“无注册”修饰符如何防止变量被保存到CPU寄存器中？

为什么？你的问题可能在其他地方。如果你用指针访问变量，编译器会把变量放在内存中。



      以下为原文

   
Why? Your problem is probably in other place.
If you access a variable by a pointer, the compiler will put the variable in memory

“神奇的单词”是“易失性”-告诉编译器重新读取变量，而不是假设它不改变。



      以下为原文

    The "magic word" is "volatile" - telling the compiler to re-read the variable instead assuming it doesn't change.

嗨，使用“易失性”修饰符。它在语言中，仅仅是为了你所要的目的。这是因为编译器无法知道的变量，比如硬件寄存器，或者中断服务访问的变量，或者上下文切换环境中的另一个线程。你是正确的，“易失性”修饰符排除了一些优化，不应该为不需要的变量指定。这正是你想要的。迈西尔



      以下为原文

    Hi,
Use "volatile" modifier.
 
It is in the language, just for the purpose you are asking for.
That is variables that may change for reasons that the compiler cannot know about, like hardware register,
or variables that are accessed by a interrupt service, or another thread in a context switching environment.
 
You are correct that "volatile" modifier preclude some optimizations, and should not be specified for variables that do not need it. That is exactly what you want.
 
   Mysil

寄存器本质上是易挥发的，不是吗？我是第三的解决方案是使用挥发性



      以下为原文

    registers are inherently volatile, do not they?
I am third that the solution is to use volatile

“ALLIN BILSENT”从长远来看可能是“固有的易失性”（尽管对于所有寄存器来说，这不一定是正确的），特别是在具有大寄存器集的机器上。短期内（特别是在当地情况下），它们可能是“非常静态的”。



      以下为原文

    @ al_bin
registers may be "inherently volatile" on the long run (although this is not necessarily true for all registers - especially on machines with big register sets). On the short run (especially in a local context) they may be "quite static".

“Dun900010001”相当静态，与它无关。如果你没有“易失性”，你会如何优化对寄存器的访问？MIPS没有累加器。任何代码示例；-查看：HTTP://www-OpenST.Org/JTC1/SC22/WG14/WWS/DOCS/N1124.PDFPGE 109 CH 7.7.3.6有机会在寄存器中用变量来破坏规则吗？对不起，删帖，第二次写。我没办法编辑。艾伯特



      以下为原文

    @ du00000001
 
"quite static"  has nothing to do with it.
How do you imagine optimizing access to the register if it is not "volatile" according to you?
MIPS has no accumulators.
 
Any code example ;-)
 
see:
http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1124.pdf
page 109 ch 6.7.3.6 
Any chance to break the rules with variable in register?
 
Sorry for delete post and write second time.
I was no able to edit it.
Albert
 
 

“易失关键词”正是你想要的，什么“副作用会大大降低效率”？没有它，代码就不起作用。比这更好的是效率的提高。你只把它应用到那些真正需要它的变量上。如果你所有的变量都是不稳定的，那么我会重新设计应用程序。



      以下为原文

   
The volatile keyword is exactly what you want.
 
What "side effects that greatly will reduce efficiency"? Without it the code doesn't work. Anything better than that is an improvement in efficiency.
 
You only apply it to those variables that really need it. If ALL your variables need to be volatile then I would be looking to redesign the application.

用于非易失性寄存器的ALLIN BIN实例？在MIPS、PPC、“更大”的ARM内核和类似的RISC体系结构上的任何合理编译器上保留了短const和短数据寻址的寄存器。您知道“7.03.6”中的“序列点”的含义吗？这仍然为在某些寄存器中存储的一些变量的拷贝执行相当多的操作提供了空间。（您是否意识到RISC机器DAN几乎不做任何操作而不在寄存器中保存一些变量的拷贝？）



      以下为原文

    @ al_bin
Example for non-volatile registers?
The registers reserved for short-const and short-data addressing by any reasonable compiler on MIPS, PPC, the "bigger" ARM cores and similar RISC architectures.
 
Do you know the meaning of "sequence point" in 6.7.3.6? This leaves still room for executing quite a number of operations on the copy of some variable currently stored in some register. (Are you aware that a RISC machine dan do hardly any operation without holding a copy of some variable in a register?)

关于消息7，MIPS有32个累加器，这是CPU中通用寄存器的全部。虽然在负载/存储体系结构中，CPU寄存器很少被标记为累加器。MySIPPs:当编辑一个在文本中有URL的帖子时，编辑窗口的菜单中有一个“unLink”按钮：指向链接，窗口标题中的“unLink”按钮将变得更高。单击它，然后将允许保存更新的消息。



      以下为原文

    @ al_bin,
about message #7, MIPS have 32 accumulators, that is what the General Purpose Registers in the CPU all are.
Although in a Load/Store architecture, the CPU registers are seldom noted as accumulators.
 
   Mysil
 
Ps.:
When editing a Post that have a URL in the text, there is a "Unlink" button in the menu of Edit window:
Point to the link, and the "Unlink" button in the window header will become highlited. Click it, and you will then be allowed to save the updated message.

有没有人知道，如果注册至少总是溢出功能退出？编译器不可能为全局变量编写一个寄存器，对吗？有官方文件吗？何时将寄存器写回变量？



      以下为原文

     
Does anybody know beyond any doubt if the register is always spilled at least upon function exit?

ie. it's impossible that the compiler devotes a register for a global variable, right?

Is there any official documentation on this? When is the register written back to the variable?

编译器不太可能将寄存器赋予全局变量（而不是短数据和短const指针）。除此之外，这是任何人的猜测…“毫无疑问”溢出？没有人可以想象一个同时使用很多寄存器的函数。我没有其他功能有这种巨大的“登记消费”，很有可能你能找到RESP。在函数调用之间保持不变。但这不值得依赖--只是巧合罢了。



      以下为原文

    It's highly unlikely the compiler devotes a register to a global variable (other than the short-data and short-const pointers). Beyond this, it's anyone's guess.
... spilled "beyond any doubt"? No. One might imagine a single function that uses a lot of registers at the same time. I no other function has this massive "register consumption", could well be that you'll be able to find the resp. register unaltered between function calls. But this is nothing to rely upon - mere coincidence.

在退出函数时，不需要将所有驻留在寄存器中的变量写入内存中。好的编译器可以进行进程间优化，即将同一个翻译单元中的多个函数优化到一起。这是一个官方的标准文件，它是指定C编程语言的ISO标准。您可以从ISO/IEC购买官方标准文档：HTTPS://www. ISO.Org/Stald/583.3.HTML。Al^ Bin在消息7中给出的链接是UNO。有效的文件，是提交标准的工作组提出的建议，因为它是为正式的标准提交的。迈西尔



      以下为原文

    It is Not required that all variables residing in registers be written back to memory when exiting a function.
Good compilers may do interprocedural optimization,
that is optimizing multiple functions in the same translation unit together.
 
It is possible to ask the compiler to keep a variable in register at all times, but it is usually Not a smart thing to do.
 
The official document specifying this is the ISO standard specifying the C programming language.
You may buy the official standard document from ISO/IEC:
https://www.iso.org/standard/57853.html
The link given by al_bin in message #7 is a unofficial document,
it is the suggestion given by the working group writing the standard, as it was submitted for formal standard.
 
   Mysil
 

编译器手册部分



      以下为原文

   
Compiler manual Section 12.3 REGISTER CONVENTIONS

为什么你还在写一个寄存器中的变量副本？因为您正在讨论寄存器中变量的情况。@ MySILY感谢您如何编辑一个帖子。



      以下为原文

   

Why are you still writing about a copy of a variable in register?
Since you are discussing with case of variable in the register.
 
@Mysil
Thanks for the hint on how to edit a post
 
 

GCC支持全局寄存器变量，但不会自动生成它们。如果你的程序是很小的，那么一些变量可以被优化只存在于一个寄存器中，但这不是一回事。只要编译器满足适用的标准（语言规范，ABI），它就非常自由地掌握它如何转换代码，这要归功于“以貌取人”的规则。如果编译器可以看到程序中变量的所有用法，那么它永远是不合法的。



      以下为原文

   
GCC supports global register variables, but it will not generate them automatically. If your program is trivially small then some variable could be optimized to exist only in a register, but that is not the same thing.
 
As long as the compiler fulfils the applicable standards (language spec, ABI), it has very free hands in how it transforms the code thanks to the "as-if" rule. If the compiler can see all uses of a variable in the program, it is perfectly legal for it to never write it to memory.

在函数入口/出口上使用寄存器通常记录在ABI（Application二进制接口）上，例如，HTTP//MATH-ATLAS.SooSurfGe.NET/DeVel/Calply/MIPSABI32.PDFIO-MIPS。例如，在寄存器丰富的体系结构中，前四个函数参数可以在寄存器R0 R3中传递，堆栈上的其余部分。不管参数是否是全局的，返回值也将在寄存器中传递。



      以下为原文

   
Use of registers on function entry/exit is usually documented somewhere as an ABI (Application Binary Interface), e.g., http://math-atlas.sourceforge.net/devel/assembly/mipsabi32.pdf for MIPS.
 
E.g. in a register rich architecture the first four function arguments may be passed in registers r0-r3, the remainder on the stack. Doesn't matter if the arguments are global or not.
 
Return value will also be passed in a register.

我建议将变量设置为易失性。在执行所有处理的函数中，将该易失性变量读取到本地变量并进行计算。这样，您可以确保变量读取一次希望读取的变量。编译器有机会优化本地副本。Nick



      以下为原文

    I suggest that you set that variable to volatile.
In the function that does all the processing, read that volatile variable to a local one and do your calculations.
This way you:
* Assure the variable is read the one time you wanted it to be read
* The compiler has a chance to optimize the local copy.
 
HTH,
Nick

在C的早期，语言的预期用途之一是在汇编程序之上一步，以允许驱动程序和OS和相关软件的其他重要部分的开发。“寄存器”关键字被包含为编译器的一个提示，该变量将被大量使用，如果可能的话，它被保存在一个合适的硬件寄存器中可能是个好主意。编译器生成的代码可能没有很多可用的实际寄存器，也不能解决OPS问题，因为它只是对编译器的一个提示，而且从来没有表示过一个要求，即只存在于硬件寄存器中的值。E，它从来没有抓住“一般”的编程，特别是当编译器在优化特定处理器的生成代码时变得更好。正如上面已经广泛讨论的，关键字“易失性”可能是最好的选择，但它确实具有MEM的每个访问的副作用。OLE（而不是硬件寄存器）。现在，这可能会给你额外的周期，但不会花费太多。OP的真正问题是，任何变量更新可能不是原子的，因此中断总是可能发生在“读取”之后，而发生在“写入”之前。在这个过程中，唯一的方法是在目标硬件的汇编程序中使用原子指令，或者在RMW序列发生时防止中断（在某些PIC设备中使用“DISI”指令），但也会导致额外的周期。



      以下为原文

    In the very early days of C, one of the intended uses of the language was to be one step above assembler to allow the development of drivers and other significant portions of an OS and related software. The 'register' keyword was included as a *hint* to the compiler that the variable was to be used a lot and that it might be a good idea if it was kept in a suitable hardware register if that was possible. 
As C was written as a hardware independent compiler, it could only be a hint as the device the compiler was generating code for may not have that many actual registers that were available for the job.
Also it would not solve the OPs problem in that it was only ever a hint to the compiler and never represented a requirement that the value exist *only* in a hardware register.
As far as I am aware, it never caught on for 'general' programming, especially as compilers got better at optimising the generated code for specific processors.
As has been discussed extensively above, the keyword 'volatile' is probably the best alternative but it does have the side effect of every access being to memory (rather than a hardware register). These days, that might cost you an extra cycle but not much more.
The real problem for the OP is any variable update is probably not atomic and so it is always possible for the interrupt to occur after the 'read' but before the 'write'. The only ways around this is to use an atomic instruction in assembler of such exists in the target hardware, or to prevent interrupts while the RMW sequence occurs (the 'disi' instruction in some PIC devices is used for this purpose) but that also incurs additional cycles.
Susan

人们常说，C只是一个美化的宏汇编程序。



      以下为原文

   


It's often said that C is just a glorified macro assembler.