Xilinx PairHMM Genomics¶
PairHMM Algorithm resides in L2/tests/pairhmm_8x2 directory.
Xilinx PairHMM Algorithm is FPGA based implementation of standard PairHMM. Xilinx implementation of PairHMM application is aimed for achieving high throughput. This Xilinx PairHMM application is developed and tested on Xilinx Alveo U250.
Results¶
Resource Utilization¶
Table below presents resource utilization of Xilinx Smithwaterman Genomics kernels. The final Fmax achieved is 300MHz.
| Flow | LUT | DSP | BRAM | URAM |
|---|---|---|---|---|
| PairHMM | 66.9K | 504 | 35 | 32 |
Performance Data¶
Table below presents kernel throughput achieved for a single compute unit.
| Topic | Results |
|---|---|
| Genomics Throughput | 4.8GCups |
Note: Overall throughput can still be increased with multiple compute units.
Software & Hardware¶
Software: Xilinx Vitis 2021.2
Hardware: xilinx_u250_gen3x16_xdma_3_1_202020_1(Xilinx Alveo U250)
Executable Usage¶
To execute pairHMM kernel : ./build/xil_pairhmm ./build/xclbin_<xsa_name>_<TARGET mode>/pairhmm.xclbin --syn <number of tests>
- Note: Default arguments are set in Makefile