Test shows the mask generated is equal even on small sizes (not in test but locally modyfied to test for an apparent difference at small sizes)
Code seems to behave the same as the Scalar version.
Benchmark
PASS : KisMaskGeneratorBenchmark::testRectangularSoftScalarMask()
RESULT : KisMaskGeneratorBenchmark::testRectangularSoftScalarMask():
35 msecs per iteration (total: 71, iterations: 2)
PASS : KisMaskGeneratorBenchmark::testRectangularSoftVectorMask()
RESULT : KisMaskGeneratorBenchmark::testRectangularSoftVectorMask():
8.9 msecs per iteration (total: 72, iterations: 8)