现代Java编译器是否会自动优化循环中的乘法中间值？

Great question—this is exactly the kind of low-level optimization modern JVM compilers (like HotSpot's C2 server compiler) are tuned to handle automatically, especially for hot code paths like your tens-of-thousands-of-iterations loop.

Let's break this down clearly:

核心优化：公共子表达式消除（CSE） #

In your example code:

arrayA[i] += doubleA * doubleB;
arrayB[i] += doubleA * doubleB;

The doubleA * doubleB calculation is a common subexpression—it's identical in both lines, and neither doubleA nor doubleB change between the two operations (assuming they're loop-invariant, which they almost certainly are here).

The C2 compiler (the aggressive optimizing compiler used in server-mode JVMs) will absolutely recognize this. It will:

• Compute doubleA * doubleB exactly once
• Store the result in a floating-point register (like an XMM register for SSE/AVX, standard for double-precision values on modern CPUs)
• Reuse that register value for both the arrayA[i] and arrayB[i] updates, instead of re-executing the multiplication instruction twice per loop iteration.

什么时候会触发这个优化？ #

This optimization kicks in once your loop becomes hot—i.e., it's executed enough times to trigger JIT compilation (the default threshold is around 10,000 iterations for most JVMs). Since you're running tens of thousands of iterations, your code will definitely hit this threshold, and the optimized machine code will replace the interpreted or lightly compiled version.

如何验证？ #

If you want to see this in action, you can enable assembly printing in the JVM with the flag:

-XX:+PrintAssembly

You'll need a debug build of the JVM or the hsdis library to view the output, but when you look at the compiled loop code, you'll only see one mulpd (or similar floating-point multiplication) instruction, followed by two addpd instructions using the same register value.

额外的 bonus 优化 #

Depending on your array sizes and CPU capabilities, the compiler might also apply further optimizations like loop unrolling or vectorization (using wider AVX registers to process multiple array elements at once), but the core register reuse for the repeated multiplication is a given.

内容的提问来源于stack exchange，提问作者user3439632