matlab-review-fi-code

Name: matlab-review-fi-code
Author: matlab

815stars105forks

How to install matlab-review-fi-code?

matlab/matlab-agentic-toolkit/matlab-review-fi-code

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-review-fi-code

Installs into the current project.

Summary

Reviews MATLAB fixed-point (fi) code for performance, code generation efficiency, and correctness. Identifies antipatterns and suggests idiomatic improvements. Use when reviewing fi, fimath, numerictype, or quantizenumeric code.

Files · 1

View on GitHub

SKILL.md

1	# fi Best Practices Review
2
3	Reviews MATLAB code for fixed-point (`fi`) best practices and suggests improvements for performance, code generation efficiency, and correctness.
4
5	## When to Use
6
7	- Reviewing MATLAB code that uses `fi`, `fimath`, `numerictype`, or `quantizenumeric`
8	- Optimizing fixed-point simulation speed
9	- Preparing fixed-point code for C or hardware code generation
10
11	## When Not to Use
12
13	- Code using only built-in integer types (`int8`, `uint16`, etc.) without `fi`
14	- Pure floating-point algorithms with no fixed-point intent
15	- Simulink-only workflows where fixed-point is configured through block dialogs (use Fixed-Point Tool instead)
16
17	## Checklist
18
19	When reviewing code, check for ALL of the following:
20
21	### 1. Vectorize fi() Calls
22
23	Problem: Scalar `fi()` in a loop is slow due to per-element object construction overhead.
24
25	Fix: Pass entire arrays to `fi()` at once.
26
27	```matlab
28	% BAD — slow: per-element fi object construction
29	for k = 1:N
30	x_fi(k) = fi(x(k), 1, 18, 16, F);
31	end
32
33	% GOOD — fast: single vectorized call, bit-true identical result
34	x_fi = fi(x, 1, 18, 16, F);
35	```
36
37	### 2. Separate Data Types from Algorithm
38
39	Problem: Hardcoding fi types inside algorithm code makes it impossible to switch between float/fixed or compare configurations.
40
41	Fix: Use a types table with empty prototypes and `cast(...,'like',...)`.
42
43	```matlab
44	% Types table (separate function)
45	function T = mytypes(dt)
46	switch dt
47	case 'double'
48	T.b = double([]); T.x = double([]); T.y = double([]);
49	case 'single'
50	T.b = single([]); T.x = single([]); T.y = single([]);
51	case 'fixed16'
52	F = fimath('RoundingMethod','Floor','OverflowAction','Wrap', ...
53	'ProductMode','KeepLSB','ProductWordLength',32, ...
54	'SumMode','KeepLSB','SumWordLength',32);
55	T.b = fi([], 1, 16, 15, F);
56	T.x = fi([], 1, 16, 15, F);
57	T.y = fi([], 1, 16, 14, F);
58	end
59	end
60
61	% Algorithm — no hardcoded types
62	function [y,z] = myfilter(b, x, z, T)
63	y = zeros(size(x), 'like', T.y);
64	for n = 1:length(x)
65	z(:) = [x(n); z(1:end-1)];
66	y(n) = b * z;
67	end
68	end
69
70	% Entrypoint — wraps types + cast + algorithm
71	function [y,z] = entrypoint(dt, b, x)
72	T = mytypes(dt);
73	b = cast(b, 'like', T.b);
74	x = cast(x, 'like', T.x);
75	z = zeros(size(b'), 'like', T.x);
76	[y,z] = myfilter(b, x, z, T);
77	end
78	```
79
80	Validation: Run with `'double'` first, then `'single'` (catches single-precision issues early — important for embedded targets where double is unavailable or slow), then `'fixed16'`.
81
82	### 3. Prevent Bit Growth with Subscripted Assignment
83
84	Problem: `acc = acc + x(n)` overwrites `acc` with a new fi object whose type may change due to FullPrecision word growth.
85
86	Fix: Use `acc(:) = acc + x(n)` to retain the original data type.
87
88	```matlab
89	% BAD — acc type may grow each iteration
90	acc = fi(0, 1, 32, 16);
91	for n = 1:numel(x)
92	acc = acc + x(n);
93	end
94
95	% GOOD — preserves acc's declared type
96	acc = fi(0, 1, 32, 16);
97	for n = 1:numel(x)
98	acc(:) = acc + x(n);
99	end
100	```
101
102	### 4. Configure fimath for Your Target
103
104	Problem: Default fimath (Nearest rounding, Saturate overflow, FullPrecision) generates bloated code. A simple `a + b` can produce many lines of C with sign-extension and overflow checks.
105
106	Fix: Choose fimath settings based on your code generation target.
107
108	```matlab
109	% For C targets (MATLAB Coder) — models integer truncation behavior
110	F_c = fimath('RoundingMethod','Floor', 'OverflowAction','Wrap', ...
111	'ProductMode','KeepLSB', 'ProductWordLength',32, ...
112	'SumMode','KeepLSB', 'SumWordLength',32);
113
114	% For DSP processor targets — models shift-right behavior
115	F_dsp = fimath('RoundingMethod','Floor', 'OverflowAction','Wrap', ...
116	'ProductMode','KeepMSB', 'ProductWordLength',32, ...
117	'SumMode','KeepMSB', 'SumWordLength',32);
118
119	% For FPGA/hardware targets — use the built-in helper
120	% hdlfimath = Floor/Wrap/FullPrecision (hardware coder manages bit widths internally)
121	F_hw = hdlfimath;
122	x_fi = fi(x, 1, 18, 16, F_hw);
123	```
124
125	Product/Sum mode selection:
126
127	\| Mode \| Behavior \| Use when \|
128	\|------\|----------\|----------\|
129	\| `KeepLSB` \| Keep least significant bits (C integer truncation) \| Targeting C/C++ (MATLAB Coder) \|
130	\| `KeepMSB` \| Keep most significant bits (shift-right) \| Targeting DSP processors \|
131	\| `FullPrecision` \| Retain all bits (word growth) \| Hardware coder (manages widths internally), or debugging \|
132	\| `SpecifyPrecision` \| Manual word/fraction lengths \| Custom precision requirements \|
133
134	Note: `hdlfimath` returns Floor/Wrap/FullPrecision. The hardware coder manages bit widths through its own pip

People who liked this also viewed

6 picks

Type

matlab-review-fi-code

815stars105forks

View on GitHub

How to install matlab-review-fi-code?

matlab/matlab-agentic-toolkit/matlab-review-fi-code

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-review-fi-code

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# fi Best Practices Review
2
3	Reviews MATLAB code for fixed-point (`fi`) best practices and suggests improvements for performance, code generation efficiency, and correctness.
4
5	## When to Use
6
7	- Reviewing MATLAB code that uses `fi`, `fimath`, `numerictype`, or `quantizenumeric`
8	- Optimizing fixed-point simulation speed
9	- Preparing fixed-point code for C or hardware code generation
10
11	## When Not to Use
12
13	- Code using only built-in integer types (`int8`, `uint16`, etc.) without `fi`
14	- Pure floating-point algorithms with no fixed-point intent
15	- Simulink-only workflows where fixed-point is configured through block dialogs (use Fixed-Point Tool instead)
16
17	## Checklist
18
19	When reviewing code, check for ALL of the following:
20
21	### 1. Vectorize fi() Calls
22
23	Problem: Scalar `fi()` in a loop is slow due to per-element object construction overhead.
24
25	Fix: Pass entire arrays to `fi()` at once.
26
27	```matlab
28	% BAD — slow: per-element fi object construction
29	for k = 1:N
30	x_fi(k) = fi(x(k), 1, 18, 16, F);
31	end
32
33	% GOOD — fast: single vectorized call, bit-true identical result
34	x_fi = fi(x, 1, 18, 16, F);
35	```
36
37	### 2. Separate Data Types from Algorithm
38
39	Problem: Hardcoding fi types inside algorithm code makes it impossible to switch between float/fixed or compare configurations.
40
41	Fix: Use a types table with empty prototypes and `cast(...,'like',...)`.
42
43	```matlab
44	% Types table (separate function)
45	function T = mytypes(dt)
46	switch dt
47	case 'double'
48	T.b = double([]); T.x = double([]); T.y = double([]);
49	case 'single'
50	T.b = single([]); T.x = single([]); T.y = single([]);
51	case 'fixed16'
52	F = fimath('RoundingMethod','Floor','OverflowAction','Wrap', ...
53	'ProductMode','KeepLSB','ProductWordLength',32, ...
54	'SumMode','KeepLSB','SumWordLength',32);
55	T.b = fi([], 1, 16, 15, F);
56	T.x = fi([], 1, 16, 15, F);
57	T.y = fi([], 1, 16, 14, F);
58	end
59	end
60
61	% Algorithm — no hardcoded types
62	function [y,z] = myfilter(b, x, z, T)
63	y = zeros(size(x), 'like', T.y);
64	for n = 1:length(x)
65	z(:) = [x(n); z(1:end-1)];
66	y(n) = b * z;
67	end
68	end
69
70	% Entrypoint — wraps types + cast + algorithm
71	function [y,z] = entrypoint(dt, b, x)
72	T = mytypes(dt);
73	b = cast(b, 'like', T.b);
74	x = cast(x, 'like', T.x);
75	z = zeros(size(b'), 'like', T.x);
76	[y,z] = myfilter(b, x, z, T);
77	end
78	```
79
80	Validation: Run with `'double'` first, then `'single'` (catches single-precision issues early — important for embedded targets where double is unavailable or slow), then `'fixed16'`.
81
82	### 3. Prevent Bit Growth with Subscripted Assignment
83
84	Problem: `acc = acc + x(n)` overwrites `acc` with a new fi object whose type may change due to FullPrecision word growth.
85
86	Fix: Use `acc(:) = acc + x(n)` to retain the original data type.
87
88	```matlab
89	% BAD — acc type may grow each iteration
90	acc = fi(0, 1, 32, 16);
91	for n = 1:numel(x)
92	acc = acc + x(n);
93	end
94
95	% GOOD — preserves acc's declared type
96	acc = fi(0, 1, 32, 16);
97	for n = 1:numel(x)
98	acc(:) = acc + x(n);
99	end
100	```
101
102	### 4. Configure fimath for Your Target
103
104	Problem: Default fimath (Nearest rounding, Saturate overflow, FullPrecision) generates bloated code. A simple `a + b` can produce many lines of C with sign-extension and overflow checks.
105
106	Fix: Choose fimath settings based on your code generation target.
107
108	```matlab
109	% For C targets (MATLAB Coder) — models integer truncation behavior
110	F_c = fimath('RoundingMethod','Floor', 'OverflowAction','Wrap', ...
111	'ProductMode','KeepLSB', 'ProductWordLength',32, ...
112	'SumMode','KeepLSB', 'SumWordLength',32);
113
114	% For DSP processor targets — models shift-right behavior
115	F_dsp = fimath('RoundingMethod','Floor', 'OverflowAction','Wrap', ...
116	'ProductMode','KeepMSB', 'ProductWordLength',32, ...
117	'SumMode','KeepMSB', 'SumWordLength',32);
118
119	% For FPGA/hardware targets — use the built-in helper
120	% hdlfimath = Floor/Wrap/FullPrecision (hardware coder manages bit widths internally)
121	F_hw = hdlfimath;
122	x_fi = fi(x, 1, 18, 16, F_hw);
123	```
124
125	Product/Sum mode selection:
126
127	\| Mode \| Behavior \| Use when \|
128	\|------\|----------\|----------\|
129	\| `KeepLSB` \| Keep least significant bits (C integer truncation) \| Targeting C/C++ (MATLAB Coder) \|
130	\| `KeepMSB` \| Keep most significant bits (shift-right) \| Targeting DSP processors \|
131	\| `FullPrecision` \| Retain all bits (word growth) \| Hardware coder (manages widths internally), or debugging \|
132	\| `SpecifyPrecision` \| Manual word/fraction lengths \| Custom precision requirements \|
133
134	Note: `hdlfimath` returns Floor/Wrap/FullPrecision. The hardware coder manages bit widths through its own pip

People who liked this also viewed

6 picks

Type