add TemporalConvolution2.lua, add test for TemporlaConvolution2, add …

…failing test for SpatialConvolution, factorize im2col out of SpatialConvolutionMM, in line with cunn

CMakeLists.txt

@@ -31,10 +31,10 @@ set(src init.cpp utils.cpp
 )
 
 set(luasrc init.lua MSECriterion.lua Pointwise.lua Threshold.lua LookupTable.lua
-  LogSoftMax.lua ClassNLLCriterion.lua StatefulTimer.lua THCLNN.lua
+  LogSoftMax.lua ClassNLLCriterion.lua StatefulTimer.lua THCLNN.lua TemporalConvolution2.lua
   Narrow.lua CMulTable.lua test.lua test/testSpatialMaxPooling.lua test/testSpatialConvolutionMM.lua
   test/testLookupTable.lua test/testMSECriterion.lua test/testSpatialUpSamplingNearest.lua
-  test/testELU.lua test/testhelpers.lua
+  test/testELU.lua test/testhelpers.lua test/testTemporalConvolution2.lua
   test/testClassNLLCriterion.lua test/testSoftMax.lua test/testLogSoftMax.lua test/testSpatialAveragePooling.lua)
 
 ADD_TORCH_PACKAGE(clnn "${src}" "${luasrc}" )

TemporalConvolution2.lua

@@ -0,0 +1,77 @@
+--[[
+
+Compared to the base TemporalConvolution, this is:
+- faster on GPUs (both in CUDA and in OpenCL)
+- slower on CPUs
+- the weights saved to file are incompatible with the original TemporalConvolution
+
+Conceptually, it's a wrapper around the highly optimized SpatialConvolutionMM class
+
+Just use it like TemporalConvolution, only with a '2' added to the class name.  Easy :-)
+
+]]--
+
+require 'torch'
+require 'nn'
+
+local TemporalConvolution2, parent = torch.class('nn.TemporalConvolution2', 'nn.Module')
+
+function TemporalConvolution2:__init(inputFrameSize, outputFrameSize, kW, dW, padW)
+   parent.__init(self)
+
+  self.inputFrameSize = inputFrameSize
+  self.outputFrameSize = outputFrameSize
+  self.kW = kW
+  self.dW = dW or 1
+  self.padW = padW or 0
+  self.sconv = nn.SpatialConvolution(inputFrameSize, outputFrameSize, 1, kW, 1, dW, 0, self.padW)
+  self.weight = self.sconv.weight
+  self.bias = self.sconv.bias
+  self.gradWeight = self.sconv.gradWeight
+  self.gradBias = self.sconv.gradBias
+end
+
+function TemporalConvolution2:clearState()
+  self.sconv:clearState()
+  parent:clearState()
+end
+
+function TemporalConvolution2:updateOutput(input)
+  assert(input:dim() == 3, 'must provide batched input')
+  local batchSize = input:size(1)
+  local numFrames = input:size(2)
+  local outFrames = numFrames - math.floor(self.kW/2)*2 + 2 * self.padW
+  if self.kW%2 == 0 then outFrames = outFrames+1 end
+
+  input = input:view(batchSize, numFrames, self.inputFrameSize, 1):transpose(2,3)
+  local output = self.sconv:updateOutput(input):transpose(2,3)
+  self.output:resize(batchSize, outFrames, self.outputFrameSize):copy(output)
+  return self.output
+end
+
+function TemporalConvolution2:updateGradInput(input, gradOutput)
+  assert(input:dim() == 3, 'must provide batched input')
+  local batchSize = input:size(1)
+  local numFrames = input:size(2)
+  local outFrames = numFrames - math.floor(self.kW/2)*2 + 2 * self.padW
+  if self.kW%2 == 0 then outFrames = outFrames+1 end
+
+  input = input:view(batchSize, numFrames, self.inputFrameSize, 1):transpose(2,3)
+  gradOutput = gradOutput:view(batchSize, outFrames, self.outputFrameSize, 1):transpose(2,3)
+  local gradInput = self.sconv:updateGradInput(input, gradOutput):transpose(2,3)
+  self.gradInput:resize(batchSize, numFrames, self.inputFrameSize):copy(gradInput)
+
+  return self.gradInput
+end
+
+function TemporalConvolution2:accGradParameters(input, gradOutput, scale)
+  assert(input:dim() == 3, 'must provide batched input')
+  local batchSize = input:size(1)
+  local numFrames = input:size(2)
+  local outFrames = numFrames - math.floor(self.kW/2)*2 + 2 * self.padW
+  if self.kW%2 == 0 then outFrames = outFrames+1 end
+
+  input = input:view(batchSize, numFrames, self.inputFrameSize, 1):transpose(2,3)
+  gradOutput = gradOutput:view(batchSize, outFrames, self.outputFrameSize, 1):transpose(2,3)
+  self.sconv:accGradParameters(input, gradOutput, scale)  
+end

init.lua

@@ -17,6 +17,8 @@ end
 
 torch.ClTensor.nn = {}
 
+include 'TemporalConvolution2.lua'
+
 include 'LookupTable.lua'
 include 'Pointwise.lua'
 include 'Threshold.lua'

lib/THCLNN/CMakeLists.txt

@@ -11,7 +11,7 @@ ENDIF()
 
 #FILE(GLOB src-cl *.cpp)
 set(src-cl Abs.cpp SpatialConvolutionMM.cpp ELU.cpp SpatialAveragePooling.cpp SpatialMaxPooling.cpp
-  SoftMax.cpp Tanh.cpp common.cpp SpatialUpSamplingNearest.cpp
+  SoftMax.cpp Tanh.cpp common.cpp SpatialUpSamplingNearest.cpp im2col.cpp
 )
 
 add_library(THCLNN MODULE ${src-cl})