vulkan tests,
"NCNN 20230517 - Target: Vulkan GPU - Model: alexnet",
Lower Results Are Better
"a",14.5,15.52,15.64
"b",16.01,15.74,14.99,15.04,16.17,15.31,14.76,15.73,14.02,15.52,15.29,15.45
"c",12.53,13.2,13.47
"NCNN 20230517 - Target: CPU - Model: resnet50",
Lower Results Are Better
"a",34.75,35.72,35.13
"b",38.09,39.91,39.09
"c",36.19,38.77,39.59
"NCNN 20230517 - Target: CPU - Model: regnety_400m",
Lower Results Are Better
"a",42.64,45.94,47.72
"b",47.25,49.43,47.42
"c",46.45,47.61,47.01
"NCNN 20230517 - Target: CPU - Model: vgg16",
Lower Results Are Better
"a",44.25,45.92,46.76
"b",46.86,47.76,47.96
"c",45.72,46.92,48.01
"NCNN 20230517 - Target: Vulkan GPU - Model: vision_transformer",
Lower Results Are Better
"a",77.48,77.14,70.89
"b",77.15,70.89,76.73,70.54,70.72,74.66,69.77,70.08,70.43,70.82,75.09,70.28
"c",71.09,70.5,76.52
"NCNN 20230517 - Target: Vulkan GPU - Model: resnet50",
Lower Results Are Better
"a",39.54,37.94,39.24
"b",40.3,39.52,39.51,40.62,38.87,38.57,40.11,39.42,39.07,41.43,40.2,40.49
"c",39.35,37.33,38.6
"NCNN 20230517 - Target: CPU - Model: mobilenet",
Lower Results Are Better
"a",25.22,25.3,25.12
"b",26.35,25.55,25.45
"c",25.68,26.38,25.83
"NCNN 20230517 - Target: Vulkan GPU - Model: yolov4-tiny",
Lower Results Are Better
"a",37.81,36.05,36.29
"b",38.02,38.43,38.11,38.87,37.36,36.52,36.73,37.38,36.19,38.9,38.56,38.44
"c",36.82,37.38,37.88
"NCNN 20230517 - Target: Vulkan GPU - Model: mobilenet",
Lower Results Are Better
"a",24.43,25.06,24.5
"b",24.94,24.29,26.83,25.66,24.62,23.89,23.55,26.81,25.17,25.68,27.73,23.81
"c",25.44,25.35,25.32
"NCNN 20230517 - Target: CPU - Model: squeezenet_ssd",
Lower Results Are Better
"a",21.87,23.24,21.78
"b",21.9,23.05,21.87
"c",22.84,22.4,23.43
"NCNN 20230517 - Target: CPU - Model: resnet18",
Lower Results Are Better
"a",18.87,19.39,18.96
"b",19.23,19.82,19.15
"c",18.84,18.95,18.93
"NCNN 20230517 - Target: CPU - Model: googlenet",
Lower Results Are Better
"a",30.87,31.24,30.5
"b",31.13,31.86,31.66
"c",30.89,32.3,31.73
"NCNN 20230517 - Target: Vulkan GPU - Model: resnet18",
Lower Results Are Better
"a",19.13,19.32,19.16
"b",19.85,19.05,18.94,19.46,19.67,19.77,19.35,19.56,18.92,19.29,19.44,19.73
"c",18.77,19.01,19.35
"NCNN 20230517 - Target: CPU - Model: yolov4-tiny",
Lower Results Are Better
"a",38.35,37.4,37.41
"b",36.49,37.2,38.46
"c",36.64,37.07,37.71
"NCNN 20230517 - Target: Vulkan GPU - Model: regnety_400m",
Lower Results Are Better
"a",46.78,47.35,46.06
"b",44.08,46.28,46.71,48.41,47.3,43.97,45.3,48.27,45.44,45.51,49.08,44.29
"c",47.39,44.87,46.03
"NCNN 20230517 - Target: Vulkan GPU - Model: vgg16",
Lower Results Are Better
"a",47.5,47.43,47.62
"b",48.35,48.74,48.37,47.78,48.22,47.21,48.22,48.73,47.23,48.22,47.9,48.64
"c",47.52,47.35,47.63
"NCNN 20230517 - Target: Vulkan GPU - Model: squeezenet_ssd",
Lower Results Are Better
"a",22.21,21.7,21.52
"b",21.48,22.21,21.87,22.08,21.77,22.05,21.89,22.24,22.14,22.41,21.63,22.3
"c",22.08,21.88,21.39
"NCNN 20230517 - Target: Vulkan GPU - Model: googlenet",
Lower Results Are Better
"a",31.35,31.77,31.18
"b",30.91,30.91,31.19,32.36,31.02,30.36,30.62,31.77,31.07,31.53,31.76,31.37
"c",31.57,31.55,31.25
"vkpeak 20230730 - fp32-scalar",
Higher Results Are Better
"a",7837.73,7836.51,7835.58
"b",7858.33,7854.8,7855.43
"c",7870.56,7865.83,7864.1
"vkpeak 20230730 - fp32-vec4",
Higher Results Are Better
"a",7796.47,7773.6,7773.94
"b",7822.87,7802.16,7800.65
"c",7825.09,7800.78,7799.57
"vkpeak 20230730 - fp16-vec4",
Higher Results Are Better
"a",12520.84,12514.7,12515.28
"b",12550.07,12549.48,12553.19
"c",12561.81,12557.9,12558.22
"VkFFT 1.2.31 - Test: FFT + iFFT R2C / C2R",
Higher Results Are Better
"a",25673,25448,25455
"b",25456,25443,25443
"c",25447,25454,25451
"vkpeak 20230730 - int16-vec4",
Higher Results Are Better
"a",12532.26,12531.51,12530.28
"b",12550.29,12549.1,12551.76
"c",12565.85,12560.5,12563.08
"VkFFT 1.2.31 - Test: FFT + iFFT C2C 1D batched in half precision",
Higher Results Are Better
"a",101097,100779,100832
"b",101147,101040,101001
"c",101034,100401,101083
"vkpeak 20230730 - fp16-scalar",
Higher Results Are Better
"a",7856.23,7851.62,7853.14
"b",7861.8,7859.54,7859.54
"c",7869.95,7865.16,7863.89
"VkFFT 1.2.31 - Test: FFT + iFFT C2C multidimensional in single precision",
Higher Results Are Better
"a",17090,16917,16971
"b",16952,16989,16987
"c",16981,16949,16972
"VkFFT 1.2.31 - Test: FFT + iFFT C2C Bluestein in single precision",
Higher Results Are Better
"a",7298,7250,7255
"b",7255,7255,7263
"c",7263,7250,7265
"vkpeak 20230730 - fp64-vec4",
Higher Results Are Better
"a",493.49,493.23,493.13
"b",493.02,492.9,492.96
"c",493.76,493.55,493.4
"vkpeak 20230730 - fp64-scalar",
Higher Results Are Better
"a",493.52,493.38,493.19
"b",492.98,492.99,493.06
"c",493.63,493.5,493.39
"vkpeak 20230730 - int32-vec4",
Higher Results Are Better
"a",1575.1,1574.85,1574.5
"b",1574.59,1574.69,1574.51
"c",1576.31,1575.81,1575.6
"vkpeak 20230730 - int32-scalar",
Higher Results Are Better
"a",1214.89,1214.39,1213.96
"b",1213.52,1213.38,1213.54
"c",1214.99,1214.3,1213.91
"VkFFT 1.2.31 - Test: FFT + iFFT C2C Bluestein benchmark in double precision",
Higher Results Are Better
"a",3075,3062,3062
"b",3066,3063,3063
"c",3067,3063,3062
"vkpeak 20230730 - int16-scalar",
Higher Results Are Better
"a",7856.14,7857.01,7855.49
"b",7852.44,7851.01,7852.64
"c",7858.9,7855.76,7856.38
"VkResample 1.0 - Upscale: 2x - Precision: Single",
Lower Results Are Better
"a",12.772,12.769,12.769
"b",12.774,12.771,12.769
"c",12.774,12.776,12.774
"VkFFT 1.2.31 - Test: FFT + iFFT C2C 1D batched in double precision",
Higher Results Are Better
"a",18419,18438,18437
"b",18430,18421,18433
"c",18422,18427,18428
"VkFFT 1.2.31 - Test: FFT + iFFT C2C 1D batched in single precision",
Higher Results Are Better
"a",58945,58941,58945
"b",58954,58944,58944
"c",58937,58936,58940
"VkFFT 1.2.31 - Test: FFT + iFFT C2C 1D batched in single precision, no reshuffling",
Higher Results Are Better
"a",64142,64150,64128
"b",64150,64132,64136
"c",64116,64138,64140
"NCNN 20230517 - Target: Vulkan GPU - Model: FastestDet",
Lower Results Are Better
"a",12.97,17.22,12.55
"b",11.73,13.14,15.76,15.8,14.63,13.69,11.69,18.58,13.98,13.93,18.06,13.34
"c",14.78,14.26,12.6
"NCNN 20230517 - Target: Vulkan GPU - Model: blazeface",
Lower Results Are Better
"a",7.04,7.84,6.77
"b",5.46,6.38,8.28,7.93,7.64,5.39,6.23,9.06,7.2,6.33,9.1,6.07
"c",6.46,6.03,5.76
"NCNN 20230517 - Target: Vulkan GPU - Model: efficientnet-b0",
Lower Results Are Better
"a",17.55,18.95,18.77
"b",15.85,19.18,22.3,23.13,20.66,16.08,18.93,22.69,19.71,16.86,22.68,15.55
"c",17,18.41,18.66
"NCNN 20230517 - Target: Vulkan GPU - Model: mnasnet",
Lower Results Are Better
"a",11.88,18.63,12.03
"b",10.7,11.83,16.17,16.34,14.06,10.62,13.46,17.28,14.85,12.32,16.78,10.25
"c",11.89,15.71,11.2
"NCNN 20230517 - Target: Vulkan GPU - Model: shufflenet-v2",
Lower Results Are Better
"a",12.86,17.09,14.31
"b",12.32,13.66,15.55,16.41,14.88,12.36,14.15,15.35,14.24,13.16,16.91,12.25
"c",12.9,13.36,13.61
"NCNN 20230517 - Target: Vulkan GPU-v3-v3 - Model: mobilenet-v3",
Lower Results Are Better
"a",11.26,15.65,12.01
"b",11.18,12.4,14.64,16.56,13.45,10.83,12.57,14.4,13.96,11.33,17.04,10.68
"c",12.23,13.22,12.27
"NCNN 20230517 - Target: Vulkan GPU-v2-v2 - Model: mobilenet-v2",
Lower Results Are Better
"a",11.89,21.63,13.62
"b",11.32,12.8,17.06,17.17,15.23,11.84,14.84,20.19,15,12.12,18.29,11.54
"c",12.2,14.67,14.01
"NCNN 20230517 - Target: CPU - Model: FastestDet",
Lower Results Are Better
"a",12.62,16.04,15.31
"b",14.51,13.9,12.69
"c",16.99,15.35,15.75
"NCNN 20230517 - Target: CPU - Model: vision_transformer",
Lower Results Are Better
"a",71.11,70.75,78.55
"b",71.68,79.45,71.2
"c",78.78,77.51,71.58
"NCNN 20230517 - Target: CPU - Model: alexnet",
Lower Results Are Better
"a",10.63,11.88,12.78
"b",13.28,13.82,14.82
"c",11.9,12.71,12.51
"NCNN 20230517 - Target: CPU - Model: blazeface",
Lower Results Are Better
"a",4.96,5.73,6.98
"b",7.38,7.97,6.02
"c",6.08,7.42,7.36
"NCNN 20230517 - Target: CPU - Model: efficientnet-b0",
Lower Results Are Better
"a",15.04,18.01,20.94
"b",22.08,20.55,16.44
"c",16.97,19.68,20.52
"NCNN 20230517 - Target: CPU - Model: mnasnet",
Lower Results Are Better
"a",10.15,13.91,15.32
"b",15,11.84,11.86
"c",13.98,15.79,16.03
"NCNN 20230517 - Target: CPU - Model: shufflenet-v2",
Lower Results Are Better
"a",11.98,14.22,16.03
"b",15.35,15.26,12.52
"c",12.86,13.32,16.17
"NCNN 20230517 - Target: CPU-v3-v3 - Model: mobilenet-v3",
Lower Results Are Better
"a",10.76,13.49,13.95
"b",13,15.32,12.03
"c",12.47,13.9,14.43
"NCNN 20230517 - Target: CPU-v2-v2 - Model: mobilenet-v2",
Lower Results Are Better
"a",10.92,13.9,16.66
"b",13.34,17.5,12.64
"c",14.07,16.56,17.41