Xilinx Snappy-Streaming Compression and Decompression¶
LZ4-Streaming example resides in L2/demos/snappy_streaming directory. To compile and test run this example execute the following commands:
Follow build instructions to generate host executable and binary.
The binary host file generated is named as “xil_snappy_streaming”, using PARALLEL_BLOCK value of 8 (default), and present in ./build directory.
Results¶
Resource Utilization¶
Table below presents resource utilization of Xilinx Snappy Streaming compress/decompress kernels (excluding data movers). It achieves Fmax of 300MHz
| Flow | LUT | LUTMem | REG | BRAM | URAM |
|---|---|---|---|---|---|
| Compress | 2.9K | 112 | 3.2K | 4 | 6 |
| DeCompress | 878 | 31 | 983 | 16 | 0 |
Performance Data¶
Table below presents best kernel throughput achieved for a single compute unit (Single Engine).
| Topic | Results |
|---|---|
| Best Compression Throughput | 260 MB/s |
| Best Decompression Throughput | 290 MB/s |
| Average Compression Ratio | 2.13x (Silesia Benchmark) |
Note: Overall throughput can still be increased with multiple compute units.
Usage¶
Execution Steps¶
To execute single file for compression :
./build/xil_snappy_streaming -cx <compress xclbin> -c <file_name>To execute single file for decompression :
./build/xil_snappy_streaming -dx <decompress xclbin> -d <file_name.snappy>To validate various files together :
./build/xil_snappy_streaming -cx <compress xclbin> -dx <decompress xclbin> -l <files.list><files.list>: Contains various file names with current path
The usage of the generated executable is as follows:
Usage: application.exe -[-h-c-l-d-B-x] --help, -h Print Help Options Default: [false] --compress, -c Compress --compress_xclbin -cx Compress binary --file_list, -l List of Input Files --decompress_xclbin -dx Decompress binary --decompress, -d Decompress --block_size, -B Compress Block Size [0-64: 1-256: 2-1024: 3-4096] Default: [0] --flow, -x Validation [0-All: 1-XcXd: 2-XcSd: 3-ScXd] Default: [1]