C++编译器与操作系统调用约定解析

需积分: 2 89 浏览量更新于2024-07-09 收藏 1.02MB PDF 举报

"调用公约，不同os下编译差异" 这篇文档《calling_conventions.pdf》主要探讨了在不同C++编译器和操作系统中调用约定（也称为调用约定或函数调用规范）的差异。作者Agner Fog详细阐述了与编译相关的各种技术细节，包括数据表示、对齐、栈对齐、寄存器使用、函数调用约定以及名字修饰（名称混淆）。以下是关键知识点的概述： 1. **调用约定标准化的需求**：调用约定是编程语言中约定如何传递参数、返回值和处理函数调用的规则。由于不同的编译器和操作系统可能有不同的实现，因此需要标准化来确保跨平台兼容性。 2. **数据表示**：不同平台和编译器可能采用不同的方式来表示数据类型，如整型、浮点型等，这会影响到函数调用中的数据传递。 3. **数据对齐**：数据对齐是指内存中的数据结构按照特定规则排列，以优化访问效率。不同的系统可能有不同的对齐策略，这对函数参数的存储和传递有直接影响。 4. **栈对齐**：栈在函数调用时用于存储临时数据和参数。栈对齐确保每次函数调用后栈指针保持在特定的边界上，以满足处理器的性能要求。 5. **寄存器使用**：不同的编译器可能会为不同类型的函数参数分配特定的寄存器，例如，某些编译器可能在64位Windows中允许使用浮点寄存器，而在其他系统中可能不支持。此外，还有对YMM向量寄存器、ZMM向量寄存器的讨论，这些是用于高速处理矢量化数据的寄存器。 6. **函数调用约定**：这部分详细介绍了如何传递和返回对象，特别是SIMD（单指令多数据）类型，这是现代处理器中用于加速计算的特性。 7. **名字修饰**（Name Mangling）：为了解决C++的命名空间、重载函数等问题，编译器会对标识符进行编码，形成机器可识别的名称。文档列举了微软、Borland、Watcom、GNU、Intel、Symantec和Digital Mars等编译器的名字修饰规则，并说明了如何通过`extern "C"`来关闭名字修饰。 8. **异常处理和栈展开**：异常处理涉及到在出现错误时如何恢复程序状态，而栈展开是异常处理的一部分，它用于回溯栈上的函数调用记录。 9. **初始化和终止**：这部分可能涉及程序启动时的初始化过程以及程序结束时的清理工作，这些操作可能受调用约定影响。这份文档对理解跨平台的C++程序设计和编译器行为至关重要，对于开发人员来说，深入理解这些概念可以帮助编写更高效、兼容性更好的代码。

pointer is always 8 modulo 16 at the entry of a procedure. A procedure must subtract an

odd multiple of 8 from the stack pointer before any call instruction. A procedure can rely on

these rules when storing XMM data that require 16-byte alignment. This applies to all 64 bit

systems (Windows, Linux, BSD).

Where at least one function parameter of type __m256 is transferred on the stack, Unix

systems (32 and 64 bit) align the parameter by 32 and the called function can rely on the

stack being aligned by 32 before the call (i.e. the stack pointer is 32 minus the word size

modulo 32 at the function entry). This does not apply if the parameter is transferred in a

Various methods for aligning the stack are described in Intel's application note AP 589

"Software Conventions for Streaming SIMD Extensions", "Data Alignment and Programming

Issues for the Streaming SIMD Extensions with the Intel® C/C++ Compiler", and "IA-32 Intel

® Architecture Optimization Reference Manual".

6 Register usage

Table 4. Register usage

16 bit

DOS,

Windows

32 bit

Windows

32 bit

Linux,

BSD, Mac

64 bit

Windows

64 bit

Linux,

BSD, Mac

scratch

registers

AX, BX, CX,

DX, ES,

ST(0)-ST(7)

EAX, ECX,

EDX,

ST(0)-

ST(7),

XMM0-XMM7,

YMM0-YMM7,

ZMM0-ZMM7,

K0-K7

EAX, ECX,

EDX,

ST(0)-

ST(7),

XMM0-XMM7,

YMM0-YMM7,

ZMM0-ZMM7,

K0-K7

RAX, RCX,

RDX, R8-R11,

ST(0)-

ST(7),

K0-K7,

XMM0-XMM5,

All YMM/ZMM

registers

except the

lower 128

bits of XMM6-

XMM15

RAX, RCX,

RDX, RSI,

RDI,

R8-R11,

ST(0)-

ST(7)

K0-K7,

XMM0-

XMM15,

YMM0-YMM15

ZMM0-ZMM31

callee-save

registers

SI, DI, BP,

EBX, ESI,

EDI, EBP

EBX, ESI,

EDI, EBP

RBX, RSI,

RDI, RBP,

R12-R15,

XMM6-XMM15,

but nothing

beyond the

lower 128

bits of vector

registers

RBX, RBP,

R12-R15

registers

for

parameter

transfer

see table 5

RCX, RDX,

R8, R9,

XMM0-XMM3,

YMM0-YMM3,

ZMM0-ZMM3

RDI, RSI,

RDX, RCX,

R8, R9,

XMM0-XMM7,

YMM0-YMM7,

ZMM0-ZMM7

registers

for return

AX, DX,

ST(0)

EAX, EDX,

ST(0),

XMM0, YMM0,

ZMM0

EAX,

ST(0),

XMM0, YMM0,

ZMM0

RAX, ST(0),

XMM0, YMM0,

ZMM0

RAX, RDX,

ST(0),

XMM0, YMM0,

ZMM0

The rules for register usage depend on the operating system, as shown in table 4. Scratch

registers are registers that can be used for temporary storage without restrictions (also

called caller-save or volatile registers). Callee-save registers are registers that you have to

save before using them and restore after using them (also called non-volatile registers). You

can rely on these registers having the same value after a call as before the call. For

example, a function using EBP may look like this

FunctionUsingEBP PROC NEAR

push ebp

mov ebp, esp

sub esp, 52

...

mov eax, [ebp+8]

push eax

call AnotherFunction

mov esp, ebp

pop ebp

ret

FunctionUsingEBP ENDP

Here, EBP is saved on the stack in the beginning of the function and restored in the end. The

code relies on EBP being unchanged after the call to AnotherFunction. EAX is also used,

but doesn't have to be saved.

It is more efficient to use registers for transferring parameters to a function and for receiving

the return value than to store these values on the stack. Some calling conventions use

certain registers for parameter transfer, but the rules for which registers to use are compiler-

specific in 16-bit and 32-bit systems. In 64-bit systems, the use of registers for parameter

transfer is standardized. All systems use registers for return values if the returned object fits

into the registers that are assigned for this purpose. See the next chapter for details.

Segment registers

You only have to care about segment registers in 16-bit mode. DS has to be saved and

restored if you change it. ES can be changed freely. In DOS programs, ES can have any

value. In 16-bit Windows, ES can only have values that are valid segment descriptors. It is

not allowed to use ES for other purposes.

In 32-bit and 64-bit mode, it is not allowed to change any segment register, not even

temporarily. CS, DS, ES and SS all point to the flat segment group. FS is used for a thread

environment block in Windows and for thread specific data in Linux. GS is used for a

processor control region in 64-bit Windows. It is unused but reserved in 32-bit Windows. It is

probably unused in 32-bit Linux.

Arithmetic flags

The rules for the arithmetic flags (zero flag, carry flag, etc.) are the same as for scratch

registers. These flags need not be saved. Some programming languages (not C++) use the

carry flag for Boolean returns.

Direction flag

The rules for the direction flag is the same in all systems. The direction flag is cleared by

default. If the direction flag is set, then it must be cleared again before any call or return.

Some compilers and subroutine libraries rely on the direction flag always being clear

(Microsoft, Watcom, Digital Mars) while other systems use the double-safe strategy of

always leaving the direction flag cleared, but not relying on receiving it cleared (Borland,

Gnu).

There is a slight possibility that some programmers may have ignored the rule for the

direction flag. Therefore, it may be wise to use the double-safe strategy and clear the

direction flag before using it if the code will be linked together with modules from unreliable

sources.

Interrupt flag

It is not allowed to turn off the interrupt flag in programs running in multi-user systems

because this would make it possible to steal unlimited amounts of CPU time from other

processes. It may be possible to turn off the interrupt flag in console mode programs

running under Windows 98 and earlier operating systems without network. But since

programs written for old operating systems are likely to be run under newer operating

systems, it is reasonable to say that it is never possible to turn off the interrupt flag in

application programs.

Floating point registers

The floating point registers ST(0)-ST(7) need not be saved. The register stack must be

emptied before any call or return, except for registers used for return values. The 64-bit

Microsoft compiler does not use ST(0)-ST(7).

MMX registers

The MM0-MM7 registers are aliased on the lower 64 bits of the floating point x87 registers

ST(0)-ST(7). There is no callee save rule, so the MMX registers can be used freely in a

function that doesn't use the floating point registers. The register set must be left in x87

mode. Therefore, it is required to issue an EMMS instruction (or FEMMS) before calling any

other (ABI compliant) function and before returning. Unfortunately, not all compilers do so.

Therefore, the use of MMX registers and the __m64 type should be avoided if possible. The

64-bit Microsoft compiler does not use MM0-MM7.

Floating point control word and MXCSR register

The floating point control word and bit 6-15 of the MXCSR register must be saved and

restored before any call or return by any procedure that needs to modify them, except for

procedures that have the purpose of changing these.

Deviating from the conventions

It is possible to deviate from the register usage conventions in an isolated section of code

as long as all interfaces to other parts of the code conform to the conventions. Some

compilers do this in a process known as whole program optimization. Any deviation from the

conventions must be well documented. Deviations from good programming practice are

justified only if a significant gain in speed can be obtained.

ABI for 64 bit Windows has been changed

Early versions of the 64 bit Windows ABI specified that only the lower 64 bits of XMM6-XMM15

have callee-save status while later versions specify that all 128 bits must be saved. An

MSDN document dated June 14, 2004 specifies the now-obsolete rule while a later version

dated February 18, 2005 specifies the new rule without comments on the change. The

change is mentioned in Intel compiler manuals. The 2005 standard is supported by Intel

C++ compiler version 8.1.015 and later. My tests show that Microsoft compiler version

14.00.2228.2 uses the obsolete convention, while version 14.00.40310.41 uses the new

convention. I have no information about Microsoft compiler versions between these two

numbers.

Microsoft 16-bit compiler

The 16-bit Microsoft compiler returns float and double through a static memory location

pointed to by AX. long double is returned in ST(0).

剩余58页未读，继续阅读

通下水道技术一流

粉丝: 107
资源: 7

C++编译器与操作系统调用约定解析

丹麦人总结的代码优化资料

程序的优化手册

file_point = select_file.rectangle（） point = （int(file_point.left+file_point.right）/2，int(file_point.top+file_point.Button）/2

Main.Func_CanDataProcess.ReadCanData(UniversalData.CanBus1, QNX_canvar.R_205_0, false , this, "setYuanKongJingKong");

最新资源