Skip to content

Instantly share code, notes, and snippets.

@t-vi
Last active October 18, 2018 12:09
Show Gist options
  • Save t-vi/d72ef71a2184b5a9e3197f9a6763105d to your computer and use it in GitHub Desktop.
Save t-vi/d72ef71a2184b5a9e3197f9a6763105d to your computer and use it in GitHub Desktop.
native batch norm
import torch
import pickle
import math
print ("git revision:", torch.__version__)
with open('/tmp/1.0.0a0-058c128_timings.pkl', "rb") as f:
reference_timings_dict = pickle.load(f)
print ("""
input shape = (bs, channels) + features
mode = (training/eval)-(forward+backward)/forward
cudnn = time normal operation (with cudnn enabled)
native = time new native code (cudnn disabled)
slowness = factor (larger = new native slower)
reference = time todayish master thnn without cudnn
slowness = factor (larger = new native slower)
evaluation = <0.8 = good , >1.1 = bad
""")
batch_sizes = [8,16,32,64,128,256,1024,10240]
#batch_sizes = [1024]
channels = [2,32,256,1024]
features = [(1,),
(4,),
(16,),
(32,),
(64,),
(128,),
(256,),
(1024,),
(10240,),
(102400,),
(32,32),
(64,64),
(128,128),
(32,32,32),
(64,64,64)]
#batch_sizes = [128]
#channels = [2]
#features = [(1024,)]
timings_dict = {}
def run_bn(input, running_mean, running_var, weight, bias, training, backward):
out = torch.nn.functional.batch_norm(input, running_mean, running_var, weight, bias, training, 0.1, 1e-5)
if backward:
grads = torch.autograd.grad(out, [input, weight, bias], torch.ones_like(out))
torch.cuda.synchronize()
return out
print ('bs | channels | features | train | cudnn | native | slowness | reference | slowness | evaluation')
print ('-- | -------- | -------- | ----- | ----- | ------ | -------- | --------- | -------- | ----------')
import timeit, gc, numpy
for bs in batch_sizes:
for c in channels:
for f in features:
shape = (bs, c)+f
size = numpy.prod(shape)
if size < 100_000_000:
running_mean = torch.randn(c, device='cuda')
running_var = torch.randn(c, device='cuda').exp()
weight = torch.randn(c, device='cuda', requires_grad=True)
bias = torch.randn(c, device='cuda', requires_grad=True)
input = torch.randn(shape, device='cuda', requires_grad=True)
gc.collect()
for training in [True, False]:
for backward in [True, False]:
out = run_bn(input, running_mean, running_var, weight, bias, training, training)
k = '{} | {} | {} | {}-{}'.format(bs, c, f, ('t' if training else 'e'), ('fb' if backward else 'f'))
torch.cuda.synchronize()
if out.grad_fn.name().startswith("Cudnn"):
res1 = timeit.timeit('run_bn(input, running_mean, running_var, weight, bias, training, backward)', number=100, globals=globals())
else:
res1 = math.nan
with torch.backends.cudnn.flags(enabled=False):
res2 = timeit.timeit('run_bn(input, running_mean, running_var, weight, bias, training, backward)', number=100, globals=globals())
ref1, ref2 = reference_timings_dict.get(k, (math.nan, math.nan))
if res2/ref2 > 1.1:
evaluation = 'bad'
elif res2/ref2 < 0.8:
evaluation = 'good'
else:
evaluation = ''
print ('{} | {:.4f} | {:.4f} | {:.2f} | {:.4f} | {:.2} | {}'.format(k, res1, res2, res2/res1, ref2, res2/ref2, evaluation))
timings_dict[k] = [res1, res2]
v = torch.__version__.replace('+','-')
if 1:
with open('/tmp/'+v+'_timings.pkl', "wb") as f:
pickle.dump(timings_dict, f)
if 0:
for k, (s1, s2) in timings_dict.items():
r1, r2 = reference_timings_dict[k]
if s2 <= r2 *0.8:
print (k, "good {:.4f} {:.4f} {:.2f}".format(s2, r2, s2/r2))
elif s2 >= r2 * 1.1:
print (k, "bad {:.4f} {:.4f} {:.2f}".format(s2, r2, s2/r2))

git revision: 1.0.0a0+3fcf190

input shape = (bs, channels) + features
mode = (training/eval)-(forward+backward)/forward
cudnn = time normal operation (with cudnn enabled)
native = time new native code (cudnn disabled)
slowness = factor (larger = new native slower)
reference = time todayish master thnn without cudnn
slowness = factor (larger = new native slower)
evaluation = <0.8 = good , >1.1 = bad
bs channels features train cudnn native slowness reference slowness evaluation
8 2 (1,) t-fb 0.0152 0.0112 0.74 0.0116 0.96
8 2 (1,) t-f 0.0047 0.0038 0.81 0.0036 1.1
8 2 (1,) e-fb 0.0110 0.0112 1.02 0.0105 1.1
8 2 (1,) e-f 0.0038 0.0031 0.82 0.0028 1.1 bad
8 2 (4,) t-fb 0.0199 0.0122 0.61 0.0113 1.1
8 2 (4,) t-f 0.0048 0.0040 0.82 0.0037 1.1
8 2 (4,) e-fb 0.0110 0.0103 0.94 0.0105 0.99
8 2 (4,) e-f 0.0037 0.0031 0.84 0.0028 1.1 bad
8 2 (16,) t-fb 0.0142 0.0114 0.80 0.0114 1.0
8 2 (16,) t-f 0.0047 0.0042 0.89 0.0037 1.1 bad
8 2 (16,) e-fb 0.0108 0.0100 0.93 0.0113 0.89
8 2 (16,) e-f 0.0037 0.0030 0.82 0.0028 1.1
8 2 (32,) t-fb 0.0135 0.0110 0.81 0.0112 0.99
8 2 (32,) t-f 0.0047 0.0042 0.89 0.0037 1.1 bad
8 2 (32,) e-fb 0.0109 0.0110 1.01 0.0107 1.0
8 2 (32,) e-f 0.0038 0.0031 0.82 0.0028 1.1 bad
8 2 (64,) t-fb 0.0145 0.0114 0.79 0.0112 1.0
8 2 (64,) t-f 0.0048 0.0043 0.89 0.0037 1.2 bad
8 2 (64,) e-fb 0.0114 0.0106 0.93 0.0105 1.0
8 2 (64,) e-f 0.0037 0.0030 0.82 0.0028 1.1
8 2 (128,) t-fb 0.0158 0.0114 0.72 0.0122 0.94
8 2 (128,) t-f 0.0048 0.0043 0.89 0.0037 1.2 bad
8 2 (128,) e-fb 0.0114 0.0104 0.92 0.0104 1.0
8 2 (128,) e-f 0.0038 0.0031 0.82 0.0028 1.1 bad
8 2 (256,) t-fb 0.0151 0.0118 0.78 0.0117 1.0
8 2 (256,) t-f 0.0048 0.0043 0.90 0.0037 1.1 bad
8 2 (256,) e-fb 0.0111 0.0101 0.91 0.0107 0.95
8 2 (256,) e-f 0.0037 0.0030 0.82 0.0029 1.1
8 2 (1024,) t-fb 0.0138 0.0119 0.86 0.0112 1.1
8 2 (1024,) t-f 0.0051 0.0047 0.92 0.0042 1.1 bad
8 2 (1024,) e-fb 0.0114 0.0111 0.98 0.0105 1.1
8 2 (1024,) e-f 0.0038 0.0031 0.81 0.0029 1.1
8 2 (10240,) t-fb 0.0182 0.0237 1.30 0.0221 1.1
8 2 (10240,) t-f 0.0064 0.0106 1.67 0.0126 0.84
8 2 (10240,) e-fb 0.0233 0.0186 0.80 0.0168 1.1 bad
8 2 (10240,) e-f 0.0039 0.0032 0.84 0.0058 0.56 good
8 2 (102400,) t-fb 0.1077 0.2198 2.04 0.2437 0.9
8 2 (102400,) t-f 0.0102 0.0966 9.51 0.1311 0.74 good
8 2 (102400,) e-fb 0.1152 0.1124 0.98 0.1386 0.81
8 2 (102400,) e-f 0.0110 0.0067 0.61 0.0462 0.14 good
8 2 (32, 32) t-fb 0.0159 0.0114 0.72 0.0113 1.0
8 2 (32, 32) t-f 0.0049 0.0044 0.89 0.0039 1.1 bad
8 2 (32, 32) e-fb 0.0117 0.0103 0.88 0.0106 0.97
8 2 (32, 32) e-f 0.0037 0.0031 0.83 0.0028 1.1
8 2 (64, 64) t-fb 0.0174 0.0140 0.80 0.0123 1.1 bad
8 2 (64, 64) t-f 0.0058 0.0059 1.02 0.0062 0.96
8 2 (64, 64) e-fb 0.0132 0.0119 0.90 0.0120 0.99
8 2 (64, 64) e-f 0.0036 0.0030 0.83 0.0035 0.86
8 2 (128, 128) t-fb 0.0160 0.0274 1.71 0.0288 0.95
8 2 (128, 128) t-f 0.0060 0.0123 2.04 0.0148 0.83
8 2 (128, 128) e-fb 0.0262 0.0255 0.97 0.0239 1.1
8 2 (128, 128) e-f 0.0039 0.0034 0.87 0.0065 0.52 good
8 2 (32, 32, 32) t-fb 0.0204 0.0600 2.95 0.0685 0.88
8 2 (32, 32, 32) t-f 0.0070 0.0255 3.62 0.0358 0.71 good
8 2 (32, 32, 32) e-fb 0.0491 0.0494 1.01 0.0486 1.0
8 2 (32, 32, 32) e-f 0.0058 0.0042 0.73 0.0165 0.25 good
8 2 (64, 64, 64) t-fb 0.3794 0.4893 1.29 0.5690 0.86
8 2 (64, 64, 64) t-f 0.1880 0.2288 1.22 0.3232 0.71 good
8 2 (64, 64, 64) e-fb 0.2647 0.2567 0.97 0.3452 0.74 good
8 2 (64, 64, 64) e-f 0.0225 0.0122 0.54 0.1142 0.11 good
8 32 (1,) t-fb 0.0156 0.0122 0.78 0.0112 1.1
8 32 (1,) t-f 0.0048 0.0038 0.80 0.0035 1.1
8 32 (1,) e-fb 0.0117 0.0106 0.90 0.0105 1.0
8 32 (1,) e-f 0.0037 0.0031 0.83 0.0028 1.1 bad
8 32 (4,) t-fb 0.0138 0.0115 0.83 0.0113 1.0
8 32 (4,) t-f 0.0048 0.0039 0.82 0.0035 1.1 bad
8 32 (4,) e-fb 0.0114 0.0112 0.98 0.0106 1.1
8 32 (4,) e-f 0.0036 0.0030 0.84 0.0028 1.1
8 32 (16,) t-fb 0.0154 0.0116 0.75 0.0111 1.0
8 32 (16,) t-f 0.0049 0.0041 0.84 0.0035 1.2 bad
8 32 (16,) e-fb 0.0113 0.0116 1.03 0.0110 1.1
8 32 (16,) e-f 0.0037 0.0031 0.82 0.0029 1.1
8 32 (32,) t-fb 0.0148 0.0116 0.78 0.0116 1.0
8 32 (32,) t-f 0.0048 0.0040 0.84 0.0035 1.1 bad
8 32 (32,) e-fb 0.0115 0.0104 0.91 0.0106 0.99
8 32 (32,) e-f 0.0038 0.0030 0.81 0.0028 1.1
8 32 (64,) t-fb 0.0145 0.0116 0.80 0.0113 1.0
8 32 (64,) t-f 0.0048 0.0041 0.85 0.0035 1.2 bad
8 32 (64,) e-fb 0.0112 0.0102 0.91 0.0106 0.96
8 32 (64,) e-f 0.0037 0.0030 0.82 0.0028 1.1
8 32 (128,) t-fb 0.0145 0.0113 0.78 0.0115 0.98
8 32 (128,) t-f 0.0048 0.0040 0.84 0.0035 1.2 bad
8 32 (128,) e-fb 0.0110 0.0102 0.93 0.0106 0.96
8 32 (128,) e-f 0.0037 0.0030 0.82 0.0028 1.1
8 32 (256,) t-fb 0.0170 0.0116 0.68 0.0113 1.0
8 32 (256,) t-f 0.0050 0.0042 0.84 0.0036 1.2 bad
8 32 (256,) e-fb 0.0113 0.0105 0.93 0.0107 0.99
8 32 (256,) e-f 0.0038 0.0031 0.82 0.0028 1.1
8 32 (1024,) t-fb 0.0157 0.0126 0.80 0.0113 1.1 bad
8 32 (1024,) t-f 0.0050 0.0049 0.98 0.0039 1.2 bad
8 32 (1024,) e-fb 0.0115 0.0105 0.91 0.0114 0.92
8 32 (1024,) e-f 0.0039 0.0033 0.84 0.0027 1.2 bad
8 32 (10240,) t-fb 0.0284 0.0404 1.42 0.0407 0.99
8 32 (10240,) t-f 0.0122 0.0199 1.63 0.0200 0.99
8 32 (10240,) e-fb 0.0307 0.0301 0.98 0.0292 1.0
8 32 (10240,) e-f 0.0093 0.0087 0.93 0.0095 0.91
8 32 (102400,) t-fb 0.2572 0.3782 1.47 0.3850 0.98
8 32 (102400,) t-f 0.0764 0.1661 2.17 0.1804 0.92
8 32 (102400,) e-fb 0.2480 0.2422 0.98 0.2559 0.95
8 32 (102400,) e-f 0.0680 0.0622 0.91 0.0779 0.8 good
8 32 (32, 32) t-fb 0.0155 0.0130 0.84 0.0113 1.2 bad
8 32 (32, 32) t-f 0.0051 0.0049 0.97 0.0040 1.2 bad
8 32 (32, 32) e-fb 0.0123 0.0112 0.91 0.0112 0.99
8 32 (32, 32) e-f 0.0040 0.0034 0.85 0.0028 1.2 bad
8 32 (64, 64) t-fb 0.0222 0.0200 0.90 0.0209 0.95
8 32 (64, 64) t-f 0.0086 0.0105 1.23 0.0101 1.0
8 32 (64, 64) e-fb 0.0203 0.0181 0.89 0.0161 1.1 bad
8 32 (64, 64) e-f 0.0059 0.0054 0.91 0.0054 0.99
8 32 (128, 128) t-fb 0.0378 0.0625 1.65 0.0627 1.0
8 32 (128, 128) t-f 0.0160 0.0296 1.85 0.0301 0.98
8 32 (128, 128) e-fb 0.0425 0.0414 0.98 0.0431 0.96
8 32 (128, 128) e-f 0.0129 0.0121 0.94 0.0139 0.87
8 32 (32, 32, 32) t-fb 0.0678 0.1199 1.77 0.1220 0.98
8 32 (32, 32, 32) t-f 0.0264 0.0550 2.09 0.0580 0.95
8 32 (32, 32, 32) e-fb 0.0791 0.0782 0.99 0.0821 0.95
8 32 (32, 32, 32) e-f 0.0222 0.0211 0.95 0.0256 0.82
8 32 (64, 64, 64) t-fb 0.7344 0.9592 1.31 0.9825 0.98
8 32 (64, 64, 64) t-f 0.2690 0.4145 1.54 0.4537 0.91
8 32 (64, 64, 64) e-fb 0.6342 0.6151 0.97 0.6514 0.94
8 32 (64, 64, 64) e-f 0.1662 0.1475 0.89 0.1952 0.76 good
8 256 (1,) t-fb 0.0151 0.0114 0.75 0.0111 1.0
8 256 (1,) t-f 0.0052 0.0039 0.75 0.0036 1.1
8 256 (1,) e-fb 0.0113 0.0109 0.97 0.0108 1.0
8 256 (1,) e-f 0.0036 0.0030 0.83 0.0029 1.0
8 256 (4,) t-fb 0.0153 0.0126 0.83 0.0116 1.1
8 256 (4,) t-f 0.0051 0.0039 0.76 0.0036 1.1
8 256 (4,) e-fb 0.0114 0.0106 0.93 0.0105 1.0
8 256 (4,) e-f 0.0038 0.0031 0.83 0.0028 1.1 bad
8 256 (16,) t-fb 0.0157 0.0133 0.85 0.0112 1.2 bad
8 256 (16,) t-f 0.0052 0.0041 0.79 0.0035 1.2 bad
8 256 (16,) e-fb 0.0114 0.0105 0.92 0.0105 0.99
8 256 (16,) e-f 0.0038 0.0031 0.82 0.0028 1.1 bad
8 256 (32,) t-fb 0.0153 0.0115 0.75 0.0115 1.0
8 256 (32,) t-f 0.0052 0.0040 0.78 0.0035 1.1 bad
8 256 (32,) e-fb 0.0122 0.0109 0.89 0.0105 1.0
8 256 (32,) e-f 0.0037 0.0031 0.82 0.0028 1.1
8 256 (64,) t-fb 0.0153 0.0116 0.76 0.0116 1.0
8 256 (64,) t-f 0.0053 0.0043 0.82 0.0036 1.2 bad
8 256 (64,) e-fb 0.0109 0.0101 0.93 0.0107 0.95
8 256 (64,) e-f 0.0036 0.0031 0.85 0.0028 1.1
8 256 (128,) t-fb 0.0143 0.0115 0.80 0.0115 0.99
8 256 (128,) t-f 0.0053 0.0047 0.89 0.0037 1.3 bad
8 256 (128,) e-fb 0.0111 0.0104 0.94 0.0105 1.0
8 256 (128,) e-f 0.0036 0.0033 0.91 0.0028 1.2 bad
8 256 (256,) t-fb 0.0156 0.0136 0.87 0.0138 0.99
8 256 (256,) t-f 0.0062 0.0063 1.02 0.0051 1.2 bad
8 256 (256,) e-fb 0.0130 0.0131 1.01 0.0122 1.1
8 256 (256,) e-f 0.0046 0.0042 0.91 0.0037 1.1 bad
8 256 (1024,) t-fb 0.0249 0.0285 1.15 0.0271 1.1
8 256 (1024,) t-f 0.0099 0.0133 1.34 0.0121 1.1 bad
8 256 (1024,) e-fb 0.0252 0.0243 0.96 0.0246 0.99
8 256 (1024,) e-f 0.0077 0.0074 0.96 0.0072 1.0
8 256 (10240,) t-fb 0.1510 0.2376 1.57 0.2336 1.0
8 256 (10240,) t-f 0.0560 0.0972 1.74 0.0988 0.98
8 256 (10240,) e-fb 0.1662 0.1662 1.00 0.1647 1.0
8 256 (10240,) e-f 0.0476 0.0478 1.00 0.0499 0.96
8 256 (32, 32) t-fb 0.0260 0.0282 1.09 0.0272 1.0
8 256 (32, 32) t-f 0.0101 0.0133 1.32 0.0121 1.1
8 256 (32, 32) e-fb 0.0260 0.0250 0.96 0.0251 1.0
8 256 (32, 32) e-f 0.0078 0.0075 0.96 0.0071 1.0
8 256 (64, 64) t-fb 0.0659 0.0974 1.48 0.0962 1.0
8 256 (64, 64) t-f 0.0259 0.0412 1.59 0.0409 1.0
8 256 (64, 64) e-fb 0.0688 0.0685 0.99 0.0685 1.0
8 256 (64, 64) e-f 0.0211 0.0207 0.98 0.0209 0.99
8 256 (128, 128) t-fb 0.2396 0.3898 1.63 0.3798 1.0
8 256 (128, 128) t-f 0.0871 0.1608 1.85 0.1600 1.0
8 256 (128, 128) e-fb 0.2701 0.2729 1.01 0.2668 1.0
8 256 (128, 128) e-f 0.0759 0.0791 1.04 0.0806 0.98
8 256 (32, 32, 32) t-fb 0.4838 0.7981 1.65 0.7843 1.0
8 256 (32, 32, 32) t-f 0.1730 0.3202 1.85 0.3305 0.97
8 256 (32, 32, 32) e-fb 0.5463 0.5541 1.01 0.5491 1.0
8 256 (32, 32, 32) e-f 0.1473 0.1505 1.02 0.1644 0.92
8 1024 (1,) t-fb 0.0183 0.0137 0.75 0.0112 1.2 bad
8 1024 (1,) t-f 0.0071 0.0051 0.72 0.0042 1.2 bad
8 1024 (1,) e-fb 0.0113 0.0106 0.93 0.0114 0.93
8 1024 (1,) e-f 0.0044 0.0034 0.76 0.0028 1.2 bad
8 1024 (4,) t-fb 0.0182 0.0116 0.64 0.0113 1.0
8 1024 (4,) t-f 0.0070 0.0050 0.72 0.0042 1.2 bad
8 1024 (4,) e-fb 0.0112 0.0105 0.93 0.0111 0.94
8 1024 (4,) e-f 0.0044 0.0033 0.74 0.0028 1.2 bad
8 1024 (16,) t-fb 0.0168 0.0138 0.83 0.0115 1.2 bad
8 1024 (16,) t-f 0.0072 0.0051 0.71 0.0042 1.2 bad
8 1024 (16,) e-fb 0.0116 0.0107 0.92 0.0107 1.0
8 1024 (16,) e-f 0.0045 0.0033 0.74 0.0029 1.2 bad
8 1024 (32,) t-fb 0.0174 0.0116 0.67 0.0117 1.0
8 1024 (32,) t-f 0.0072 0.0050 0.70 0.0042 1.2 bad
8 1024 (32,) e-fb 0.0136 0.0106 0.78 0.0113 0.94
8 1024 (32,) e-f 0.0044 0.0033 0.75 0.0029 1.1 bad
8 1024 (64,) t-fb 0.0191 0.0137 0.72 0.0128 1.1
8 1024 (64,) t-f 0.0078 0.0064 0.81 0.0052 1.2 bad
8 1024 (64,) e-fb 0.0142 0.0126 0.89 0.0128 0.99
8 1024 (64,) e-f 0.0047 0.0042 0.89 0.0037 1.1 bad
8 1024 (128,) t-fb 0.0210 0.0172 0.82 0.0160 1.1
8 1024 (128,) t-f 0.0088 0.0086 0.99 0.0070 1.2 bad
8 1024 (128,) e-fb 0.0171 0.0172 1.00 0.0147 1.2 bad
8 1024 (128,) e-f 0.0056 0.0052 0.92 0.0048 1.1
8 1024 (256,) t-fb 0.0245 0.0270 1.10 0.0264 1.0
8 1024 (256,) t-f 0.0103 0.0131 1.27 0.0102 1.3 bad
8 1024 (256,) e-fb 0.0248 0.0233 0.94 0.0235 0.99
8 1024 (256,) e-f 0.0077 0.0074 0.96 0.0070 1.1
8 1024 (1024,) t-fb 0.0621 0.0930 1.50 0.0929 1.0
8 1024 (1024,) t-f 0.0247 0.0391 1.58 0.0384 1.0
8 1024 (1024,) e-fb 0.0681 0.0660 0.97 0.0661 1.0
8 1024 (1024,) e-f 0.0225 0.0207 0.92 0.0202 1.0
8 1024 (10240,) t-fb 0.5787 0.8887 1.54 0.8952 0.99
8 1024 (10240,) t-f 0.2066 0.3623 1.75 0.3676 0.99
8 1024 (10240,) e-fb 0.6233 0.6259 1.00 0.6288 1.0
8 1024 (10240,) e-f 0.1798 0.1831 1.02 0.1872 0.98
8 1024 (32, 32) t-fb 0.0622 0.0939 1.51 0.0930 1.0
8 1024 (32, 32) t-f 0.0247 0.0419 1.69 0.0385 1.1
8 1024 (32, 32) e-fb 0.0681 0.0661 0.97 0.0663 1.0
8 1024 (32, 32) e-f 0.0226 0.0208 0.92 0.0201 1.0
8 1024 (64, 64) t-fb 0.2352 0.3555 1.51 0.3591 0.99
8 1024 (64, 64) t-f 0.0839 0.1456 1.74 0.1472 0.99
8 1024 (64, 64) e-fb 0.2501 0.2500 1.00 0.2518 0.99
8 1024 (64, 64) e-f 0.0739 0.0740 1.00 0.0741 1.0
16 2 (1,) t-fb 0.0166 0.0114 0.68 0.0120 0.94
16 2 (1,) t-f 0.0048 0.0039 0.81 0.0039 1.0
16 2 (1,) e-fb 0.0110 0.0105 0.95 0.0105 0.99
16 2 (1,) e-f 0.0037 0.0031 0.85 0.0028 1.1 bad
16 2 (4,) t-fb 0.0142 0.0114 0.80 0.0113 1.0
16 2 (4,) t-f 0.0048 0.0039 0.81 0.0040 0.98
16 2 (4,) e-fb 0.0111 0.0104 0.94 0.0105 0.99
16 2 (4,) e-f 0.0037 0.0031 0.83 0.0028 1.1 bad
16 2 (16,) t-fb 0.0145 0.0119 0.82 0.0112 1.1
16 2 (16,) t-f 0.0048 0.0046 0.94 0.0040 1.1 bad
16 2 (16,) e-fb 0.0111 0.0103 0.93 0.0106 0.97
16 2 (16,) e-f 0.0037 0.0031 0.83 0.0028 1.1 bad
16 2 (32,) t-fb 0.0137 0.0112 0.82 0.0112 1.0
16 2 (32,) t-f 0.0048 0.0046 0.94 0.0040 1.1 bad
16 2 (32,) e-fb 0.0124 0.0107 0.86 0.0106 1.0
16 2 (32,) e-f 0.0038 0.0031 0.82 0.0028 1.1 bad
16 2 (64,) t-fb 0.0156 0.0115 0.74 0.0112 1.0
16 2 (64,) t-f 0.0049 0.0046 0.94 0.0040 1.2 bad
16 2 (64,) e-fb 0.0114 0.0105 0.93 0.0106 1.0
16 2 (64,) e-f 0.0037 0.0031 0.82 0.0028 1.1
16 2 (128,) t-fb 0.0164 0.0113 0.69 0.0112 1.0
16 2 (128,) t-f 0.0048 0.0046 0.95 0.0040 1.1 bad
16 2 (128,) e-fb 0.0110 0.0103 0.93 0.0106 0.97
16 2 (128,) e-f 0.0037 0.0031 0.85 0.0028 1.1 bad
16 2 (256,) t-fb 0.0143 0.0112 0.78 0.0112 1.0
16 2 (256,) t-f 0.0048 0.0046 0.98 0.0040 1.2 bad
16 2 (256,) e-fb 0.0114 0.0103 0.90 0.0107 0.96
16 2 (256,) e-f 0.0037 0.0031 0.85 0.0028 1.1 bad
16 2 (1024,) t-fb 0.0163 0.0119 0.73 0.0114 1.0
16 2 (1024,) t-f 0.0052 0.0052 1.00 0.0048 1.1
16 2 (1024,) e-fb 0.0117 0.0120 1.03 0.0107 1.1 bad
16 2 (1024,) e-f 0.0037 0.0031 0.85 0.0030 1.0
16 2 (10240,) t-fb 0.0177 0.0397 2.24 0.0404 0.98
16 2 (10240,) t-f 0.0059 0.0148 2.49 0.0180 0.82
16 2 (10240,) e-fb 0.0318 0.0294 0.93 0.0288 1.0
16 2 (10240,) e-f 0.0038 0.0034 0.89 0.0076 0.44 good
16 2 (102400,) t-fb 0.2112 0.3890 1.84 0.4513 0.86
16 2 (102400,) t-f 0.0832 0.1823 2.19 0.2563 0.71 good
16 2 (102400,) e-fb 0.2068 0.2062 1.00 0.2741 0.75 good
16 2 (102400,) e-f 0.0127 0.0107 0.85 0.0909 0.12 good
16 2 (32, 32) t-fb 0.0165 0.0147 0.89 0.0115 1.3 bad
16 2 (32, 32) t-f 0.0052 0.0052 1.00 0.0048 1.1
16 2 (32, 32) e-fb 0.0119 0.0110 0.93 0.0109 1.0
16 2 (32, 32) e-f 0.0037 0.0031 0.85 0.0030 1.1
16 2 (64, 64) t-fb 0.0164 0.0161 0.98 0.0164 0.98
16 2 (64, 64) t-f 0.0061 0.0083 1.36 0.0092 0.9
16 2 (64, 64) e-fb 0.0184 0.0144 0.78 0.0154 0.94
16 2 (64, 64) e-f 0.0037 0.0032 0.87 0.0045 0.71 good
16 2 (128, 128) t-fb 0.0189 0.0616 3.26 0.0692 0.89
16 2 (128, 128) t-f 0.0071 0.0266 3.73 0.0361 0.74 good
16 2 (128, 128) e-fb 0.0492 0.0485 0.99 0.0487 0.99
16 2 (128, 128) e-f 0.0049 0.0045 0.92 0.0167 0.27 good
16 2 (32, 32, 32) t-fb 0.0232 0.1282 5.52 0.1485 0.86
16 2 (32, 32, 32) t-f 0.0081 0.0613 7.60 0.0842 0.73 good
16 2 (32, 32, 32) e-fb 0.0802 0.0786 0.98 0.0914 0.86
16 2 (32, 32, 32) e-f 0.0063 0.0054 0.85 0.0307 0.17 good
16 2 (64, 64, 64) t-fb 0.5270 0.9801 1.86 1.1401 0.86
16 2 (64, 64, 64) t-f 0.2044 0.4561 2.23 0.6485 0.7 good
16 2 (64, 64, 64) e-fb 0.5102 0.5050 0.99 0.6899 0.73 good
16 2 (64, 64, 64) e-f 0.0264 0.0214 0.81 0.2281 0.094 good
16 32 (1,) t-fb 0.0153 0.0113 0.74 0.0113 1.0
16 32 (1,) t-f 0.0047 0.0038 0.81 0.0041 0.93
16 32 (1,) e-fb 0.0112 0.0105 0.94 0.0105 0.99
16 32 (1,) e-f 0.0036 0.0030 0.83 0.0028 1.1
16 32 (4,) t-fb 0.0137 0.0122 0.89 0.0112 1.1
16 32 (4,) t-f 0.0047 0.0038 0.81 0.0040 0.96
16 32 (4,) e-fb 0.0113 0.0104 0.92 0.0106 0.98
16 32 (4,) e-f 0.0036 0.0030 0.84 0.0028 1.1
16 32 (16,) t-fb 0.0141 0.0112 0.79 0.0112 1.0
16 32 (16,) t-f 0.0047 0.0046 0.97 0.0040 1.1 bad
16 32 (16,) e-fb 0.0115 0.0110 0.96 0.0108 1.0
16 32 (16,) e-f 0.0036 0.0031 0.86 0.0028 1.1
16 32 (32,) t-fb 0.0142 0.0112 0.79 0.0112 1.0
16 32 (32,) t-f 0.0048 0.0046 0.97 0.0040 1.1 bad
16 32 (32,) e-fb 0.0119 0.0105 0.88 0.0112 0.93
16 32 (32,) e-f 0.0038 0.0032 0.85 0.0029 1.1 bad
16 32 (64,) t-fb 0.0145 0.0112 0.77 0.0115 0.97
16 32 (64,) t-f 0.0047 0.0046 0.97 0.0041 1.1 bad
16 32 (64,) e-fb 0.0110 0.0102 0.93 0.0109 0.94
16 32 (64,) e-f 0.0037 0.0031 0.85 0.0029 1.1
16 32 (128,) t-fb 0.0149 0.0114 0.76 0.0116 0.98
16 32 (128,) t-f 0.0047 0.0046 0.98 0.0041 1.1 bad
16 32 (128,) e-fb 0.0113 0.0102 0.91 0.0108 0.95
16 32 (128,) e-f 0.0036 0.0031 0.85 0.0029 1.1
16 32 (256,) t-fb 0.0134 0.0111 0.83 0.0111 1.0
16 32 (256,) t-f 0.0047 0.0047 0.99 0.0041 1.1 bad
16 32 (256,) e-fb 0.0109 0.0104 0.96 0.0104 1.0
16 32 (256,) e-f 0.0038 0.0032 0.85 0.0028 1.1 bad
16 32 (1024,) t-fb 0.0166 0.0139 0.84 0.0132 1.1
16 32 (1024,) t-f 0.0064 0.0069 1.08 0.0063 1.1
16 32 (1024,) e-fb 0.0145 0.0136 0.94 0.0136 1.0
16 32 (1024,) e-f 0.0046 0.0041 0.90 0.0039 1.0
16 32 (10240,) t-fb 0.0448 0.0769 1.72 0.0774 0.99
16 32 (10240,) t-f 0.0185 0.0359 1.94 0.0370 0.97
16 32 (10240,) e-fb 0.0506 0.0505 1.00 0.0527 0.96
16 32 (10240,) e-f 0.0146 0.0143 0.98 0.0167 0.85
16 32 (102400,) t-fb 0.5550 0.7627 1.37 0.7768 0.98
16 32 (102400,) t-f 0.1853 0.3325 1.79 0.3603 0.92
16 32 (102400,) e-fb 0.4969 0.4884 0.98 0.5166 0.95
16 32 (102400,) e-f 0.1298 0.1196 0.92 0.1569 0.76 good
16 32 (32, 32) t-fb 0.0215 0.0146 0.68 0.0132 1.1 bad
16 32 (32, 32) t-f 0.0066 0.0073 1.12 0.0063 1.2 bad
16 32 (32, 32) e-fb 0.0170 0.0139 0.82 0.0135 1.0
16 32 (32, 32) e-f 0.0047 0.0044 0.95 0.0040 1.1 bad
16 32 (64, 64) t-fb 0.0279 0.0339 1.22 0.0339 1.0
16 32 (64, 64) t-f 0.0106 0.0170 1.61 0.0168 1.0
16 32 (64, 64) e-fb 0.0285 0.0269 0.94 0.0263 1.0
16 32 (64, 64) e-f 0.0080 0.0076 0.96 0.0082 0.93
16 32 (128, 128) t-fb 0.0677 0.1202 1.78 0.1214 0.99
16 32 (128, 128) t-f 0.0263 0.0554 2.11 0.0577 0.96
16 32 (128, 128) e-fb 0.0782 0.0781 1.00 0.0817 0.96
16 32 (128, 128) e-f 0.0218 0.0214 0.98 0.0255 0.84
16 32 (32, 32, 32) t-fb 0.1294 0.2386 1.84 0.2412 0.99
16 32 (32, 32, 32) t-f 0.0479 0.1066 2.22 0.1140 0.94
16 32 (32, 32, 32) e-fb 0.1542 0.1533 0.99 0.1605 0.96
16 32 (32, 32, 32) e-f 0.0399 0.0392 0.98 0.0491 0.8 good
16 256 (1,) t-fb 0.0156 0.0113 0.73 0.0113 1.0
16 256 (1,) t-f 0.0051 0.0039 0.76 0.0041 0.95
16 256 (1,) e-fb 0.0111 0.0115 1.03 0.0106 1.1
16 256 (1,) e-f 0.0040 0.0032 0.79 0.0028 1.1 bad
16 256 (4,) t-fb 0.0149 0.0125 0.84 0.0112 1.1 bad
16 256 (4,) t-f 0.0052 0.0040 0.77 0.0041 0.99
16 256 (4,) e-fb 0.0113 0.0106 0.93 0.0106 1.0
16 256 (4,) e-f 0.0040 0.0031 0.79 0.0028 1.1 bad
16 256 (16,) t-fb 0.0160 0.0116 0.73 0.0113 1.0
16 256 (16,) t-f 0.0052 0.0047 0.90 0.0041 1.2 bad
16 256 (16,) e-fb 0.0114 0.0105 0.92 0.0108 0.97
16 256 (16,) e-f 0.0039 0.0032 0.81 0.0028 1.1 bad
16 256 (32,) t-fb 0.0153 0.0114 0.74 0.0113 1.0
16 256 (32,) t-f 0.0052 0.0047 0.90 0.0041 1.1 bad
16 256 (32,) e-fb 0.0111 0.0101 0.91 0.0108 0.94
16 256 (32,) e-f 0.0039 0.0031 0.80 0.0028 1.1
16 256 (64,) t-fb 0.0136 0.0118 0.87 0.0113 1.0
16 256 (64,) t-f 0.0053 0.0051 0.95 0.0041 1.2 bad
16 256 (64,) e-fb 0.0112 0.0107 0.95 0.0105 1.0
16 256 (64,) e-f 0.0040 0.0034 0.85 0.0028 1.2 bad
16 256 (128,) t-fb 0.0171 0.0148 0.87 0.0129 1.2 bad
16 256 (128,) t-f 0.0062 0.0064 1.03 0.0052 1.2 bad
16 256 (128,) e-fb 0.0140 0.0125 0.90 0.0122 1.0
16 256 (128,) e-f 0.0046 0.0041 0.90 0.0037 1.1 bad
16 256 (256,) t-fb 0.0193 0.0190 0.99 0.0168 1.1 bad
16 256 (256,) t-f 0.0074 0.0093 1.26 0.0081 1.1 bad
16 256 (256,) e-fb 0.0187 0.0148 0.79 0.0153 0.97
16 256 (256,) e-f 0.0056 0.0055 0.97 0.0050 1.1
16 256 (1024,) t-fb 0.0374 0.0515 1.38 0.0511 1.0
16 256 (1024,) t-f 0.0145 0.0227 1.56 0.0220 1.0
16 256 (1024,) e-fb 0.0378 0.0368 0.97 0.0367 1.0
16 256 (1024,) e-f 0.0120 0.0120 1.00 0.0116 1.0
16 256 (10240,) t-fb 0.2981 0.4834 1.62 0.4732 1.0
16 256 (10240,) t-f 0.1068 0.1973 1.85 0.1994 0.99
16 256 (10240,) e-fb 0.3345 0.3359 1.00 0.3327 1.0
16 256 (10240,) e-f 0.0918 0.0935 1.02 0.0996 0.94
16 256 (32, 32) t-fb 0.0384 0.0518 1.35 0.0511 1.0
16 256 (32, 32) t-f 0.0145 0.0227 1.56 0.0219 1.0
16 256 (32, 32) e-fb 0.0403 0.0380 0.94 0.0373 1.0
16 256 (32, 32) e-f 0.0120 0.0121 1.00 0.0116 1.0
16 256 (64, 64) t-fb 0.1247 0.1904 1.53 0.1868 1.0
16 256 (64, 64) t-f 0.0466 0.0795 1.71 0.0788 1.0
16 256 (64, 64) e-fb 0.1323 0.1325 1.00 0.1314 1.0
16 256 (64, 64) e-f 0.0386 0.0394 1.02 0.0395 1.0
16 256 (128, 128) t-fb 0.4851 0.8003 1.65 0.7808 1.0
16 256 (128, 128) t-f 0.1730 0.3218 1.86 0.3296 0.98
16 256 (128, 128) e-fb 0.5475 0.5539 1.01 0.5485 1.0
16 256 (128, 128) e-f 0.1467 0.1503 1.02 0.1655 0.91
16 1024 (1,) t-fb 0.0210 0.0141 0.67 0.0118 1.2 bad
16 1024 (1,) t-f 0.0071 0.0064 0.90 0.0052 1.2 bad
16 1024 (1,) e-fb 0.0128 0.0118 0.92 0.0113 1.0
16 1024 (1,) e-f 0.0056 0.0039 0.70 0.0032 1.2 bad
16 1024 (4,) t-fb 0.0163 0.0129 0.79 0.0118 1.1
16 1024 (4,) t-f 0.0071 0.0064 0.89 0.0052 1.2 bad
16 1024 (4,) e-fb 0.0133 0.0119 0.89 0.0113 1.1
16 1024 (4,) e-f 0.0056 0.0039 0.70 0.0032 1.2 bad
16 1024 (16,) t-fb 0.0199 0.0133 0.67 0.0127 1.0
16 1024 (16,) t-f 0.0072 0.0063 0.88 0.0052 1.2 bad
16 1024 (16,) e-fb 0.0129 0.0119 0.92 0.0119 1.0
16 1024 (16,) e-f 0.0055 0.0039 0.70 0.0031 1.2 bad
16 1024 (32,) t-fb 0.0182 0.0155 0.85 0.0134 1.2 bad
16 1024 (32,) t-f 0.0079 0.0070 0.89 0.0062 1.1 bad
16 1024 (32,) e-fb 0.0140 0.0128 0.91 0.0128 1.0
16 1024 (32,) e-f 0.0055 0.0040 0.73 0.0040 1.0
16 1024 (64,) t-fb 0.0196 0.0189 0.96 0.0198 0.95
16 1024 (64,) t-f 0.0090 0.0094 1.05 0.0082 1.2 bad
16 1024 (64,) e-fb 0.0175 0.0147 0.84 0.0148 1.0
16 1024 (64,) e-f 0.0058 0.0055 0.94 0.0050 1.1
16 1024 (128,) t-fb 0.0247 0.0284 1.15 0.0281 1.0
16 1024 (128,) t-f 0.0104 0.0135 1.30 0.0121 1.1 bad
16 1024 (128,) e-fb 0.0257 0.0244 0.95 0.0241 1.0
16 1024 (128,) e-f 0.0077 0.0075 0.97 0.0071 1.1
16 1024 (256,) t-fb 0.0353 0.0507 1.43 0.0499 1.0
16 1024 (256,) t-f 0.0154 0.0224 1.46 0.0211 1.1
16 1024 (256,) e-fb 0.0390 0.0366 0.94 0.0364 1.0
16 1024 (256,) e-f 0.0121 0.0119 0.98 0.0113 1.1
16 1024 (1024,) t-fb 0.1184 0.1828 1.54 0.1822 1.0
16 1024 (1024,) t-f 0.0417 0.0755 1.81 0.0750 1.0
16 1024 (1024,) e-fb 0.1320 0.1290 0.98 0.1282 1.0
16 1024 (1024,) e-f 0.0416 0.0386 0.93 0.0380 1.0
16 1024 (32, 32) t-fb 0.1186 0.1830 1.54 0.1823 1.0
16 1024 (32, 32) t-f 0.0420 0.0757 1.80 0.0751 1.0
16 1024 (32, 32) e-fb 0.1326 0.1293 0.97 0.1283 1.0
16 1024 (32, 32) e-f 0.0421 0.0387 0.92 0.0380 1.0
16 1024 (64, 64) t-fb 0.4733 0.7102 1.50 0.7145 0.99
16 1024 (64, 64) t-f 0.1682 0.2895 1.72 0.2930 0.99
16 1024 (64, 64) e-fb 0.4962 0.4988 1.01 0.5002 1.0
16 1024 (64, 64) e-f 0.1431 0.1470 1.03 0.1467 1.0
32 2 (1,) t-fb 0.0179 0.0122 0.68 0.0115 1.1
32 2 (1,) t-f 0.0048 0.0039 0.81 0.0047 0.82
32 2 (1,) e-fb 0.0124 0.0127 1.02 0.0117 1.1
32 2 (1,) e-f 0.0039 0.0032 0.82 0.0029 1.1
32 2 (4,) t-fb 0.0178 0.0128 0.72 0.0121 1.1
32 2 (4,) t-f 0.0049 0.0039 0.81 0.0050 0.78 good
32 2 (4,) e-fb 0.0122 0.0112 0.92 0.0114 0.99
32 2 (4,) e-f 0.0038 0.0031 0.82 0.0031 0.99
32 2 (16,) t-fb 0.0151 0.0120 0.80 0.0116 1.0
32 2 (16,) t-f 0.0048 0.0057 1.18 0.0050 1.1 bad
32 2 (16,) e-fb 0.0120 0.0110 0.92 0.0109 1.0
32 2 (16,) e-f 0.0037 0.0034 0.93 0.0031 1.1
32 2 (32,) t-fb 0.0141 0.0121 0.85 0.0116 1.0
32 2 (32,) t-f 0.0048 0.0056 1.18 0.0049 1.1 bad
32 2 (32,) e-fb 0.0127 0.0109 0.86 0.0109 1.0
32 2 (32,) e-f 0.0037 0.0034 0.94 0.0031 1.1 bad
32 2 (64,) t-fb 0.0158 0.0119 0.76 0.0115 1.0
32 2 (64,) t-f 0.0047 0.0056 1.18 0.0050 1.1 bad
32 2 (64,) e-fb 0.0116 0.0110 0.95 0.0109 1.0
32 2 (64,) e-f 0.0037 0.0035 0.93 0.0031 1.1 bad
32 2 (128,) t-fb 0.0148 0.0122 0.82 0.0116 1.1
32 2 (128,) t-f 0.0049 0.0058 1.16 0.0050 1.2 bad
32 2 (128,) e-fb 0.0118 0.0108 0.91 0.0118 0.91
32 2 (128,) e-f 0.0036 0.0034 0.95 0.0031 1.1 bad
32 2 (256,) t-fb 0.0152 0.0137 0.90 0.0115 1.2 bad
32 2 (256,) t-f 0.0050 0.0059 1.17 0.0049 1.2 bad
32 2 (256,) e-fb 0.0122 0.0112 0.92 0.0108 1.0
32 2 (256,) e-f 0.0038 0.0035 0.92 0.0031 1.1 bad
32 2 (1024,) t-fb 0.0165 0.0163 0.99 0.0127 1.3 bad
32 2 (1024,) t-f 0.0058 0.0068 1.17 0.0065 1.0
32 2 (1024,) e-fb 0.0144 0.0126 0.87 0.0120 1.0
32 2 (1024,) e-f 0.0037 0.0035 0.95 0.0036 0.96
32 2 (10240,) t-fb 0.0172 0.0776 4.52 0.0881 0.88
32 2 (10240,) t-f 0.0073 0.0345 4.73 0.0469 0.74 good
32 2 (10240,) e-fb 0.0573 0.0569 0.99 0.0596 0.96
32 2 (10240,) e-f 0.0049 0.0046 0.94 0.0204 0.23 good
32 2 (102400,) t-fb 0.4176 0.7744 1.85 0.8997 0.86
32 2 (102400,) t-f 0.1616 0.3610 2.23 0.5106 0.71 good
32 2 (102400,) e-fb 0.3995 0.3991 1.00 0.5440 0.73 good
32 2 (102400,) e-f 0.0190 0.0188 0.99 0.1796 0.1 good
32 2 (32, 32) t-fb 0.0173 0.0162 0.94 0.0126 1.3 bad
32 2 (32, 32) t-f 0.0059 0.0069 1.16 0.0065 1.1
32 2 (32, 32) e-fb 0.0169 0.0130 0.77 0.0123 1.1
32 2 (32, 32) e-f 0.0038 0.0036 0.95 0.0036 0.99
32 2 (64, 64) t-fb 0.0170 0.0287 1.69 0.0299 0.96
32 2 (64, 64) t-f 0.0061 0.0130 2.15 0.0154 0.84
32 2 (64, 64) e-fb 0.0271 0.0258 0.95 0.0248 1.0
32 2 (64, 64) e-f 0.0037 0.0036 0.98 0.0069 0.53 good
32 2 (128, 128) t-fb 0.0240 0.1300 5.43 0.1494 0.87
32 2 (128, 128) t-f 0.0081 0.0623 7.66 0.0847 0.74 good
32 2 (128, 128) e-fb 0.0804 0.0793 0.99 0.0917 0.87
32 2 (128, 128) e-f 0.0059 0.0062 1.06 0.0309 0.2 good
32 2 (32, 32, 32) t-fb 0.1029 0.2517 2.45 0.2916 0.86
32 2 (32, 32, 32) t-f 0.0108 0.1187 11.03 0.1656 0.72 good
32 2 (32, 32, 32) e-fb 0.1404 0.1391 0.99 0.1774 0.78 good
32 2 (32, 32, 32) e-f 0.0082 0.0079 0.97 0.0591 0.13 good
32 2 (64, 64, 64) t-fb 1.0514 1.9589 1.86 2.2840 0.86
32 2 (64, 64, 64) t-f 0.4052 0.9108 2.25 1.2975 0.7 good
32 2 (64, 64, 64) e-fb 1.0096 1.0073 1.00 1.3787 0.73 good
32 2 (64, 64, 64) e-f 0.0423 0.0403 0.95 0.4543 0.089 good
32 32 (1,) t-fb 0.0165 0.0120 0.72 0.0117 1.0
32 32 (1,) t-f 0.0047 0.0038 0.81 0.0050 0.76 good
32 32 (1,) e-fb 0.0115 0.0111 0.96 0.0108 1.0
32 32 (1,) e-f 0.0037 0.0030 0.82 0.0031 0.99
32 32 (4,) t-fb 0.0158 0.0121 0.77 0.0131 0.93
32 32 (4,) t-f 0.0047 0.0039 0.83 0.0050 0.79 good
32 32 (4,) e-fb 0.0119 0.0109 0.92 0.0110 0.99
32 32 (4,) e-f 0.0037 0.0031 0.83 0.0031 0.98
32 32 (16,) t-fb 0.0136 0.0149 1.10 0.0118 1.3 bad
32 32 (16,) t-f 0.0047 0.0057 1.21 0.0050 1.1 bad
32 32 (16,) e-fb 0.0115 0.0107 0.93 0.0112 0.96
32 32 (16,) e-f 0.0036 0.0035 0.97 0.0031 1.1 bad
32 32 (32,) t-fb 0.0160 0.0118 0.74 0.0118 1.0
32 32 (32,) t-f 0.0048 0.0057 1.19 0.0050 1.1 bad
32 32 (32,) e-fb 0.0124 0.0107 0.86 0.0108 0.99
32 32 (32,) e-f 0.0037 0.0035 0.96 0.0031 1.1 bad
32 32 (64,) t-fb 0.0134 0.0117 0.88 0.0115 1.0
32 32 (64,) t-f 0.0048 0.0057 1.20 0.0050 1.1 bad
32 32 (64,) e-fb 0.0114 0.0107 0.93 0.0108 0.99
32 32 (64,) e-f 0.0037 0.0035 0.95 0.0031 1.1 bad
32 32 (128,) t-fb 0.0139 0.0118 0.85 0.0117 1.0
32 32 (128,) t-f 0.0047 0.0057 1.21 0.0050 1.1 bad
32 32 (128,) e-fb 0.0115 0.0107 0.93 0.0109 0.99
32 32 (128,) e-f 0.0037 0.0035 0.96 0.0031 1.1 bad
32 32 (256,) t-fb 0.0136 0.0135 0.99 0.0121 1.1 bad
32 32 (256,) t-f 0.0049 0.0058 1.19 0.0050 1.1 bad
32 32 (256,) e-fb 0.0123 0.0113 0.92 0.0111 1.0
32 32 (256,) e-f 0.0039 0.0035 0.90 0.0031 1.1 bad
32 32 (1024,) t-fb 0.0217 0.0205 0.95 0.0200 1.0
32 32 (1024,) t-f 0.0086 0.0111 1.30 0.0103 1.1
32 32 (1024,) e-fb 0.0211 0.0178 0.85 0.0161 1.1 bad
32 32 (1024,) e-f 0.0058 0.0054 0.93 0.0054 1.0
32 32 (10240,) t-fb 0.0830 0.1497 1.80 0.1515 0.99
32 32 (10240,) t-f 0.0326 0.0683 2.09 0.0717 0.95
32 32 (10240,) e-fb 0.0965 0.0967 1.00 0.1014 0.95
32 32 (10240,) e-f 0.0257 0.0259 1.01 0.0312 0.83
32 32 (32, 32) t-fb 0.0226 0.0206 0.91 0.0201 1.0
32 32 (32, 32) t-f 0.0086 0.0112 1.31 0.0103 1.1
32 32 (32, 32) e-fb 0.0221 0.0162 0.73 0.0171 0.95
32 32 (32, 32) e-f 0.0067 0.0055 0.82 0.0054 1.0
32 32 (64, 64) t-fb 0.0385 0.0628 1.63 0.0628 1.0
32 32 (64, 64) t-f 0.0161 0.0302 1.88 0.0304 0.99
32 32 (64, 64) e-fb 0.0422 0.0415 0.98 0.0428 0.97
32 32 (64, 64) e-f 0.0124 0.0125 1.01 0.0139 0.89
32 32 (128, 128) t-fb 0.1294 0.2388 1.85 0.2408 0.99
32 32 (128, 128) t-f 0.0479 0.1075 2.24 0.1134 0.95
32 32 (128, 128) e-fb 0.1533 0.1541 1.00 0.1607 0.96
32 32 (128, 128) e-f 0.0394 0.0402 1.02 0.0489 0.82
32 32 (32, 32, 32) t-fb 0.3236 0.4803 1.48 0.4887 0.98
32 32 (32, 32, 32) t-f 0.0918 0.2114 2.30 0.2273 0.93
32 32 (32, 32, 32) e-fb 0.3072 0.3073 1.00 0.3232 0.95
32 32 (32, 32, 32) e-f 0.0764 0.0764 1.00 0.0971 0.79 good
32 256 (1,) t-fb 0.0155 0.0125 0.81 0.0115 1.1
32 256 (1,) t-f 0.0051 0.0039 0.76 0.0050 0.77 good
32 256 (1,) e-fb 0.0128 0.0121 0.95 0.0109 1.1 bad
32 256 (1,) e-f 0.0045 0.0032 0.70 0.0031 1.0
32 256 (4,) t-fb 0.0154 0.0130 0.85 0.0116 1.1 bad
32 256 (4,) t-f 0.0052 0.0041 0.79 0.0050 0.82
32 256 (4,) e-fb 0.0126 0.0113 0.90 0.0108 1.0
32 256 (4,) e-f 0.0045 0.0031 0.69 0.0031 1.0
32 256 (16,) t-fb 0.0156 0.0126 0.81 0.0133 0.95
32 256 (16,) t-f 0.0052 0.0058 1.11 0.0050 1.2 bad
32 256 (16,) e-fb 0.0127 0.0114 0.90 0.0108 1.1
32 256 (16,) e-f 0.0044 0.0035 0.80 0.0031 1.1 bad
32 256 (32,) t-fb 0.0142 0.0133 0.94 0.0119 1.1 bad
32 256 (32,) t-f 0.0053 0.0058 1.09 0.0050 1.2 bad
32 256 (32,) e-fb 0.0131 0.0118 0.90 0.0123 0.96
32 256 (32,) e-f 0.0045 0.0036 0.80 0.0032 1.1 bad
32 256 (64,) t-fb 0.0161 0.0156 0.97 0.0136 1.1 bad
32 256 (64,) t-f 0.0062 0.0075 1.21 0.0064 1.2 bad
32 256 (64,) e-fb 0.0162 0.0134 0.83 0.0128 1.0
32 256 (64,) e-f 0.0047 0.0046 0.99 0.0040 1.1 bad
32 256 (128,) t-fb 0.0216 0.0202 0.94 0.0173 1.2 bad
32 256 (128,) t-f 0.0074 0.0096 1.29 0.0082 1.2 bad
32 256 (128,) e-fb 0.0188 0.0149 0.79 0.0149 1.0
32 256 (128,) e-f 0.0056 0.0054 0.96 0.0048 1.1 bad
32 256 (256,) t-fb 0.0250 0.0301 1.20 0.0298 1.0
32 256 (256,) t-f 0.0100 0.0148 1.48 0.0133 1.1 bad
32 256 (256,) e-fb 0.0262 0.0253 0.97 0.0255 0.99
32 256 (256,) e-f 0.0078 0.0081 1.05 0.0076 1.1
32 256 (1024,) t-fb 0.0662 0.0993 1.50 0.0969 1.0
32 256 (1024,) t-f 0.0258 0.0423 1.64 0.0415 1.0
32 256 (1024,) e-fb 0.0689 0.0694 1.01 0.0685 1.0
32 256 (1024,) e-f 0.0206 0.0211 1.03 0.0210 1.0
32 256 (10240,) t-fb 0.6028 0.9980 1.66 0.9826 1.0
32 256 (10240,) t-f 0.2193 0.4036 1.84 0.4154 0.97
32 256 (10240,) e-fb 0.6822 0.6892 1.01 0.6853 1.0
32 256 (10240,) e-f 0.1800 0.1869 1.04 0.2050 0.91
32 256 (32, 32) t-fb 0.0666 0.0995 1.50 0.0970 1.0
32 256 (32, 32) t-f 0.0259 0.0424 1.64 0.0416 1.0
32 256 (32, 32) e-fb 0.0695 0.0698 1.00 0.0686 1.0
32 256 (32, 32) e-f 0.0206 0.0212 1.03 0.0210 1.0
32 256 (64, 64) t-fb 0.2406 0.3812 1.58 0.3727 1.0
32 256 (64, 64) t-f 0.0878 0.1567 1.78 0.1578 0.99
32 256 (64, 64) e-fb 0.2639 0.2655 1.01 0.2606 1.0
32 256 (64, 64) e-f 0.0742 0.0760 1.02 0.0778 0.98
32 1024 (1,) t-fb 0.0187 0.0157 0.84 0.0145 1.1
32 1024 (1,) t-f 0.0071 0.0089 1.25 0.0072 1.2 bad
32 1024 (1,) e-fb 0.0178 0.0140 0.78 0.0133 1.0
32 1024 (1,) e-f 0.0078 0.0050 0.64 0.0039 1.3 bad
32 1024 (4,) t-fb 0.0176 0.0160 0.91 0.0134 1.2 bad
32 1024 (4,) t-f 0.0071 0.0089 1.25 0.0071 1.2 bad
32 1024 (4,) e-fb 0.0181 0.0156 0.87 0.0138 1.1 bad
32 1024 (4,) e-f 0.0077 0.0049 0.64 0.0038 1.3 bad
32 1024 (16,) t-fb 0.0179 0.0176 0.98 0.0159 1.1 bad
32 1024 (16,) t-f 0.0080 0.0098 1.23 0.0086 1.1 bad
32 1024 (16,) e-fb 0.0181 0.0149 0.83 0.0146 1.0
32 1024 (16,) e-f 0.0077 0.0051 0.66 0.0051 1.0
32 1024 (32,) t-fb 0.0214 0.0202 0.95 0.0217 0.93
32 1024 (32,) t-f 0.0090 0.0109 1.20 0.0099 1.1
32 1024 (32,) e-fb 0.0195 0.0165 0.85 0.0195 0.85
32 1024 (32,) e-f 0.0078 0.0054 0.70 0.0056 0.97
32 1024 (64,) t-fb 0.0254 0.0300 1.18 0.0299 1.0
32 1024 (64,) t-f 0.0106 0.0149 1.40 0.0133 1.1 bad
32 1024 (64,) e-fb 0.0264 0.0250 0.95 0.0252 0.99
32 1024 (64,) e-f 0.0083 0.0081 0.98 0.0076 1.1
32 1024 (128,) t-fb 0.0369 0.0522 1.42 0.0514 1.0
32 1024 (128,) t-f 0.0153 0.0236 1.54 0.0223 1.1
32 1024 (128,) e-fb 0.0394 0.0375 0.95 0.0368 1.0
32 1024 (128,) e-f 0.0121 0.0124 1.03 0.0117 1.1
32 1024 (256,) t-fb 0.0626 0.0970 1.55 0.0946 1.0
32 1024 (256,) t-f 0.0249 0.0416 1.67 0.0397 1.0
32 1024 (256,) e-fb 0.0680 0.0682 1.00 0.0736 0.93
32 1024 (256,) e-f 0.0207 0.0211 1.02 0.0204 1.0
32 1024 (1024,) t-fb 0.2354 0.3611 1.53 0.3604 1.0
32 1024 (1024,) t-f 0.0852 0.1533 1.80 0.1482 1.0
32 1024 (1024,) e-fb 0.2593 0.2542 0.98 0.2522 1.0
32 1024 (1024,) e-f 0.0806 0.0753 0.93 0.0742 1.0
32 1024 (32, 32) t-fb 0.2364 0.3611 1.53 0.3608 1.0
32 1024 (32, 32) t-f 0.0844 0.1492 1.77 0.1485 1.0
32 1024 (32, 32) e-fb 0.2590 0.2538 0.98 0.2523 1.0
32 1024 (32, 32) e-f 0.0799 0.0753 0.94 0.0743 1.0
64 2 (1,) t-fb 0.0158 0.0155 0.98 0.0134 1.2 bad
64 2 (1,) t-f 0.0048 0.0039 0.81 0.0064 0.6 good
64 2 (1,) e-fb 0.0179 0.0146 0.82 0.0127 1.2 bad
64 2 (1,) e-f 0.0037 0.0031 0.83 0.0033 0.93
64 2 (4,) t-fb 0.0146 0.0155 1.06 0.0134 1.2 bad
64 2 (4,) t-f 0.0049 0.0040 0.81 0.0069 0.57 good
64 2 (4,) e-fb 0.0161 0.0150 0.93 0.0138 1.1
64 2 (4,) e-f 0.0038 0.0032 0.83 0.0038 0.84
64 2 (16,) t-fb 0.0151 0.0148 0.98 0.0137 1.1
64 2 (16,) t-f 0.0049 0.0078 1.58 0.0070 1.1 bad
64 2 (16,) e-fb 0.0162 0.0140 0.86 0.0130 1.1
64 2 (16,) e-f 0.0038 0.0043 1.12 0.0038 1.1 bad
64 2 (32,) t-fb 0.0163 0.0159 0.98 0.0136 1.2 bad
64 2 (32,) t-f 0.0049 0.0078 1.58 0.0070 1.1 bad
64 2 (32,) e-fb 0.0154 0.0134 0.87 0.0126 1.1
64 2 (32,) e-f 0.0038 0.0042 1.11 0.0038 1.1 bad
64 2 (64,) t-fb 0.0157 0.0149 0.94 0.0133 1.1 bad
64 2 (64,) t-f 0.0048 0.0078 1.62 0.0069 1.1 bad
64 2 (64,) e-fb 0.0140 0.0126 0.90 0.0128 0.98
64 2 (64,) e-f 0.0037 0.0042 1.14 0.0038 1.1
64 2 (128,) t-fb 0.0141 0.0154 1.09 0.0137 1.1 bad
64 2 (128,) t-f 0.0049 0.0079 1.61 0.0070 1.1 bad
64 2 (128,) e-fb 0.0145 0.0124 0.86 0.0127 0.98
64 2 (128,) e-f 0.0037 0.0042 1.14 0.0038 1.1 bad
64 2 (256,) t-fb 0.0148 0.0152 1.02 0.0140 1.1
64 2 (256,) t-f 0.0052 0.0080 1.54 0.0069 1.2 bad
64 2 (256,) e-fb 0.0149 0.0134 0.90 0.0131 1.0
64 2 (256,) e-f 0.0037 0.0042 1.14 0.0038 1.1 bad
64 2 (1024,) t-fb 0.0179 0.0208 1.16 0.0181 1.2 bad
64 2 (1024,) t-f 0.0061 0.0102 1.68 0.0099 1.0
64 2 (1024,) e-fb 0.0196 0.0153 0.78 0.0144 1.1
64 2 (1024,) e-f 0.0038 0.0042 1.12 0.0048 0.88
64 2 (10240,) t-fb 0.0241 0.1642 6.82 0.1851 0.89
64 2 (10240,) t-f 0.0095 0.0766 8.08 0.1052 0.73 good
64 2 (10240,) e-fb 0.0954 0.0944 0.99 0.1134 0.83
64 2 (10240,) e-f 0.0066 0.0064 0.96 0.0379 0.17 good
64 2 (102400,) t-fb 0.8262 1.5397 1.86 1.7904 0.86
64 2 (102400,) t-f 0.3178 0.7174 2.26 1.0169 0.71 good
64 2 (102400,) e-fb 0.7969 0.7934 1.00 1.0821 0.73 good
64 2 (102400,) e-f 0.0387 0.0354 0.91 0.3569 0.099 good
64 2 (32, 32) t-fb 0.0175 0.0185 1.06 0.0178 1.0
64 2 (32, 32) t-f 0.0061 0.0101 1.66 0.0099 1.0
64 2 (32, 32) e-fb 0.0208 0.0160 0.77 0.0146 1.1
64 2 (32, 32) e-f 0.0038 0.0042 1.11 0.0048 0.88
64 2 (64, 64) t-fb 0.0174 0.0640 3.69 0.0708 0.9
64 2 (64, 64) t-f 0.0071 0.0283 3.98 0.0369 0.77 good
64 2 (64, 64) e-fb 0.0518 0.0502 0.97 0.0494 1.0
64 2 (64, 64) e-f 0.0047 0.0056 1.18 0.0170 0.33 good
64 2 (128, 128) t-fb 0.1025 0.2539 2.48 0.2923 0.87
64 2 (128, 128) t-f 0.0108 0.1205 11.18 0.1659 0.73 good
64 2 (128, 128) e-fb 0.1406 0.1400 1.00 0.1778 0.79 good
64 2 (128, 128) e-f 0.0086 0.0095 1.11 0.0591 0.16 good
64 2 (32, 32, 32) t-fb 0.2678 0.4965 1.85 0.5757 0.86
64 2 (32, 32, 32) t-f 0.1050 0.2326 2.21 0.3273 0.71 good
64 2 (32, 32, 32) e-fb 0.2601 0.2598 1.00 0.3487 0.74 good
64 2 (32, 32, 32) e-f 0.0138 0.0130 0.94 0.1153 0.11 good
64 2 (64, 64, 64) t-fb 2.0976 3.9118 1.86 4.5779 0.85
64 2 (64, 64, 64) t-f 0.8034 1.8172 2.26 2.5942 0.7 good
64 2 (64, 64, 64) e-fb 2.0266 2.0121 0.99 2.7602 0.73 good
64 2 (64, 64, 64) e-f 0.0930 0.0789 0.85 0.9074 0.087 good
64 32 (1,) t-fb 0.0182 0.0147 0.81 0.0138 1.1
64 32 (1,) t-f 0.0048 0.0039 0.80 0.0069 0.56 good
64 32 (1,) e-fb 0.0141 0.0128 0.91 0.0124 1.0
64 32 (1,) e-f 0.0038 0.0031 0.81 0.0037 0.83
64 32 (4,) t-fb 0.0154 0.0144 0.94 0.0145 0.99
64 32 (4,) t-f 0.0049 0.0040 0.83 0.0070 0.58 good
64 32 (4,) e-fb 0.0162 0.0140 0.86 0.0126 1.1 bad
64 32 (4,) e-f 0.0039 0.0031 0.81 0.0038 0.82
64 32 (16,) t-fb 0.0143 0.0149 1.04 0.0143 1.0
64 32 (16,) t-f 0.0048 0.0079 1.63 0.0069 1.1 bad
64 32 (16,) e-fb 0.0153 0.0125 0.82 0.0129 0.97
64 32 (16,) e-f 0.0037 0.0043 1.16 0.0038 1.1 bad
64 32 (32,) t-fb 0.0143 0.0140 0.98 0.0134 1.0
64 32 (32,) t-f 0.0047 0.0079 1.66 0.0069 1.1 bad
64 32 (32,) e-fb 0.0137 0.0133 0.98 0.0126 1.1
64 32 (32,) e-f 0.0038 0.0044 1.16 0.0038 1.2 bad
64 32 (64,) t-fb 0.0140 0.0142 1.01 0.0134 1.1
64 32 (64,) t-f 0.0048 0.0079 1.66 0.0069 1.1 bad
64 32 (64,) e-fb 0.0141 0.0127 0.90 0.0129 0.99
64 32 (64,) e-f 0.0039 0.0043 1.10 0.0038 1.2 bad
64 32 (128,) t-fb 0.0150 0.0166 1.10 0.0147 1.1 bad
64 32 (128,) t-f 0.0050 0.0080 1.59 0.0069 1.2 bad
64 32 (128,) e-fb 0.0159 0.0134 0.84 0.0129 1.0
64 32 (128,) e-f 0.0040 0.0044 1.10 0.0037 1.2 bad
64 32 (256,) t-fb 0.0183 0.0180 0.98 0.0168 1.1
64 32 (256,) t-f 0.0064 0.0099 1.55 0.0091 1.1
64 32 (256,) e-fb 0.0202 0.0158 0.78 0.0144 1.1
64 32 (256,) e-f 0.0047 0.0056 1.20 0.0050 1.1 bad
64 32 (1024,) t-fb 0.0277 0.0356 1.29 0.0347 1.0
64 32 (1024,) t-f 0.0105 0.0180 1.71 0.0174 1.0
64 32 (1024,) e-fb 0.0284 0.0274 0.97 0.0266 1.0
64 32 (1024,) e-f 0.0080 0.0079 0.99 0.0083 0.95
64 32 (10240,) t-fb 0.1675 0.3000 1.79 0.3032 0.99
64 32 (10240,) t-f 0.0689 0.1339 1.94 0.1420 0.94
64 32 (10240,) e-fb 0.1911 0.1931 1.01 0.2010 0.96
64 32 (10240,) e-f 0.0476 0.0496 1.04 0.0607 0.82
64 32 (32, 32) t-fb 0.0276 0.0356 1.29 0.0348 1.0
64 32 (32, 32) t-f 0.0107 0.0180 1.69 0.0174 1.0
64 32 (32, 32) e-fb 0.0290 0.0277 0.95 0.0266 1.0
64 32 (32, 32) e-f 0.0081 0.0079 0.98 0.0084 0.95
64 32 (64, 64) t-fb 0.0682 0.1212 1.78 0.1211 1.0
64 32 (64, 64) t-f 0.0265 0.0566 2.13 0.0580 0.98
64 32 (64, 64) e-fb 0.0775 0.0793 1.02 0.0812 0.98
64 32 (64, 64) e-f 0.0210 0.0226 1.08 0.0254 0.89
64 32 (128, 128) t-fb 0.3232 0.4804 1.49 0.4865 0.99
64 32 (128, 128) t-f 0.0925 0.2142 2.32 0.2261 0.95
64 32 (128, 128) e-fb 0.3049 0.3091 1.01 0.3218 0.96
64 32 (128, 128) e-f 0.0761 0.0799 1.05 0.0963 0.83
64 32 (32, 32, 32) t-fb 0.7041 0.9727 1.38 0.9904 0.98
64 32 (32, 32, 32) t-f 0.2333 0.4252 1.82 0.4581 0.93
64 32 (32, 32, 32) e-fb 0.6174 0.6239 1.01 0.6544 0.95
64 32 (32, 32, 32) e-f 0.1477 0.1560 1.06 0.1958 0.8 good
64 256 (1,) t-fb 0.0153 0.0178 1.16 0.0142 1.3 bad
64 256 (1,) t-f 0.0052 0.0039 0.75 0.0070 0.56 good
64 256 (1,) e-fb 0.0168 0.0134 0.80 0.0131 1.0
64 256 (1,) e-f 0.0056 0.0032 0.57 0.0038 0.84
64 256 (4,) t-fb 0.0146 0.0157 1.08 0.0146 1.1
64 256 (4,) t-f 0.0052 0.0047 0.90 0.0069 0.67 good
64 256 (4,) e-fb 0.0156 0.0140 0.90 0.0124 1.1 bad
64 256 (4,) e-f 0.0055 0.0035 0.63 0.0038 0.92
64 256 (16,) t-fb 0.0138 0.0172 1.25 0.0154 1.1 bad
64 256 (16,) t-f 0.0054 0.0081 1.50 0.0070 1.1 bad
64 256 (16,) e-fb 0.0173 0.0132 0.76 0.0128 1.0
64 256 (16,) e-f 0.0055 0.0043 0.79 0.0038 1.1 bad
64 256 (32,) t-fb 0.0177 0.0184 1.04 0.0158 1.2 bad
64 256 (32,) t-f 0.0062 0.0097 1.57 0.0088 1.1 bad
64 256 (32,) e-fb 0.0195 0.0183 0.94 0.0167 1.1
64 256 (32,) e-f 0.0056 0.0056 0.99 0.0050 1.1 bad
64 256 (64,) t-fb 0.0191 0.0218 1.14 0.0201 1.1
64 256 (64,) t-f 0.0075 0.0123 1.64 0.0106 1.2 bad
64 256 (64,) e-fb 0.0211 0.0193 0.92 0.0162 1.2 bad
64 256 (64,) e-f 0.0058 0.0064 1.10 0.0055 1.2 bad
64 256 (128,) t-fb 0.0255 0.0302 1.18 0.0288 1.0
64 256 (128,) t-f 0.0100 0.0151 1.51 0.0133 1.1 bad
64 256 (128,) e-fb 0.0264 0.0247 0.94 0.0246 1.0
64 256 (128,) e-f 0.0077 0.0079 1.02 0.0072 1.1
64 256 (256,) t-fb 0.0368 0.0548 1.49 0.0559 0.98
64 256 (256,) t-f 0.0146 0.0262 1.80 0.0247 1.1
64 256 (256,) e-fb 0.0389 0.0388 1.00 0.0404 0.96
64 256 (256,) e-f 0.0120 0.0136 1.13 0.0127 1.1
64 256 (1024,) t-fb 0.1246 0.1949 1.56 0.1889 1.0
64 256 (1024,) t-f 0.0466 0.0817 1.75 0.0802 1.0
64 256 (1024,) e-fb 0.1335 0.1354 1.01 0.1327 1.0
64 256 (1024,) e-f 0.0379 0.0399 1.05 0.0399 1.0
64 256 (32, 32) t-fb 0.1247 0.1953 1.57 0.1886 1.0
64 256 (32, 32) t-f 0.0468 0.0819 1.75 0.0800 1.0
64 256 (32, 32) e-fb 0.1334 0.1354 1.01 0.1326 1.0
64 256 (32, 32) e-f 0.0379 0.0399 1.05 0.0398 1.0
64 256 (64, 64) t-fb 0.4918 0.7860 1.60 0.7708 1.0
64 256 (64, 64) t-f 0.1754 0.3193 1.82 0.3274 0.98
64 256 (64, 64) e-fb 0.5373 0.5451 1.01 0.5364 1.0
64 256 (64, 64) e-f 0.1426 0.1513 1.06 0.1591 0.95
64 1024 (1,) t-fb 0.0180 0.0259 1.44 0.0209 1.2 bad
64 1024 (1,) t-f 0.0071 0.0140 1.97 0.0111 1.3 bad
64 1024 (1,) e-fb 0.0253 0.0241 0.96 0.0161 1.5 bad
64 1024 (1,) e-f 0.0122 0.0070 0.58 0.0053 1.3 bad
64 1024 (4,) t-fb 0.0170 0.0263 1.55 0.0226 1.2 bad
64 1024 (4,) t-f 0.0072 0.0140 1.93 0.0110 1.3 bad
64 1024 (4,) e-fb 0.0259 0.0252 0.97 0.0188 1.3 bad
64 1024 (4,) e-f 0.0120 0.0071 0.59 0.0052 1.4 bad
64 1024 (16,) t-fb 0.0228 0.0319 1.40 0.0300 1.1
64 1024 (16,) t-f 0.0091 0.0170 1.87 0.0155 1.1
64 1024 (16,) e-fb 0.0276 0.0264 0.96 0.0254 1.0
64 1024 (16,) e-f 0.0120 0.0073 0.61 0.0076 0.96
64 1024 (32,) t-fb 0.0258 0.0360 1.40 0.0354 1.0
64 1024 (32,) t-f 0.0107 0.0178 1.67 0.0167 1.1
64 1024 (32,) e-fb 0.0299 0.0278 0.93 0.0273 1.0
64 1024 (32,) e-f 0.0120 0.0079 0.65 0.0087 0.91
64 1024 (64,) t-fb 0.0363 0.0557 1.53 0.0566 0.98
64 1024 (64,) t-f 0.0155 0.0265 1.71 0.0252 1.1
64 1024 (64,) e-fb 0.0393 0.0398 1.01 0.0400 0.99
64 1024 (64,) e-f 0.0130 0.0136 1.05 0.0127 1.1
64 1024 (128,) t-fb 0.0633 0.1012 1.60 0.0982 1.0
64 1024 (128,) t-f 0.0253 0.0448 1.77 0.0422 1.1
64 1024 (128,) e-fb 0.0700 0.0714 1.02 0.0690 1.0
64 1024 (128,) e-f 0.0208 0.0226 1.09 0.0212 1.1
64 1024 (256,) t-fb 0.1194 0.1898 1.59 0.1840 1.0
64 1024 (256,) t-f 0.0419 0.0795 1.90 0.0770 1.0
64 1024 (256,) e-fb 0.1325 0.1341 1.01 0.1294 1.0
64 1024 (256,) e-f 0.0380 0.0398 1.05 0.0387 1.0
64 1024 (1024,) t-fb 0.4735 0.7219 1.52 0.7196 1.0
64 1024 (1024,) t-f 0.1691 0.2987 1.77 0.2953 1.0
64 1024 (1024,) e-fb 0.5146 0.5079 0.99 0.5031 1.0
64 1024 (1024,) e-f 0.1572 0.1505 0.96 0.1476 1.0
64 1024 (32, 32) t-fb 0.4740 0.7226 1.52 0.7177 1.0
64 1024 (32, 32) t-f 0.1695 0.2992 1.77 0.2952 1.0
64 1024 (32, 32) e-fb 0.5129 0.5071 0.99 0.5018 1.0
64 1024 (32, 32) e-f 0.1565 0.1510 0.97 0.1474 1.0
128 2 (1,) t-fb 0.0198 0.0210 1.06 0.0195 1.1
128 2 (1,) t-f 0.0048 0.0038 0.81 0.0099 0.39 good
128 2 (1,) e-fb 0.0224 0.0199 0.89 0.0150 1.3 bad
128 2 (1,) e-f 0.0037 0.0031 0.82 0.0043 0.71 good
128 2 (4,) t-fb 0.0138 0.0208 1.51 0.0199 1.0
128 2 (4,) t-f 0.0047 0.0039 0.81 0.0108 0.36 good
128 2 (4,) e-fb 0.0213 0.0194 0.91 0.0152 1.3 bad
128 2 (4,) e-f 0.0037 0.0031 0.83 0.0051 0.6 good
128 2 (16,) t-fb 0.0147 0.0213 1.45 0.0198 1.1
128 2 (16,) t-f 0.0048 0.0120 2.52 0.0108 1.1 bad
128 2 (16,) e-fb 0.0191 0.0169 0.89 0.0152 1.1 bad
128 2 (16,) e-f 0.0036 0.0057 1.55 0.0051 1.1 bad
128 2 (32,) t-fb 0.0143 0.0212 1.48 0.0201 1.1
128 2 (32,) t-f 0.0047 0.0120 2.55 0.0107 1.1 bad
128 2 (32,) e-fb 0.0213 0.0166 0.78 0.0149 1.1 bad
128 2 (32,) e-f 0.0037 0.0057 1.54 0.0051 1.1 bad
128 2 (64,) t-fb 0.0140 0.0213 1.52 0.0199 1.1
128 2 (64,) t-f 0.0049 0.0119 2.45 0.0107 1.1 bad
128 2 (64,) e-fb 0.0204 0.0162 0.79 0.0149 1.1
128 2 (64,) e-f 0.0037 0.0057 1.54 0.0051 1.1 bad
128 2 (128,) t-fb 0.0154 0.0212 1.38 0.0209 1.0
128 2 (128,) t-f 0.0052 0.0122 2.35 0.0108 1.1 bad
128 2 (128,) e-fb 0.0205 0.0155 0.76 0.0149 1.0
128 2 (128,) e-f 0.0036 0.0057 1.57 0.0051 1.1 bad
128 2 (256,) t-fb 0.0160 0.0225 1.41 0.0205 1.1
128 2 (256,) t-f 0.0058 0.0123 2.14 0.0108 1.1 bad
128 2 (256,) e-fb 0.0206 0.0168 0.82 0.0151 1.1 bad
128 2 (256,) e-f 0.0036 0.0057 1.56 0.0051 1.1 bad
128 2 (1024,) t-fb 0.0165 0.0339 2.06 0.0321 1.1
128 2 (1024,) t-f 0.0060 0.0166 2.78 0.0168 0.99
128 2 (1024,) e-fb 0.0294 0.0273 0.93 0.0258 1.1
128 2 (1024,) e-f 0.0039 0.0057 1.45 0.0072 0.79 good
128 2 (10240,) t-fb 0.1706 0.3169 1.86 0.3641 0.87
128 2 (10240,) t-f 0.0679 0.1494 2.20 0.2068 0.72 good
128 2 (10240,) e-fb 0.1706 0.1703 1.00 0.2211 0.77 good
128 2 (10240,) e-f 0.0092 0.0099 1.07 0.0734 0.13 good
128 2 (102400,) t-fb 1.6522 3.0860 1.87 3.5714 0.86
128 2 (102400,) t-f 0.6332 1.4363 2.27 2.0299 0.71 good
128 2 (102400,) e-fb 1.5878 1.5912 1.00 2.1575 0.74 good
128 2 (102400,) e-f 0.0669 0.0693 1.04 0.7112 0.097 good
128 2 (32, 32) t-fb 0.0198 0.0338 1.71 0.0323 1.0
128 2 (32, 32) t-f 0.0061 0.0167 2.74 0.0169 0.99
128 2 (32, 32) e-fb 0.0292 0.0276 0.94 0.0255 1.1
128 2 (32, 32) e-f 0.0040 0.0057 1.44 0.0072 0.8
128 2 (64, 64) t-fb 0.0227 0.1359 5.99 0.1517 0.9
128 2 (64, 64) t-f 0.0081 0.0661 8.11 0.0862 0.77 good
128 2 (64, 64) e-fb 0.0812 0.0818 1.01 0.0933 0.88
128 2 (64, 64) e-f 0.0058 0.0084 1.45 0.0314 0.27 good
128 2 (128, 128) t-fb 0.2691 0.5035 1.87 0.5765 0.87
128 2 (128, 128) t-f 0.1056 0.2379 2.25 0.3277 0.73 good
128 2 (128, 128) e-fb 0.2600 0.2615 1.01 0.3489 0.75 good
128 2 (128, 128) e-f 0.0128 0.0165 1.29 0.1153 0.14 good
128 2 (32, 32, 32) t-fb 0.5313 0.9906 1.86 1.1454 0.86
128 2 (32, 32, 32) t-f 0.2059 0.4621 2.24 0.6516 0.71 good
128 2 (32, 32, 32) e-fb 0.5096 0.5103 1.00 0.6923 0.74 good
128 2 (32, 32, 32) e-f 0.0222 0.0231 1.04 0.2286 0.1 good
128 2 (64, 64, 64) t-fb 4.1995 7.8576 1.87 9.1219 0.86
128 2 (64, 64, 64) t-f 1.6182 3.6529 2.26 5.1926 0.7 good
128 2 (64, 64, 64) e-fb 4.0444 4.0363 1.00 5.5181 0.73 good
128 2 (64, 64, 64) e-f 0.1669 0.1615 0.97 1.8197 0.089 good
128 32 (1,) t-fb 0.0158 0.0237 1.50 0.0198 1.2 bad
128 32 (1,) t-f 0.0048 0.0039 0.80 0.0107 0.36 good
128 32 (1,) e-fb 0.0214 0.0173 0.81 0.0181 0.95
128 32 (1,) e-f 0.0042 0.0030 0.73 0.0051 0.6 good
128 32 (4,) t-fb 0.0164 0.0202 1.23 0.0191 1.1
128 32 (4,) t-f 0.0048 0.0040 0.84 0.0108 0.37 good
128 32 (4,) e-fb 0.0193 0.0157 0.81 0.0151 1.0
128 32 (4,) e-f 0.0042 0.0031 0.73 0.0051 0.6 good
128 32 (16,) t-fb 0.0148 0.0214 1.45 0.0202 1.1
128 32 (16,) t-f 0.0047 0.0122 2.60 0.0108 1.1 bad
128 32 (16,) e-fb 0.0204 0.0169 0.83 0.0165 1.0
128 32 (16,) e-f 0.0041 0.0060 1.44 0.0051 1.2 bad
128 32 (32,) t-fb 0.0139 0.0225 1.62 0.0194 1.2 bad
128 32 (32,) t-f 0.0047 0.0122 2.57 0.0108 1.1 bad
128 32 (32,) e-fb 0.0214 0.0153 0.71 0.0160 0.96
128 32 (32,) e-f 0.0042 0.0059 1.42 0.0051 1.2 bad
128 32 (64,) t-fb 0.0139 0.0229 1.65 0.0210 1.1
128 32 (64,) t-f 0.0050 0.0122 2.47 0.0108 1.1 bad
128 32 (64,) e-fb 0.0208 0.0167 0.80 0.0157 1.1
128 32 (64,) e-f 0.0045 0.0059 1.30 0.0051 1.2 bad
128 32 (128,) t-fb 0.0174 0.0268 1.53 0.0257 1.0
128 32 (128,) t-f 0.0065 0.0151 2.33 0.0136 1.1 bad
128 32 (128,) e-fb 0.0254 0.0243 0.96 0.0236 1.0
128 32 (128,) e-f 0.0046 0.0079 1.69 0.0070 1.1 bad
128 32 (256,) t-fb 0.0228 0.0323 1.41 0.0305 1.1
128 32 (256,) t-f 0.0086 0.0180 2.10 0.0164 1.1
128 32 (256,) e-fb 0.0272 0.0259 0.95 0.0248 1.0
128 32 (256,) e-f 0.0059 0.0083 1.41 0.0075 1.1 bad
128 32 (1024,) t-fb 0.0384 0.0659 1.71 0.0641 1.0
128 32 (1024,) t-f 0.0162 0.0320 1.97 0.0315 1.0
128 32 (1024,) e-fb 0.0457 0.0454 0.99 0.0435 1.0
128 32 (1024,) e-f 0.0126 0.0130 1.03 0.0142 0.91
128 32 (10240,) t-fb 0.4407 0.6069 1.38 0.6146 0.99
128 32 (10240,) t-f 0.1486 0.2683 1.81 0.2855 0.94
128 32 (10240,) e-fb 0.3823 0.3897 1.02 0.4062 0.96
128 32 (10240,) e-f 0.0915 0.0989 1.08 0.1216 0.81
128 32 (32, 32) t-fb 0.0389 0.0662 1.70 0.0641 1.0
128 32 (32, 32) t-f 0.0165 0.0321 1.95 0.0315 1.0
128 32 (32, 32) e-fb 0.0456 0.0449 0.99 0.0435 1.0
128 32 (32, 32) e-f 0.0126 0.0130 1.03 0.0142 0.91
128 32 (64, 64) t-fb 0.1295 0.2411 1.86 0.2414 1.0
128 32 (64, 64) t-f 0.0483 0.1106 2.29 0.1142 0.97
128 32 (64, 64) e-fb 0.1517 0.1563 1.03 0.1598 0.98
128 32 (64, 64) e-f 0.0384 0.0431 1.12 0.0486 0.89
128 32 (128, 128) t-fb 0.7012 0.9766 1.39 0.9857 0.99
128 32 (128, 128) t-f 0.2339 0.4320 1.85 0.4558 0.95
128 32 (128, 128) e-fb 0.6142 0.6299 1.03 0.6518 0.97
128 32 (128, 128) e-f 0.1467 0.1625 1.11 0.1947 0.83
128 256 (1,) t-fb 0.0181 0.0236 1.30 0.0205 1.2 bad
128 256 (1,) t-f 0.0052 0.0040 0.77 0.0108 0.37 good
128 256 (1,) e-fb 0.0223 0.0178 0.80 0.0153 1.2 bad
128 256 (1,) e-f 0.0077 0.0030 0.39 0.0052 0.59 good
128 256 (4,) t-fb 0.0196 0.0234 1.19 0.0202 1.2 bad
128 256 (4,) t-f 0.0052 0.0055 1.05 0.0108 0.5 good
128 256 (4,) e-fb 0.0217 0.0176 0.81 0.0156 1.1 bad
128 256 (4,) e-f 0.0077 0.0039 0.51 0.0051 0.76 good
128 256 (16,) t-fb 0.0167 0.0278 1.66 0.0264 1.1
128 256 (16,) t-f 0.0062 0.0148 2.38 0.0135 1.1
128 256 (16,) e-fb 0.0260 0.0248 0.96 0.0244 1.0
128 256 (16,) e-f 0.0077 0.0079 1.03 0.0071 1.1 bad
128 256 (32,) t-fb 0.0179 0.0314 1.75 0.0301 1.0
128 256 (32,) t-f 0.0075 0.0179 2.37 0.0163 1.1
128 256 (32,) e-fb 0.0264 0.0255 0.97 0.0249 1.0
128 256 (32,) e-f 0.0078 0.0084 1.08 0.0075 1.1 bad
128 256 (64,) t-fb 0.0258 0.0380 1.48 0.0356 1.1
128 256 (64,) t-f 0.0101 0.0204 2.01 0.0180 1.1 bad
128 256 (64,) e-fb 0.0301 0.0293 0.97 0.0271 1.1
128 256 (64,) e-f 0.0082 0.0099 1.20 0.0086 1.1 bad
128 256 (128,) t-fb 0.0374 0.0550 1.47 0.0529 1.0
128 256 (128,) t-f 0.0145 0.0262 1.81 0.0235 1.1 bad
128 256 (128,) e-fb 0.0388 0.0382 0.98 0.0383 1.0
128 256 (128,) e-f 0.0120 0.0129 1.07 0.0120 1.1
128 256 (256,) t-fb 0.0666 0.1080 1.62 0.1098 0.98
128 256 (256,) t-f 0.0258 0.0504 1.95 0.0492 1.0
128 256 (256,) e-fb 0.0717 0.0773 1.08 0.0770 1.0
128 256 (256,) e-f 0.0206 0.0259 1.26 0.0234 1.1 bad
128 256 (1024,) t-fb 0.2405 0.3909 1.63 0.3787 1.0
128 256 (1024,) t-f 0.0879 0.1620 1.84 0.1605 1.0
128 256 (1024,) e-fb 0.2669 0.2724 1.02 0.2641 1.0
128 256 (1024,) e-f 0.0722 0.0776 1.07 0.0787 0.99
128 256 (32, 32) t-fb 0.2411 0.3910 1.62 0.3789 1.0
128 256 (32, 32) t-f 0.0886 0.1624 1.83 0.1611 1.0
128 256 (32, 32) e-fb 0.2679 0.2722 1.02 0.2640 1.0
128 256 (32, 32) e-f 0.0730 0.0779 1.07 0.0786 0.99
128 1024 (1,) t-fb 0.0179 0.0450 2.51 0.0342 1.3 bad
128 1024 (1,) t-f 0.0072 0.0241 3.34 0.0188 1.3 bad
128 1024 (1,) e-fb 0.0398 0.0338 0.85 0.0270 1.3 bad
128 1024 (1,) e-f 0.0207 0.0112 0.54 0.0079 1.4 bad
128 1024 (4,) t-fb 0.0182 0.0494 2.71 0.0442 1.1 bad
128 1024 (4,) t-f 0.0079 0.0260 3.30 0.0226 1.2 bad
128 1024 (4,) e-fb 0.0421 0.0374 0.89 0.0336 1.1 bad
128 1024 (4,) e-f 0.0206 0.0111 0.54 0.0111 1.0
128 1024 (16,) t-fb 0.0255 0.0617 2.42 0.0587 1.0
128 1024 (16,) t-f 0.0108 0.0298 2.77 0.0282 1.1
128 1024 (16,) e-fb 0.0480 0.0439 0.92 0.0416 1.1
128 1024 (16,) e-f 0.0206 0.0114 0.55 0.0128 0.89
128 1024 (32,) t-fb 0.0372 0.0689 1.85 0.0691 1.0
128 1024 (32,) t-f 0.0156 0.0317 2.03 0.0325 0.98
128 1024 (32,) e-fb 0.0536 0.0460 0.86 0.0471 0.98
128 1024 (32,) e-f 0.0206 0.0129 0.62 0.0151 0.86
128 1024 (64,) t-fb 0.0636 0.1109 1.74 0.1141 0.97
128 1024 (64,) t-f 0.0251 0.0515 2.05 0.0515 1.0
128 1024 (64,) e-fb 0.0741 0.0778 1.05 0.0767 1.0
128 1024 (64,) e-f 0.0224 0.0260 1.16 0.0234 1.1 bad
128 1024 (128,) t-fb 0.1192 0.2032 1.70 0.1963 1.0
128 1024 (128,) t-f 0.0418 0.0867 2.07 0.0844 1.0
128 1024 (128,) e-fb 0.1372 0.1422 1.04 0.1357 1.0
128 1024 (128,) e-f 0.0379 0.0430 1.13 0.0404 1.1
128 1024 (256,) t-fb 0.2371 0.3753 1.58 0.3627 1.0
128 1024 (256,) t-f 0.0849 0.1570 1.85 0.1511 1.0
128 1024 (256,) e-fb 0.2600 0.2647 1.02 0.2541 1.0
128 1024 (256,) e-f 0.0730 0.0779 1.07 0.0759 1.0
256 2 (1,) t-fb 0.0152 0.0294 1.94 0.0314 0.94
256 2 (1,) t-f 0.0047 0.0038 0.80 0.0169 0.22 good
256 2 (1,) e-fb 0.0287 0.0279 0.97 0.0264 1.1
256 2 (1,) e-f 0.0040 0.0033 0.83 0.0062 0.54 good
256 2 (4,) t-fb 0.0164 0.0301 1.83 0.0327 0.92
256 2 (4,) t-f 0.0049 0.0039 0.79 0.0185 0.21 good
256 2 (4,) e-fb 0.0290 0.0278 0.96 0.0265 1.0
256 2 (4,) e-f 0.0037 0.0030 0.83 0.0078 0.39 good
256 2 (16,) t-fb 0.0191 0.0356 1.86 0.0330 1.1
256 2 (16,) t-f 0.0048 0.0205 4.30 0.0186 1.1
256 2 (16,) e-fb 0.0295 0.0284 0.96 0.0274 1.0
256 2 (16,) e-f 0.0037 0.0087 2.33 0.0078 1.1 bad
256 2 (32,) t-fb 0.0153 0.0347 2.27 0.0324 1.1
256 2 (32,) t-f 0.0049 0.0204 4.13 0.0185 1.1 bad
256 2 (32,) e-fb 0.0284 0.0272 0.96 0.0259 1.1
256 2 (32,) e-f 0.0037 0.0087 2.33 0.0078 1.1 bad
256 2 (64,) t-fb 0.0181 0.0349 1.93 0.0323 1.1
256 2 (64,) t-f 0.0054 0.0205 3.82 0.0184 1.1 bad
256 2 (64,) e-fb 0.0285 0.0281 0.99 0.0259 1.1
256 2 (64,) e-f 0.0037 0.0087 2.33 0.0078 1.1 bad
256 2 (128,) t-fb 0.0212 0.0356 1.68 0.0324 1.1
256 2 (128,) t-f 0.0059 0.0209 3.57 0.0185 1.1 bad
256 2 (128,) e-fb 0.0285 0.0272 0.96 0.0256 1.1
256 2 (128,) e-f 0.0037 0.0087 2.35 0.0078 1.1 bad
256 2 (256,) t-fb 0.0170 0.0369 2.17 0.0327 1.1 bad
256 2 (256,) t-f 0.0060 0.0210 3.49 0.0186 1.1 bad
256 2 (256,) e-fb 0.0296 0.0280 0.95 0.0261 1.1
256 2 (256,) e-f 0.0038 0.0087 2.29 0.0078 1.1 bad
256 2 (1024,) t-fb 0.0171 0.0753 4.40 0.0753 1.0
256 2 (1024,) t-f 0.0071 0.0361 5.11 0.0394 0.92
256 2 (1024,) e-fb 0.0548 0.0542 0.99 0.0519 1.0
256 2 (1024,) e-f 0.0046 0.0119 2.60 0.0178 0.67 good
256 2 (10240,) t-fb 0.3342 0.6268 1.88 0.7223 0.87
256 2 (10240,) t-f 0.1304 0.2937 2.25 0.4101 0.72 good
256 2 (10240,) e-fb 0.3220 0.3240 1.01 0.4368 0.74 good
256 2 (10240,) e-f 0.0146 0.0167 1.15 0.1442 0.12 good
256 2 (102400,) t-fb 3.2930 6.1654 1.87 7.1333 0.86
256 2 (102400,) t-f 1.2685 2.8756 2.27 4.0608 0.71 good
256 2 (102400,) e-fb 3.1717 3.1867 1.00 4.3148 0.74 good
256 2 (102400,) e-f 0.1296 0.1433 1.11 1.4241 0.1 good
256 2 (32, 32) t-fb 0.0188 0.0756 4.01 0.0756 1.0
256 2 (32, 32) t-f 0.0071 0.0362 5.08 0.0394 0.92
256 2 (32, 32) e-fb 0.0551 0.0528 0.96 0.0521 1.0
256 2 (32, 32) e-f 0.0046 0.0119 2.58 0.0178 0.67 good
256 2 (64, 64) t-fb 0.1026 0.2649 2.58 0.2970 0.89
256 2 (64, 64) t-f 0.0108 0.1279 11.87 0.1685 0.76 good
256 2 (64, 64) e-fb 0.1436 0.1431 1.00 0.1807 0.79 good
256 2 (64, 64) e-f 0.0079 0.0140 1.77 0.0600 0.23 good
256 2 (128, 128) t-fb 0.5310 1.0012 1.89 1.1472 0.87
256 2 (128, 128) t-f 0.2058 0.4715 2.29 0.6521 0.72 good
256 2 (128, 128) e-fb 0.5098 0.5195 1.02 0.6932 0.75 good
256 2 (128, 128) e-f 0.0217 0.0313 1.44 0.2287 0.14 good
256 2 (32, 32, 32) t-fb 1.0555 1.9757 1.87 2.2879 0.86
256 2 (32, 32, 32) t-f 0.4059 0.9204 2.27 1.3025 0.71 good
256 2 (32, 32, 32) e-fb 1.0138 1.0172 1.00 1.3801 0.74 good
256 2 (32, 32, 32) e-f 0.0398 0.0435 1.09 0.4545 0.096 good
256 32 (1,) t-fb 0.0156 0.0293 1.88 0.0326 0.9
256 32 (1,) t-f 0.0048 0.0039 0.81 0.0185 0.21 good
256 32 (1,) e-fb 0.0285 0.0276 0.97 0.0262 1.1
256 32 (1,) e-f 0.0050 0.0030 0.61 0.0078 0.39 good
256 32 (4,) t-fb 0.0173 0.0290 1.67 0.0328 0.88
256 32 (4,) t-f 0.0048 0.0040 0.84 0.0185 0.22 good
256 32 (4,) e-fb 0.0286 0.0276 0.97 0.0261 1.1
256 32 (4,) e-f 0.0050 0.0031 0.61 0.0078 0.39 good
256 32 (16,) t-fb 0.0159 0.0359 2.25 0.0331 1.1
256 32 (16,) t-f 0.0048 0.0211 4.37 0.0187 1.1 bad
256 32 (16,) e-fb 0.0293 0.0277 0.95 0.0265 1.0
256 32 (16,) e-f 0.0051 0.0093 1.85 0.0078 1.2 bad
256 32 (32,) t-fb 0.0158 0.0377 2.38 0.0345 1.1
256 32 (32,) t-f 0.0049 0.0210 4.27 0.0185 1.1 bad
256 32 (32,) e-fb 0.0300 0.0279 0.93 0.0265 1.1
256 32 (32,) e-f 0.0050 0.0093 1.85 0.0078 1.2 bad
256 32 (64,) t-fb 0.0174 0.0464 2.66 0.0439 1.1
256 32 (64,) t-f 0.0065 0.0256 3.97 0.0234 1.1
256 32 (64,) e-fb 0.0364 0.0331 0.91 0.0321 1.0
256 32 (64,) e-f 0.0066 0.0123 1.87 0.0108 1.1 bad
256 32 (128,) t-fb 0.0230 0.0534 2.32 0.0504 1.1
256 32 (128,) t-f 0.0086 0.0304 3.55 0.0277 1.1
256 32 (128,) e-fb 0.0393 0.0346 0.88 0.0343 1.0
256 32 (128,) e-f 0.0058 0.0129 2.20 0.0114 1.1 bad
256 32 (256,) t-fb 0.0274 0.0598 2.18 0.0560 1.1
256 32 (256,) t-f 0.0106 0.0323 3.05 0.0298 1.1
256 32 (256,) e-fb 0.0416 0.0401 0.96 0.0377 1.1
256 32 (256,) e-f 0.0080 0.0139 1.73 0.0126 1.1
256 32 (1024,) t-fb 0.0683 0.1287 1.89 0.1249 1.0
256 32 (1024,) t-f 0.0266 0.0606 2.28 0.0603 1.0
256 32 (1024,) e-fb 0.0830 0.0838 1.01 0.0830 1.0
256 32 (1024,) e-f 0.0227 0.0236 1.04 0.0261 0.9
256 32 (10240,) t-fb 0.8891 1.2443 1.40 1.2563 0.99
256 32 (10240,) t-f 0.2918 0.5511 1.89 0.5788 0.95
256 32 (10240,) e-fb 0.7736 0.8028 1.04 0.8303 0.97
256 32 (10240,) e-f 0.1798 0.2100 1.17 0.2470 0.85
256 32 (32, 32) t-fb 0.0689 0.1288 1.87 0.1249 1.0
256 32 (32, 32) t-f 0.0266 0.0606 2.28 0.0604 1.0
256 32 (32, 32) e-fb 0.0824 0.0840 1.02 0.0832 1.0
256 32 (32, 32) e-f 0.0218 0.0237 1.08 0.0261 0.91
256 32 (64, 64) t-fb 0.3287 0.4899 1.49 0.4928 0.99
256 32 (64, 64) t-f 0.0931 0.2204 2.37 0.2294 0.96
256 32 (64, 64) e-fb 0.3047 0.3161 1.04 0.3231 0.98
256 32 (64, 64) e-f 0.0733 0.0844 1.15 0.0962 0.88
256 256 (1,) t-fb 0.0188 0.0322 1.71 0.0332 0.97
256 256 (1,) t-f 0.0055 0.0046 0.84 0.0186 0.25 good
256 256 (1,) e-fb 0.0313 0.0297 0.95 0.0269 1.1 bad
256 256 (1,) e-f 0.0120 0.0033 0.27 0.0079 0.42 good
256 256 (4,) t-fb 0.0172 0.0347 2.02 0.0355 0.98
256 256 (4,) t-f 0.0056 0.0078 1.40 0.0187 0.42 good
256 256 (4,) e-fb 0.0313 0.0307 0.98 0.0274 1.1 bad
256 256 (4,) e-f 0.0120 0.0050 0.42 0.0078 0.64 good
256 256 (16,) t-fb 0.0207 0.0555 2.68 0.0525 1.1
256 256 (16,) t-f 0.0075 0.0312 4.17 0.0284 1.1
256 256 (16,) e-fb 0.0395 0.0363 0.92 0.0355 1.0
256 256 (16,) e-f 0.0120 0.0130 1.09 0.0116 1.1 bad
256 256 (32,) t-fb 0.0255 0.0580 2.27 0.0552 1.0
256 256 (32,) t-f 0.0101 0.0319 3.16 0.0296 1.1
256 256 (32,) e-fb 0.0406 0.0387 0.95 0.0380 1.0
256 256 (32,) e-f 0.0120 0.0139 1.15 0.0125 1.1 bad
256 256 (64,) t-fb 0.0378 0.0708 1.87 0.0661 1.1
256 256 (64,) t-f 0.0146 0.0368 2.52 0.0327 1.1 bad
256 256 (64,) e-fb 0.0447 0.0483 1.08 0.0450 1.1
256 256 (64,) e-f 0.0129 0.0170 1.31 0.0147 1.2 bad
256 256 (128,) t-fb 0.0669 0.1066 1.60 0.1023 1.0
256 256 (128,) t-f 0.0262 0.0487 1.86 0.0448 1.1
256 256 (128,) e-fb 0.0708 0.0736 1.04 0.0714 1.0
256 256 (128,) e-f 0.0206 0.0233 1.13 0.0219 1.1
256 256 (256,) t-fb 0.1249 0.2141 1.71 0.2171 0.99
256 256 (256,) t-f 0.0468 0.1002 2.14 0.0963 1.0
256 256 (256,) e-fb 0.1350 0.1471 1.09 0.1501 0.98
256 256 (256,) e-f 0.0379 0.0497 1.31 0.0460 1.1
256 256 (1024,) t-fb 0.4899 0.8167 1.67 0.7898 1.0
256 256 (1024,) t-f 0.1757 0.3348 1.91 0.3367 0.99
256 256 (1024,) e-fb 0.5478 0.5651 1.03 0.5477 1.0
256 256 (1024,) e-f 0.1413 0.1581 1.12 0.1633 0.97
256 256 (32, 32) t-fb 0.4908 0.8149 1.66 0.7919 1.0
256 256 (32, 32) t-f 0.1756 0.3346 1.91 0.3374 0.99
256 256 (32, 32) e-fb 0.5465 0.5639 1.03 0.5490 1.0
256 256 (32, 32) e-f 0.1409 0.1582 1.12 0.1634 0.97
256 1024 (1,) t-fb 0.0185 0.0877 4.75 0.0667 1.3 bad
256 1024 (1,) t-f 0.0074 0.0445 6.02 0.0345 1.3 bad
256 1024 (1,) e-fb 0.0759 0.0579 0.76 0.0464 1.2 bad
256 1024 (1,) e-f 0.0379 0.0195 0.52 0.0135 1.4 bad
256 1024 (4,) t-fb 0.0231 0.0994 4.30 0.0871 1.1 bad
256 1024 (4,) t-f 0.0092 0.0527 5.70 0.0472 1.1 bad
256 1024 (4,) e-fb 0.0803 0.0622 0.77 0.0576 1.1
256 1024 (4,) e-f 0.0379 0.0197 0.52 0.0198 0.99
256 1024 (16,) t-fb 0.0375 0.1241 3.31 0.1210 1.0
256 1024 (16,) t-f 0.0159 0.0567 3.56 0.0578 0.98
256 1024 (16,) e-fb 0.0954 0.0777 0.81 0.0776 1.0
256 1024 (16,) e-f 0.0378 0.0199 0.53 0.0236 0.85
256 1024 (32,) t-fb 0.0640 0.1395 2.18 0.1404 0.99
256 1024 (32,) t-f 0.0251 0.0627 2.49 0.0664 0.94
256 1024 (32,) e-fb 0.1042 0.0898 0.86 0.0929 0.97
256 1024 (32,) e-f 0.0378 0.0233 0.61 0.0284 0.82
256 1024 (64,) t-fb 0.1205 0.2282 1.89 0.2369 0.96
256 1024 (64,) t-f 0.0422 0.1034 2.45 0.1061 0.97
256 1024 (64,) e-fb 0.1497 0.1577 1.05 0.1595 0.99
256 1024 (64,) e-f 0.0416 0.0500 1.20 0.0473 1.1
256 1024 (128,) t-fb 0.2391 0.4173 1.75 0.4073 1.0
256 1024 (128,) t-f 0.0859 0.1771 2.06 0.1759 1.0
256 1024 (128,) e-fb 0.2782 0.2904 1.04 0.2795 1.0
256 1024 (128,) e-f 0.0725 0.0839 1.16 0.0830 1.0
256 1024 (256,) t-fb 0.4747 0.7602 1.60 0.7321 1.0
256 1024 (256,) t-f 0.1697 0.3187 1.88 0.3081 1.0
256 1024 (256,) e-fb 0.5205 0.5364 1.03 0.5134 1.0
256 1024 (256,) e-f 0.1412 0.1577 1.12 0.1527 1.0
1024 2 (1,) t-fb 0.0157 0.0752 4.79 0.1105 0.68 good
1024 2 (1,) t-f 0.0047 0.0038 0.82 0.0593 0.065 good
1024 2 (1,) e-fb 0.0720 0.0709 0.98 0.0662 1.1
1024 2 (1,) e-f 0.0037 0.0029 0.80 0.0178 0.17 good
1024 2 (4,) t-fb 0.0162 0.0740 4.57 0.1164 0.64 good
1024 2 (4,) t-f 0.0046 0.0039 0.86 0.0654 0.06 good
1024 2 (4,) e-fb 0.0715 0.0704 0.99 0.0728 0.97
1024 2 (4,) e-f 0.0038 0.0030 0.81 0.0242 0.13 good
1024 2 (16,) t-fb 0.0158 0.1285 8.13 0.1175 1.1
1024 2 (16,) t-f 0.0051 0.0722 14.18 0.0658 1.1
1024 2 (16,) e-fb 0.0722 0.0800 1.11 0.0730 1.1
1024 2 (16,) e-f 0.0037 0.0268 7.19 0.0243 1.1 bad
1024 2 (32,) t-fb 0.0176 0.1262 7.19 0.1149 1.1
1024 2 (32,) t-f 0.0057 0.0715 12.54 0.0652 1.1
1024 2 (32,) e-fb 0.0701 0.0777 1.11 0.0707 1.1
1024 2 (32,) e-f 0.0037 0.0266 7.24 0.0240 1.1 bad
1024 2 (64,) t-fb 0.0180 0.1264 7.01 0.1153 1.1
1024 2 (64,) t-f 0.0060 0.0716 12.01 0.0653 1.1
1024 2 (64,) e-fb 0.0704 0.0780 1.11 0.0710 1.1
1024 2 (64,) e-f 0.0041 0.0267 6.57 0.0241 1.1 bad
1024 2 (128,) t-fb 0.0167 0.1359 8.12 0.1220 1.1 bad
1024 2 (128,) t-f 0.0059 0.0735 12.43 0.0653 1.1 bad
1024 2 (128,) e-fb 0.0744 0.0914 1.23 0.0829 1.1 bad
1024 2 (128,) e-f 0.0039 0.0269 6.93 0.0241 1.1 bad
1024 2 (256,) t-fb 0.0179 0.1685 9.39 0.1544 1.1
1024 2 (256,) t-f 0.0071 0.0898 12.63 0.0836 1.1
1024 2 (256,) e-fb 0.0936 0.1126 1.20 0.1043 1.1
1024 2 (256,) e-f 0.0046 0.0379 8.27 0.0350 1.1
1024 2 (1024,) t-fb 0.1033 0.3146 3.05 0.3181 0.99
1024 2 (1024,) t-f 0.0107 0.1628 15.18 0.1802 0.9
1024 2 (1024,) e-fb 0.1576 0.1798 1.14 0.1935 0.93
1024 2 (1024,) e-f 0.0084 0.0392 4.68 0.0640 0.61 good
1024 2 (10240,) t-fb 1.3214 2.4976 1.89 2.8826 0.87
1024 2 (10240,) t-f 0.5072 1.1662 2.30 1.6395 0.71 good
1024 2 (10240,) e-fb 1.2759 1.2879 1.01 1.7395 0.74 good
1024 2 (10240,) e-f 0.0476 0.0595 1.25 0.5721 0.1 good
1024 2 (32, 32) t-fb 0.1034 0.3146 3.04 0.3180 0.99
1024 2 (32, 32) t-f 0.0108 0.1629 15.11 0.1802 0.9
1024 2 (32, 32) e-fb 0.1578 0.1798 1.14 0.1935 0.93
1024 2 (32, 32) e-f 0.0081 0.0393 4.83 0.0640 0.61 good
1024 2 (64, 64) t-fb 0.5333 1.0438 1.96 1.1692 0.89
1024 2 (64, 64) t-f 0.2064 0.4993 2.42 0.6650 0.75 good
1024 2 (64, 64) e-fb 0.5224 0.5482 1.05 0.7064 0.78 good
1024 2 (64, 64) e-f 0.0210 0.0484 2.31 0.2327 0.21 good
1024 2 (128, 128) t-fb 2.1064 3.9906 1.89 4.6073 0.87
1024 2 (128, 128) t-f 0.8058 1.8800 2.33 2.6038 0.72 good
1024 2 (128, 128) e-fb 2.0245 2.0744 1.02 2.7686 0.75 good
1024 2 (128, 128) e-f 0.0761 0.1277 1.68 0.9096 0.14 good
1024 2 (32, 32, 32) t-fb 4.2096 7.9008 1.88 9.1695 0.86
1024 2 (32, 32, 32) t-f 1.6098 3.6785 2.29 5.2038 0.71 good
1024 2 (32, 32, 32) e-fb 4.0552 4.0815 1.01 5.5401 0.74 good
1024 2 (32, 32, 32) e-f 0.1478 0.1781 1.21 1.8242 0.098 good
1024 32 (1,) t-fb 0.0158 0.0744 4.71 0.1160 0.64 good
1024 32 (1,) t-f 0.0048 0.0040 0.84 0.0653 0.061 good
1024 32 (1,) e-fb 0.0716 0.0709 0.99 0.0723 0.98
1024 32 (1,) e-f 0.0096 0.0031 0.32 0.0241 0.13 good
1024 32 (4,) t-fb 0.0158 0.0742 4.68 0.1165 0.64 good
1024 32 (4,) t-f 0.0048 0.0056 1.15 0.0651 0.086 good
1024 32 (4,) e-fb 0.0701 0.0708 1.01 0.0721 0.98
1024 32 (4,) e-f 0.0097 0.0039 0.41 0.0241 0.16 good
1024 32 (16,) t-fb 0.0183 0.1672 9.12 0.1538 1.1
1024 32 (16,) t-f 0.0063 0.0867 13.69 0.0816 1.1
1024 32 (16,) e-fb 0.0933 0.1137 1.22 0.1052 1.1
1024 32 (16,) e-f 0.0097 0.0393 4.04 0.0348 1.1 bad
1024 32 (32,) t-fb 0.0223 0.1832 8.23 0.1712 1.1
1024 32 (32,) t-f 0.0085 0.1051 12.33 0.0971 1.1
1024 32 (32,) e-fb 0.0921 0.1131 1.23 0.1046 1.1
1024 32 (32,) e-f 0.0117 0.0393 3.36 0.0350 1.1 bad
1024 32 (64,) t-fb 0.0278 0.1873 6.74 0.1746 1.1
1024 32 (64,) t-f 0.0105 0.1066 10.15 0.0981 1.1
1024 32 (64,) e-fb 0.0921 0.1166 1.27 0.1075 1.1
1024 32 (64,) e-f 0.0118 0.0407 3.45 0.0358 1.1 bad
1024 32 (128,) t-fb 0.0392 0.1980 5.05 0.1858 1.1
1024 32 (128,) t-f 0.0162 0.1103 6.79 0.1021 1.1
1024 32 (128,) e-fb 0.0974 0.1250 1.28 0.1159 1.1
1024 32 (128,) e-f 0.0126 0.0433 3.45 0.0383 1.1 bad
1024 32 (256,) t-fb 0.0689 0.2235 3.24 0.2079 1.1
1024 32 (256,) t-f 0.0269 0.1184 4.41 0.1106 1.1
1024 32 (256,) e-fb 0.1202 0.1457 1.21 0.1352 1.1
1024 32 (256,) e-f 0.0226 0.0482 2.13 0.0448 1.1
1024 32 (1024,) t-fb 0.3313 0.5225 1.58 0.5132 1.0
1024 32 (1024,) t-f 0.0932 0.2385 2.56 0.2406 0.99
1024 32 (1024,) e-fb 0.3219 0.3380 1.05 0.3355 1.0
1024 32 (1024,) e-f 0.0763 0.0927 1.22 0.1005 0.92
1024 32 (32, 32) t-fb 0.3322 0.5242 1.58 0.5085 1.0
1024 32 (32, 32) t-f 0.0935 0.2383 2.55 0.2398 0.99
1024 32 (32, 32) e-fb 0.3235 0.3395 1.05 0.3334 1.0
1024 32 (32, 32) e-f 0.0763 0.0925 1.21 0.0999 0.93
1024 256 (1,) t-fb 0.0187 0.0909 4.87 0.1237 0.74 good
1024 256 (1,) t-f 0.0064 0.0072 1.13 0.0654 0.11 good
1024 256 (1,) e-fb 0.0965 0.0831 0.86 0.0846 0.98
1024 256 (1,) e-f 0.0378 0.0049 0.13 0.0241 0.2 good
1024 256 (4,) t-fb 0.0231 0.1027 4.44 0.1753 0.59 good
1024 256 (4,) t-f 0.0078 0.0193 2.46 0.0978 0.2 good
1024 256 (4,) e-fb 0.1161 0.0959 0.83 0.1083 0.89
1024 256 (4,) e-f 0.0379 0.0117 0.31 0.0354 0.33 good
1024 256 (16,) t-fb 0.0399 0.2083 5.22 0.1963 1.1
1024 256 (16,) t-f 0.0153 0.1129 7.40 0.1060 1.1
1024 256 (16,) e-fb 0.1265 0.1325 1.05 0.1232 1.1
1024 256 (16,) e-f 0.0378 0.0438 1.16 0.0396 1.1 bad
1024 256 (32,) t-fb 0.0744 0.2201 2.96 0.2088 1.1
1024 256 (32,) t-f 0.0300 0.1170 3.90 0.1106 1.1
1024 256 (32,) e-fb 0.1332 0.1438 1.08 0.1350 1.1
1024 256 (32,) e-f 0.0378 0.0481 1.27 0.0443 1.1
1024 256 (64,) t-fb 0.1299 0.2736 2.11 0.2597 1.1
1024 256 (64,) t-f 0.0500 0.1347 2.69 0.1250 1.1
1024 256 (64,) e-fb 0.1638 0.1812 1.11 0.1688 1.1
1024 256 (64,) e-f 0.0416 0.0593 1.43 0.0516 1.1 bad
1024 256 (128,) t-fb 0.2442 0.4439 1.82 0.4388 1.0
1024 256 (128,) t-f 0.0902 0.1930 2.14 0.1892 1.0
1024 256 (128,) e-fb 0.2847 0.3035 1.07 0.2937 1.0
1024 256 (128,) e-f 0.0723 0.0913 1.26 0.0846 1.1
1024 256 (256,) t-fb 0.4950 0.9391 1.90 0.9444 0.99
1024 256 (256,) t-f 0.1770 0.4468 2.52 0.4199 1.1
1024 256 (256,) e-fb 0.5655 0.6484 1.15 0.6456 1.0
1024 256 (256,) e-f 0.1410 0.2222 1.58 0.1939 1.1 bad
1024 1024 (1,) t-fb 0.0298 0.3699 12.43 0.3077 1.2 bad
1024 1024 (1,) t-f 0.0132 0.1934 14.63 0.1695 1.1 bad
1024 1024 (1,) e-fb 0.3011 0.2287 0.76 0.1974 1.2 bad
1024 1024 (1,) e-f 0.1416 0.0692 0.49 0.0671 1.0
1024 1024 (4,) t-fb 0.0455 0.4015 8.83 0.3617 1.1 bad
1024 1024 (4,) t-f 0.0172 0.2030 11.78 0.1935 1.0
1024 1024 (4,) e-fb 0.3229 0.2503 0.78 0.2292 1.1
1024 1024 (4,) e-f 0.1416 0.0691 0.49 0.0729 0.95
1024 1024 (16,) t-fb 0.2117 0.5145 2.43 0.5203 0.99
1024 1024 (16,) t-f 0.0591 0.2312 3.91 0.2511 0.92
1024 1024 (16,) e-fb 0.3802 0.3123 0.82 0.3188 0.98
1024 1024 (16,) e-f 0.1415 0.0738 0.52 0.0932 0.79 good
1024 1024 (32,) t-fb 0.2822 0.5643 2.00 0.5765 0.98
1024 1024 (32,) t-f 0.1095 0.2498 2.28 0.2701 0.92
1024 1024 (32,) e-fb 0.4155 0.3651 0.88 0.3756 0.97
1024 1024 (32,) e-f 0.1415 0.0915 0.65 0.1118 0.82
1024 1024 (64,) t-fb 0.4978 0.9969 2.00 1.0256 0.97
1024 1024 (64,) t-f 0.1828 0.4532 2.48 0.4546 1.0
1024 1024 (64,) e-fb 0.6173 0.6831 1.11 0.6905 0.99
1024 1024 (64,) e-f 0.1571 0.2261 1.44 0.2075 1.1
10240 2 (1,) t-fb 0.0161 0.5653 35.14 1.0533 0.54 good
10240 2 (1,) t-f 0.0049 0.0043 0.87 0.5647 0.0076 good
10240 2 (1,) e-fb 0.5325 0.5326 1.00 0.6176 0.86
10240 2 (1,) e-f 0.0073 0.0030 0.41 0.1567 0.019 good
10240 2 (4,) t-fb 0.0186 0.5645 30.28 1.1171 0.51 good
10240 2 (4,) t-f 0.0060 0.0071 1.17 0.6282 0.011 good
10240 2 (4,) e-fb 0.5326 0.5325 1.00 0.6814 0.78 good
10240 2 (4,) e-f 0.0073 0.0031 0.43 0.2201 0.014 good
10240 2 (16,) t-fb 0.0181 1.4222 78.46 1.3142 1.1
10240 2 (16,) t-f 0.0060 0.6833 114.69 0.6310 1.1
10240 2 (16,) e-fb 0.6067 0.9322 1.54 0.8625 1.1
10240 2 (16,) e-f 0.0073 0.2425 33.32 0.2214 1.1
10240 2 (32,) t-fb 0.0201 1.6177 80.67 1.5032 1.1
10240 2 (32,) t-f 0.0073 0.8707 119.43 0.8191 1.1
10240 2 (32,) e-fb 0.7264 1.0539 1.45 0.9740 1.1
10240 2 (32,) e-f 0.0074 0.3464 46.91 0.3225 1.1
10240 2 (64,) t-fb 0.0241 1.7402 72.33 1.6200 1.1
10240 2 (64,) t-f 0.0094 0.9936 105.77 0.9349 1.1
10240 2 (64,) e-fb 0.7271 1.0565 1.45 0.9766 1.1
10240 2 (64,) e-f 0.0087 0.3471 39.72 0.3225 1.1
10240 2 (128,) t-fb 0.1892 1.7523 9.26 1.6235 1.1
10240 2 (128,) t-f 0.0801 1.0015 12.51 0.9365 1.1
10240 2 (128,) e-fb 0.7301 1.0620 1.45 0.9795 1.1
10240 2 (128,) e-f 0.0093 0.3496 37.68 0.3240 1.1
10240 2 (256,) t-fb 0.3509 1.7810 5.08 1.6454 1.1
10240 2 (256,) t-f 0.1410 1.0054 7.13 0.9415 1.1
10240 2 (256,) e-fb 0.7493 1.0852 1.45 0.9928 1.1
10240 2 (256,) e-f 0.0150 0.3515 23.46 0.3265 1.1
10240 2 (1024,) t-fb 1.3235 3.0706 2.32 3.1075 0.99
10240 2 (1024,) t-f 0.5106 1.5799 3.09 1.7625 0.9
10240 2 (1024,) e-fb 1.4310 1.7480 1.22 1.8753 0.93
10240 2 (1024,) e-f 0.0495 0.3669 7.41 0.6154 0.6 good
10240 2 (32, 32) t-fb 1.3249 3.0698 2.32 3.1069 0.99
10240 2 (32, 32) t-f 0.5096 1.5796 3.10 1.7615 0.9
10240 2 (32, 32) e-fb 1.4311 1.7486 1.22 1.8758 0.93
10240 2 (32, 32) e-f 0.0496 0.3667 7.39 0.6150 0.6 good
10240 2 (64, 64) t-fb 5.2565 10.2989 1.96 11.6242 0.89
10240 2 (64, 64) t-f 2.0033 4.9179 2.45 6.6181 0.74 good
10240 2 (64, 64) e-fb 5.1502 5.4393 1.06 7.0290 0.77 good
10240 2 (64, 64) e-f 0.1775 0.4669 2.63 2.3127 0.2 good
10240 32 (1,) t-fb 0.0231 0.7231 31.31 1.2876 0.56 good
10240 32 (1,) t-f 0.0082 0.0083 1.02 0.6260 0.013 good
10240 32 (1,) e-fb 0.6416 0.6116 0.95 0.8436 0.72 good
10240 32 (1,) e-f 0.0659 0.0059 0.09 0.2178 0.027 good
10240 32 (4,) t-fb 0.0289 0.7893 27.33 1.6233 0.49 good
10240 32 (4,) t-f 0.0103 0.0367 3.57 0.9246 0.04 good
10240 32 (4,) e-fb 0.7830 0.7345 0.94 0.9858 0.75 good
10240 32 (4,) e-f 0.0662 0.0174 0.26 0.3196 0.054 good
10240 32 (16,) t-fb 0.0660 1.7906 27.12 1.6633 1.1
10240 32 (16,) t-f 0.0251 1.0163 40.43 0.9457 1.1
10240 32 (16,) e-fb 0.8014 1.0992 1.37 1.0090 1.1
10240 32 (16,) e-f 0.0663 0.3647 5.50 0.3259 1.1 bad
10240 32 (32,) t-fb 0.0889 1.7850 20.07 1.6551 1.1
10240 32 (32,) t-f 0.0356 1.0179 28.63 0.9434 1.1
10240 32 (32,) e-fb 0.7904 1.0916 1.38 1.0002 1.1
10240 32 (32,) e-f 0.0663 0.3685 5.56 0.3280 1.1 bad
10240 32 (64,) t-fb 0.1692 1.8294 10.82 1.7015 1.1
10240 32 (64,) t-f 0.0686 1.0341 15.08 0.9570 1.1
10240 32 (64,) e-fb 0.8312 1.1256 1.35 1.0322 1.1
10240 32 (64,) e-f 0.0886 0.3834 4.33 0.3370 1.1 bad
10240 32 (128,) t-fb 0.4425 1.9475 4.40 1.8193 1.1
10240 32 (128,) t-f 0.1478 1.0779 7.29 0.9990 1.1
10240 32 (128,) e-fb 0.9026 1.2257 1.36 1.1301 1.1
10240 32 (128,) e-f 0.0944 0.4195 4.44 0.3719 1.1 bad
10240 32 (256,) t-fb 0.9001 2.2193 2.47 2.0623 1.1
10240 32 (256,) t-f 0.2917 1.1707 4.01 1.0916 1.1
10240 32 (256,) e-fb 1.1562 1.4463 1.25 1.3411 1.1
10240 32 (256,) e-f 0.1851 0.4755 2.57 0.4387 1.1
10240 256 (1,) t-fb 0.1381 0.8149 5.90 1.6261 0.5 good
10240 256 (1,) t-f 0.0348 0.0442 1.27 0.9237 0.048 good
10240 256 (1,) e-fb 1.0798 0.7589 0.70 0.9896 0.77 good
10240 256 (1,) e-f 0.3487 0.0274 0.08 0.3198 0.086 good
10240 256 (4,) t-fb 0.4444 1.1810 2.66 1.7252 0.68 good
10240 256 (4,) t-f 0.1744 0.3615 2.07 0.9629 0.38 good
10240 256 (4,) e-fb 1.1237 0.8868 0.79 1.0451 0.85
10240 256 (4,) e-f 0.3496 0.1240 0.35 0.3334 0.37 good
10240 256 (16,) t-fb 0.4750 2.1131 4.45 2.0139 1.0
10240 256 (16,) t-f 0.1920 1.1202 5.83 1.0709 1.0
10240 256 (16,) e-fb 1.2583 1.3443 1.07 1.2507 1.1
10240 256 (16,) e-f 0.3488 0.4338 1.24 0.3975 1.1
10240 256 (32,) t-fb 0.8112 2.2043 2.72 2.0746 1.1
10240 256 (32,) t-f 0.3024 1.1620 3.84 1.0876 1.1
10240 256 (32,) e-fb 1.3070 1.4351 1.10 1.3396 1.1
10240 256 (32,) e-f 0.3488 0.4768 1.37 0.4380 1.1
10240 1024 (1,) t-fb 0.8186 3.8406 4.69 3.3326 1.2 bad
10240 1024 (1,) t-f 0.2301 1.9879 8.64 1.8707 1.1
10240 1024 (1,) e-fb 3.0802 2.3614 0.77 2.0270 1.2 bad
10240 1024 (1,) e-f 1.3847 0.6659 0.48 0.6505 1.0
10240 1024 (4,) t-fb 2.2558 4.1393 1.83 3.9954 1.0
10240 1024 (4,) t-f 0.9924 2.0945 2.11 2.1120 0.99
10240 1024 (4,) e-fb 3.3391 2.6455 0.79 2.4329 1.1
10240 1024 (4,) e-f 1.3869 0.6709 0.48 0.7558 0.89
@t-vi
Copy link
Author

t-vi commented Oct 17, 2018

So these are particularly terrible in the update:

bs channels features mode cudnn native slowness reference slowness eval
8 256 (128,) t-f 0,0053 0,0047 0,89 0,0037 1,3 bad
64 1024 (4,) e-f 0,012 0,0071 0,59 0,0052 1,4 bad
128 1024 (1,) e-f 0,0207 0,0112 0,54 0,0079 1,4 bad
256 1024 (1,) e-f 0,0379 0,0195 0,52 0,0135 1,4 bad
64 1024 (1,) e-fb 0,0253 0,0241 0,96 0,0161 1,5 bad

While the factor is large, these cases belong to the faster ones (including measurement problems etc.)

Filtering for time > 0.01, I have

bs channels features mode cudnn native slowness reference slowness eval
1024 1024 (1,) t-fb 0,0298 0,3699 12,43 0,3077 1,2 bad
1024 1024 (1,) e-fb 0,3011 0,2287 0,76 0,1974 1,2 bad
10240 1024 (1,) t-fb 0,8186 3,8406 4,69 3,3326 1,2 bad
10240 1024 (1,) e-fb 3,0802 2,3614 0,77 2,027 1,2 bad

I'd say that it is not too worrisome.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment