2019.1 SDAccel™ Development Environment Tutorials

Mixing C++ and RTL Kernels

Introduction

In the SDAccel™ development environment framework, an application program is split between a host application and hardware accelerated kernels. The host application is developed in C/C++ with OpenCL™ API calls. The hardware kernels, on the other hand, can be developed in C/C++, OpenCL C, or RTL. In fact, SDAccel environment applications can use any combination of kernels developed in the different languages. The host code is indifferent to how the kernel was developed; it uses the same function call.

Tutorial Overview

This tutorial demonstrates an application using two kernels, one designed in C++ and the other designed in RTL, with the host code accessing the kernels in an identical manner.

This tutorial is split into two parts:

In the first part, you will build an application (host and kernel) with one C++ based kernel. The host code, including the kernel function call, is reviewed.
In the second part, an RTL-based kernel will be added to the application. The updated host code, including the function call to the additional kernel, is reviewed.

During both parts, the application is built using a Makefile. Software emulation is run in step one and hardware emulation is run in step two. In both steps the generated Application Timeline will be reviewed to highlight the kernels being called and run by the host application.

The host code and C++ kernel code are supplied. The RTL code will be generated using the RTL Kernel Wizard.

Before You Begin

This tutorial uses:

BASH Linux shell commands.
2019.1 SDx™ release and the xilinx_u200_xdma_201830_1 platform. If necessary, it can be easily extended to other versions and platforms.

IMPORTANT:

Before running any of the examples, make sure you have installed Xilinx Runtime (XRT) and the SDAccel development environment as described in the SDAccel Development Environment Release Notes, Installation, and Licensing Guide (UG1238).

If you will run applications on the Alveo™ card, ensure the card and software drivers have been correctly installed by following the instructions in the Getting Started with Alveo Data Center Accelerator Cards Guide (UG1301).

Before running any of the examples, ensure you have setup the SDx tools, platform, and runtime by running the following commands:

  #setup Xilinx SDx tools, XILINX_SDX and XILINX_VIVADO will be set in this step. source <SDX install path>/settings64.sh. for example:
  source /opt/Xilinx/SDx/2019.1/settings64.sh
  #Setup runtime. XILINX_XRT will be set in this step
  source /opt/xilinx/xrt/setup.sh

Accessing the Tutorial Reference Files

To access the reference files, type the following into a terminal: git clone https://github.com/Xilinx/SDAccel-Tutorials.
Navigate to SDAccel-Tutorials-master/docs/mixing-c-rtl-kernels/reference-files.

Building an Application with C++ Based Kernel

In this step, you will build an application, consisting of host code and a C++ kernel using a makefile.

For an overview on building an application, refer to Building An Application.

C++ Based Kernel

The C++ based kernel adds two input vectors, and generates the output result. The source code is found in the following directory.

./reference-files/src/kernel_cpp/

The makefile will build this kernel, and add it to the hardware platform (xclbin), which can be accessed by the host code.

Host Code

The host code for step 1 (host_step1.cpp) can be found in the following directory.

./reference-files/src/host/

It sets up the platform, and defines global memory buffers and connections to the kernel. The four important sets of OpenCL API calls in the host code are described below. To view these calls, open the host_step1.cpp file.

The first set of code, on lines 189-191, creates the program to execute. It uses the binary container, which contains only the C++-based kernel.
```
cl::Program::Binaries bins;
bins.push_back({buf,nb});
cl::Program program(context, devices, bins);
```
The second set of code, on line 195, gets the C++ krnl_vadd kernel object from the program, and assigns the name krnl_vector_add. It allows the kernel to be used by the host.
```
cl::Kernel krnl_vector_add(program,"krnl_vadd");
```

The third set of code, on lines 213-216, assigns the krnl_vector_add kernel arguments to the buffers.

krnl_vector_add.setArg(0,buffer_a);
krnl_vector_add.setArg(1,buffer_b);
krnl_vector_add.setArg(2,buffer_result);
krnl_vector_add.setArg(3,DATA_SIZE);

The argument numbers 0, 1, 2, and 3 match the argument order in the krnl_vadd definition found in krnl_vadd.cpp as follows.

NOTE: Arguments a and b are inputs, and c is an output.
```
void krnl_vadd(
             int* a,
             int* b,
             int* c,
             const int n_elements)
```
Finally, on line 220, the following OpenCL API launches the krnl_vector_add kernel.
```
q.enqueueTask(krnl_vector_add);
```

For complete details on host code programming, refer to the SDAccel Environment Programmers Guide (UG1277).

Build the Application

To build the application targeting software emulation, run the following makefile from the ./reference-files/run1 directory.
```
make all TARGET=sw_emu
```
This will build both the host software and hardware binary targeted to software emulation. The makefile will also generate the platform JSON emulation file to use during emulation.

Run Emulation

During emulation, you will gather application timeline data, consisting of host and device events, which can be reviewed after emulation has completed on a common timeline. The Application Timeline data collection must be enabled before running the emulation by setting timeline_trace=true option in an xrt.ini file.

[Debug]
profile=true
timeline_trace=true

This file has already been created and is located in the run1 and run2 directories.

To run software emulation on the design, set the XCL_EMULATION_MODE environment variable to run the following command.
```
export XCL_EMULATION_MODE=sw_emu
```
From within the run1 directory, to run software emulation, use the following command.
```
./host krnl_vadd.sw_emu.xilinx_u200_xdma_201830_1.xclbin
```
When the application successfully completes, the following message is displayed in the Console window.

TEST WITH ONE KERNEL PASSED

Review the Application Timeline

Review the Application Timeline generated during software emulation to visualize the host events and the kernel running.

To view the Application Timeline, create a waveform file from the CSV file by running the following command from within the run1 directory.
```
sdx_analyze trace -f wdb -i ./timeline_trace.csv
```
Open the generated timeline trace waveform file using SDx with the following command from within the run1 directory.
```
sdx -workspace workspace -report timeline_trace.wdb
```
Under the Device>Binary Container, you will see a line called Compute Unit krnl_vadd_1.
Traverse along the timeline, and zoom in on the compute unit krnl_vadd_1 to confirm it shows as running.
After reviewing, close the Application Timeline window.

NOTE: A compute unit is an instantiation of the kernel on the FPGA.

Putting it All Together

For step 1, the following steps were performed. All commands are run in the ./reference-files/run1 directory:

   # Build the application
   make all TARGET=sw_emu

   # Set XCL_EMULATION_MODE environment variable for software emulation
   export XCL_EMULATION_MODE=sw_emu

   # Run software emulation
   ./host krnl_vadd.sw_emu.xilinx_u200_xdma_201830_1.xclbin

   # Create timeline waveform and view in SDx
   sdx_analyze trace -f wdb -i ./timeline_trace.csv
   sdx -workspace workspace -report timeline_trace.wdb

Building an Application with C++ and RTL-Based Kernels

Now that you have successfully built and run an application with a C++ based kernel, update the application to include an RTL-based kernel.

Similar to the previous section, you will build, emulate, and review the generated Application Timeline. Regardless of how the kernels were designed, after the kernel has been built, the host code accesses the kernels through similar function calls.

RTL-Based Kernel

First, you will create and package an RTL-based kernel using the RTL Kernel Wizard. By default, the wizard creates a kernel to increment by one. This kernel will be used in this tutorial. Additionally, the wizard automates the steps needed to package the RTL design into a kernel object file (XO).

You will generate the RTL-based kernel by quickly going through the RTL Kernel Wizard steps without much detail. Review the Getting Started with RTL Kernels RTL Kernel Wizard for a thorough tutorial. For complete details on the RTL Kernel Wizard, refer to the SDAccel Environment User Guide (UG1023).

Create the SDAccel Project

To open the SDAccel environment GUI, enter sdx in the command line.
Select ./mixing-c-rtl-kernels/workspace as the workspace directory, and click Launch.
Select Create Application Project, and name it rtl_project, and click Next.
Select the xilinx_u200_xdma_201830_1 platform, and click Next.
Under Templates, select Empty Application, and click Finish. This creates an SDAccel environment project.

Next, generate a RTL-based kernel from within the SDAccel environment project.

Under the Xilinx menu, select RTL Kernel Wizard.
This opens the RTL Kernel Wizard Welcome page.
Click Next.
In the General Settings dialog box, keep all the default settings, and click Next.
In the Scalars dialog box, set the number of scalar arguments to 0, and click Next.
In the Global Memory dialog box, keep all the default settings, and click Next.
In the Streaming Interfaces dialog box, keep all the default settings, and click Next.
The Summary dialog box is displayed and provides a summary of the RTL kernel settings and includes a function prototype which conveys what a kernel call would look like as a C function.
Click OK.

The RTL Kernel source files have now been created.

The Vivado Design Suite Project

At this point, the Vivado Design Suite opens a project automatically with the generated RTL code corresponding to the default A = A + 1 function. You can navigate to review the source files, or even run RTL simulation. However, for this tutorial, we will not be modifying the default RTL Kernel and will only package into an object file (XO).

In Flow Navigator, click Generate RTL Kernel.
In the Generate RTL Kernel dialog box, select the Sources-only packaging option.
For Software Emulation Sources, you can add a C++ model of the RTL kernel which is used for Software Emulation.
The C++ model must be coded by the design engineer. Typically, there is no C++ model available, and Hardware Emulation is used to test the design.

Because the RTL Wizard creates a C++ model of the vadd design, the steps to add this file are also given below.
Click … (the browser button).
Double-click the imports directory.
Select the only CPP file (.cpp), and click OK.
To generate the RTL kernel, click OK.
After the RTL kernel has been generated successfully, click Yes to exit the Vivado Design Suite, and return to the SDAccel environment.

RTL Kernel has been imported will be displayed.
Click OK.
Exit the SDAccel tool.

At this point, you have packaged the RTL kernel into the following object file, sdx_kernel_wizard_0.xo found in the following directory.

./mixing-c-rtl-kernels/workspace/rtl_project/src/sdx_rtl_kernel/sdx_kernel_wizard_0

Host Code Updates

To access the RTL-based kernel, the host code needs to be updated. The updates have been done in the host_step2.cpp file located in the following directory.

./reference-files/src/host/

The updates includes additional OpenCL API calls briefly described below. The additional OpenCL API calls are identical to the ones used for the C++ based kernel with the arguments changed for the RTL-based kernel.

cl::Program::Binaries bins;
bins.push_back({buf,nb});
cl::Program program(context, devices, bins);

The following code gets the sdx_kernel_wizard_0 object from the program, and assigns the name krnl_const_add on line 198. The sdx_kernel_wizard_0 object name matches the name of the kernel you generated with the RTL Wizard.

cl::Kernel krnl_const_add(program,"sdx_kernel_wizard_0");

Next, define the krnl_const_add kernel arguments on line 219.

NOTE: In the host code, the buffer buffer_result is passed directly from the C kernel to the RTL kernel through DDR without being moved back to the host memory.

krnl_const_add.setArg(0,buffer_result);

Launch the krnl_const_add kernel on line 222.

q.enqueueTask(krnl_const_add);

Build and Emulation with C++ and RTL Based Kernels

With the RTL-based kernel added and host code updated, build the application, targeting hardware emulation through the updated makefile in the run2 directory. The makefile has been update to add both the CPP and RTL-based kernels to the hardware platform file (xclbin).

Navigate to the ./mixing-c-rtl-kernels/reference-files/run2 directory.
To build the application targeting hardware emulation, run the following makefile from the ./reference-files/run2 directory.
```
make all TARGET=hw_emu
```
As before, run emulation, and generate and review the Application Timeline by running the following commands from within the run2 directory.
Set XCL_EMULATION_MODE environment variable for hardware emulation.
```
export XCL_EMULATION_MODE=hw_emu
```

Run hardware emulation.

./host krnl_vadd.hw_emu.xilinx_u200_xdma_201830_1.xclbin

Create a waveform file from the CSV file.

sdx_analyze trace -f wdb -i ./timeline_trace.csv

Open the generated timeline trace waveform file.

sdx -workspace workspace -report timeline_trace.wdb

Under Device>Binary Container, traverse along the timeline and zoom in.
You will now see both compute units, krnl_vadd_1 and sdx_kernel_wizard_0_1, show as running.
After reviewing, close the Application Timeline window.

SDAccel applications can use any combination of kernels, regardless of the language they were developed in.

Because a CPP emulation file was packaged with the RTL Kernel (through the RTL Wizard), you can also run software emulation. To run software emulation, you can use the following steps:

   # Build the application
   make all TARGET=sw_emu

   # Set XCL_EMULATION_MODE environment variable for software emulation
   export XCL_EMULATION_MODE=sw_emu

   # Run software emulation
   ./host krnl_vadd.sw_emu.xilinx_u200_xdma_201830_1.xclbin

   # Create timeline waveform and view in SDx
   sdx_analyze trace -f wdb -i ./timeline_trace.csv
   sdx -workspace workspace -report timeline_trace.wdb

Next Steps

For a better understanding of building applications, running emulation, and generating profile and trace reports, see the Essential Concepts for Building and Running the Accelerated Application tutorial.
For details about the RTL Kernel Wizard:
- Review the Getting Started with RTL Kernels tutorial.
- Review the RTL Kernel Wizard section in the SDAccel Environment User Guide (UG1023).

Return to Main Page

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

2019.1 SDAccel™ Development Environment Tutorials

Mixing C++ and RTL Kernels

Introduction

Tutorial Overview

Before You Begin

Accessing the Tutorial Reference Files

Building an Application with C++ Based Kernel

C++ Based Kernel

Host Code

Build the Application

Run Emulation

Review the Application Timeline

Putting it All Together

Building an Application with C++ and RTL-Based Kernels

RTL-Based Kernel

Create the SDAccel Project

The Vivado Design Suite Project

Host Code Updates

Build and Emulation with C++ and RTL Based Kernels

Next Steps

Files

README.md

Latest commit

History

README.md

File metadata and controls

2019.1 SDAccel™ Development Environment Tutorials

Mixing C++ and RTL Kernels

Introduction

Tutorial Overview

Before You Begin

Accessing the Tutorial Reference Files

Building an Application with C++ Based Kernel

C++ Based Kernel

Host Code

Build the Application

Run Emulation

Review the Application Timeline

Putting it All Together

Building an Application with C++ and RTL-Based Kernels

RTL-Based Kernel

Create the SDAccel Project

The Vivado Design Suite Project

Host Code Updates

Build and Emulation with C++ and RTL Based Kernels

Next Steps