blocks|key|4112580|text|来自Accelerate+documentation|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|4112581|4112582|+Accelerate通过利用其向量处理能力，在CPU上提供高性能、高能效的计算。下面的加速库对该功能进行抽象，以便为它们编写的代码在运行时为可用的处理器执行适当的指令。|blockquote|4112583|4112584|entityMap|0|LINK|mutability|MUTABLE|url|https://developer.apple.com/documentation/accelerate^0|D|D|0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|Q|8|@]|9|@$A|R|B|S|1|T]]|C|$]]|$1|D|3|-4|5|6|7|U|8|@]|9|@]|C|$]]|$1|E|3|F|5|G|7|V|8|@]|9|@]|C|$]]|$1|H|3|-4|5|6|7|W|8|@]|9|@]|C|$]]|$1|I|3|-4|5|6|7|X|8|@]|9|@]|C|$]]]|J|$K|$5|L|M|N|C|$O|P]]]]

From Accelerate <a href="https://developer.apple.com/documentation/accelerate" rel="nofollow noreferrer">documentation</a>:

<blockquote>
 Accelerate provides high-performance, energy-efficient computation on the CPU by leveraging its vector-processing capability. The following Accelerate libraries abstract that capability so that code written for them executes appropriate instructions for the processor available at runtime.
</blockquote>

blocks|key|793865|text|SIMD代表“单指令，多数据”。这意味着某些指令可以同时对多组数据进行操作。不同类型的硬件支持不同的SIMD指令。在您的示例中，您能够同时执行三个浮点加法操作。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|793866|https://developer.apple.com/documentation/accelerate/simd|offset|length|793867|https://en.wikipedia.org/wiki/SIMD|793868|entityMap|0|LINK|mutability|MUTABLE|url|1^0|0|0|1L|0|0|0|Y|1|0^^$0|@$1|2|3|4|5|6|7|P|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|Q|8|@]|9|@$D|R|E|S|1|T]]|A|$]]|$1|F|3|G|5|6|7|U|8|@]|9|@$D|V|E|W|1|X]]|A|$]]|$1|H|3|-4|5|6|7|Y|8|@]|9|@]|A|$]]]|I|$J|$5|K|L|M|A|$N|C]]|O|$5|K|L|M|A|$N|G]]]]

SIMD stands for "Single instruction, multiple data". What that means is that certain instructions can operate on multiple sets of data simultaneously. Different types of hardware support different SIMD instructions. In your case, you were able to perform three floating point addition operations simultaneously.

<a href="https://developer.apple.com/documentation/accelerate/simd" rel="nofollow noreferrer">https://developer.apple.com/documentation/accelerate/simd</a>

<a href="https://en.wikipedia.org/wiki/SIMD" rel="nofollow noreferrer">https://en.wikipedia.org/wiki/SIMD</a>

blocks|key|4112664|text|首先，这些是非常不同的操作，所以在微基准中比较它们并不是特别有用。您的第一个示例可以添加在运行时确定的任意数量的值。第二个示例恰好添加了3对。这意味着第一个必须为存储分配内存，这可能是非常昂贵的。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|4112665|此外，请确保仔细分析，并打开优化。第一个例子非常容易被优化。第二个不是，像这样的非常小的操作的时间是非常困难的。测量的行为可能会使你的数字发生很大的偏差。(也就是说，我对这些数字并不感到特别惊讶。)|4112666|但正确的比较应该是这样：|4112667|let+p+=+(1.0,+2.0,+3.0)
let+q+=+(10.0,+20.0,+3.0)

let+result+=+(p.0+%2B+q.0,+p.1+%2B+q.1,+p.2+%2B+q.2)|code-block|syntax|javascript|4112668|这与double3代码做了同样的事情，我希望它的速度大致一样快。它可能会更快或更慢，但它应该很接近，因为优化器很可能能够将其转换为非常类似于simd代码的内容。|offset|length|style|CODE|4112669|有关SIMD代码真正在做什么的更多信息，请参阅Rob的答案，但关键功能是它是一个硬件级别的功能，可以将相同的操作同时应用于多个值。不是多线程意义上的并发，而是由硬件同时进行的。但要使用此功能，输入和输出必须采用非常精确的格式和布局。(实际上，您正在使用的double3实际上在内部转换为double4，因为操作一次在2或4个Double上完成；最后一个被忽略。)|4112670|entityMap^0|0|0|0|0|2|7|1Y|4|0|3J|7|3Z|7|0^^$0|@$1|2|3|4|5|6|7|U|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|V|8|@]|9|@]|A|$]]|$1|D|3|E|5|6|7|W|8|@]|9|@]|A|$]]|$1|F|3|G|5|H|7|X|8|@]|9|@]|A|$I|J]]|$1|K|3|L|5|6|7|Y|8|@$M|Z|N|10|O|P]|$M|11|N|12|O|P]]|9|@]|A|$]]|$1|Q|3|R|5|6|7|13|8|@$M|14|N|15|O|P]|$M|16|N|17|O|P]]|9|@]|A|$]]|$1|S|3|-4|5|6|7|18|8|@]|9|@]|A|$]]]|T|$]]

First, these are wildly different operations, so comparing them in a microbenchmark is not particularly informative. Your first example can add an arbitrary number of values, determined at runtime. The second example adds precisely 3 pairs. This means the first has to allocate memory for storage, which can be extremely expensive.

Also, make sure that you're profiling carefully, and with optimizations turned on. The first example is very subject to optimizations. The second isn't. Timing very small operations like this is extremely difficult. The act of measurement can skew your numbers a lot. (That said, I'm not particularly surprised by these numbers.)

But the correct thing to compare to would be this:

<pre><code>let p = (1.0, 2.0, 3.0)
let q = (10.0, 20.0, 3.0)

let result = (p.0 + q.0, p.1 + q.1, p.2 + q.2)
</code></pre>

That does the same thing as the <code>double3</code> code, and I would expect it to be roughly as fast. It might be a little faster or slower, but it should be close, since the optimizer will likely be able to convert it into something very similar to the <code>simd</code> code.

See Rob's answer for more what the SIMD code is really doing, but the key feature is that it is a hardware level feature that can apply the same operation to multiple values simultaneously. Not in the multi-threaded sense of concurrently, but literally at the same time by the hardware. But to use this feature, the input and output must be in very precise formats and layouts. (In fact, the <code>double3</code> you're using is actually converted to a <code>double4</code> internally, because the operations are done on 2 or 4 Doubles at a time; the last one is just ignored.)

When benchmarking both of the vector and normal method there is difference in time.

Normal method

<pre><code>let arrayA: [Float] = [1,2,3]
let arrayB: [Float] = [10,20,30]

let arrayResult:[Float] = zip(arrayA,arrayB).map() {$0 + $1}

Time required: 0.0025310516357421875
</code></pre>

with Accelerate

<pre><code>let p = double3(1, 2, 3)
let q = double3(10, 20, 30)

print(p + q)

Time required: 0.0009039640426635742
</code></pre>

How it is working internally. It's become 2.79x faster in new method.

Accelerate vs Parallel Addition in Swift

翻译质量差，导致语言生硬或混乱。

没有提供实际的解决方法或示例。

解答不清晰，无法理解或解决问题。

页面排版不美观，阅读体验差。

文章

问答

视频

学习中心

腾讯云实验室

直播

竞赛

腾讯云代码分析专区

腾讯iOA零信任安全管理系统专区

腾讯云架构师技术同盟交流圈

腾讯云数据库专区

腾讯云顾问专区

腾讯云原生专区

腾讯混元专区

腾讯云TCE专区

腾讯云Lighthouse专区

腾讯云HAI专区

腾讯云Edgeone专区

腾讯云存储专区

腾讯云智能专区

腾讯轻联专区 

腾讯云开发专区

TAPD专区

腾讯轻量云游戏服专区

腾讯云最具价值专家

腾讯云架构师技术同盟

腾讯云创作之星

腾讯云开发者先锋

腾讯云代码助手

云原生构建

TAPD 敏捷项目管理

Cloud Studio

SDK中心

API中心

命令行工具

涵盖代码开发、场景应用、自动测试全流程，助你从零构建专属AI助手

一站式MCP教程库，解锁AI应用新玩法

当对向量法和正态法都进行基准测试时，存在时间上的差异。正态方法let arrayA: [Float] = [1,2,3]let arrayB: [Float] = [10,20,30]let arrayResult:[Float] = zip(arrayA,arrayB).map() {$0 + $1}Time req...

问在Swift中加速vs并行加法
EN

Stack Overflow用户

社区

活动

圈层

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问在Swift中加速vs并行加法EN