Version: Vitis 2022.2
This tutorial is divided in 3 sections.
Section 1 (~10 mins):
An overview of Vitis and the host/kernel paradigm
See how Vitis takes care of the heavy lifting to let you focus on the application code
Section 2 (a couple of minutes if Alveo U50 card is already installed):
Setup the Vitis development tools
Detect and check the card installed on the server
Section 3 (a few hours):
Understand the Cholesky algorithm and run it on the CPU first
Re-organize the code to create both a host and a kernel program
Review the APIs that bind the host and the kernel
Apply incremental optimizations to the kernel across several modules to improve throughput. Finally program the Alveo card with the fully optimized accelerator to verify the performance
-
Meet the Cholesky algorithm!
Run a CPU version
-
Get a performance baseline as a reference point with a first kernel design
Run Vitis in GUI mode or via
make
Run Vitis Analyzer to visualize the application timeline
Run Vitis HLS to study kernel code performance and resource metrics
Vitis Module 2 (short module to focus on the impact of
PIPELINE
andINTERFACE
)Understanding instruction parallelism with the HLS
PIPELINE
pragmaApplying the
INTERFACE
pragma to manage physical ports adapters
-
Modify design to use the more hardware efficient C++
float
data types (compared todouble
)Run Vitis, Vitis Analyzer and Vitis HLS
-
Apply the
DATAFLOW
task parallelism optimization pragmaRun Vitis, Vitis Analyzer and Vitis HLS (including viewing specific dataflow waveforms)
Run on the U50 card
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
Copyright© 2020–2022 Xilinx
XD018