277. Wireless device for Xeon SP 83xx/63xx-D (bbdev) for Turbo decoding/encoding

277.1. Description

The Wireless Baseband library provides a common programming framework that abstracts HW accelerators based on FPGA and/or Fixed Function Accelerators that assist with 3gpp Physical Layer processing. Furthermore, it decouples the application from the compute-intensive wireless functions by abstracting their optimized libraries to appear as virtual bbdev devices. The functional scope of the BBDEV library are those functions in relation to the 3gpp Layer 1 signal processing (channel coding, modulation, ...). The framework currently only supports Turbo Code FEC function. The Wireless Baseband library follows the same ideology of DPDK’s Ethernet Device and Crypto Device frameworks. Wireless Baseband provides a generic acceleration abstraction framework which supports both physical (hardware) and virtual (software) wireless acceleration functions.

Physical bbdev devices are discovered during the PCI probe/enumeration of the EAL function which is executed at DPDK initialization, based on their PCI device identifier, each unique PCI BDF (bus/bridge, device, function). Virtual devices can be created by two mechanisms, either using the EAL command line options or from within the application using an EAL API directly. so it is required to perform validation of the framework divided into 2 stages: Stage 1: Validation of the SW-only solution (turbo_sw) Stage 2: Validation of the HW-accelerated solution (Xeon SP 83xx/63xx-D TIP) on an Xeon SP 83xx/63xx-D platform. We now only support stage 1.

277.2. Prerequisites

  1. OS and Hardware

    1. An AVX2 supporting machine
    2. Windriver TS 2 or CentOS 7 operating systems (Fedora 25 and Ubuntu 16.04 is ok.)
    3. Intel ICC compiler installed

  2. FlexRAN SDK Libraries

    To build DPDK with the turbo_sw PMD the user is required to download the export controlled FlexRAN SDK Libraries. An account at Intel Resource Design Center needs to be registered from https://www.intel.com/content/www/us/en/design/resource-design-center.html Direct download link from Intel RDC -> https://cdrd.intel.com/v1/dl/getContent/575367 After download is complete, the user needs to unpack and compile on their system before building DPDK. You can refer to the file dpdk/doc/guides/bbdevs/turbo_sw.rst.

  3. PMD setting

    Current BBDEV framework is en-suited with two vdev PMD drivers: null and turbo_sw. 1) Null PMD is similar to cryptodev Null PMD, which is an empty driver to measure the overhead added by the framework. 2) Turbo_sw is a sw-only driver wrapper for FlexRAN SDK optimized Turbo coding libraries.

  4. Test tool

    A test suite for BBDEV is packaged with the framework to ease the validation needs for various functions and use cases. The tool to use for validation and testing is called: test-bbdev, that is packaged with test vectors that are ready-to-use. Test-bbdev tool is located at this location: app/test-bbdev/ The command-line options you can refer to: dpdk/doc/guides/tools/testbbdev.rst

277.3. Test case 1: bbdev null device

Executing bbdev null device with bbdev_null.data helps in measuring the overhead introduced by the bbdev framework:

./test-bbdev.py -e="--vdev=baseband_null0"
-v ./test_vectors/bbdev_null.data

bbdev_null device does not have to be defined explicitly as it is created by default. so the command-line can be written as:

./test-bbdev.py -v ./test_vectors/bbdev_null.data

The case only cover unittest, all the other cases are skipped. The “bbdev_null.data” can be omitted:

./test-bbdev.py

or you can define the specific test defined:

./test-bbdev.py -c validation

or:

./test-bbdev.py -v test_vectors/bbdev_null.data

All above test cases run with “–vdev=baseband_null0”.

277.4. Test case 2: Turbo encoding validation

It runs validation test for Turbo encode vector file Number of operations to process on device is set to 64 and operations timeout is set to 120s and enqueue/dequeue burst size is set to 8 and to 32. Moreover a bbdev (turbo_sw) device will be created:

./test-bbdev.py -p ../../x86_64-native-linuxapp-icc/app/dpdk-test-bbdev \
-e="--vdev=baseband_turbo_sw" -t 120 -c validation \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -b 8 32

you can check the result from the detailed printing, and compare the two results from different burst size.

277.5. Test case 3: Turbo decoding validation

It runs validation test for Turbo decode vector file we use default options:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c validation \
-v ./test_vectors/turbo_dec_*

you can check the result from the detailed printing.

277.6. Test case 4: Turbo encoding latency

It runs latency test for Turbo encode vector file:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c latency \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 512 -b 64 512

Test calculates three latency metrics: 1) offload_latency_tc 2) offload_latency_empty_q_tc 3) operation_latency_tc You can compare the three latency from different burst size.

277.7. Test case 5: Turbo decoding latency

It runs latency test for Turbo decode vector file:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c latency \
-v ./test_vectors/turbo_dec_c1_k40_r0_e17280_sbd_negllr.data -n 512 -b 64

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c latency \
-v ./test_vectors/turbo_dec_c1_k40_r0_e17280_sbd_negllr.data -n 128 -b 64

You can compare the three latency from different number of operations.

277.8. Test case 6: Turbo encoding throughput

It runs throughput test for Turbo encode vector file:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -l 16 -b 64

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -l 8 -b 64

You can compare the turbo encode throughput from different number of lcores. then different burst size:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -l 16 -b 64

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -l 16 -b 32

You can compare the turbo encode throughput from different burst size. then different number of operations:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 64 -l 16 -b 32

./test-bbdev.py -e="--vdev=baseband_turbo_sw" -c throughput \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data -n 32 -l 16 -b 32

You can compare the turbo encode throughput from different number of operations.

277.9. Test case 7: Turbo decoding throughput

It runs throughput test for Turbo decode vector file. you can run the three subcases of test case 6 which just needed to change the test vector file to decode data file, and compare the results.

277.10. Test case 8: Turbo encoding and decoding offload and latency

It runs offload ** and **latency test for Turbo encode vector file:

./test-bbdev.py -p ../../x86_64-native-linuxapp-icc/app/dpdk-test-bbdev \
-e="--vdev=baseband_turbo_sw" -t 120 -c offload latency \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data \
./test_vectors/turbo_dec_c1_k40_r0_e17280_sbd_negllr.data -n 64 -l 16 -b 8 32

Compare the results.

277.11. Test case 9: Scrum all tests and all vector files

It runs all tests and all vector files:

./test-bbdev.py -e="--vdev=baseband_turbo_sw" \
-v ./test_vectors/turbo_enc_c1_k40_r0_e1196_rm.data

Then go through all the .date files.