MIGraphX Fundamentals#
2023-05-25
8 min read time
MIGraphX provides an optimized execution engine for deep learning neural networks.
We will cover some simple operations in the MIGraphX framework here.
For a quick start guide to using MIGraphX, look in the examples directory: https://github.com/ROCmSoftwarePlatform/AMDMIGraphX/tree/develop/examples/migraphx.
Location of the Examples#
The ref_dev_examples.cpp can be found in the test directory (/test).
The executable file test_ref_dev_examples based on this file will be created in the bin/ of the build directory after running make -j$(nproc) test_ref_dev_examples.
The executable will also be created when running make -j$(nproc) check, alongside with all the other tests.
Directions for building MIGraphX from source can be found in the main README file: https://github.com/ROCmSoftwarePlatform/AMDMIGraphX#readme.
Adding Two Literals#
A program is a collection of modules, which are collections of instructions to be executed when calling eval <migraphx::program::eval>. Each instruction has an associated operation <migraphx::operation> which represents the computation to be performed by the instruction.
We start with a snippet of the simple add_two_literals() function:
// create the program and get a pointer to the main module
migraphx::program p;
auto* mm = p.get_main_module();
// add two literals to the program
auto one = mm->add_literal(1);
auto two = mm->add_literal(2);
// make the add operation between the two literals and add it to the program
mm->add_instruction(migraphx::make_op("add"), one, two);
// compile the program on the reference device
p.compile(migraphx::ref::target{});
// evaulate the program and retreive the result
auto result = p.eval({}).back();
std::cout << "add_two_literals: 1 + 2 = " << result << "\n";
We start by creating a simple migraphx::program object and then getting a pointer to the main module of it.
The program is a collection of modules that start executing from the main module, so instructions are added to the modules rather than directly onto the program object.
We then use the add_literal <migraphx::program::add_literal> function to add an instruction that stores the literal number 1 while returning an instruction_ref <migraphx::instruction_ref>.
The returned instruction_ref <migraphx::instruction_ref> can be used in another instruction as an input.
We use the same add_literal <migraphx::program::add_literal> function to add a 2 to the program.
After creating the literals, we then create the instruction to add the numbers together.
This is done by using the add_instruction <migraphx::program::add_instruction> function with the "add" operation <migraphx::program::operation> created by make_op <migraphx::program::make_op> along with the previous add_literal instruction_ref <migraphx::instruction_ref> for the input arguments of the instruction.
Finally, we can run this program <migraphx::program> by compiling it for the reference target (CPU) and then running it with eval <migraphx::program::eval>
The result is then retreived and printed to the console.
We can compile the program for the GPU as well, but the file will have to be moved to the test/gpu/ directory and the correct target must be included:
#include <migraphx/gpu/target.hpp>
Using Parameters#
The previous program will always produce the same value of adding 1 and 2.
In the next program we want to pass an input to a program and compute a value based on the input.
We can modify the program to take an input parameter x, as seen in the add_parameter() function:
migraphx::program p;
auto* mm = p.get_main_module();
migraphx::shape s{migraphx::shape::int32_type, {1}};
// add a "x" parameter with the shape s
auto x = mm->add_parameter("x", s);
auto two = mm->add_literal(2);
// add the "add" instruction between the "x" parameter and "two" to the module
mm->add_instruction(migraphx::make_op("add"), x, two);
p.compile(migraphx::ref::target{});
This adds a parameter of type int32, and compiles it for the CPU.
To run the program, we need to pass the parameter as a parameter_map when we call eval <migraphx::program::eval>.
We create the parameter_map by setting the x key to an argument <migraphx::argument> object with an int data type:
// create a parameter_map object for passing a value to the "x" parameter
std::vector<int> data = {4};
migraphx::parameter_map params;
params["x"] = migraphx::argument(s, data.data());
auto result = p.eval(params).back();
std::cout << "add_parameters: 4 + 2 = " << result << "\n";
EXPECT(result.at<int>() == 6);
Handling Tensor Data#
In the previous examples we have only been dealing with scalars, but the shape <migraphx::shape> class can describe multi-dimensional tensors. For example, we can compute a simple convolution:
migraphx::program p;
auto* mm = p.get_main_module();
// create shape objects for the input tensor and weights
migraphx::shape input_shape{migraphx::shape::float_type, {2, 3, 4, 4}};
migraphx::shape weights_shape{migraphx::shape::float_type, {3, 3, 3, 3}};
// create the parameters and add the "convolution" operation to the module
auto input = mm->add_parameter("X", input_shape);
auto weights = mm->add_parameter("W", weights_shape);
mm->add_instruction(migraphx::make_op("convolution", {{"padding", {1, 1}}, {"stride", {2, 2}}}), input, weights);
Here we create two parameters for both the input and weights.
In the previous examples, we created simple literals, however, most programs will take data from allocated buffers (usually on the GPU).
In this case, we can create argument <migraphx::argument> objects directly from the pointers to the buffers:
// Compile the program
p.compile(migraphx::ref::target{});
// Allocated buffers by the user
std::vector<float> a = ...;
std::vector<float> c = ...;
// Solution vector
std::vector<float> sol = ...;
// Create the arguments in a parameter_map
migraphx::parameter_map params;
params["X"] = migraphx::argument(input_shape, a.data());
params["W"] = migraphx::argument(weights_shape, c.data());
// Evaluate and confirm the result
auto result = p.eval(params).back();
std::vector<float> results_vector(64);
result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
EXPECT(migraphx::verify_range(results_vector, sol));
An argument <migraphx::argument> can handle memory buffers from either the GPU or the CPU.
By default when running the program <migraphx::program>, buffers are allocated on the corresponding target.
When compiling for the CPU, the buffers by default will be allocated on the CPU.
When compiling for the GPU, the buffers by default will be allocated on the GPU.
With the option offload_copy=true set while compiling for the GPU, the buffers will be located on the CPU.
Importing From ONNX#
A program <migraphx::program> can be built directly from an onnx file using the MIGraphX ONNX parser.
This makes it easier to use neural networks directly from other frameworks.
In this case, there is an parse_onnx function:
program p = migraphx::parse_onnx("model.onnx");
p.compile(migraphx::gpu::target{});