This guide is designed to help you start using Intel® Distribution for GDB* for debugging Data Parallel C++ (DPC++) and OpenCL™ applications. It describes how to set up the debugger to debug applications with kernels offloaded to CPU and GPU devices.
Intel® Distribution for GDB* is available as part of the Intel® oneAPI Base Toolkit. For more information on Intel® oneAPI Toolkits, visit the product page.
Visit the Release Notes page for the Known Issues and most up-to-date information.
Ensure that gdb-oneapi and gdbserver-gt binaries are present in the PATH variable. Otherwise, update environment variables by sourcing the <oneAPI_root_dir>/setvars.sh script.
To set up the GPU debugger, you must have root access.
Check that you have installed and loaded the debug companion driver (igfxdcd). You can check whether the module is installed with:
modinfo igfxdcd
If the module is not found, install it with:
For APT-based systems:
sudo dpkg -i $ONEAPI_ROOT/debugger/latest/igfxdcd-*-Linux.deb
For RPM-based systems:
sudo rpm -i $ONEAPI_ROOT/debugger/latest/igfxdcd-*-Linux.rpm
where <install_dir> is the oneAPI install location.
Load the debug companion driver with:
sudo modprobe igfxdcd
Test if you now can attach to the GPU and listen to debug events by executing the following command:
Linux (sudo): source /opt/intel/oneapi/setvars.sh Linux (user): source ~/intel/oneapi/setvars.sh
gdbserver-gt --attach :1234 1
The output below indicates successful attachment:
Attached; pid = 1 Listening on port 1234
To exit testing, press Ctrl+C.
The output below indicates that the companion driver is not installed or loaded properly. Please, review the GPU installation and configuration instructions to ensure that you set up the device correctly.
failed to initialize GT; result: Failure Exiting
You can use the sample project, Array Transform, to quickly get started with the application debugger.
To get the sample, choose any of the following ways:
Navigate to the src of the sample project:
cd array-transform/src
Compile the DPC++ application by enabling the debug info (-g flag) and turning off optimizations (-O0 flag).
dpcpp -g -O0 array-transform.cpp -o array-transform
If compilation and linking is done separately, retain the -g -O0 flags at the link step. The link step is when dpcpp translates these flags to be passed to the device compiler at runtime. Example:
dpcpp -g -O0 -c array-transform.cpp
dpcpp -g -O0 array-transform.o -o array-transform
Ahead-of-Time (AOT) compilation is the recommended method for typical sized application since compiling with debug information takes considerably longer time. To use Ahead-of-Time compilation mode:
For debugging on GPU:
Pass the -cl-opt-disable and -cl-kernel-debug-enable flags to the device compiler. For example:
dpcpp -g -O0 -fsycl-targets=spir64_gen-unknown-unknown-sycldevice \
-Xs "-device kbl -internal_options -cl-kernel-debug-enable -options -cl-opt-disable"\
array-transform.cpp -o array-transform
Ahead-of-Time compilation requires the OpenCL™ Offline (OC Compiler LOC). For more information, refer the section "Install OpenCL™ Offline Compiler (OCLOC)" of the Installation Guide.
-cl-kernel-debug-enable flag is required to run the application under the debugger. If it is not passed, the kernel cannot be offloaded to GPU during the debugging session, resulting in a runtime error.
For debugging on CPU, no additional actions are required.
For more information on AOT compilation, please refer to the Intel® oneAPI DPC++ Compiler Developer Guide and Reference.
When the setup above is completed, start the debug session:
INTELGT_AUTO_ATTACH_DISABLE=1
The Auto-Attach feature enables listening to debug events from the GPU. This feature is enabled by default and does not affect the debugging capability on the CPU device. You can turn it off to eliminate the extra output the feature creates. For more information about Auto-Attach, refer to the Tutorial.
gdb-oneapi array-transform
You should see the (gdb) prompt.
To make sure that the kernel is offloaded to the right device, do the following steps:
When you execute the run command from the (gdb) prompt, pass the cpu or gpu argument:
run cpu
run gpu
Define the environment variables as follows:
export CL_CONFIG_DEVICES=fpga-emu
Ensure that Intel_FPGA_SSG_Emulator is installed by running:
cat /etc/OpenCL/vendors/Intel_FPGA_SSG_Emulator.icd
The returned value must be
libintelocl_emu.so.
gdb-oneapi array-transform
Execute the run command from the (gdb) prompt, pass the accelerator argument:
run accelerator
Document |
Description |
---|---|
This document describes the basic scenarios to follow while debugging DPC++ and OpenCL with Intel® Distribution for GDB*. |
|
This document describes all common tasks that you can complete with Intel® Distribution for GDB* and provides necessary technical details.. |
|
The notes contain the most up-to-date information about Intel® Distribution for GDB* as part of the Intel® oneAPI Base Toolkit. |
|
This page contains brief introduction on Intel® oneAPI Toolkits and links to useful resources. |
Intel technologies may require enabled hardware, software or service activation.
No product or component can be absolutely secure.
Your costs and results may vary.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos.