Benchmarking model described in Real-time Portrait Segmentation on Smartphones by Prisma on MNN framework.
Mi 9T Pto; Snapdragon 855
Forward type: **Vulkan** thread=4** precision=2
--------> Benchmarking... loop = 8
Resize error for batchnorm/mul_1, code=2
[ - ] mdepth.mnn max = 276.845ms min = 274.095ms avg = 275.366ms
Vulkan don't support for 151, type=ReLU, Special case
Vulkan don't support for 156, type=ReLU, Special case
Vulkan don't support for 161, type=ReLU, Special case
Vulkan don't support for 166, type=ReLU, Special case
Vulkan don't support for 197, type=ReLU, Special case
Vulkan don't support for 202, type=ReLU, Special case
Vulkan don't support for 207, type=ReLU, Special case
Vulkan don't support for 244, type=ReLU, Special case
Vulkan don't support for 249, type=ReLU, Special case
Vulkan don't support for 286, type=ReLU, Special case
Vulkan don't support for 291, type=ReLU, Special case
Vulkan don't support for 328, type=ReLU, Special case
Vulkan don't support for 333, type=ReLU, Special case
[ - ] prisma.mnn max = 317.457ms min = 277.697ms avg = 297.406ms
Forward type: **OpenCL** thread=4** precision=2
--------> Benchmarking... loop = 8
Don't support type 45, Pad
Don't support type 45, Pad_1
[ - ] mdepth.mnn max = 4484.374ms min = 4456.479ms avg = 4469.937ms
[ - ] prisma.mnn max = 32.153ms min = 26.017ms avg = 28.700ms
Forward type: **N/A** thread=4** precision=2
--------> Benchmarking... loop = 8
gpu type : Adreno (TM) 640
gl version : OpenGL ES 3.2 [email protected] (GIT@e0436c5, Ib165d46056, 1568166573) (Date:09/10/19)
gpu type : Adreno (TM) 640
gl version : OpenGL ES 3.2 [email protected] (GIT@e0436c5, Ib165d46056, 1568166573) (Date:09/10/19)
gpu type : Adreno (TM) 640
gl version : OpenGL ES 3.2 [email protected] (GIT@e0436c5, Ib165d46056, 1568166573) (Date:09/10/19)
[ - ] prisma.mnn max = 86.119ms min = 72.489ms avg = 76.170ms
Forward type: **CPU** thread=4** precision=2
--------> Benchmarking... loop = 8
[ - ] prisma.mnn max = 28.842ms min = 26.734ms avg = 27.242ms
Mi A1; Snapdragon 625
Forward type: **Vulkan** thread=4** precision=2
--------> Benchmarking... loop = 8
Segmentation fault
Forward type: **OpenCL** thread=4** precision=2
--------> Benchmarking... loop = 8
[ - ] prisma.mnn max = 174.921ms min = 173.746ms avg = 174.329ms
Forward type: **N/A** thread=4** precision=2
--------> Benchmarking... loop = 8
gpu type : Adreno (TM) 506
gl version : OpenGL ES 3.2 [email protected] (GIT@2df12b3, I07da2d9908) (Date:10/04/18)
gpu type : Adreno (TM) 506
gl version : OpenGL ES 3.2 [email protected] (GIT@2df12b3, I07da2d9908) (Date:10/04/18)
gpu type : Adreno (TM) 506
gl version : OpenGL ES 3.2 [email protected] (GIT@2df12b3, I07da2d9908) (Date:10/04/18)
[ - ] prisma.mnn max = 1136.323ms min = 1133.633ms avg = 1135.320ms
Forward type: **CPU** thread=4** precision=2
--------> Benchmarking... loop = 8
[ - ] prisma.mnn max = 157.517ms min = 153.530ms avg = 154.697ms