you want to use an image? Is the memory used
for images faster than memory used for buffers?
1D Image stays the image, so it has all advantages that Image has against Buffer. That is:
Image IO operations are usually well-cached.
Samplers can be used, which gives benefits like computationally cheap interpolation, hardware-resolved out-ouf-bound access, etc.
Though, you should remember that Image has some constraints in comparison to regular Buffer:
Single Image can be used either for reading or for writing within one kernel.
You can't use vloadN / vstoreN operations, which can handle up to 16 values per call. Your best option is read_imageX & write_imageX functions, which can load / store up to 4 values per one call. That can be serious issue on GPU, with vector architecture.
If you are not using 4-component format, usually, you are loosing part of performance as many functions process samples from color planes simultaneously. So, payload is decreasing.
If we talk about GPU, different parts of hardware are involved into processing of Images & Buffers, so it's difficult to draw up, how one is better than another. Carefull benchmarking & algorithm optimizations are needed.
Related
VkImageCreateInfo has the following member:
VkFormat format;
And VkImageViewCreateInfo has the same member.
What I don't understand why you would ever have a different format in the VkImageView from the VkImage needed to create it.
I understand some formats are compatible with one another, but I don't know why you would use one of the alternate formats
The canonical use case and primary original motivation (in D3D10, where this idea originated) is using a single image as either R8G8B8A8_UNORM or R8G8B8A8_SRGB -- either because it holds different content at different times, or because sometimes you want to operate in sRGB-space without linearization.
More generally, it's useful sometimes to have different "types" of content in an image object at different times -- this gives engines a limited form of memory aliasing, and was introduced to graphics APIs several years before full-featured memory aliasing was a thing.
Like a lot of Vulkan, the API is designed to expose what the hardware can do. Memory layout (image) and the interpretation of that memory as data (image view) are different concepts in the hardware, and so the API exposes that. The API exposes it simply because that's how the hardware works and Vulkan is designed to be a thin abstraction; just because the API can do it doesn't mean you need to use it ;)
As you say, in most cases it's not really that useful ...
I think there are some cases where it could be more efficient, for example getting a compute shader to generate integer data for some types of image processing can be more energy efficient than either float computation or manually normalizing integer data to create unorm data. Using aliasing you the compute shader can directly write e.g. uint8 integers and a fragment shader can read the same data as unorm8 data
in these days, i'm trying program on mobile gpu(adreno)
the algorithm what i use for image processing has 'randomness' for memory access.
it refers some pixels in 'fixed' range for filtering.
BUT, i cant know exactly which pixel will be referred(depends on image)
as far as i understood. if multiple thread access local memory bank
it causes bank conflict. so in my case it should make bank conflict.
MY question: Can i eliminate bank conflict at random memory access?
or can i reduce them?
Assuming that the distances of your randomly accessed pixels is somehow normal distributed, you could think of tiling your image into subimages.
What I mean: instead of working with a (lets say) 1024x1024 image, you might have 4x4 images of size 256x256. Each of them is kept together in memory, so "near" pixel access stays within the same image object. Only the far distance operations need to access different subimages.
A second option: instead of using CLImage objects, try to save your data into an array. The data in the array can be stored in a Z-order curve sorting. This also leads to a reduced spatially distribution (compared to row-order-sorting)
But of course, this depends strongly on your image size.
There are a variety of ways to deal with bank conflicts - the size of the elements you are working with, the strides between lines and shifting the coordinates around to different memory addresses. It's never going to be as good as non-random / conflict free though and so what you will notice is depending on the image - you will see significantly different compute times.
See http://cuda-programming.blogspot.com/2013/02/bank-conflicts-in-shared-memory-in-cuda.html
I am implementing a solution using OpenCL and I want to do the following thing, say for example you have a large array of data that you want to copy in the GPU once and have many kernels process batches of it and store the results in their specific output buffers.
The actual question is here which way is faster? En-queue each kernel with the portion of the array it needs to have or pass out the whole array before hand an let each kernel (in the same context) process the required batch, since they would have the same address space and could each map the array concurrently. Of course the said array is read-only but is not constant as it changes every time I execute the kernel(s)... (so I could cache it using a global memory buffer).
Also if the second way is actually faster could you point me with direction on how this could be implemented, as I haven't found anything concrete yet (although I am still searching :)).
Cheers.
I use the second memory normally. Sharing the memory is easy. Just pass the same buffer to each kernel. I do this in my real-time ray-tracer. I render with one kernel and post-process (image process) with another.
Using the C++ bindings it looks something like this
cl_input_mem = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(cl_uchar4)*npixels, NULL, &err);
kernel_render.setArg(0, cl_input_mem);
kernel_postprocess.setArg(0, cl_input_mem);
If you want one kernel to operate on a different segment of the array/memory you can pass an offset value to the kernel arguments and add that to e.g. the global memory pointer for each kernel.
I would use the first method if the array (actually the sum of each buffer - including output) does not fit in memory. Another reason to use the first method is if you're running on multiple devices. In my ray tracer I use the first method when I render on multiple devices. For example I have one GTX 580 render the upper half of the screen and the other GTX 580 rendering the lower half (actually I do this dynamically so one device may render 30% while the other 70% but that's besides the point). I have each device only render it's fraction of the output and then I assemble the output on the CPU. With PCI 3.0 the transfer back and forth between CPU and GPU (multiple times) has a negligible effect on the frame rate even for 1920x1080 images.
I have an OpenCL kernel that calculates total force on a particle exerted by other particles in the system, and then another one that integrates the particle position/velocity. I would like to parallelize these kernels across multiple GPUs, basically assigning some amount of particles to each GPU. However, I have to run this kernel multiple times, and the result from each GPU is used on every other. Let me explain that a little further:
Say you have particle 0 on GPU 0, and particle 1 on GPU 1. The force on particle 0 is changed, as is the force on particle 1, and then their positions and velocities are changed accordingly by the integrator. Then, these new positions need to be placed on each GPU (both GPUs need to know where both particle 0 and particle 1 are) and these new positions are used to calculate the forces on each particle in the next step, which is used by the integrator, whose results are used to calculate forces, etc, etc. Essentially, all the buffers need to contain the same information by the time the force calculations roll around.
So, the question is: What is the best way to synchronize buffers across GPUs, given that each GPU has a different buffer? They cannot have a single shared buffer if I want to keep parallelism, as per my last question (though, if there is a way to create a shared buffer and still keep multiple GPUs, I'm all for that). I suspect that copying the results each step will cause more slowdown than it's worth to parallelize the algorithm across GPUs.
I did find this thread, but the answer was not very definitive and applied only to a single buffer across all GPUs. I would like to know, specifically, for Nvidia GPUs (more specifically, the Tesla M2090).
EDIT: Actually, as per this thread on the Khronos forums, a representative from the OpenCL working group says that a single buffer on a shared context does indeed get spread across multiple GPUs, with each one making sure that it has the latest info in memory. However, I'm not seeing that behavior on Nvidia GPUs; when I use watch -n .5 nvidia-smi while my program is running in the background, I see one GPU's memory usage go up for a while, and then go down while another GPU's memory usage goes up. Is there anyone out there that can point me in the right direction with this? Maybe it's just their implementation?
It sounds like you are having implementation trouble.
There's a great presentation from SIGGRAPH that shows a few different ways to utilize multiple GPUs with shared memory. The slides are here.
I imagine that, in your current setup, you have a single context containing multiple devices with multiple command queues. This is probably the right way to go, for what you're doing.
Appendix A of the OpenCL 1.2 specification says that:
OpenCL memory objects, [...] are created using a context and can be shared across multiple command-queues created using the same context.
Further:
The application needs to implement appropriate synchronization across threads on the host processor to ensure that the changes to the state of a shared object [...] happen in the correct order [...] when multiple command-queues in multiple threads are making changes to the state of a shared object.
So it would seem to me that your kernel that calculates particle position and velocity needs to depend on your kernel that calculates the inter-particle forces. It sounds like you already know that.
To put things more in terms of your question:
What is the best way to synchronize buffers across GPUs, given that each GPU has a different buffer?
... I think the answer is "don't have the buffers be separate." Use the same cl_mem object between two devices by having that cl_mem object come from the same context.
As for where the data actually lives... as you pointed out, that's implementation-defined (at least as far as I can tell from the spec). You probably shouldn't worry about where the data is living, and just access the data from both command queues.
I realize this could create some serious performance concerns. Implementations will likely evolve and get better, so if you write your code according to the spec now, it'll probably run better in the future.
Another thing you could try in order to get a better (or a least different) buffer-sharing behavior would be to make the particle data a map.
If it's any help, our setup (a bunch of nodes with dual C2070s) seem to share buffers fairly optimally. Sometimes, the data is kept on only one device, other times it might have the data exist in both places.
All in all, I think the answer here is to do it in the best way the spec provides and hope for the best in terms of implementation.
I hope I was helpful,
Ryan
Suppose that I've two big functions. Is it better to write them in a separate kernels and call them sequentially, or is better to write only one kernel? (I don't want to read the data back and force form between host and device in between). What about the speed up if I want to call the kernel many times?
One thing to consider is the effect of register pressure on hardware utilization and performance.
As a general rule, big kernels have big register footprints. Typical OpenCL devices (ie. GPUs) have very finite register file sizes and large kernels can result in lower concurrency (fewer concurrent warps/wavefronts), less opportunities for latency hiding, and poorer overall performance. On the other hand, kernel launch overheads are pretty low on most platforms, so if your algorithm doesn't have an enormous amount of state to save between "phases" of execution, the penalty of using multiple kernels can be rather low.
Using multiple kernels also has another side benefit -- you get implicit synchronization between all work units for free. Often that can eliminate the need for atomic memory operations and synchronization primitives which can have a negative impact on code performance.
The ultimate guide should be measured performance. There is no universal rule-of-thumb for this sort of things. Benchmarking is the only way to know for sure.
In general this is a question of (maybe) slightly better performance vs. readibility of your code. Copying buffers is no issue as long as you keep them within the same context. E.g. you could set one output buffer of a kernel to be an input buffer of the next kernel, which would not involve any copying.
The proper way to code in OpenCL is to separate your code into parallel tasks, and each of them is a kernel. This is, each "for loop" should be a kernel. Some times one single CPU code function could result in a 4 kernel implementation in OCL.
If you need to store data between kernel executions just use OpenCL buffers and do not copy to host (this solves the DEVICE<->HOST bottleneck).
If both functions act to different data you could propably write a single kernel, but that depends on the complexity of the operation being run.