关于浮点数出现在条件控制语句中的问题

今天修改程序，发现程序中有如下语句

TFloat32 RateColor1[6] = {0}; TFloat32 RateColor2[6] = {0}; ... if ( m_BlNatureImage && (RateColor1[3]>=0.5 || RateColor2[3]>=0.5) ) { ... }

其中条件控制语句中出现了两个浮点数相比较的情况。

记得MISRA中是明令禁止比较浮点数是否相等问题（此问题答案见C-FAQ Q14.5：http://c-faq.com/fp/fpequal.html），又禁止浮点数出现在for循环的条件控制语句上，因为浮点数比较结果会有不确定性，且会累积一些不确定的值。

 

那么是否能出现if语句的条件表达式中呢？

反正看着不舒服，就去网上找答案。

 

目前有两张方法，一个是将浮点数转换为“可以接受的精度”的定点数以后再进行比较。

比如二目运算符>,<两边同时左移10位（放大1024倍），嫌精度不够可以继续左移10位（放大1024*1024倍）。

即：

TSint32 RateColor1[6] = {0}; TSint32 RateColor2[6] = {0}; ... if ( m_BlNatureImage && ((RateColor1[3]<<10)>=512 || (RateColor2[3]<<10)>=512) ) { ... }

还有一种可以推荐的方法，我在misra的论坛上找到了。

其实它并没有回避浮点数比较，只是避免了不确定值的累积（没有验证）。

http://www.misra.org.uk/forum/viewtopic.php?f=69&t=873

 

问题：Rule 13.4 clarification

by Alexandre Langenieux » Mon Sep 01, 2008 4:38 pm

Dear Misra committee,

The rule 13.4 says:
The controlling expression of a for statement shall not contain any objects of floating type.

Does it apply only to loop counter, as suggest in the text explaining the rule ( ... floating-point variables shall not be used for this purpose [loop counter] ...). or to all operand of a controlling expression ?

I mean, can we consider the following for loop as compliant ?

extern float32_t foo(void); int32_t i; float32_t f; for (i=0; i<187 && (f > 1.0 && f < 3.0) ; i++) { ... f = foo() ... }

回答：

by misra-c » Mon Sep 08, 2008 2:52 pm

Rule 13.4 states that no objects of floating point type may appear in the controlling expression (i.e. the 2nd expression) of a for statement.
Therefore, the example given in the question violates rule 13.4. The recommended method for introducing additional conditions into the controlling expression is to use a boolean flag. So, the example given could be re-written as:

extern float32_t foo(void); int32_t i; float32_t f; bool_t flag; flag = (f > 1.0F && f < 3.0F);//标志初始化 for (i=0; i<187 && flag ; i++) { ... f = foo() ... flag = (f > 1.0F && f < 3.0F); }

Using a floating point object to count loop iterations is not recommended. For some values of the loop limits and increment, the accumulation of rounding errors could result in a number of iterations that differs from the expected number.

However, there are contexts in which use of floating point objects in a controlling expression might not cause any issues. MISRA-C:2004 does not permit any such other uses but MISRA will give due consideration to such contexts in a future version of the MISRA C rules. [ID0000010]

MISRA C Committee

看来只要是不用在循环的条件控制语句上就行！if的控制语句还是可以用的。。。