public FloatTensor Pow(FloatTensor x, bool inline = true)
{
// Check if both tensors are compatible for sum
SameSizeDimensionsShapeAndLocation(ref x);
FloatTensor result = inline ? this : this.emptyTensorCopy();
if (dataOnGpu)
{
result.Gpu(this.shader);
if (inline)
{
result.PowElemGPU_(x); return(this);
}
else
{
return(PowElemGPU(x, result));
}
}
else
{
var nCpu = SystemInfo.processorCount;
Parallel.For(0, nCpu, workerId => {
var max = size * (workerId + 1) / nCpu;
for (var i = size * workerId / nCpu; i < max; i++)
{
result.Data [i] = (float)Math.Pow((double)Data [i], x.Data [i]);
}
});
}
return(result);
}
public FloatTensor Pow(FloatTensor x, bool inline = false, FloatTensor result = null)
{
if (!IsContiguous() || !x.IsContiguous())
{
throw new InvalidOperationException("All tensors must be contiguous, call Contiguous() to convert");
}
// Check if both tensors are compatible for sum
SameSizeDimensionsShapeAndLocation(ref x);
result = HookAutograd(ref result, ref x, "pow_elem", inline);
if (dataOnGpu)
{
result.Gpu(shader);
if (!inline)
{
return(PowElemGPU(x, result));
}
result.PowElemGPU_(x);
return(this);
}
result.Data = data.AsParallel().Zip(x.Data.AsParallel(), (a, b) => (float)Math.Pow((double)a, b))
.ToArray();
return(result);
}
public FloatTensor Pow(FloatTensor x, bool inline = false)
{
// Check if both tensors are compatible for sum
SameSizeDimensionsShapeAndLocation(ref x);
if (inline & autograd)
{
throw new InvalidOperationException("Cannot call inline functions if you intend to run backprop.");
}
FloatTensor result = inline ? this : this.emptyTensorCopy();
if (dataOnGpu)
{
result.Gpu(shader);
if (inline)
{
result.PowElemGPU_(x);
return(this);
}
else
{
return(PowElemGPU(x, result));
}
}
else
{
var nCpu = SystemInfo.processorCount;
Parallel.For(0, nCpu, workerId => {
var max = size * (workerId + 1) / nCpu;
for (var i = size * workerId / nCpu; i < max; i++)
{
result.Data [i] = (float)Math.Pow((double)Data [i], x.Data [i]);
}
});
}
HookAutograd(ref result, ref x, "pow_elem");
return(result);
}