`ChainedScanDecoupledLookbackDecoupledFallback` in Unity package throws `AssertionException` if size is exactly maxElements

[minghinshi]

Example code that produces the bug:

ComputeBuffer tempBuffer0 = null;
ComputeBuffer tempBuffer1 = null;

// 4 elements
ComputeBuffer prefixSumIn = new(1, 16);
ComputeBuffer prefixSumOut = new(1, 16);

ComputeShader compute = [code that gets compute shader];
ChainedScanDecoupledLookbackDecoupledFallback prefixSum = new(compute, 4, ref tempBuffer0, ref tempBuffer1);
prefixSum.PrefixSumInclusive(4, prefixSumIn, prefixSumOut, tempBuffer0, tempBuffer1);

This throws AssertionException from line 127 in ChainedScanDecoupledLookbackDecoupledFallback.

GPUPrefixSums/GPUPrefixSumsUnity/Runtime/ChainedScanDecoupledLookbackDecoupledFallback.cs

Lines 127 to 132 in 98d93a4

	Assert.IsTrue(
	m_isValid &&
	IsAligned(prefixSumIn.stride) &&
	IsAligned(prefixSumOut.stride) &&
	size < m_allocatedSize &&
	size > k_minSize);

I believe this is caused by the size being equal to the maxElements parameter in the constructor of ChainedScanDecoupledLookbackDecoupledFallback. In the constructor, m_allocatedSize is set to 4...

GPUPrefixSums/GPUPrefixSumsUnity/Runtime/ChainedScanDecoupledLookbackDecoupledFallback.cs

Line 53 in 98d93a4

m_allocatedSize = AlignFour(maxElements);

When PrefixSumInclusive is called, the check in line 131 fails (4 < 4 is false), so an exception is raised.

But I expected maxElements to really be the maximum number of elements. If the input size is 4, the code will run. If the size is 5, it throws an error.