diff options
author | Boris Fomitchev <bfomitchev@nvidia.com> | 2015-10-14 04:19:11 +0300 |
---|---|---|
committer | Boris Fomitchev <bfomitchev@nvidia.com> | 2015-10-14 04:19:11 +0300 |
commit | 651e612640928ac5259e6131ac481cb2e5094aec (patch) | |
tree | 1cc4a1105020821bcc09c80843a12187c39d43ac /SpatialConvolution.lua | |
parent | ae62fd12fd8bd7d53569855b68d70a2f320ab2f0 (diff) | |
parent | 350243991bf7c1f76648c668df2df6be644a2b31 (diff) |
Fixing conflicts
Diffstat (limited to 'SpatialConvolution.lua')
-rw-r--r-- | SpatialConvolution.lua | 14 |
1 files changed, 14 insertions, 0 deletions
diff --git a/SpatialConvolution.lua b/SpatialConvolution.lua index bbdffe2..f1597f1 100644 --- a/SpatialConvolution.lua +++ b/SpatialConvolution.lua @@ -263,6 +263,7 @@ function SpatialConvolution:createIODescriptors(input) algSearchMode = 'CUDNN_CONVOLUTION_BWD_DATA_PREFER_FASTEST' end if cudnn.benchmark then -- the manual auto-tuner is run +<<<<<<< HEAD if autotunerCache[3][autotunerHash] then algType[0] = autotunerCache[3][autotunerHash] else @@ -284,6 +285,19 @@ function SpatialConvolution:createIODescriptors(input) shape(self.weight), shape(input[input_slice]), shape(self.output[output_slice]))) end +======= + local perfResults = ffi.new("cudnnConvolutionBwdDataAlgoPerf_t[?]", 1) + local intt = torch.IntTensor(1); + errcheck('cudnnFindConvolutionBackwardDataAlgorithm', + cudnn.getHandle(), + self.weightDesc[0], self.oDesc[0], + self.convDesc[0], self.iDesc[0], + 1, intt:data(), perfResults) + algType[0] = perfResults[0].algo + if cudnn.verbose then + print('cudnnConvolutionBwdDataAlgoPerf_t\tAutoTuning:', perfResults[0].time, '\t', + tonumber(perfResults[0].memory), '\t', tonumber(perfResults[0].algo)) +>>>>>>> 350243991bf7c1f76648c668df2df6be644a2b31 end else errcheck('cudnnGetConvolutionBackwardDataAlgorithm', |