Welcome to mirror list, hosted at ThFree Co, Russian Federation.

github.com/soumith/cudnn.torch.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--Pointwise.lua4
-rw-r--r--Pooling.lua4
-rw-r--r--SpatialConvolution.lua18
-rw-r--r--SpatialSoftMax.lua4
-rw-r--r--VolumetricConvolution.lua14
-rw-r--r--init.lua26
6 files changed, 45 insertions, 25 deletions
diff --git a/Pointwise.lua b/Pointwise.lua
index bf1358a..9b3e71a 100644
--- a/Pointwise.lua
+++ b/Pointwise.lua
@@ -28,7 +28,7 @@ function Pointwise:updateOutput(input)
self:createIODescriptors(input)
if self.inplace then self.output = input end
errcheck('cudnnActivationForward',
- cudnn.handle[cutorch.getDevice()-1], self.mode,
+ cudnn.getHandle(), self.mode,
one:data(),
self.iDesc[0], input:data(),
zero:data(),
@@ -46,7 +46,7 @@ function Pointwise:updateGradInput(input, gradOutput)
self:createIODescriptors(input)
if self.inplace then self.output = input; self.gradInput = gradOutput end
errcheck('cudnnActivationBackward',
- cudnn.handle[cutorch.getDevice()-1], self.mode,
+ cudnn.getHandle(), self.mode,
one:data(),
self.iDesc[0], self.output:data(),
self.iDesc[0], gradOutput:data(),
diff --git a/Pooling.lua b/Pooling.lua
index 085e3f3..7190059 100644
--- a/Pooling.lua
+++ b/Pooling.lua
@@ -86,7 +86,7 @@ local zero = torch.FloatTensor({0});
function Pooling:updateOutput(input)
if not self.poolDesc then self:resetPoolDescriptors() end
self:createIODescriptors(input)
- errcheck('cudnnPoolingForward', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnPoolingForward', cudnn.getHandle(),
self.poolDesc[0],
one:data(),
self.iDesc[0], input:data(),
@@ -105,7 +105,7 @@ function Pooling:updateGradInput(input, gradOutput)
if not self.poolDesc then self:resetPoolDescriptors() end
self:createIODescriptors(input)
errcheck('cudnnPoolingBackward',
- cudnn.handle[cutorch.getDevice()-1], self.poolDesc[0],
+ cudnn.getHandle(), self.poolDesc[0],
one:data(),
self.oDesc[0], self.output:data(),
self.oDesc[0], gradOutput:data(),
diff --git a/SpatialConvolution.lua b/SpatialConvolution.lua
index 861d62f..78aa056 100644
--- a/SpatialConvolution.lua
+++ b/SpatialConvolution.lua
@@ -105,14 +105,14 @@ function SpatialConvolution:createIODescriptors(input)
local algWorkspaceLimit = self.nInputPlane * self.kH * self.kW * 4 -- 4 = sizeof int.
if self.fastest_mode then algSearchMode = 'CUDNN_CONVOLUTION_FWD_PREFER_FASTEST' end
errcheck('cudnnGetConvolutionForwardAlgorithm',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.iDesc[0], self.weightDesc[0],
self.convDesc[0], self.oDesc[0],
algSearchMode, algWorkspaceLimit, algType)
self.algType = algType
local bufSize = torch.LongTensor(1)
errcheck('cudnnGetConvolutionForwardWorkspaceSize',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.iDesc[0], self.weightDesc[0],
self.convDesc[0], self.oDesc[0],
algType[0], bufSize:data())
@@ -144,7 +144,7 @@ function SpatialConvolution:updateOutput(input)
if not self.weightDesc then self:resetWeightDescriptors() end
self:createIODescriptors(input)
for g=0,self.groups-1 do
- errcheck('cudnnConvolutionForward', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionForward', cudnn.getHandle(),
one:data(),
self.iDesc[0], input:data() + g*self.input_offset,
self.weightDesc[0], self.weight:data() + g*self.weight_offset,
@@ -152,7 +152,7 @@ function SpatialConvolution:updateOutput(input)
self.extraBuffer:data(), self.extraBuffer:nElement(),
zero:data(),
self.oDesc[0], self.output:data() + g*self.output_offset);
- errcheck('cudnnAddTensor', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnAddTensor', cudnn.getHandle(),
'CUDNN_ADD_SAME_C',
one:data(), self.biasDesc[0], self.bias:data() + g*self.bias_offset,
one:data(), self.oDesc[0], self.output:data() + g*self.output_offset);
@@ -167,7 +167,7 @@ function SpatialConvolution:updateGradInput(input, gradOutput)
if not self.weightDesc then self:resetWeightDescriptors() end
self:createIODescriptors(input)
for g=0,self.groups-1 do
- errcheck('cudnnConvolutionBackwardData', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionBackwardData', cudnn.getHandle(),
one:data(),
self.weightDesc[0], self.weight:data() + g*self.weight_offset,
self.oDesc[0], gradOutput:data() + g*self.output_offset,
@@ -190,13 +190,13 @@ function SpatialConvolution:accGradParameters(input, gradOutput, scale)
if not self.weightDesc then self:resetWeightDescriptors() end
for g=0,self.groups-1 do
-- gradBias
- errcheck('cudnnConvolutionBackwardBias', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionBackwardBias', cudnn.getHandle(),
self.scaleT:data(),
self.oDesc[0], gradOutput:data() + g*self.output_offset,
one:data(),
self.biasDesc[0], self.gradBias:data() + g*self.bias_offset);
-- gradWeight
- errcheck('cudnnConvolutionBackwardFilter', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionBackwardFilter', cudnn.getHandle(),
self.scaleT:data(),
self.iDesc[0], input:data() + g*self.input_offset,
self.oDesc[0], gradOutput:data() + g*self.output_offset,
@@ -209,9 +209,9 @@ end
--[[
function SpatialConvolution:zeroGradParameters()
-- gradWeight, gradBias to zero
- errcheck('cudnnSetTensor', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnSetTensor', cudnn.getHandle(),
self.weightDesc, self.gradWeight:data(), zero:data());
- errcheck('cudnnSetTensor', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnSetTensor', cudnn.getHandle(),
self.biasDesc, self.gradBias:data(), zero:data());
end
]]--
diff --git a/SpatialSoftMax.lua b/SpatialSoftMax.lua
index 97c1e38..adbb044 100644
--- a/SpatialSoftMax.lua
+++ b/SpatialSoftMax.lua
@@ -58,7 +58,7 @@ local zero = torch.FloatTensor({0});
function SpatialSoftMax:updateOutput(input)
self:createIODescriptors(input)
errcheck('cudnnSoftmaxForward',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.algorithm, self.mode,
one:data(),
self.iDesc[0], input:data(),
@@ -71,7 +71,7 @@ function SpatialSoftMax:updateGradInput(input, gradOutput)
assert(gradOutput:isContiguous());
self:createIODescriptors(input)
errcheck('cudnnSoftmaxBackward',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.algorithm, self.mode,
one:data(),
self.oDesc[0], self.output:data(),
diff --git a/VolumetricConvolution.lua b/VolumetricConvolution.lua
index 74857e2..1c71477 100644
--- a/VolumetricConvolution.lua
+++ b/VolumetricConvolution.lua
@@ -88,14 +88,14 @@ function VolumetricConvolution:createIODescriptors(input)
-- create forwardAlgorithm descriptors for
local algType = ffi.new("cudnnConvolutionFwdAlgo_t[?]", 1)
errcheck('cudnnGetConvolutionForwardAlgorithm',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.iDesc[0], self.weightDesc[0], self.convDesc[0],
self.oDesc[0], 'CUDNN_CONVOLUTION_FWD_PREFER_FASTEST',
-1, algType)
self.algType = algType
local bufSize = torch.LongTensor(1)
errcheck('cudnnGetConvolutionForwardWorkspaceSize',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.iDesc[0], self.weightDesc[0],
self.convDesc[0], self.oDesc[0],
algType[0], bufSize:data())
@@ -121,7 +121,7 @@ local zero = torch.FloatTensor({0});
function VolumetricConvolution:updateOutput(input)
if not self.weightDesc then self:resetWeightDescriptors() end
self:createIODescriptors(input)
- errcheck('cudnnConvolutionForward', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionForward', cudnn.getHandle(),
one:data(),
self.iDesc[0], input:data(),
self.weightDesc[0], self.weight:data(),
@@ -129,7 +129,7 @@ function VolumetricConvolution:updateOutput(input)
self.extraBuffer:data(), self.extraBuffer:nElement(),
zero:data(),
self.oDesc[0], self.output:data());
- errcheck('cudnnAddTensor', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnAddTensor', cudnn.getHandle(),
'CUDNN_ADD_SAME_C', one:data(),
self.biasDesc[0], self.bias:data(), one:data(),
self.oDescBias[0], self.output:data());
@@ -142,7 +142,7 @@ function VolumetricConvolution:updateGradInput(input, gradOutput)
and gradOutput:isContiguous());
if not self.weightDesc then self:resetWeightDescriptors() end
self:createIODescriptors(input)
- errcheck('cudnnConvolutionBackwardData', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionBackwardData', cudnn.getHandle(),
one:data(),
self.weightDesc[0], self.weight:data(),
self.oDesc[0], gradOutput:data(),
@@ -164,14 +164,14 @@ function VolumetricConvolution:accGradParameters(input, gradOutput, scale)
self:createIODescriptors(input)
if not self.weightDesc then self:resetWeightDescriptors() end
-- gradBias
- errcheck('cudnnConvolutionBackwardBias', cudnn.handle[cutorch.getDevice()-1],
+ errcheck('cudnnConvolutionBackwardBias', cudnn.getHandle(),
self.scaleT:data(),
self.oDescBias[0], gradOutput:data(),
one:data(),
self.biasDesc[0], self.gradBias:data());
-- gradWeight
errcheck('cudnnConvolutionBackwardFilter',
- cudnn.handle[cutorch.getDevice()-1],
+ cudnn.getHandle(),
self.scaleT:data(),
self.iDesc[0], input:data(),
self.oDesc[0], gradOutput:data(),
diff --git a/init.lua b/init.lua
index 5867a99..6eeef67 100644
--- a/init.lua
+++ b/init.lua
@@ -5,7 +5,21 @@ include 'ffi.lua'
local C = cudnn.C
local ffi = require 'ffi'
+local initialized = false
+local maxStreamsPerDevice = 100
+
+function cudnn.getHandle()
+ local curStream = cutorch.getStream()
+ assert(curStream < maxStreamsPerDevice, 'cudnn bindings only support max of : '
+ .. maxStreamsPerDevice .. ' streams per device')
+ return cudnn.handle[(((cutorch.getDevice()-1)*maxStreamsPerDevice) + curStream)]
+end
+
local errcheck = function(f, ...)
+ if initialized then
+ C.cudnnSetStream(cudnn.getHandle(),
+ ffi.C.THCState_getCurrentStream(cutorch.getState()))
+ end
local status = C[f](...)
if status ~= 'CUDNN_STATUS_SUCCESS' then
local str = ffi.string(C.cudnnGetErrorString(status))
@@ -16,11 +30,13 @@ cudnn.errcheck = errcheck
local numDevices = cutorch.getDeviceCount()
local currentDevice = cutorch.getDevice()
-cudnn.handle = ffi.new('struct cudnnContext*[?]', numDevices)
+cudnn.handle = ffi.new('struct cudnnContext*[?]', numDevices*maxStreamsPerDevice)
-- create handle
for i=1,numDevices do
cutorch.setDevice(i)
- errcheck('cudnnCreate', cudnn.handle+i-1)
+ for j=0,maxStreamsPerDevice-1 do
+ errcheck('cudnnCreate', cudnn.handle+(((i-1)*maxStreamsPerDevice) + j))
+ end
end
cutorch.setDevice(currentDevice)
@@ -28,12 +44,16 @@ local function destroy(handle)
local currentDevice = cutorch.getDevice()
for i=1,numDevices do
cutorch.setDevice(i)
- errcheck('cudnnDestroy', handle[i-1]);
+ for j=0,maxStreamsPerDevice-1 do
+ errcheck('cudnnDestroy', handle[(((i-1)*maxStreamsPerDevice) + j)]);
+ end
end
cutorch.setDevice(currentDevice)
end
ffi.gc(cudnn.handle, destroy)
+initialized = true
+
function cudnn.toDescriptor(t)
assert(torch.typename(t) == 'torch.CudaTensor')
local descriptor = ffi.new('struct cudnnTensorStruct*[1]')
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-25 19:13:11 +0300