*)xrtBOMap(out_bohdl);
// s2mm ip
xrtKernelHandle s2mm_khdl = xrtPLKernelOpen(dhdl, uuid, "s2mm"); // Open kernel handle
xrtRunHandle s2mm_rhdl = xrtRunOpen(s2mm_khdl);
xrtRunSetArg(s2mm_rhdl, 0, out_bohdl); // set kernel arg
xrtRunSetArg(s2mm_rhdl, 2, OUTPUT_SIZE); // set kernel arg
xrtRunStart(s2mm_rhdl); //launch s2mm kernel
// random_noise ip
xrtKernelHandle random_noise_khdl = xrtPLKernelOpen(dhdl, uuid, "random_noise");
xrtRunHandle random_noise_rhdl = xrtRunOpen(random_noise_khdl);
xrtRunSetArg(random_noise_rhdl, 1, OUTPUT_SIZE);
xrtRunStart(random_noise_rhdl);
printf("run random_noise\n");
// update graph parameters (RTP) & run
...
// wait for s2mm done
auto state = xrtRunWait(s2mm_rhdl);
// Transfer data from global memory in device back to host memory
xrtBOSync(out_bohdl, XCL_BO_SYNC_BO_FROM_DEVICE , output_size_in_bytes,/*OFFSET=*/ 0);
// Post-processing of data
After post-processing the data, release the allocated objects:
gr.end();
xrtRunClose(s2mm_rhdl);
xrtKernelClose(s2mm_khdl);
xrtRunClose(random_noise_rhdl);
xrtKernelClose(random_noise_khdl);
xrtBOFree(out_bohdl);
xrtDeviceClose(dhdl);
The XRT API version to control graph exectuion is as follows:
auto ghdl=xrtGraphOpen(dhdl,uuid,"gr");
ret=xrtGraphUpdateRTP(ghdl,"gr.fir24.in[1]",(char*)narrow_filter,12*sizeof(int));
ret=xrtGraphRun(ghdl,16);
ret=xrtGraphWait(ghdl,0);
ret=xrtGraphUpdateRTP(ghdl,"gr.fir24.in[1]",(char*)wide_filter,12*sizeof(int));
ret=xrtGraphRun(ghdl,16);
You can see that the XRT API to open, run, wait, and RTP update graph are: xrtGraphOpen(), xrtGraphRun(), xrtGraphWait(), xrtGraphUpdateRTP(). For more information about XRT APIs on graphs, see the *Versal ACAP AI Engine Programming Environment User Guide* (UG1076).
Run the following `make` command to build the host exectuable file:
make host
Notice the following linker script links the libraries `adf_api_xrt` and `xrt_coreutil`. these are necessary for the `adf` API to work together with the XRT API.
${CXX} -o ../host.exe aie_control_xrt.o host.o -ladf_api_xrt -lgcc -lc -lxilinxopencl -lxrt_coreutil -lpthread -lrt -ldl -lcrypt -lstdc++ -L${SDKTARGETSYSROOT}/usr/lib/ --sysroot=${SDKTARGETSYSROOT} -L$(XILINX_VITIS)/aietools/lib/aarch64.o
Run the following `make` command to build all necessary files and launch HW cosimulation:
make run_hw_emu
In the Linux prompt, run following commands:
mount /dev/mmcblk0p1 /mnt
cd /mnt
export XILINX_XRT=/usr
export XCL_EMULATION_MODE=hw_emu
./host.exe a.xclbin
To exit QEMU press Ctrl+A, x
For hw mode, run following `make` command to generate an SD card package:
make package TARGET=hw
In hardware, after booting Linux from the SD card, run following commands in the Linux prompt:
export XILINX_XRT=/usr
cd /mnt/sd-mmcblk0p1
./host.exe a.xclbin
The host code is self-checking. It will check the output data against the golden data. If the output matches the golden data, after the run is complete, it will print the following:
TEST PASSED
### Conclusion
In this step you learned about:
* Asynchronous array RTP
* Launching AI Engine simulator, HW cosimulation, and HW
Next, review [Asynchronous Array RTP Update and Read for AI Engine Kernel](./step5_async_array_update_read.md).
XD001 | © Copyright 2021 Xilinx, Inc.
