uadk supports heterogeneous computing

[Liulongfang]

    After uadk supports hardware acceleration and instruction acceleration functions. Users expect to be able to

use both hardware acceleration and instruction acceleration. It is used to use instructions to continue to improve
and accelerate business performance after the hardware business is full. And it can automatically adapt to a variety
of acceleration devices.
The current patchset was developed for this purpose. And it has been fully adapted to all algorithm types of uadk.

   When using the updated framework, compared with separate hardware acceleration, the performance of

hybrid acceleration is significantly higher, and the acceleration effect has been significantly improved.

sm3 test cmd:
numactl --cpunodebind=0 --membind=0 uadk_tool benchmark --alg sm3 --mode sva --opt 0 --sync --pktlen 1024 --seconds 10 --thread 1 --multi 1 --ctxnum 1 --prefetch
numactl --cpunodebind=0 --membind=0 uadk_tool benchmark --alg sm3 --mode sva --opt 0 --sync --pktlen 1024 --seconds 10 --thread 1 --multi 1 --ctxnum 1 --prefetch --init2

SM3 1024B Performance(MB/s)                

tds------init1(HW)-----init2(HW + CE)----increase
1-----------393.3--------437.1-------------11.14%
2----------762.1---------823.4------------8.04%
4----------1508.4-------1564.1------------3.69%
8----------3007.4------3074.9-----------2.24%
16---------4851.8-------5429.2-----------11.90%
32--------4854.1-------8698.8------------79.21%

sm4 test cmd:
numactl --cpunodebind=0 --membind=0 uadk_tool benchmark --alg sm4-128-ecb --mode sva --opt 0 --sync --pktlen 1024 --seconds 10 --thread 1 --multi 1 --ctxnum 1 --prefetch
numactl --cpunodebind=0 --membind=0 uadk_tool benchmark --alg sm4-128-ecb --mode sva --opt 0 --sync --pktlen 1024 --seconds 10 --thread 1 --multi 1 --ctxnum 1 --prefetch --init2
numactl --cpunodebind=0 --membind=0 uadk_tool benchmark --alg sm4-128-ecb --mode sva --opt 0 --async --pktlen 1024 --seconds 10 --thread 1 --multi 1 --ctxnum 1 --prefetch --init2

SM4 1024B Performance(MB/s)                

tds-------init1(HW)----init2(HW + CE)---------increase
1-------------461----------1482.5---------------221.58%
2------------914----------2575.4---------------181.77%
4-----------1699.9--------4737.6---------------178.70%
8-----------3301.5--------7327.8---------------121.95%
16----------5837.5--------9737.4---------------66.81%
32----------8897.7-------10432.4--------------17.25%

SM4 1024B async Performance(MB/s)

tds-------init1(HW)----init2(HW + CE)---------increase
1-----------1368.3--------1683.9---------------23.07%
2------------2652---------3235.5---------------22.00%
4-----------3979.5--------5094.5---------------28.02%
8-----------6667.7---------8587----------------28.79%
16----------8900.9-------11067.8---------------24.34%
32----------8905.9-------10209.1--------------14.63%

[gaozhangfei]

有单侧ce的数据么
还有测试命令，要是方便也贴下

[Liulongfang]

单侧ce的数据如下：
SM4 1024B CE Performance(MB/s)
tds-------init1(CE)
1-----------2955.9
2-----------3446.6
4-----------5774.3
8-----------8399.2
16----------10035.8
32----------10638.9

SM3 1024B CE Performance(MB/s)
tds-------init1(CE)
1-----------436.2
2-----------824.9
4-----------1571.6
8-----------3107.2
16----------5571.2
32----------9071.6

[gaozhangfei]

硬件性能偏低，可有测过 --thread 8 --ctxnum 8？
有1+1>2的情形么
可以选择是否打开调度吧。