性能优化 #

% 计时
tic;
result = my_function();
elapsed = toc;
fprintf('耗时: %.4f 秒\n', elapsed);

% 更精确的计时
t = timeit(@() my_function(x));
fprintf('平均耗时: %.4f 秒\n', t);

% 启动Profiler
profile on;

% 运行代码
my_function();

% 查看结果
profile viewer;
profile off;

% 获取数据
p = profile('info');

% 不推荐：循环
n = 1e6;
result = zeros(1, n);
for i = 1:n
    result(i) = sin(i);
end

% 推荐：向量化
result = sin(1:n);

% 元素运算
A = rand(1000, 1000);
B = rand(1000, 1000);

% 矩阵运算（向量化）
C = A .* B;    % 元素乘法
C = A + B;     % 加法
C = A .^ 2;    % 元素平方

% 避免循环
% 不推荐
for i = 1:size(A, 1)
    for j = 1:size(A, 2)
        C(i, j) = A(i, j) * B(i, j);
    end
end

% 推荐
C = A .* B;

% 不推荐：手动实现
function s = my_sum(A)
    s = 0;
    for i = 1:length(A)
        s = s + A(i);
    end
end

% 推荐：使用内置函数
s = sum(A);

% 不推荐：动态扩展
tic;
data = [];
for i = 1:10000
    data = [data, rand()];
end
toc;  % 较慢

% 推荐：预分配
tic;
data = zeros(1, 10000);
for i = 1:10000
    data(i) = rand();
end
toc;  % 较快

% 预分配元胞数组
data = cell(1, 10000);
for i = 1:10000
    data{i} = rand();
end

% 预分配结构体数组
data = struct('x', cell(1, 10000), 'y', cell(1, 10000));
for i = 1:10000
    data(i).x = rand();
    data(i).y = rand();
end

% 使用合适的数据类型
data_int8 = int8(randi([0 255], 1000, 1000));   % 1 MB
data_double = double(randi([0 255], 1000, 1000)); % 8 MB

% 使用single
data_single = single(rand(1000, 1000));  % 4 MB
data_double = double(rand(1000, 1000));  % 8 MB

% 不推荐：创建副本
A = rand(10000);
B = A;  % 复制
B(1) = 0;  % 触发写时复制

% 推荐：直接操作
A = rand(10000);
A(1) = 0;

% 处理大数据后清除
large_data = rand(10000);
result = sum(large_data(:));
clear large_data;  % 释放内存

% 普通for循环
tic;
result = zeros(1, 1000);
for i = 1:1000
    result(i) = sum(rand(1000));
end
toc;

% 并行for循环
tic;
result = zeros(1, 1000);
parfor i = 1:1000
    result(i) = sum(rand(1000));
end
toc;

% 启动并行池
pool = parpool;

% 关闭并行池
delete(pool);

% 检查并行池状态
pool = gcp('nocreate');
if isempty(pool)
    parpool;
end

% parfor限制
parfor i = 1:1000
    % 不能使用break、continue
    % 变量分类要明确
    result(i) = compute(i);
end

% 广播变量
param = 0.5;
parfor i = 1:1000
    result(i) = compute(i, param);  % param是广播变量
end

% 稀疏矩阵
A = sparse(10000, 10000);
A(1, 1) = 1;
A(10000, 10000) = 1;

% 转换
A_full = full(A);  % 稀疏转全
A_sparse = sparse(A_full);  % 全转稀疏

% 对于计算密集型任务，可以使用C/C++编写MEX文件
% mex my_function.c

% MATLAB的JIT编译器会自动优化循环
% 确保循环体简单，避免复杂数据结构

function compare_performance()
    n = 1e6;
    
    % 方法1：循环
    tic;
    result1 = zeros(1, n);
    for i = 1:n
        result1(i) = sin(i);
    end
    t1 = toc;
    
    % 方法2：向量化
    tic;
    result2 = sin(1:n);
    t2 = toc;
    
    fprintf('循环: %.4f 秒\n', t1);
    fprintf('向量化: %.4f 秒\n', t2);
    fprintf('加速比: %.2fx\n', t1/t2);
end