gistfile1.lua

#!/usr/bin/env torch
require 'nn'
require 'image'
require 'xlua'
require 'pl'


opt = lapp[[
		 -t,--threads            (default 8)           number of threads
   -p,--type               (default float)       float or cuda
   -i,--devid              (default 1)           device ID (if using CUDA)
]]

p = xlua.Profiler()
torch.setnumthreads(opt.threads)
torch.manualSeed(1)
torch.setdefaulttensortype('torch.FloatTensor')


if opt.type == 'cuda' then
   print('==> switching to CUDA')
   require 'cunn'
   cutorch.setDevice(opt.devid)
   print('==> using GPU #' .. cutorch.getDevice())

   nn.SpatialConvolutionMM = nn.SpatialConvolution
end

iH = 192
iW = 192
fin = 16
fout = 32
kH = 10
kW = 10
batchSize = 128

-- input:
lena1 = torch.Tensor(batchSize,fin,iH,iW)

-- model to test:
model = nn.SpatialConvolution(fin, fout, kW, kW)

-- copy to GPU if desired:
if opt.type == 'cuda' then
   model = nn.SpatialConvolutionCUDA(fin, fout, kW, kW):cuda()
   lena1 = torch.CudaTensor(fin,iH,iW,batchSize)
end

-- test speed:
p:start('spatialconv')
lena2 = model:forward(lena1)
if opt.type == 'cuda' then cutorch.synchronize() end
p:lap('spatialconv')

p:printAll{}


print('Gops/s:', ( batchSize*fin*fout*kH*kW*((iH-kH)+1)*((iW-kW)+1)*2 ) / p:cpu('spatialconv') / 1e9 ) -- 2 operations MUL, ACC