search

Tencent / ncnn

ncnn is a high-performance neural network inference framework optimized for the mobile platform
Other
20.15k stars 4.14k forks source link

ncnn+vulkan benchmark on xiaomi-mix2s(sdm845+adreno630)-armv8 #980

Closed GaussRieman closed 5 years ago

GaussRieman commented 5 years ago

polaris:/data/local/tmp $ ./benchncnn 4 8 0 -1
[0 Adreno (TM) 630] queueC=0 queueT=0 memU=2 memDL=2 memHV=2 [0 Adreno (TM) 630] fp16s=0 fp16a=0 int8s=0 int8a=0 loop_count = 4 num_threads = 8 powersave = 0 gpu_device = -1 squeezenet min = 21.29 max = 23.88 avg = 22.51 squeezenet-int8 min = 19.84 max = 24.12 avg = 22.45 mobilenet min = 26.33 max = 30.23 avg = 28.12 mobilenet-int8 min = 32.68 max = 34.84 avg = 33.66 mobilenet_v2 min = 24.94 max = 26.65 avg = 25.58 shufflenet min = 15.59 max = 20.51 avg = 17.90 mnasnet min = 22.68 max = 25.07 avg = 23.54 proxylessnasnet min = 26.60 max = 29.16 avg = 27.61 googlenet min = 78.21 max = 80.24 avg = 79.11 googlenet-int8 min = 74.48 max = 80.33 avg = 78.01 resnet18 min = 69.21 max = 70.60 avg = 69.94 resnet18-int8 min = 64.66 max = 69.81 avg = 67.01 alexnet min = 105.64 max = 111.90 avg = 108.49 vgg16 min = 322.05 max = 335.43 avg = 326.09 resnet50 min = 331.46 max = 350.55 avg = 338.39 resnet50-int8 min = 150.48 max = 156.43 avg = 153.72 squeezenet-ssd min = 60.20 max = 64.76 avg = 62.96 squeezenet-ssd-int8 min = 59.75 max = 63.49 avg = 61.29 mobilenet-ssd min = 56.82 max = 58.47 avg = 57.76 mobilenet-ssd-int8 min = 66.83 max = 72.32 avg = 69.17 mobilenet-yolo min = 124.87 max = 127.09 avg = 126.10 mobilenet-yolov3 min = 130.62 max = 132.28 avg = 131.36

polaris:/data/local/tmp $ ./benchncnn 4 4 2 -1
[0 Adreno (TM) 630] queueC=0 queueT=0 memU=2 memDL=2 memHV=2 [0 Adreno (TM) 630] fp16s=0 fp16a=0 int8s=0 int8a=0 loop_count = 4 num_threads = 4 powersave = 2 gpu_device = -1 squeezenet min = 16.60 max = 17.62 avg = 17.25 squeezenet-int8 min = 12.54 max = 12.75 avg = 12.64 mobilenet min = 21.44 max = 21.55 avg = 21.50 mobilenet-int8 min = 19.72 max = 19.91 avg = 19.81 mobilenet_v2 min = 15.54 max = 15.61 avg = 15.57 shufflenet min = 9.51 max = 9.61 avg = 9.58 mnasnet min = 14.74 max = 15.01 avg = 14.85 proxylessnasnet min = 17.26 max = 17.46 avg = 17.37 googlenet min = 61.28 max = 62.05 avg = 61.65 googlenet-int8 min = 49.91 max = 50.02 avg = 49.95 resnet18 min = 59.36 max = 60.61 avg = 59.76 resnet18-int8 min = 41.69 max = 42.03 avg = 41.80 alexnet min = 96.46 max = 96.86 avg = 96.68 vgg16 min = 305.06 max = 306.28 avg = 305.97 resnet50 min = 304.90 max = 308.08 avg = 306.77 resnet50-int8 min = 97.94 max = 98.71 avg = 98.30 squeezenet-ssd min = 47.25 max = 50.42 avg = 48.46 squeezenet-ssd-int8 min = 36.13 max = 37.17 avg = 36.51 mobilenet-ssd min = 45.76 max = 45.90 avg = 45.85 mobilenet-ssd-int8 min = 39.57 max = 39.69 avg = 39.64 mobilenet-yolo min = 101.24 max = 101.93 avg = 101.48 mobilenet-yolov3 min = 105.47 max = 105.78 avg = 105.65

polaris:/data/local/tmp $ ./benchncnn 4 4 2 0
[0 Adreno (TM) 630] queueC=0 queueT=0 memU=2 memDL=2 memHV=2 [0 Adreno (TM) 630] fp16s=0 fp16a=0 int8s=0 int8a=0 loop_count = 4 num_threads = 4 powersave = 2 gpu_device = 0 squeezenet min = 106.33 max = 108.36 avg = 107.70 mobilenet min = 166.99 max = 167.89 avg = 167.39 mobilenet_v2 min = 77.64 max = 128.68 avg = 90.76 shufflenet min = 37.26 max = 37.75 avg = 37.41 mnasnet min = 81.79 max = 82.51 avg = 82.22 proxylessnasnet min = 85.97 max = 87.07 avg = 86.62 googlenet min = 375.65 max = 377.53 avg = 376.64 resnet18 min = 451.76 max = 453.45 avg = 452.64 alexnet min = 499.08 max = 506.94 avg = 502.15 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vgg16 min = 1.24 max = 1.32 avg = 1.28 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 resnet50 min = 2.56 max = 2.61 avg = 2.59 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 squeezenet-ssd min = 7.84 max = 9.26 avg = 8.23 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 mobilenet-ssd min = 4.76 max = 4.91 avg = 4.83 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 mobilenet-yolo min = 5.66 max = 6.17 avg = 5.97 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 vkQueueSubmit failed -4 vkWaitForFences failed 2 mobilenet-yolov3 min = 3.94 max = 4.90 avg = 4.59

nihui commented 5 years ago

thank you ! vkQueueSubmit failed -4 error has been fixed in the latest code