Discrepancy in fft2 output?
AndyGreenwell opened this issue · comments
In performing a few numerical experiments with fft2 between various environments, I have noticed what seem to be discrepancies in output values that might not be attributable just to floating point round-off error.
I would appreciate someone's take on whether I have set up this test properly for evaluating behavior of ArrayFire's fft2 function.
Setup
Configuration
- Compare numerical output from Matlab 2017b, Anaconda 5.0.1/Python 3.6.3/Scipy 0.19.1, Julia 0.6.2, and ArrayFire 3.5.1
- All GPU functionality executed against CUDA 8.0.61.2
- Windows 10 operating system
Test performed
- In all environments, construct a Float32 matrix of zeros of size (2048, 4096)
- Fill the four 50x50 corners of the matrix with the value 1.0
- Take the real to complex fft2 on the resulting array
Results observed
- CPU outputs from Matlab, Scipy and Julia all match to within reasonable floating point differences
- GPU output in Matlab matches the CPU output from all three above environments to within reasonable floating point differences
- GPU output in arrayfire-python seems to match in one column of output, but has significant differences in the remaining columns.
** As a specific numerical difference, notice the value in the first row, second column of each output.
** For the CPU executions and Matlab GPU, the entry has a value with a real part of 9990.2 and an imaginary part of 7.6623.
** For the ArrayFire execution, the entry has a value with a real part of 9960.82031250 and an imaginary part 15.27970791, or if the result is transposed below a real part of 9960.82617188 and an imaginary part of 7.63974667j.
Question
- Have I structured this test properly for execution in the arrayfire-python environment?
Request
- Provide a utility function for gathering an ArrayFire array into a numpy array if numpy is present on the system.
Commands and Output
Matlab 2017b
>> a = zeros(2048,4096,'single');
>> ad = gpuArray(a);
>> a(1:50,1:50) = 1;
>> a(1:50,end-49:end) = 1;
>> a(end-49:end,1:50) = 1;
>> a(end-49:end,end-49:end) = 1;
>> ad = gpuArray(a);
>> bd = fft2(ad);
>> b = fft2(a);
>> b(1:10,1:10)
ans =
10×10 single matrix
Columns 1 through 6
1.0000000e+04 + 0.0000000e+00i 9.9901953e+03 + 7.6623907e+00i 9.9608184e+03 + 1.5279749e+01i 9.9119746e+03 + 2.2807228e+01i 9.8438311e+03 + 3.0200647e+01i 9.7566348e+03 + 3.7416420e+01i
9.9608193e+03 + 1.5279821e+01i 9.9510430e+03 + 2.2897202e+01i 9.9217695e+03 + 3.0439827e+01i 9.8731055e+03 + 3.7863113e+01i 9.8052178e+03 + 4.5123505e+01i 9.7183516e+03 + 5.2177799e+01i
9.8438320e+03 + 3.0200367e+01i 9.8341582e+03 + 3.7713478e+01i 9.8052168e+03 + 4.5123173e+01i 9.7571133e+03 + 5.2385574e+01i 9.6900127e+03 + 5.9457695e+01i 9.6041543e+03 + 6.6297554e+01i
9.6506875e+03 + 4.4412247e+01i 9.6411904e+03 + 5.1763443e+01i 9.6128066e+03 + 5.8984249e+01i 9.5656348e+03 + 6.6031860e+01i 9.4998398e+03 + 7.2864410e+01i 9.4156563e+03 + 7.9440842e+01i
9.3840957e+03 + 5.7580696e+01i 9.3748516e+03 + 6.4714691e+01i 9.3472412e+03 + 7.1693748e+01i 9.3013613e+03 + 7.8476349e+01i 9.2373730e+03 + 8.5022018e+01i 9.1555049e+03 + 9.1291229e+01i
9.0478047e+03 + 6.9397179e+01i 9.0388809e+03 + 7.6261902e+01i 9.0122490e+03 + 8.2950027e+01i 8.9680020e+03 + 8.9421822e+01i 8.9062969e+03 + 9.5638336e+01i 8.8273525e+03 + 1.0156185e+02i
8.6465059e+03 + 7.9583786e+01i 8.6379688e+03 + 8.6131058e+01i 8.6125068e+03 + 9.2483597e+01i 8.5702129e+03 + 9.8603455e+01i 8.5112344e+03 + 1.0445392e+02i 8.4357813e+03 + 1.0999908e+02i
8.1857832e+03 + 8.7900818e+01i 8.1776904e+03 + 9.4086906e+01i 8.1535771e+03 + 1.0006406e+02i 8.1135269e+03 + 1.0579652e+02i 8.0576821e+03 + 1.1124962e+02i 7.9862402e+03 + 1.1638995e+02i
7.6720059e+03 + 9.4154678e+01i 7.6644121e+03 + 9.9940964e+01i 7.6418032e+03 + 1.0550838e+02i 7.6042583e+03 + 1.1082356e+02i 7.5519102e+03 + 1.1585425e+02i 7.4849438e+03 + 1.2056924e+02i
7.1122246e+03 + 9.8196442e+01i 7.1051763e+03 + 1.0354982e+02i 7.0842085e+03 + 1.0867901e+02i 7.0493945e+03 + 1.1355305e+02i 7.0008579e+03 + 1.1814230e+02i 6.9387700e+03 + 1.2241808e+02i
Columns 7 through 10
9.6506855e+03 + 4.4412262e+01i 9.5263613e+03 + 5.1146973e+01i 9.3840957e+03 + 5.7580811e+01i 9.2243906e+03 + 6.3676044e+01i
9.6128057e+03 + 5.8984318e+01i 9.4889590e+03 + 6.5502625e+01i 9.3472422e+03 + 7.1693794e+01i 9.1881523e+03 + 7.7521202e+01i
9.4998398e+03 + 7.2864082e+01i 9.3774365e+03 + 7.9118256e+01i 9.2373730e+03 + 8.5021957e+01i 9.0801426e+03 + 9.0539658e+01i
9.3133770e+03 + 8.5721733e+01i 9.1933662e+03 + 9.1669067e+01i 9.0560420e+03 + 9.7246353e+01i 8.9018877e+03 + 1.0241939e+02i
9.0560410e+03 + 9.7246155e+01i 8.9393359e+03 + 1.0285026e+02i 8.8057949e+03 + 1.0806866e+02i 8.6558906e+03 + 1.1286887e+02i
8.7314434e+03 + 1.0715637e+02i 8.6189102e+03 + 1.1238715e+02i 8.4901475e+03 + 1.1722084e+02i 8.3456074e+03 + 1.2162746e+02i
8.3441182e+03 + 1.1520484e+02i 8.2365664e+03 + 1.2003874e+02i 8.1135059e+03 + 1.2446959e+02i 7.9753682e+03 + 1.2846884e+02i
7.8994521e+03 + 1.2118526e+02i 7.7976235e+03 + 1.2560530e+02i 7.6811123e+03 + 1.2962128e+02i 7.5503271e+03 + 1.3320709e+02i
7.4035938e+03 + 1.2493880e+02i 7.3081489e+03 + 1.2893518e+02i 7.1989424e+03 + 1.3253171e+02i 7.0763594e+03 + 1.3570422e+02i
6.8633486e+03 + 1.2635329e+02i 6.7748604e+03 + 1.2992242e+02i 6.6736143e+03 + 1.3310135e+02i 6.5599688e+03 + 1.3586807e+02i
>> bd(1:10,1:10)
ans =
10×10 single gpuArray matrix
Columns 1 through 6
1.0000000e+04 + 0.0000000e+00i 9.9901953e+03 + 7.6623096e+00i 9.9608193e+03 + 1.5279715e+01i 9.9119746e+03 + 2.2807190e+01i 9.8438320e+03 + 3.0200514e+01i 9.7566348e+03 + 3.7416363e+01i
9.9608203e+03 + 1.5279708e+01i 9.9510430e+03 + 2.2896988e+01i 9.9217695e+03 + 3.0439701e+01i 9.8731055e+03 + 3.7863056e+01i 9.8052188e+03 + 4.5123158e+01i 9.7183506e+03 + 5.2177612e+01i
9.8438330e+03 + 3.0200592e+01i 9.8341592e+03 + 3.7713570e+01i 9.8052188e+03 + 4.5123596e+01i 9.7571133e+03 + 5.2385696e+01i 9.6900127e+03 + 5.9457623e+01i 9.6041543e+03 + 6.6297523e+01i
9.6506855e+03 + 4.4412113e+01i 9.6411914e+03 + 5.1763298e+01i 9.6128066e+03 + 5.8984150e+01i 9.5656348e+03 + 6.6031898e+01i 9.4998398e+03 + 7.2864044e+01i 9.4156563e+03 + 7.9440460e+01i
9.3840967e+03 + 5.7580814e+01i 9.3748525e+03 + 6.4714752e+01i 9.3472422e+03 + 7.1693832e+01i 9.3013613e+03 + 7.8476456e+01i 9.2373740e+03 + 8.5022049e+01i 9.1555039e+03 + 9.1291382e+01i
9.0478037e+03 + 6.9397087e+01i 9.0388809e+03 + 7.6261871e+01i 9.0122480e+03 + 8.2950119e+01i 8.9680020e+03 + 8.9421753e+01i 8.9062969e+03 + 9.5638115e+01i 8.8273516e+03 + 1.0156184e+02i
8.6465059e+03 + 7.9583710e+01i 8.6379688e+03 + 8.6131195e+01i 8.6125078e+03 + 9.2483505e+01i 8.5702139e+03 + 9.8603386e+01i 8.5112344e+03 + 1.0445370e+02i 8.4357822e+03 + 1.0999883e+02i
8.1857837e+03 + 8.7901047e+01i 8.1776909e+03 + 9.4087189e+01i 8.1535776e+03 + 1.0006441e+02i 8.1135278e+03 + 1.0579699e+02i 8.0576826e+03 + 1.1124986e+02i 7.9862412e+03 + 1.1638994e+02i
7.6720059e+03 + 9.4154358e+01i 7.6644121e+03 + 9.9940605e+01i 7.6418032e+03 + 1.0550810e+02i 7.6042588e+03 + 1.1082333e+02i 7.5519102e+03 + 1.1585382e+02i 7.4849434e+03 + 1.2056885e+02i
7.1122246e+03 + 9.8196213e+01i 7.1051772e+03 + 1.0354974e+02i 7.0842095e+03 + 1.0867896e+02i 7.0493955e+03 + 1.1355289e+02i 7.0008584e+03 + 1.1814211e+02i 6.9387705e+03 + 1.2241788e+02i
Columns 7 through 10
9.6506855e+03 + 4.4412270e+01i 9.5263613e+03 + 5.1146660e+01i 9.3840977e+03 + 5.7580605e+01i 9.2243916e+03 + 6.3676071e+01i
9.6128066e+03 + 5.8984367e+01i 9.4889580e+03 + 6.5502228e+01i 9.3472422e+03 + 7.1693573e+01i 9.1881523e+03 + 7.7521240e+01i
9.4998398e+03 + 7.2864265e+01i 9.3774355e+03 + 7.9118210e+01i 9.2373740e+03 + 8.5021736e+01i 9.0801445e+03 + 9.0539764e+01i
9.3133770e+03 + 8.5721619e+01i 9.1933652e+03 + 9.1668533e+01i 9.0560420e+03 + 9.7246140e+01i 8.9018867e+03 + 1.0241924e+02i
9.0560420e+03 + 9.7246361e+01i 8.9393340e+03 + 1.0285011e+02i 8.8057969e+03 + 1.0806861e+02i 8.6558916e+03 + 1.1286905e+02i
8.7314434e+03 + 1.0715643e+02i 8.6189092e+03 + 1.1238669e+02i 8.4901475e+03 + 1.1722072e+02i 8.3456064e+03 + 1.2162760e+02i
8.3441172e+03 + 1.1520473e+02i 8.2365664e+03 + 1.2003819e+02i 8.1135068e+03 + 1.2446938e+02i 7.9753682e+03 + 1.2846857e+02i
7.8994521e+03 + 1.2118549e+02i 7.7976230e+03 + 1.2560516e+02i 7.6811128e+03 + 1.2962163e+02i 7.5503271e+03 + 1.3320723e+02i
7.4035938e+03 + 1.2493860e+02i 7.3081484e+03 + 1.2893465e+02i 7.1989434e+03 + 1.3253122e+02i 7.0763594e+03 + 1.3570404e+02i
6.8633481e+03 + 1.2635315e+02i 6.7748594e+03 + 1.2992197e+02i 6.6736152e+03 + 1.3310097e+02i 6.5599688e+03 + 1.3586806e+02i
Julia 0.6.2
_ _ _(_)_ | A fresh approach to technical computing
(_) | (_) (_) | Documentation: https://docs.julialang.org
_ _ _| |_ __ _ | Type "?help" for help.
| | | | | | |/ _` | |
| | |_| | | | (_| | | Version 0.6.2 (2017-12-13 18:08 UTC)
_/ |\__'_|_|_|\__'_| | Official http://julialang.org/ release
|__/ | x86_64-w64-mingw32
julia> a = zeros(Float32,2048,4096);
julia> a[1:50,1:50] = 1;
julia> a[end-49:end,end-49:end] = 1;
julia> a[end-49:end,1:50] = 1;
julia> a[1:50,end-49:end] = 1;
julia> b = fft(a);
julia> b
2048×4096 Array{Complex{Float32},2}:
10000.0+0.0im 9990.2+7.66239im 9960.82+15.2797im … 9960.82-15.2797im 9990.2-7.6622im
9960.82+15.2797im 9951.04+22.8974im 9921.77+30.4397im 9921.82+0.000190735im 9951.06+7.63267im
9843.83+30.2006im 9834.16+37.714im 9805.22+45.1234im 9805.31+15.0411im 9834.21+22.6289im
9650.69+44.4121im 9641.19+51.7632im 9612.81+58.9841im 9612.94+29.4921im 9641.26+36.9738im
9384.1+57.5809im 9374.85+64.7148im 9347.24+71.6939im 9347.42+43.0165im 9374.94+50.334im
9047.8+69.3972im 9038.88+76.2618im 9012.25+82.9501im … 9012.46+55.3005im 9038.99+62.3966im
8646.51+79.5837im 8637.97+86.1308im 8612.51+92.4835im 8612.75+66.0603im 8638.09+72.8803im
8185.78+87.9012im 8177.69+94.0872im 8153.58+100.064im 8153.85+75.0492im 8177.83+81.5428im
7672.01+94.1544im 7664.41+99.9406im 7641.8+105.508im 7642.09+82.063im 7664.56+88.1835im
7112.22+98.1964im 7105.18+103.55im 7084.21+108.679im 7084.51+86.9443im 7105.33+92.6503im
6514.06+99.9322im 6507.59+104.825im 6488.38+109.494im … 6488.69+89.5875im 6507.75+94.843im
5885.56+99.321im 5879.72+103.733im 5862.35+107.925im 5862.66+89.9389im 5879.87+94.7138im
5235.11+96.3775im 5229.9+100.294im 5214.45+103.999im 5214.75+88.0008im 5230.05+92.2716im
⋮ ⋱ ⋮
5235.11-96.3775im 5230.05-92.2717im 5214.75-88.0008im 5214.45-103.999im 5229.9-100.294im
5885.56-99.321im 5879.87-94.7138im 5862.66-89.9389im 5862.35-107.925im 5879.72-103.733im
6514.06-99.9323im 6507.75-94.8429im 6488.69-89.5875im 6488.38-109.494im 6507.59-104.826im
7112.22-98.1963im 7105.33-92.6503im 7084.51-86.9443im 7084.21-108.679im 7105.18-103.55im
7672.01-94.1543im 7664.56-88.1834im 7642.09-82.0628im … 7641.8-105.508im 7664.41-99.9406im
8185.78-87.9011im 8177.83-81.5428im 8153.85-75.0491im 8153.58-100.064im 8177.69-94.0871im
8646.51-79.5836im 8638.09-72.8803im 8612.75-66.0602im 8612.51-92.4834im 8637.97-86.1309im
9047.8-69.3972im 9038.99-62.3965im 9012.46-55.3005im 9012.25-82.9501im 9038.88-76.2619im
9384.1-57.5808im 9374.94-50.3339im 9347.42-43.0165im 9347.24-71.6937im 9374.85-64.7149im
9650.69-44.4122im 9641.26-36.974im 9612.94-29.4923im … 9612.81-58.9841im 9641.19-51.7635im
9843.83-30.2006im 9834.21-22.628im 9805.31-15.0411im 9805.22-45.1234im 9834.16-37.7132im
9960.82-15.2798im 9951.07-7.63271im 9921.82-0.000112057im 9921.77-30.4398im 9951.04-22.8975im
Anaconda 5.0.1 / Python 3.6.3 / Scipy 0.19.1 / ArrayFire 3.5.1
Python 3.6.3 |Anaconda custom (64-bit)| (default, Oct 15 2017, 03:27:45) [MSC v.1900 64 bit (AMD64)]
Type 'copyright', 'credits' or 'license' for more information
IPython 6.1.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: import scipy.fftpack
In [2]: import arrayfire as af
In [3]: import numpy as np
In [4]: a = np.zeros((2048,4096),dtype='float32')
In [5]: a[0:50,0:50] = 1;
In [6]: a[-50:,0:50] = 1;
In [7]: a[0:50,-50:] = 1;
In [8]: a[-50:,-50:] = 1;
In [9]: ad = af.Array(a.ctypes.data, a.shape, a.dtype.char)
In [10]: bd = af.signal.fft2(ad);
In [11]: b = scipy.fftpack.fft2(a);
In [12]: b[0:10,0:10]
Out[12]:
array([[ 10000.00000000 +0.j , 9990.19531250 +7.66216707j,
9960.81933594 +15.27928734j, 9911.97460938 +22.80728149j,
9843.83203125 +30.20039368j, 9756.63476562 +37.41600418j,
9650.68750000 +44.41161346j, 9526.36132812 +51.14674377j,
9384.09765625 +57.58042908j, 9224.39160156 +63.67562866j],
[ 9960.82031250 +15.27928543j, 9951.04296875 +22.89645386j,
9921.76953125 +30.4388485j , 9873.10449219 +37.86270905j,
9805.21875000 +45.1227417j , 9718.35156250 +52.17685699j,
9612.80664062 +58.98316193j, 9488.95800781 +65.50195312j,
9347.24218750 +71.69306183j, 9188.15234375 +77.52036285j],
[ 9843.83300781 +30.20038986j, 9834.15917969 +37.71330643j,
9805.21875000 +45.12274551j, 9757.11230469 +52.38564301j,
9690.01269531 +59.45753479j, 9604.15429688 +66.29711151j,
9499.83984375 +72.86351013j, 9377.43554688 +79.11798859j,
9237.37402344 +85.02155304j, 9080.14453125 +90.53924561j],
[ 9650.68554688 +44.41159058j, 9641.19140625 +51.76257324j,
9612.80664062 +58.98316956j, 9565.63378906 +66.03126526j,
9499.83984375 +72.8635025j , 9415.65625000 +79.43978882j,
9313.37695312 +85.72050476j, 9193.36523438 +91.66822052j,
9056.04296875 +97.24534607j, 8901.88769531+102.41835785j],
[ 9384.09570312 +57.58043289j, 9374.85253906 +64.71422577j,
9347.24218750 +71.69306946j, 9301.36132812 +78.47616577j,
9237.37304688 +85.02154541j, 9155.50488281 +91.29064941j,
9056.04199219 +97.24536133j, 8939.33398438+102.84980774j,
8805.79687500+108.0680542j , 8655.89160156+112.86830139j],
[ 9047.80371094 +69.39637756j, 9038.88085938 +76.26094055j,
9012.24804688 +82.94894409j, 8968.00195312 +89.42112732j,
8906.29687500 +95.63734436j, 8827.35156250+101.5608139j ,
8731.44335938+107.15504456j, 8618.91015625+112.38607025j,
8490.14746094+117.21989441j, 8345.60644531+121.62640381j],
[ 8646.50585938 +79.58358765j, 8637.96875000 +86.13063049j,
8612.50781250 +92.48301697j, 8570.21289062 +98.60340118j,
8511.23535156+104.4535675j , 8435.78222656+109.99856567j,
8344.11914062+115.20417786j, 8236.56640625+120.0382843j ,
8113.50683594+124.46896362j, 7975.36816406+128.46818542j],
[ 8185.78369141 +87.90079498j, 8177.69140625 +94.0867157j ,
8153.57763672+100.06369019j, 8113.52734375+105.79656982j,
8057.68310547+111.24942017j, 7986.24121094+116.38951874j,
7899.45312500+121.18466187j, 7797.62402344+125.60507965j,
7681.11328125+129.6210022j , 7550.32714844+133.20664978j],
[ 7672.00585938 +94.15422058j, 7664.41210938 +99.94036865j,
7641.80322266+105.50764465j, 7604.25781250+110.82318115j,
7551.91015625+115.85353851j, 7484.94335938+120.56845093j,
7403.59472656+124.93786621j, 7308.14843750+128.93453979j,
7198.94335938+132.53105164j, 7076.35888672+135.70358276j],
[ 7112.22558594 +98.19590759j, 7105.17773438+103.54916382j,
7084.20947266+108.67822266j, 7049.39550781+113.55259705j,
7000.85839844+118.14163208j, 6938.77050781+122.4172821j ,
6863.34960938+126.35232544j, 6774.86035156+129.92172241j,
6673.61572266+133.10049438j, 6559.96972656+135.86732483j]], dtype=complex64)
In [13]: np.reshape(af.real(bd[0:10,0:10]).to_array(),(10,10)) + 1j* np.reshape(af.imag(bd[0:10,0:10]).to_array(),(10,10))
Out[13]:
array([[ 10000.00000000 +0.j , 9960.82031250 +15.27970791j,
9843.83300781 +30.20059204j, 9650.68554688 +44.41211319j,
9384.09667969 +57.58081436j, 9047.80371094 +69.3970871j ,
8646.50585938 +79.58370972j, 8185.78369141 +87.90104675j,
7672.00585938 +94.15435791j, 7112.22460938 +98.19621277j],
[ 9960.82617188 +7.63974667j, 9922.37304688 +22.83113861j,
9806.41015625 +37.60722351j, 9614.57324219 +51.62036133j,
9349.56250000 +64.54022217j, 9015.09570312 +76.06126404j,
8615.84765625 +85.9103241j , 8157.37207031 +93.85339355j,
7646.00781250 +99.70072174j, 7088.77539062+103.31070709j],
[ 9843.85546875 +15.10028267j, 9806.41992188 +30.08589172j,
9692.37890625 +44.60408783j, 9503.34277344 +58.31238174j,
9241.97070312 +70.88645172j, 8911.93554688 +82.02682495j,
8517.84765625 +91.46707153j, 8065.19189453 +98.97968292j,
7560.22851562+104.38186646j, 7009.89208984+107.53892517j],
[ 9650.73828125 +22.20603943j, 9614.59179688 +36.87163544j,
9503.33691406 +51.02319336j, 9318.54492188 +64.32604218j,
9062.81933594 +76.46383667j, 8739.75000000 +87.14578247j,
8353.85546875 +96.11399078j, 7910.50878906+103.14953613j,
7415.84082031+108.0773468j , 6876.64550781+110.77018738j],
[ 9384.18457031 +28.79040909j, 9349.57714844 +43.02650833j,
9241.92968750 +56.7084198j , 9062.76269531 +69.51176453j,
8814.60253906 +81.13082886j, 8500.93554688 +91.28530884j,
8126.15039062 +99.72776031j, 7695.46435547+106.24877167j,
7214.83642578+110.68280029j, 6690.86962891+112.91079712j],
[ 9047.93750000 +34.69852066j, 9015.08886719 +48.40181732j,
8911.81347656 +61.51851654j, 8739.56835938 +73.73666382j,
8500.78613281 +84.76317596j, 8198.82128906 +94.33047485j,
7837.89843750+102.20306396j, 7423.04492188+108.18251801j,
6960.00292969+112.11325073j, 6455.13378906+113.88554382j],
[ 8646.68847656 +39.79174805j, 8615.79687500 +52.86640167j,
8517.59179688 +65.33032227j, 8353.46679688 +76.88651276j,
8125.73632812 +87.25634766j, 7837.60400391 +96.18662262j,
7493.10156250+103.45495605j, 7097.02929688+108.87561035j,
6654.86914062+112.30345154j, 6172.69873047+113.6383667j ],
[ 8186.01953125 +43.95049286j, 8157.24414062 +56.30944061j,
8064.73730469 +68.04281616j, 7909.81054688 +78.87002563j,
7694.65234375 +88.52960205j, 7422.28906250 +96.78328705j,
7096.53320312+103.42358398j, 6721.92529297+108.27767944j,
6303.65136719+111.21266174j, 5847.46191406+112.13720703j],
[ 7672.29492188 +47.0770874j , 7645.76708984 +58.64332581j,
7559.50195312 +69.57875061j, 7414.72460938 +79.62129974j,
7213.47998047 +88.52716827j, 6958.60009766 +96.07530212j,
6653.65625000+102.0738678j , 6302.89697266+106.36385345j,
5911.18115234+108.82404327j, 5483.89501953+109.3730011j ],
[ 7112.56347656 +49.09799957j, 7088.37988281 +59.80516052j,
7008.81201172 +69.88614655j, 6874.99218750 +79.09902954j,
6688.81152344 +87.21903229j, 6452.88964844 +94.04324341j,
6170.53320312 +99.39584351j, 5845.67919922+103.13275146j,
5482.82714844+105.14454651j, 5086.96777344+105.35946655j]], dtype=complex64)
In [14]: af.display(bd[:10,:10])
bd[:10,:10]
[10 10 1 1]
(10000.0000,0.0000) (9960.8262,7.6397) (9843.8555,15.1003) (9650.7383,22.2060) (9384.1846,28.7904) (9047.9375,34.6985) (8646.6885,39.7917) (8186.0195,43.9505) (7672.2949,47.0771) (7112.5635,49.0980)
(9960.8203,15.2797) (9922.3730,22.8311) (9806.4199,30.0859) (9614.5918,36.8716) (9349.5771,43.0265) (9015.0889,48.4018) (8615.7969,52.8664) (8157.2441,56.3094) (7645.7671,58.6433) (7088.3799,59.8052)
(9843.8330,30.2006) (9806.4102,37.6072) (9692.3789,44.6041) (9503.3369,51.0232) (9241.9297,56.7084) (8911.8135,61.5185) (8517.5918,65.3303) (8064.7373,68.0428) (7559.5020,69.5788) (7008.8120,69.8861)
(9650.6855,44.4121) (9614.5732,51.6204) (9503.3428,58.3124) (9318.5449,64.3260) (9062.7627,69.5118) (8739.5684,73.7367) (8353.4668,76.8865) (7909.8105,78.8700) (7414.7246,79.6213) (6874.9922,79.0990)
(9384.0967,57.5808) (9349.5625,64.5402) (9241.9707,70.8865) (9062.8193,76.4638) (8814.6025,81.1308) (8500.7861,84.7632) (8125.7363,87.2563) (7694.6523,88.5296) (7213.4800,88.5272) (6688.8115,87.2190)
(9047.8037,69.3971) (9015.0957,76.0613) (8911.9355,82.0268) (8739.7500,87.1458) (8500.9355,91.2853) (8198.8213,94.3305) (7837.6040,96.1866) (7422.2891,96.7833) (6958.6001,96.0753) (6452.8896,94.0432)
(8646.5059,79.5837) (8615.8477,85.9103) (8517.8477,91.4671) (8353.8555,96.1140) (8126.1504,99.7278) (7837.8984,102.2031) (7493.1016,103.4550) (7096.5332,103.4236) (6653.6563,102.0739) (6170.5332,99.3958)
(8185.7837,87.9010) (8157.3721,93.8534) (8065.1919,98.9797) (7910.5088,103.1495) (7695.4644,106.2488) (7423.0449,108.1825) (7097.0293,108.8756) (6721.9253,108.2777) (6302.8970,106.3639) (5845.6792,103.1328)
(7672.0059,94.1544) (7646.0078,99.7007) (7560.2285,104.3819) (7415.8408,108.0773) (7214.8364,110.6828) (6960.0029,112.1133) (6654.8691,112.3035) (6303.6514,111.2127) (5911.1812,108.8240) (5482.8271,105.1445)
(7112.2246,98.1962) (7088.7754,103.3107) (7009.8921,107.5389) (6876.6455,110.7702) (6690.8696,112.9108) (6455.1338,113.8855) (6172.6987,113.6384) (5847.4619,112.1372) (5483.8950,109.3730) (5086.9678,105.3595)
Can you display MATLAB results at a larger precision ?
You are also not displaying bd directly. Try to use af.display(bd[:10, :10]). The numbers should be similar to what you see in matlab.
Edited the original comment to include long e output format for Matlab and the af.display output in python.
@AndyGreenwell I think it is the way you are converting numpy array to arrayfire array. I am getting correct results if I use ad = af.to_array(a) instead of the method you are using.
@pavanky Thanks, using af.to_array is giving me the proper result after subsequently performing the fft2.
I was following the third example on this page of the docs to construct the ArrayFire array from a numpy array, so possibly that example might need updating.
Proper expected output shown below.
In [15]: ad2 = af.to_array(a);
In [16]: bd2 = af.signal.fft2(ad2);
In [17]: af.display(bd2[:10,:10])
bd2[:10,:10]
[10 10 1 1]
(10000.0000,0.0000) (9990.1953,7.6623) (9960.8193,15.2797) (9911.9746,22.8072) (9843.8320,30.2005) (9756.6348,37.4164) (9650.6855,44.4123) (9526.3613,51.1467) (9384.0977,57.5806) (9224.3916,63.6761)
(9960.8203,15.2797) (9951.0430,22.8970) (9921.7695,30.4397) (9873.1055,37.8631) (9805.2188,45.1232) (9718.3506,52.1776) (9612.8066,58.9844) (9488.9580,65.5022) (9347.2422,71.6936) (9188.1523,77.5212)
(9843.8330,30.2006) (9834.1592,37.7136) (9805.2188,45.1236) (9757.1133,52.3857) (9690.0127,59.4576) (9604.1543,66.2975) (9499.8398,72.8643) (9377.4355,79.1182) (9237.3740,85.0217) (9080.1445,90.5398)
(9650.6855,44.4121) (9641.1914,51.7633) (9612.8066,58.9841) (9565.6348,66.0319) (9499.8398,72.8640) (9415.6563,79.4405) (9313.3770,85.7216) (9193.3652,91.6685) (9056.0420,97.2461) (8901.8867,102.4192)
(9384.0967,57.5808) (9374.8525,64.7148) (9347.2422,71.6938) (9301.3613,78.4765) (9237.3740,85.0220) (9155.5039,91.2914) (9056.0420,97.2464) (8939.3340,102.8501) (8805.7969,108.0686) (8655.8916,112.8690)
(9047.8037,69.3971) (9038.8809,76.2619) (9012.2480,82.9501) (8968.0020,89.4218) (8906.2969,95.6381) (8827.3516,101.5618) (8731.4434,107.1564) (8618.9092,112.3867) (8490.1475,117.2207) (8345.6064,121.6276)
(8646.5059,79.5837) (8637.9688,86.1312) (8612.5078,92.4835) (8570.2139,98.6034) (8511.2344,104.4537) (8435.7822,109.9988) (8344.1172,115.2047) (8236.5664,120.0382) (8113.5068,124.4694) (7975.3682,128.4686)
(8185.7837,87.9010) (8177.6909,94.0872) (8153.5776,100.0644) (8113.5278,105.7970) (8057.6826,111.2499) (7986.2412,116.3899) (7899.4521,121.1855) (7797.6230,125.6052) (7681.1128,129.6216) (7550.3271,133.2072)
(7672.0059,94.1544) (7664.4121,99.9406) (7641.8032,105.5081) (7604.2588,110.8233) (7551.9102,115.8538) (7484.9434,120.5688) (7403.5938,124.9386) (7308.1484,128.9346) (7198.9434,132.5312) (7076.3594,135.7040)
(7112.2246,98.1962) (7105.1772,103.5497) (7084.2095,108.6790) (7049.3955,113.5529) (7000.8584,118.1421) (6938.7705,122.4179) (6863.3481,126.3531) (6774.8594,129.9220) (6673.6152,133.1010) (6559.9688,135.8681)
I'll fix the docs thanks.