Linux环境下O3优化与O0的差异性
|
4,483
|
0
|
踩坑日记
|
Zero

947 字
|
12 分钟

Linux环境下O3优化与O0的差异性

概述：在实现RISCV_ISA的P拓展指令描述仿真时，遇到了奇怪的问题，在Linux环境下O3与O0编译器优化的程序结果不同，经过调查发现，问题出在short类型指针的相关优化问题，至于这算Bug还是过度优化不好定义，但作为编译器， 无论做什么优化，至少应该保证得到的结果是正确的，下面开始描述问题。

Linux测试环境：

$ uname -a
Linux oberon 5.15.0-56-generic #62-Ubuntu SMP Tue Nov 22 19:54:14 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
$ c++ --version 
c++ (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

测试用例：

#include <iostream>

int main() {
  uint64_t rd = 0;
  uint64_t rs1 = 0x0101010101;
  uint64_t rs2 = 0x1010101010;
  int16_t *rs1_p = (int16_t *)&rs1;
  int16_t *rs2_p = (int16_t *)&rs2;
  int16_t *rd_p = (int16_t *)&rd;
  for (uint32_t i = 0; i < 4; ++i) {
    *(rd_p + i) = *(rs1_p + i) + *(rs2_p + i);
    printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
    printf("---------------- rd_p= 0x%x ------------\n", *(rd_p + i));
    printf("---------------- rd = 0x%lx -------------\n", rd);
    printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
  }
  return 0;
}

稍稍解释一下上面代码，因为P拓展是SIMD指令，功能是将uint64_t 类型的两个操作数，分别拆成uint8_t uint16_t 或 uint32_t 的几个数字来进行计算，在这里选择使用一个for循环处理逻辑相同的计算部分。

需要承认这种写法是有点怪异，其实是可以使用x86的SSE指令集的builtin函数实现，但可能有点麻烦，所以暂时就用For循环吧。 这也就造成了，这里需要使用指针获取int16_t 大小的内存地址之后用来加法运算。

补充，至少在我看开C++允许这么干，如果不行欢迎纠正。

O3&O0的编译测试：

Linux 直接开搞，无非编译O3与O0两份ELF。

$ c++ -O3 test3.cpp -o test_O3
$ c++ -O0 test3.cpp -o test_O0

结果：

$ ./test_O0 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x1111 ------------
---------------- rd = 0x1111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x1111 ------------
---------------- rd = 0x11111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
$ ./test_O3 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x3e0a ------------
---------------- rd = 0x0 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x5db6 ------------
---------------- rd = 0x0 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x5dcf ------------
---------------- rd = 0x0 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0xffffd128 ------------
---------------- rd = 0x0 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

分析：

可以很明显的看到，O3编译的结果与O0不同，原因猜测是因为，O3优化减少中间变量，尽量使用寄存器中保存的值，而不是内存中的值。

short类型的指针操作会被编译器优化掉，那么其他类型的是什么结果？？？将int16_t 换成int8_t ，具体代码就不贴了，无非就是指针类型改变一下。结果如下：

$ ./test_O3 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x11 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x11111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
$ ./test_O0 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x11 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x11111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

解决方案之volatile:

此时，O3与O0的结果已经一致，没有出现过渡优化问题，由于不是专业人士，无法评判是编译器的Bug，还是由于自身专业水平有限，不会添加编译参数造成这种现象。

那么，出问题是要解决呀，毕竟，毕竟，它和我都不想跑！！！

叮，C++神奇的关键字volatile。详细的可以看相关介绍。

简单的说，当他修饰某个对象时，就是告诉编译器，不要尝试对它进行奇怪的优化了，因为它是易变的，每次用它的时候都去内存（这里所说的内存，包括内存与缓存，由于缓存的一致性可以保障内存与缓存一致，所以不需要真的去内存取，除非Cache Miss）里面取吧。

修改后的代码：

#include <iostream>

int main() {
  volatile uint64_t rd = 0;
  volatile uint64_t rs1 = 0x0101010101;
  volatile uint64_t rs2 = 0x1010101010;
  int16_t *rs1_p = (int16_t *)&rs1;
  int16_t *rs2_p = (int16_t *)&rs2;
  int16_t *rd_p = (int16_t *)&rd;
  for (uint32_t i = 0; i < 4; ++i) {
    *(rd_p + i) = *(rs1_p + i) + *(rs2_p + i);
    printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
    printf("---------------- rd_p= 0x%x ------------\n", *(rd_p + i));
    printf("---------------- rd = 0x%lx -------------\n", rd);
    printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
  }
  return 0;
}
$ ./test_O3 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x1111 ------------
---------------- rd = 0x1111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x1111 ------------
---------------- rd = 0x11111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x11 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
---------------- rd_p= 0x0 ------------
---------------- rd = 0x1111111111 -------------
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Linuxvolatile编译器优化
暂无评论

发送评论 编辑评论 


				

			

						

				

					

						

							

								
							

							
						

					

				

				

					

						

							

								
							

							
						

					

				

				

					

						

							

								
							

							
							
															
													

					

				

			

			
			

				

					
				

			

				

											

							
							
						

																

							
							
						

																

							
							
						

										
					
											
						

	

		
|´・ω・)ノ
ヾ(≧∇≦*)ゝ
(☆ω☆)
（╯‵□′）╯︵┴─┴
￣﹃￣
(/ω＼)
∠( ᐛ 」∠)＿
(๑•̀ㅁ•́ฅ)
→_→
୧(๑•̀⌄•́๑)૭
٩(ˊᗜˋ*)و
(ノ°ο°)ノ
(´இ皿இ｀)
⌇●﹏●⌇
(ฅ´ω`ฅ)
(╯°A°)╯︵○○○
φ(￣∇￣o)
ヾ(´･ ･｀｡)ノ"
( ง ᵒ̌皿ᵒ̌)ง⁼³₌₃
(ó﹏ò｡)
Σ(っ °Д °;)っ
( ,,´･ω･)ﾉ"(´っω･｀｡)
╮(╯▽╰)╭
o(*////▽////*)q
＞﹏＜
( ๑´•ω•) "(ㆆᴗㆆ)
😂
😀
😅
😊
🙂
🙃
😌
😍
😘
😜
😝
😏
😒
🙄
😳
😡
😔
😫
😱
😭
💩
👻
🙌
🖕
👍
👫
👬
👭
🌚
🌝
🙈
💊
😶
🙏
🍦
🍉
😣
Source: github.com/k4yt3x/flowerhd
	

	

		
颜文字
Emoji
小恐龙
花!
	


									

			

			
			
		

	






上一篇