matlab-review-fi-object-code

Vectorize fi() calls by passing entire arrays instead of constructing objects element-by-element in loops, and preallocate output arrays with zeros(...,'like',prototype). For further acceleration, use fiaccel to compile to MEX or quantizenumeric to stay in double.

For C targets with MATLAB Coder, use Floor rounding, Wrap overflow, and KeepLSB product and sum modes with 32-bit word lengths to model integer truncation. Default fimath (Nearest, Saturate, FullPrecision) generates bloated code with sign-extension and overflow checks.

Use quantizenumeric when your algorithm stays in double and you only need to inject quantization effects at specific points, avoiding fi object overhead. Use fi when you need full fixed-point arithmetic rules, instrumentation, or code generation.

The assignment acc = acc + x(n) creates a new fi object whose type grows under FullPrecision fimath. Use subscripted assignment acc(:) = acc + x(n) to preserve the accumulator's declared numerictype across iterations.

For C targets, use FunctionApproximation.Problem from Fixed-Point Designer to generate optimized lookup tables. For FPGA or hardware targets, use CORDIC functions like cordicsin, which map efficiently to shift-add logic without large ROM.

Skip this review for code using only built-in integer types like int8 without fi objects, pure floating-point algorithms with no fixed-point intent, or Simulink-only workflows where fixed-point is configured through block dialogs.