forked from jiuyuan/InfiniTensor
add batch_norm
This commit is contained in:
parent
3d122aebfe
commit
514666591e
|
@ -28,9 +28,11 @@ class BatchNormCudnn : public CudaKernelWithoutConfig {
|
|||
for (size_t i = 0; i < dims.size(); ++i) {
|
||||
dimArray[i] = dims[i];
|
||||
strideArray[i] = op->getInputs(0)->getStride()[i];
|
||||
dimPArray[i] = op->getInputs(1)->getDims()[i];
|
||||
stridePArray[i] = op->getInputs(1)->getStride()[i];
|
||||
dimPArray[i] = 1;
|
||||
stridePArray[i] = 1;
|
||||
}
|
||||
dimPArray[1] = op->getInputs(0)->getDims()[1];
|
||||
stridePArray[1] = op->getInputs(0)->getStride()[1];
|
||||
// get inputs
|
||||
cudnnTensorDescriptor_t inDesc;
|
||||
checkCudnnError(cudnnCreateTensorDescriptor(&inDesc));
|
||||
|
|
Loading…
Reference in New Issue