The function call overhead for an SVD-based matrix inverse should be tiny. Maybe if they are inverting 4x4 matrices, it's measurable. Inlining can give huge spe

The function call overhead for an SVD-based matrix inverse should be tiny. Maybe if they are inverting 4x4 matrices, it's measurable. Inlining can give huge speedups but it should not be applied to every function without thought.