search

HPCE / hpce-2018-cw3

1 stars 5 forks source link

readme

Overal Goals

The deadline for this coursework is:

22:00 Mon 11 Feb

Submission of deliverables is via github only, with only a hash going into blackboard to prove when you submitted.

The overall goals of this coursework are to:

This coursework alone is not intended to make you a GPU expert. You should know how to create an OpenCL program from scratch by the end, but the performance you get may not be be as high as a TBB version.

This builds on your existing knowledge of iteration spaces, and your experience with tbb::parallel_for. There are also links to the dependencies you've been seeing when implementing things using tbb::task_group, but now there is much less safety.

Checking your OpenCL environment

Before writing any code, it is worth checking your OpenCL environment and SDK, just to check that you have the right devices and things installed. There is a file called src/test_opencl.cpp, which is a simple, but complete, OpenCL program. It doesn't do a lot, but it does allocate GPU buffers and try to execute a kernel, which is enough to tell whether things are up and running.

Compile the program:

make bin/test_opencl

and make sure you can both build and execute it. You may need to fiddle with your environment to make it build, and find/download an SDK. While an OpenCL run-time is installed in most systems, there is often no SDK installed.

SDKs and Run-times

As with TBB, you need two parts: the OpenCL SDK, which gives the headers and libraries you need to compile your programs; and the OpenCL run-time, which provides access to the underlying compute resource and does all the scheduling.

Cross-platform

In 2014 @farrell236 was nice enough to contribute the CMakeLists.txt, which is a way of getting cross-platform builds with CMake. I've kept it here, as exposure to CMake is a good thing (I've seen quite a few students (EIE2 2016 onwards) using CLion, which is based on CMake).

Linux

You should be able to get a working OpenCL driver via your GPU vendor - often that will come with an SDK as well. If you don't have a GPU, or it isn't supported, you can try using a software OpenCL implementation. Intel provide a multi-core version, and the AMD OpenCL run-time also contains a software version, though you'll need to be root to install it.

Many DoC machines have NVidia cards, with the OpenCL drivers installed.

If you want to use the Intel OpenCL provided under Ubuntu, then @TheoFranquet had some suggestions on how to get it setup.

Mac users

As far as I know all Macs come with OpenCL installed these days, but the SDK may not be installed. I would imagine you can bring the SDK over with whatever package managed OSX has, as Apple are very integrated into the whole OpenCL thing.

Apparently, adding the option:

-framework OpenCL

to your compilation may help when trying to sort out header and linker directories.

AWS

I've created an image for AWS which has OpenCL set up, both for software and the GPU. The title of the image is HPCE-2018, which can be selected when you launch an AWS instance. The instance is located in the "N. Virginia" region - if you can't find it, check whether AWS has defaulted you into a different region.

This is an ubuntu image, so the default login is ubuntu So you would login with ssh as:

ssh -A -i [your-key] admin@[server-ip-address]

I will be using a g3.4xlarge instance for assessment. It's probably a good idea for you to test your work on one of these prior to submission.

Debugging

Because the program you're writing works as part of a pipeline, you may find it difficult to launch in a debugger. You have a couple of options here:

The Heat World code

To make this somewhat realistic, you are working with an existing proprietary application. The overall structure is a classic example of a finite-difference or stencil operation, where you are applying some sort of local smoothing operation over huge numbers of time-steps. Such stencil calculations are very common, but different enough that one often can't use an off-the-shelf piece of software. Application areas where they arise include:

The core of the code is in the two files include/heat.hpp and src/heat.cpp, with the header giving the API for creating, storing, loading, and stepping a model of a heat system.

As well as the core src/heat.cpp, there are also a number of driver programs:

Each of these source files must be compiled together with heat.cpp to produce an executable. There are no other dependencies on external libraries.

Use your chosen development environment to compile the three files, resulting in bin/make_world, bin/step_world and bin/render_world (you can have them build in whatever directory you like). If you now do:

bin/make_world 10 0.1

You'll see it create a text description of a heat world (you may want to make your terminal larger, or look at it in a text file. At the top is the header, describing the dimensions of the world, and the propagation constant (0.1 in this case). The propagation constant measures how fast heat migrates between conductive cells in this world, so a constant of 0.01 would result in slower movement of heat.

Below the header you'll see a grid of 0, 1, and 2, defining the shape of the world. Any 2s represent insulator, where heat cannot flow, while any 1s represents fixed temperature cells, such as heat-sinks and heat-sources. All the 0s are "normal" conductive cells.

Below the world grid is the current temperature or state of every cell in the grid, with values between 0 and 1. After creating the world it is cold, except for a row of 1s near the top of the state which are all hot. If you compare these cells with the world grid, you'll see these 1s are fixed, so they will keep their value of 1 as time progresses. However, surrounding cells will change temperature in response to the cells around them.

Now advance the world by a single time-step of length 0.1 seconds:

bin/make_world 10 0.1 | bin/step_world 0.1 1

You'll see that the the cells next to the heat source row have warmed up. The warming is not uniform, as at the far left and far right the cells are up against insulators, so according to the simple model using here, they end up slightly warmer.

You can now advance the world by a few seconds:

bin/make_world 10 0.1 | bin/step_world 0.1 100

You should see that some of the heat is making it's way around the right hand side, but it can't come through the insulator across the middle-left. Step forward by 1000 seconds:

bin/make_world 10 0.1 | bin/step_world 0.1 10000

and you'll see the world has largely reached a steady-state, with high temperatures around the top, then slowly decreasing temperatures through to the single point heat-sink at the bottom.

In such problems we are usually interested in looking at very high spatial resolutions (i.e. lots of cells), and very fine temporal resolutions. This sort of discretisation is most accurate when both time and space are very fine. To visualise much finer-grain scenarios, it is convenient to view it as a bitmap:

bin/make_world 100 0.1 | bin/step_world 0.1 100 | bin/render_world dump.bmp

If you look at dump.bmp, you should see insulators rendered in green, and the temperature shown from blue (0) to red (1). However, over a total time of 0.1*100=10 seconds, very little will have happened. To see anything happen, you'll need to up the time significantly:

bin/make_world 100 0.1 | bin/step_world 0.1 100000 | bin/render_world dump.bmp

It is getting fairly slow, even at a resolution of 100x100, but in practise we'd like to increase resolution 10x in all three dimensions (both spatial axes + time), which increases the compute time by a factor of 1000.

This type of computation is ideal for a GPU, as it maps very well to the grid-structure provided by both the hardware and the programming model.

Preparing the code for OpenCL

Create a new folder within src called src/your_login. This is where your additional code will live (as before, in principle I should be able to compile everyone in the classes's code together at the end).

We're going to convert the original C++ code to OpenCL quite incrementally, keeping a trail of working versions as we go. You should keep track of this as labels within git, so that you can always get back to a working version, and work out what you changed that broke it (however, I don't require you to do this). If all you have is the original sequential and the highly optimised parallel GPU version, then it can be very difficult to see how the two are actually related.

Examining the source code

The first thing to do is to look at the source code, and to identify opportunities for parallelism. If you open src/heat.cpp, you'll see a number of functions, but the computationally intensive part is the function hpce::StepWorld. This is responsible for moving the world forwards by n steps which each increment time by dt.

Within StepWorld there is some initial setup which allocates buffers and calculates constants, followed three nested loops. These loops represent the bulk of the time in the algorithm, and it should be clear that the execution time complexity is O(n x w x h). So broadly speaking if we double any one of n, w, or h then the execution time will double, and if we double all of n, w, and h then the execution time will be eight times longer.

Recalling your experience with TBB, you should immediately see opportunities for parallelism here: all the iterations of the two inner loops are independent, and so can be parallelised. However, there is a loop carried dependency across the time loop, similar to the erode filter we looked at in lectures. However, unlike the erode example, where there were multiple frames which could be pipelined, here we have a single world to update, so pipelining will not help.

Extracting the inner loop to a lambda

First we want to isolate the part of the program that will be moved to the GPU. Because the GPU works in a different memory space, it is useful to emulate that hard boundary in software first, before starting to convert to kernels.

Writing the code

Take the file src/heat.cpp, and copy it to a new file called src/your_login/step_world_v1_lambda.cpp. The original file contains multiple support and IO functions which we aren't interested in (we'll still be linking with the original src/heat.cpp, so cut out all the functions which aren't StepWorld. Rename the remaining StepWorld to StepWorldV1Lambda, and put it in a namespace called your_login. So you should be left with a file which looks something like:

#include "heat.hpp"

... // More include files

namespace hpce{
namespace your_login{

void StepWorldV1Lambda(world_t &world, float dt, unsigned n)
{
    ... // Existing implementation
}

}; // namespace your_login
}; // namespace hpce

Our first step is to pull the two inner loops out into a lambda, starting to isolate the core loops. Pull the body of the inner-most loop into a seperate lambda called kernel_xy, which should take an x and y parameter, but capture everything else by reference (i.e. use the [&] lambda form) - this is an intermediate step, so we won't worry too much about shared dynamic state.

The resulting code should look something like:

std::vector<float> buffer(w*h);

auto kernel_xy = [&](unsigned x, unsigned y)
{
    ... // Original loop body
};

for(unsigned t=0;t<n;t++){
    for(unsigned y=0;y<h;y++){
        for(unsigned x=0;x<w;x++){
            kernel_xy(x,y);
        }  // end of for(x...
    } // end of for(y...

    ... // Rest of code

Note that the lambda has to be declared after buffer, as if you try to declare it before, then it is impossible to close over the variable - you can't refer to something that hasn't been named yet.

Building the test framework

To compile this, we need some sort of driver program to load the world, call our function, and save the world. Fortunately, we already have such a program in src/step_world.cpp. Take the main from that program (src/step_world.cpp) and paste it in below your existing code. Note that it should appear outside the namespaces, as main goes in the global namespace. Modify the call to hpce::StepWorld to hpce::your_login::StepWorldV1Lambda, to ensure it uses your new function.

Modify your build environment so that src/heat.cpp is compiled together with src/your_login/step_world_v1_lambda.cpp to produce a single executable called step_world_v1_lambda. Again, it is up to you where you choose to have it built.

_If you are using the default makefile, then doing:

make bin/your_login/step_world_v1_lambda

should build it through the magic of pattern matching, then you can run it as:

bin/your_login/step_world_v1_lambda

_

By making sure the executables have different names we can still run old versions, and we retain the ability to select between different implementations. Often students - I'm sure not you - resort to #ifdef and commented out code as they develop new versions of existing code. This is an alternative approach to that taking in CW1 and CW2: instead of having a switch that allows one program to have multiple implementations, we have multiple programs that can be switched in and out of a pipeline.

Test your program still does the same as the original implementation in whatever way you think best. Note that you have both step_world and step_world_v1_lambda available at the same time, and that there are programs like diff (posix) which will tell you if two text files are different. In bash this is particularly easy to do, and it can be incorporated into your makefile if you wish - make test is often a good target to have.

Making the isolation clean

Your lambda version captures everything by reference, so within kernel_xy there are references to world, buffer, and so on. For the GPU version the kernel cannot make any references to things which are not explictly passed as parameters, and the best way of emulating this is to move the kernel outside the stepping function into a pure function. In this context, "pure function" means a function that is only able to interact with the world via its parameters, and accesses no global state or captured variables.

OpenCL also doesn't allow complex C++ data-types to cross the CPU-GPU boundary, so we'll also need to convert those to primitive types or pointers to primitive types. For example, the kernel currently closes over (captures) world which is of type world_t, and buffer, which has type std::vector<float>. These need to be turned into parameters with primitive types and known size, such as uint32_t, float, and/or pointers to those types.

Based on your step_world_v1_lambda.cpp code, create a new file called step_world_v2_function.cpp in the same directory, and rename the inner function to hpce::your_login::StepWorldV2Function. Make sure you can build and run the file before proceeding.

Currently it will have a lambda called kernel_xy; take the lambda outside the StepWorldV2Function, and make it a new function called kernel_xy:

// Note the change from unsigned -> uint32_t to get a known size type
void kernel_xy(uint32_t x, uint32_t y)
{
    ... // Original lambda body
}

void StepWorldV2Function(world_t &world, float dt, unsigned n)
{
    ... // Original code, now without lambda

If you try compiling this, you'll see that it doesn't work - you'll get undefined errors about w, world, and so on. This is telling you all the information/data/memory/structures which will eventually have to be manually copied to the GPU, and if necessary back again afterwards. When accelerating for real, this is a good point to think about communication costs - have you really chosen the best point to cut the program at? Are there any things being copied across which could more cheaply be recalculated within the kernel?

Resolve all these errors by adding parameters to the definition of kernel_xy, and by adding arguments to the point where kernel_xy is called. For example, to resolve the error with w, you would modify the declaration:

void kernel_xy(uint32_t x, uint32_t y, uint32_t w);

and then use:

for(unsigned x=0;x<w;x++){
    kernel_xy(x,y, w);
}  // end of for(x...

Note that some things will require conversion from C++ types to primitive types. For example to convert world.state, you'll need to pass a pointer to the data:

void kernel_xy(uint32_t x, uint32_t y, uint32_t w, const float *world_state);

and:

for(unsigned x=0;x<w;x++){
    kernel_xy(x,y, w, &world.state[0]);
}  // end of for(x...

I used "const" on the world_state parameter to indicate that it is never modified inside the kernel. However, the converted buffer parameter cannot be const, as the kernel must write to the array to produce output. This process helps show which arrays must be copied both to and from the GPU, and which only need to be copied one way.

When it comes to passing the cell properties, note that cell_flags_t was originally defined to have the underlying type uint32_t, so you can safely cast to a (const uint32_t *) when converting parameters.

Keep adding parameters until you have removed all errors, making sure that you convert all of them to primitive types or pointers to arrays of primitive types.

Test the function to make sure it still works. If so, we are now in a position to map it into a CPU kernel.

Initial conversion to OpenCL (v3)

We've now got a version of the code which has separated the computationally intensive part out into a function. This function has:

To convert this to OpenCL, first we'll handle the host code (the long boring bit), then look at the GPU kernel. We'll do all this in a file called src/your_login/step_world_v3_opencl.cpp, with a function called src::your_login::StepWorldV3OpenCL.

Create the new file based on the src/your_login/step_world_v2_function.cpp code, and check you can still build and run it. While we're setting up the OpenCL parts, don't worry too much about the data going in and out. You can simply test it using something equivalent to:

bin/make_world | bin/step_world_v3_opencl > /dev/null

as for a while we'll just look at stderr.

Note that you'll have to uncomment

# LDLIBS += -lOpenCL

in the makefile to enable linking against the OpenCL libraries.

Opening the context, compiling kernels

Because OpenCL supports many devices, there are a number of stages involved in starting things up and connecting to various devices. Before we even think about executing kernels, we need to:

  1. Choose an OpenCL platform, as there might be multiple platforms available in one machine. For example, there may be both Intel and NVidia platforms installed.

  2. Select one of the devices exposed by the platform. As an example, the AMD platform may expose both CPU and GPU OpenCL devices.

  3. Create an OpenCL context for the platform, which describes the environment within which we can launch kernels and manage memory buffers.

  4. Create and build our kernels. It is common to load and build the GPU kernel code at run-time, just before the kernel is first used, though it is also possible to pre-compile them (pre-compilation is required for FPGA-based OpenCL providers).

These steps can often be simplified with library code, but here we'll rely on just the standard OpenCL C++ wrappers. These bindings come from CL/cl.hpp, and are installed in most recent OpenCL SDKs, but just in case, there is a version in the include directory too. The functions provided by this header are all documented at the Khronos site.

Add the include at the top of src/your_login/step_world_v3_opencl.cpp:

#define CL_USE_DEPRECATED_OPENCL_1_2_APIS
#define __CL_ENABLE_EXCEPTIONS 
#include "CL/cl.hpp"

The __CL_ENABLE_EXCEPTIONS definition tells the C++ wrapper that it should throw exceptions when things go wrong, rather than just returning an error code. This makes it much harder to accidentally ignore steps which went wrong. The deprecation macro is because NVidia still does not seem to officially support OpenCL 2.0, so we are working with 1.2 as a lowest common denominator.

At this point you may hit an error about CL/opencl.h not being found. Make sure that OpenCL is available, and that you have appropriately modified your include directories to point towards the include files in the OpenCL SDK. Assuming the initial test_opencl.cpp program compiled this should work fine.

Choosing a platform

We'll add the generic OpenCL initialisation at the start of StepWorldV3OpenCL, before doing anything with the world-specific code. First get a list of platforms from OpenCL:

std::vector<cl::Platform> platforms;

cl::Platform::get(&platforms);
if(platforms.size()==0)
    throw std::runtime_error("No OpenCL platforms found.");

After this code executes we'll have a list of installed platforms, which we can print the names of (in production code you would comment this out or make it optional):

std::cerr<<"Found "<<platforms.size()<<" platforms\n";
for(unsigned i=0;i<platforms.size();i++){
    std::string vendor=platforms[i].getInfo<CL_PLATFORM_VENDOR>();
    std::cerr<<"  Platform "<<i<<" : "<<vendor<<"\n";
}

At this point we could select just the first platform, but it is nice to be able to provide a way to select a specific platform. Often this would be by a configuration file, but we'll do it by an environment variable:

int selectedPlatform=0;
if(getenv("HPCE_SELECT_PLATFORM")){
    selectedPlatform=atoi(getenv("HPCE_SELECT_PLATFORM"));
}
std::cerr<<"Choosing platform "<<selectedPlatform<<"\n";
cl::Platform platform=platforms.at(selectedPlatform);

This means that by default it will select the first platform, but if we change an environment variable (e.g. export HPCE_SELECT_PLATFORM=1 in bash), then the executable will do something different. While not as good as a configuration file, this makes it possible to select what happens without recompiling. Again, I'm sure you don't use #ifdef and commenting to change functionality, but many less enlightened students do.

We now have a handle to our chosen platform in platform. This handle will automatically be deleted when the object goes out of scope, so there is no explicit handle management needed.

Choose a device

Within the platform there might be multiple devices, which we can enumerate:

std::vector<cl::Device> devices;
platform.getDevices(CL_DEVICE_TYPE_ALL, &devices);  
if(devices.size()==0){
    throw std::runtime_error("No opencl devices found.\n");
}

std::cerr<<"Found "<<devices.size()<<" devices\n";
for(unsigned i=0;i<devices.size();i++){
    std::string name=devices[i].getInfo<CL_DEVICE_NAME>();
    std::cerr<<"  Device "<<i<<" : "<<name<<"\n";
}

Depending on your system you may now see two devices. On my desktop, I get:

dt10@TOTO /cygdrive/e/_dt10_/documents/teaching/2016/hpce/cw/CW3
$ bin/make_world.exe 64 | bin/step_world_v3_opencl > /dev/null
Loaded world with w=64, h=64
Stepping by dt=0.1 for n=1
Found 1 platforms
  Platform 0 : Advanced Micro Devices, Inc.
Choosing platform 0
Found 2 devices
  Device 0 : Caicos
  Device 1 :         Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz

This shows I have one GPU device (Caicos), and one CPU device. Within each device there may be multiple CPUs, but here OpenCL is just listing the physical clusters available.

Despite one of these devices being a CPU, remember that it is just as valid OpenCL device as a GPU. It is often not as fast or powerful, but it can still be programmed in the same way. However, on a multi-core machine the CPU OpenCL device may in fact be very powerful, with the ability to exploit wide SIMD instructions. Do not under-estimate the power of a c5.9xlarge running all 36 cores over vectorisable code, as it can often beat a GPU.

As before we'll select device 0 by default, but give the option of changing that via an environment variable:

int selectedDevice=0;
if(getenv("HPCE_SELECT_DEVICE")){
    selectedDevice=atoi(getenv("HPCE_SELECT_DEVICE"));
}
std::cerr<<"Choosing device "<<selectedDevice<<"\n";
cl::Device device=devices.at(selectedDevice);

So for example I can switch between device 0 and 1 without re-compiling:

dt10@TOTO /cygdrive/e/_dt10_/documents/teaching/2016/hpce/cw/CW3
$ bin/make_world.exe 64 | bin/step_world_v3_opencl > /dev/null
Loaded world with w=64, h=64
Stepping by dt=0.1 for n=1
Found 1 platforms
  Platform 0 : Advanced Micro Devices, Inc.
Choosing platform 0
Found 2 devices
  Device 0 : Caicos
  Device 1 :         Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
Choosing device 0

dt10@TOTO /cygdrive/e/_dt10_/documents/teaching/2016/hpce/cw/CW3
$ export HPCE_SELECT_DEVICE=1

dt10@TOTO /cygdrive/e/_dt10_/documents/teaching/2016/hpce/cw/CW3
$ bin/make_world.exe 64 | bin/step_world_v3_opencl > /dev/null
Loaded world with w=64, h=64
Stepping by dt=0.1 for n=1
Found 1 platforms
  Platform 0 : Advanced Micro Devices, Inc.
Choosing platform 0
Found 2 devices
  Device 0 : Caicos
  Device 1 :         Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
Choosing device 1

Creating a context

To actually make use of the device, we need to create an OpenCL context, which represents a domain within which we can create and use kernels and memory buffers. Each object can only be used within the context it was created for, and it is unusual to have more than one context in existence at a time (though it is legal).

Contexts can actually cover multiple devices (though we won't use that ability here), so to create a context we pass a vector of devices. Conveniently, we still have such a vector in devices from when we enumerated the devices, which we can use to construct the context:

cl::Context context(devices);

That's about it for creating the context - it doesn't do a lot, but we need it as a parameter to all the kernel and memory buffer functions coming up.

Loading and building a kernel

The system is now pretty much initialised, and we can start doing application specific stuff. The first thing we want to do is create and compile our kernels. Each kernel is a piece of code compiled for the GPU, which can then be launched over a particular iteration space. But before writing the kernel, we'll write the host code to load and compile it.

Kernel code often exists as text files outside the program, and it is largely up to the user to get the text of the source code - OpenCL doesn't know or care where it comes from. Finding and locating this code at run-time can be a pain, and it is not uncommon to embed them as resources within the executable. For the purposes of this coursework we will assume that all cl code is in the src/your_login directory. So for example, the opencl code for this particular executable will eventually be placed at src/your_login/step_world_v3_kernel.cl. However, as a get-out, we will again define an environment variable called HPCE_CL_SRC_DIR which can be used to override that at run-time.

Having defined where the code will be found, the first thing to do is load the file contents. This is trivial, but worth knowing how to do simply, so briefly read this, then integrate it into your file:

// To go at the top of the file
#include <fstream>
#include <streambuf>

// To go in hpce::your_login, just above StepWorldV3OpenCL
std::string LoadSource(const char *fileName)
{
    // TODO : Don't forget to change your_login here
    std::string baseDir="src/your_login";
    if(getenv("HPCE_CL_SRC_DIR")){
        baseDir=getenv("HPCE_CL_SRC_DIR");
    }

    std::string fullName=baseDir+"/"+fileName;

    // Open a read-only binary stream over the file
    std::ifstream src(fullName, std::ios::in | std::ios::binary);
    if(!src.is_open())
        throw std::runtime_error("LoadSource : Couldn't load cl file from '"+fullName+"'.");

    // Read all characters of the file into a string
    return std::string(
        (std::istreambuf_iterator<char>(src)), // Node the extra brackets.
        std::istreambuf_iterator<char>()
    );
}

We need to collect together sources into a program, which is the unit of compilation in OpenCL. In our case it will only contain one kernel, but in principle there could be many kernels, spread across multiple sources. So, returning to where we left off, just after creating the context:

std::string kernelSource=LoadSource("step_world_v3_kernel.cl");

cl::Program::Sources sources;   // A vector of (data,length) pairs
sources.push_back(std::make_pair(kernelSource.c_str(), kernelSource.size()+1)); // push on our single string

cl::Program program(context, sources);
program.build(devices);

At this point you should be able to compile and run the overall program, but it should crash because there is no such kernel.

Writing a kernel

OpenCL host code (what we were just doing in C++) and OpenCL kernel code is written in a different language, so we need to create a new file to represent it. Create a file called src/your_login/step_world_v3_kernel.cl, and copy the entire kernel_xy function into it. You can leave the original kernel_xy function where it is, so that the software path still works.

Your file src/your_login/step_world_v3_kernel.cl should now look something like (your parameters may be in a different order):

void kernel_xy(unsigned x, unsigned y,
    unsigned w,
    float inner, float outer,
    const uint32_t *properties,
    const float *world_state,
    float *buffer
    )
{
    unsigned index=y*w + x;

    ... // Rest of code
}

Kernel functions are prefixed with __kernel, so as a minimum step towards creating a working kernel, change the function type in the cl file to:

__kernel void kernel_xy(unsigned x, unsigned y, // Rest of kernel...

Try running your C++ program again. It should now be able to find the file, but then OpenCL will throw an exception when it tries to compile it. The underlying cause is syntax errors, just as with a malformed C program, but here we can't see them.

To expose the specific error messages, go back to the C++ file and modify the build statement to the following:

cl::Program program(context, sources);
try{
    program.build(devices);
}catch(...){
    for(unsigned i=0;i<devices.size();i++){
        std::cerr<<"Log for device "<<devices[i].getInfo<CL_DEVICE_NAME>()<<":\n\n";
        std::cerr<<program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[i])<<"\n\n";
    }
    throw;
}

This code is catching the error being thrown, and then extracting the build log for each device. If you run it again, you should see a whole bunch of complaints.

You'll see a number of errors, most likely including:

After dealing with the errors, the code should now happily build. However, we are not done yet, as unlike tbb::parallel_for, the iteration space for OpenCL is embedded directly in the language. So rather than passing x and y as parameters, we will need to get the indices out of special indexing registers.

There are two types of builtin indices:

The dimensionality of each id is determined by the dimension of the iteration space used at launch-time (which is coming up soon). In this specific case we are using a 2D iteration space, and our memory pointers are global, so we can extract the x and y co-ordinates from the global work-group. As well as the co-ordinates, the run-time and hardware also know the sizes of the groups via get_local_size and get_global_size, so we can also remove the w parameter:

__kernel void kernel_xy(
    float inner, float outer,
    __global const uint *properties,
    __global const float *world_state,
    __global float *buffer
    )
{
    uint x=get_global_id(0);
    uint y=get_global_id(1);
    uint w=get_global_size(0);

    unsigned index=y*w + x;

    ... // Original code

In principle, this should now be a fully working OpenCL kernel!

Creating and executing kernels

We now have all the static infrastructure in place, with our devices and context open, and our OpenCL code compiled. At this point we have the choice of starting up multiple tasks which make use of the program we compiled, as the OpenCL interface is inherently multi-threaded, and there is nothing to stop multiple threads or tasks sharing a device.

However, within each thread we would have to make sure that all OpenCL tasks are synchronised and co-ordinated with each other (regardless of what other threads are doing). The mechanism for doing that is called a command queue, and as it suggests, it is a queueing or scheduling structure for issuing commands to the device. These commands can relate both to data movement between CPU and GPU, and to the actual execution of kernels on the GPU. It is essentially a way of building dependency graphs at run-time, then letting the GPU execute those graphs in the most efficient way possible.

To actually execute our kernel, we'll need to follow these steps:

  1. Allocate on-GPU memory buffers for the non-scalar parameters.

  2. Bind any fixed input parameters (e.g. inner) to the kernel.

  3. Create a command queue associated with the context.

  4. Copy any fixed input data (e.g. properties) to the GPU memory buffers.

  5. Bind any dynamic input parameters to the kernel.

  6. Copy any input data to the GPU memory buffers.

  7. Schedule the kernel onto the command queue. This (finally!) is where the actual execution happens.

  8. Copy any results back from GPU memory buffers.

Allocating buffers

We have three arrays in use while stepping the world:

Each of these arrays uses one float or uint32_t per cell, so the total bytes for each buffer is 4*world.w*world.h. We can allocate space for those arrays on the GPU's local memory using cl::Buffer. First we'll create the properties array:

size_t cbBuffer=4*world.w*world.h;
cl::Buffer buffProperties(context, CL_MEM_READ_ONLY, cbBuffer);
cl::Buffer buffState(context, CL_MEM_READ_ONLY, cbBuffer);
cl::Buffer buffBuffer(context, CL_MEM_WRITE_ONLY, cbBuffer);

So we have now allocated the buffers on the GPU, but don't have any idea about what data is on them. The flags mentioning READ and WRITE are hints to the GPU about how the buffers will be used from the kernel - you can always pass CL_MEM_READ_WRITE and not think about it, but if you can determine how memory will be accessed, it is possible the OpenCL runtime can optimise memory transfers.

Setting kernel parameters

It is often the case that some kernel parameters are constant through multiple uses of the kernel, in which case we can create the kernel and bind those parameters while setting things up. In our specific case, all our parameters are actually fixed, so we can set them all up before entering the loop (in a later optimisation we will have to modify some of them as we execute).

To set the kernel parameters, we first have to create an instance of our kernel which we can set the parameters on and schedule for execution. Just after the buffers, add:

cl::Kernel kernel(program, "kernel_xy");

This will find the compiled kernel called kernel_xy in the program, and create a new instance of it. We may wish to have multiple instances of the same kernel in order to schedule parallel copies with different parameters, but here we need just one.

Parameters are bound to the kernel using the setArg member function, and are set by index, not name. So we can quickly go and look in our kernel definition to count off the indices:

__kernel void kernel_xy(
    float inner,    // 0
    float outer,    // 1
    __global const uint *properties,    // 2
    __global const float *world_state,  // 3
    __global float *buffer  // 4
);

Don't worry if your kernel parameters are in a different order, just note which order they occur in.

Given the order, you can then bind the arguments in the C++ code:

kernel.setArg(0, inner);
kernel.setArg(1, outer);
kernel.setArg(2, buffProperties);
kernel.setArg(3, buffState);
kernel.setArg(4, buffBuffer);

You'll need to make sure that this occurs after the inner and outer variables are declared, but before the loop is entered.

The setArg function is overloaded, so it knows how to deal with scalar parameters such as inner as well as non-scalar cl::Buffers like buffState. However, you cannot bind a raw c++ pointer, such as a float *: at best you will get a compile-time error, and at worst it will crash unpredictably at run-time.

Creating a command queue

This is our final part of our overall application setup before moving inside the execution loop. As with the context, this one is quite simple. After setting the kernel arguments, simply create a command queue:

cl::CommandQueue queue(context, device);

Note that the command queue is bound to a single specific device, even though there may be multiple devices in our context. That is fine for our purposes, and is fine as long as you only want to work with one OpenCL device, but if we wanted to use multiple devices, we would need multiple command queues.

Copying over fixed data

The properties array is constant across all iterations, so we may as well copy it over to the GPU just once, before starting the iterations. Copies are scheduled to the command queue using enqueueWriteBuffer, which copies a specified area from host (CPU) memory to device (GPU) memory:

queue.enqueueWriteBuffer(buffProperties, CL_TRUE, 0, cbBuffer, &world.properties[0]);

The parameters to the function are:

Check that your program still compiles and runs. At this point it should still perform its normal function using software, but we are about to rip that out.

So assuming everything compiles and runs, we will now move into the loop over t. The part we are going to replace is the double loop over y and x, so:

  1. Check you know what the two loops are doing, and their loop bounds.

  2. Delete both inner loops.

You should be left with just a loop over t, with a call to std::swap and the time increment at the end. At this point the program is now broken, until we get the kernel working.

Copying memory buffers

The first thing to do within the loop is to copy the current state over to the GPU:

cl::Event evCopiedState;
queue.enqueueWriteBuffer(buffState, CL_FALSE, 0, cbBuffer, &world.state[0], NULL, &evCopiedState);

Unlike the previous copy, we are doing this one asynchronously, shown by the second parameter being CL_FALSE rather than CL_TRUE. That means that the call to enqueueWriteBuffer can return even before the copy has completed. We will have other operations which need the copy to complete before they can start, so we pass a pointer to evCopiedState to the function. This event can then be passed to any future function which must wait for this copy, explicitly indicating the dependency to the OpenCL run-time. Once this write has finished, the evCopiedState even will be signalled, releasing any waiting tasks.

Executing the kernel

Finally! A lot of GPU programming involves setting up a whole load of infrastructure before actually executing a kernel. Most of it is boiler-plate, and can be hidden by libraries, but it is all leading up to this:

cl::NDRange offset(0, 0);               // Always start iterations at x=0, y=0
cl::NDRange globalSize(w, h);   // Global size must match the original loops
cl::NDRange localSize=cl::NullRange;    // We don't care about local size

std::vector<cl::Event> kernelDependencies(1, evCopiedState);
cl::Event evExecutedKernel;
queue.enqueueNDRangeKernel(kernel, offset, globalSize, localSize, &kernelDependencies, &evExecutedKernel);

The first part is setting up the iteration space, with the offset specifying that both x and y start and zero, and globalSize saying that we should loop up to x<w and y<h. The localSize parameter determines clustering of threads into local work-groups, which we don't yet care about here (as we're only using global work-groups).

The second part establishes the dependencies of the kernel, by creating a vector of all the things that must complete before the kernel can run. For this case, there is only one dependency: the copy must have completed, so the event evCopiedState must have completed. We also produce an event called evExecutedKernel, so that we can tell when this kernel finishes.

After enqueueing the kernel, we don't know whether the copy has finished, or the kernel has finished. Maybe both have, maybe neither has. We cannot assume anything until we synchronise.

Copying the results back

The final job is to copy the results back after the kernel finishes:

std::vector<cl::Event> copyBackDependencies(1, evExecutedKernel);
queue.enqueueReadBuffer(buffBuffer, CL_TRUE, 0, cbBuffer, &buffer[0], &copyBackDependencies);

The copy cannot start until the kernel has finished, so we have to indicate the depdendency on evExecutedKernel. However, after this completes we need to interact with buffer, so we make this call synchronous using CL_TRUE.

And that's it. You should now have (laboriously) accelerated and existing C++ program. If all has gone well, the program should correctly spawn the calculation on the CPU, and perform all steps correctly. But as always: check to be sure.

Is it fast?

The whole point of this exercise was to make things faster, and if you time things as they stand, you may not be that impresssed:

time (bin/make_world | bin/step_world 0.1 1 > /dev/null)
time (bin/make_world | bin/$USER/step_world_v3_opencl 0.1 1 > /dev/null)

You may find that the opencl version is slower, possibly by a large amount. This can be attributed to the balance between two factors:

  1. There is a lot of overhead querying device drivers and setting up OS level information.
  2. The kernel must be compiled from scratch before it can execute anything.
  3. The amount of compute for the first world is very small, as the world is tiny and only iterated one step.

However, what we want to deal with in practise are:

  1. Large worlds. The greater the resolution of the world, the more precise the model, and the less the effect of spatial discretisaion.

  2. Small timesteps. Smaller values of dt will give less discretisation error, resulting in a more accurate model.

  3. Large periods. We want to see the steady state of system, which means running it until it stops changing, which could be many tens of thousands of iterations.

There is also a large hidden cost at the moment due to the formatting of the world data, as it is being laboriously converted to and from text when it is generated in make_world, and then when it is input and output in step_world. The reason for the text format is to make it easy to debug, but there is also a binary format, supported by both programs. To compare the two, try generating a larger world in text then binary:

time (bin/make_world 1000 0.1   0  > /dev/null)   # text format
time (bin/make_world 1000 0.1   1  > /dev/null)   # binary format

Depending on your platform, you'll see between one and two orders of magnitude in difference between the two. I would recommend using the binary format when not debugging, as otherwise your improvements in speed will be swamped by conversions.

Let's try to eliminate the overheads, by reading from a binary file and trying to increase the world size a bit:

bin/make_world 1000 0.1 1 > /tmp/world.bin  # Save binary world to temp file

Now look at the raw cost of execution with no steps at all:

time (cat /tmp/world.bin | bin/step_world 0.1 0  1 > /dev/null) 
time (cat /tmp/world.bin | bin/$USER/step_world_v3_opencl 0.1 0  1 > /dev/null)

The first command measures the cost of getting the data in and out, which should be down to tens of milliseconds, while the second command measures the overhead of getting the GPU up and running. This will vary depending on the system, but tells us something useful: if the current execution time of your program is less than the GPU startup cost, you will not be able to make it faster using a GPU.

We can check this by measuring the cost of one step:

time (cat /tmp/world.bin | bin/step_world 0.1 1  1 > /dev/null) 
time (cat /tmp/world.bin | bin/$USER/step_world_v3_opencl 0.1 1  1 > /dev/null)

The difference between this pair of times and the previous pair of times tells us about the marginal cost of each frame, after performing setup. Depending on your platform, the GPU time per frame will be similar to or, more likely, quite a bit slower than the original CPU.

So what is wrong? We have a number of problems here:

  1. Unecessary copying : we are copying the new state from GPU buffer back to CPU buffer, doing nothing with it, then copying the same data back again.

  2. Too little work per thread : there is a fixed scheduling cost in order to spin a GPU thread up. If there is not much work to do, then the scheduling overhead outweighs the benefits.

  3. No memory optimisations : the current approach places all data in global memory, which is the slowest memory type in the GPU. We should be taking advantage of private and shared memory as well.

Optimising CPU to GPU transfers (v4)

The first thing to try is to remove the redundant memory copies: if you look at StepWorldV3OpenCL the enqueue read at the bottom of the time loop will usually end up getting copied straight back over to the GPU at the top of the next iteration. We only look at the data right at the end, so there is no point bringing it back to the CPU in-between. So the only time the data needs to be transferred is:

  1. Before the very first execution of the kernel

  2. After the very last execution of the kernel.

Create a new file called src/your_login/step_world_v4_double_buffered.cpp, based on src/your_login/step_world_v3_opencl.cpp, and name the new function hpce::your_login::StepWorldV4DoubleBuffered.

We can minimise the transfers by moving the write of buffState off the top of the loop, and the read of buffBuffer off the bottom. Note that as you move the read over the call to std::swap, the target of the read should change: instead of reading into &buffer[0], it should read directly into &world.state[0]. To keep things simple, make both the copies synchronous, so pass CL_TRUE instead of CL_FALSE, and remove any input and output events.

The only thing that will be left in the time loop is the enqueue of the kernel, and the swap to flip the world state with the buffer. First let's fix the buffer, by observing that swapping the host buffers now has no effect, and we need to swap the GPU buffers being passed to the kernel instead.

Every time we call the kernel we want to flip the order of the buffers passed to the kernel, so we need to take the two setArg calls for the buffers:

kernel.setArg(3, buffState);
kernel.setArg(4, buffBuffer);

and bring them inside the loop, just above the point where the kernel is executed.

We also need to swap the two device pointers, so replace the std::swap of the host buffers with a swap of the device buffers:

std::swap(buffState, buffBuffer);   // Used to be std::swap(world.state, buffer)

Now we need to fix the kernel, as currently it has dependencies on events that no longer exist, so remove the input dependency list and the output event. You'll now have a loop which just enqueues kernels, without knowing when they have executed. The problem is that there is no fixed ordering between the kernels: the run-time could queue up 10 or 20 calls to enqueue kernel, then issue them all in parallel. So we need one kernel to finish before the next one starts, in order to make sure the output buffer is completely written before the next kernel call uses it as an input buffer.

There are multiple solutions to this, but the easiest is to add a call to enqueueBarrierWithWaitList just after the kernel is enqueue'd:

queue.enqueueNDRangeKernel(kernel, offset, globalSize, localSize);

queue.enqueueBarrierWithWaitList(); // <- new barrier here

std::swap(buffState, buffBuffer);

This creates a synchronisation point within the command queue, so anything before the barrier must complete before anything after the barrier can start. However, it doesn't say anything about the relationship with the C++ program. It is entirely possible that the host program could queue up hundreds of kernel calls and barriers before the OpenCL run-time gets round to executing them. Indeed, it may only be the synchronised memory copy after the loop that forces the chain of kernels and barriers to run at all.

So the inner loop should now look something like:

kernel.setArg(3, buffState);
kernel.setArg(4, buffBuffer);
queue.enqueueNDRangeKernel(kernel, offset, globalSize, localSize);

queue.enqueueBarrierWithWaitList();

std::swap(buffState, buffBuffer);

The kernel code itself is identical to that for the previous implementation, so you don't need to create a new .cl file. Sometimes you can optimise execution purely by modifying the host code, and other times tweaks to the kernel code are all that is needed.

A final problem is that our memory buffers were originally declared with read-only and write-only flags, but now we are violating that requirements. Go back to your buffer declarations, and modify them to:

cl::Buffer buffState(context, CL_MEM_READ_WRITE, cbBuffer);
cl::Buffer buffBuffer(context, CL_MEM_READ_WRITE, cbBuffer);

There is a lot going on here, so to summarise what the program should now look like, it should be something like:

Create buffProperties : read only
Create buffState, buffBuffer : read/write

Create kernel
Set inner, outer, and properties arguments

write properties data to buffProperties

write initial state to buffState

for t = 0 .. n-1 begin
   Set state argument to buffState
   Set output argument to buffBuffer

   enqueue kernel

   barrier

   swap(buffState, buffBuffer);
end

read buffState into output

You may find it confusing due to the abstraction of the cl::Buffer, but it helps if you think of them as a special kind of pointer. For example, imagine buffState and buffBuffer have notional addresses 0x4000 and 0x8000 on the GPU. Then what happens for n=2 is:

buffState = (void*)0x4000;
buffBuffer = (void*)0x8000;

memcpy(originalState , buffState, ...);

///////////////////////////////////////////////////

// loop iteration t=0
// buffState==0x4000, buffBuffer==0x8000

// Reading from 0x4000, writing to 0x8000
kernel(..., input = buffState (0x4000), output = buffBuffer (0x8000))

swap(buffState, buffBuffer); // swaps the values of the pointers

//////////////////////////////////////////////////

// loop iteration t=1
// buffState==0x8000, buffBuffer==0x4000

// reading from 0x8000, writing to 0x4000
kernel(..., input = buffState (0x8000), output = buffBuffer (0x4000))

swap(buffState, buffBuffer);

//////////////////////////////////////////////////

// After loop
// buffState==0x4000, buffBuffer==0x8000

memcpy(buffState, outputState, ...);

I highly recommend you try this with n=1, n=2, and n=3, as it is easy to think this is working for n=1 when it fails for larger numbers.

Depending on your platform, you may now start to see a reasonable speed-up over software (though maybe still not over TBB - depends a lot on the hardware).

Optimising global to GPU memory accesses (v5)

One of the biggest problems in GPU programming is managing the different memory spaces. Slight differences in memory layout can cause large changes in execution time, while moving arrays and variables from global to private memory can have huge performance implications. In general, the best memory accesses are those which never happen, so it is worth-while trying to optimise them out. GPU compilers can be more conservative then CPU compilers, so it is a good idea to help them out.

Looking at our current kernel, we can see that there are five reads to the properties array for a normal cell (non insulator). However, four of those reads are getting very little information back, as we only depend on a single bit of information from the four neighbours. A good approach in such scenarios is to try to pack data into spare bit-fields, increasing the information density and reducing the number of memory reads. In this case, we already have one 32-bit integer describing the properties, of which only 2 bits are currently being used. So as well as the 2 bits describing the properties of the current cell, we could quite reasonably include four bits describing whether the four neigbouring cells are insulators or not, saving a memory access.

This requires two modifications: one in the host code to set up the more complex flags, and another in the kernel code to take advantage of the bits.

Kernel code

Create a new kernel called src/your_login/step_world_v5_packed_properties.cl based on the v4 kernel.

At the top of the code, read the properties for the current cell into a private uint variable. This value will be read once into fast private memory, and then from then on it can be accessed very cheaply. For each of the branches, rewrite it to check bit-flags in the variable. For example, if your private properties variable is called myProps, and you decided bit 3 of the properties indicated whether the cell above is an insulator, the first branch could be re-written to:

// Cell above
if(myProps & 0x4) {
    contrib += outer;
    acc += outer * world_state[index-w];
}

The other neighbours will need to depend on different bits in the properties.

Host code

Create a new implementation called src/your_login/step_world_v5_packed_properties.cpp based on the v4 host code.

In the host code, you need to make sure the the correct flags have been inserted into the properties buffer. This should only have local effect, so we cannot modify world.properties directly. Instead create a temporary array in host memory:

std::vector<uint32_t> packed(w*h, 0);

and fill it with the appropriate bits. This will involve looping over all the co-ordinates, using the following process at each (x,y) co-ordinate:

packed(x,y) = world.properties(x,y)
if world.properties(x,y) is normal:
    if world.properties(x,y-1) is insulator:
        packed(x,y) = packed(x,y) + 4
    if world.properties(x,y+1) is insulator:
        packed(x,y) = packed(x,y) + 8
    # Handle left and right cases

This process takes some time (though it could be parallellised), but we don't care too much, as it only happens once for multiple time iterations. Once the array is prepared, it can be transferred to the GPU instead of the properties array.

This got my laptop up to about 2.5x speedup, so faster than the two cores in my device can go. On an AWS g2 GPU instance I looked at a 5000x5000 grid, stepped over 1000 time steps, which is more at the resolution we typically use:

time (bin/make_world 5000 0.1 1 | bin/step_world 0.1 1000 1 > /dev/null)
time (bin/make_world 5000 0.1 1 | bin/dt10/step_world_v3_opencl 0.1 1000 1 > /dev/null)
time (bin/make_world 5000 0.1 1 | bin/dt10/step_world_v4_double_buffered 0.1 1000 1 > /dev/null)
time (bin/make_world 5000 0.1 1 | bin/dt10/step_world_v5_packed_properties 0.1 1000 1 > /dev/null)

For my code, I found:

Method time (secs) speedup (total) speedup (incremental)
software 164.6 1.0 1.0
opencl 66.2 2.5 2.5
doublebuff 9.0 18.3 7.5
packing 6.5 25.3 1.4

I strongly encourage you to also try the software OpenCL provider. The AWS GPU instance has both a GPU and software provider installed, and you can use HPCE_SELECT_PLATFORM to choose which one you want. The GPU instance only has 16 cores, and the code is not optimised for CPU-based OpenCL providers, but it should still be less than half the time of the original software.

Further optimisations

In this context of this particular coursework, we won't go any further, but the next issue to tackle would be the small amount of work being done by each thread. At each co-ordinate there are only a handful of operations performed, so it would be good to agglomerate (batch up) some of those operations, to reduce scheduling overhead. This agglomeration could be temporal, by partially unrolling multiple iterations, or spatial, by grouping together consecutive iterations into one larger iteration.

We are also not using shared memory at all, which is wasting a precious resource. For this problem we want to have all processors working on the same problem, which means they need to communicate via global memory and perform global barriers (which don't formally exist). However, there are certain locking mechanisms that can be used to co-ordinate this, and a hierarchical approach could be used:

Such an approach would rely on hardware level knowledge to avoid deadlock, but is the kind of thing that might be done in practise.

Another optimisation is to cache the compiled binary for the kernel code. Depending on the device, the run-time, and other system characteristics, compiling the binary may be very fast, or quite slow. For example, on my desktop it takes well under a second, but on my laptop it takes around 10 seconds, both using the AMD OpenCL implementation. For large numbers of time-steps this cost will shrink, but for short-lived programs it is a potentially significant cost. The NVidia provider actually does this by default, so you may see less of this effect on such platforms.

Deliverables

At this point I'm going to stop requiring you to create graphs, and trust you to use the techniques you have to understand the various scaling properties. Another thing I highly reccommend doing is to actually create a tbb::parallel_for version and compare against that.

Make sure all your source files are checked into your git repository:

Each implementation should perform the same step as the original step_world. While there may be minor numerical differences due to the single-precision arithmetic, there should be no major differences in the output over time.

Also include any useful documents or supporting material, your repository does not have to be clean. However, note that:

Submission

Submission of code is via your git repository. Make sure you have committed and pushed to github - you are strongly encouraged to clone it into a different directory and then test it, just in case you are relying on something that wasn't commited.

Once your code is committed, note the commit hash of your current revision:

git log -1

This will produce something like:

ubuntu@ubuntu-xenial:/vagrant$ git log -1
commit 94d8419b20c78da86415bea7236d3719915977a3
Author: James Davis <m8pple@github.com>
Date:   Fri Jan 02 14:26:40 2018 +0000

    All tests passing.

You can also see the commit hash in the github web-site, and it is a good way of checking you have actually commited.

The commit hash of this revision is 94d8419b20c78da86415bea7236d3719915977a3 which is a cryptographic hash over all the files in your repository, including the history of all the files. Because the hash is cryptographically secure, it is impossible to take one commit hash, then come up with a different set of files which produces the same hash. The hash produced on your local machine will also match the hash calculated by github.

So take your hash (and just the hash), and submit it via blackboard. This is proof of existence - even if github goes down (and it does), you can later on prove that the existence of your hash in blackboard means you must have done the work associated with the hash. The hash in blackboard will also be the exact revision of your repository that will get checked out of github and tested. So you can carry on editing and modifying the repository, but only the commit with the hash submitted to blackboard is the one tested.

To summarise:

1 - Test your code.

2 - Commit your code to your local repo.

3 - Note the commit hash (git log -1).

4 - Submit the hash via blackboard.

5 - Push the code to your github repo.

You can repeat this process as many times as you want, up until the deadline.

Compatibility patches

Missing alloca.h

Windows doesn't have a version of , so if necessary just comment it out. See this issue.

Debugging tips

Debug output

In this particular exercise, you shouldn't encounter too many problems, but one useful feature of modern GPUs is that they allow debugging output, even from within kernels running in the GPU. In OpenCL 1.2 printf support is enabled by default, so you can use it from within kernels. Note that the printf output from different GPU threads may become interleaved - they are executing truly in parallel, after all.

In NVidia platforms you may be less lucky, as they try to keep some functionality CUDA-only for business reasons. However, you can always try using a software OpenCL implementation for testing purposes.

Software implementations

Another useful trick when debugging is to fall back on a software OpenCL implementation if possible. This can sometimes make it more obvious where a particular memory address is causing problems, as the exception occurs on the processor, rather than in the GPU.

Dynamically changing kernels

When OpenCL kernels are held as seperate files, it makes it very easy to play around with them without recompiling. For example, when trying to debug a particular crash or bug in a new or optimised kernel, you can easily comment out particular lines and re-run the program. Each kernel gets compiled anyway, so there is very little overhead to experimenting.

Locked up and crashing machines

Some systems are more or less stable than others when it comes to OpenCL. In principle it should be impossible to cause any problems using OpenCL code, and each process should still be isolated from each other. However, in practise the drivers can be a bit buggy, and it is not unheard of to see a blue-screen or kernel panic, The AMD drivers and hardware are particularly problematic in this regard, at least historically.

However, any such problems are usually due to incorrect kernel code, which is reading or writing where it shouldn't. A good starting point is to comment out suspicious memory writes, or to try working within a software OpenCL implementation as suggested above.

Even in a correct program, you may find that the OpenCL code can cause the machine to stutter or lag, in a way that is quite different to standard CPU load. This is usually caused due to time-slicing of the GPU with the graphics display. Older cards cannot run multiple kernels at the same time, so if the system is running your code, it cannot update the windows and display. It is generally a good idea to limit the execution time of any one kernel, so if possible tune things so that any given thread does not take more a few hundred milli-seconds. If you are on a server this is less of an issue, but even there you may find that the OS aborts kernel calls which take multiple seconds.