PRB: Bad Code Generated for "a+b == a" FP Comparison Expression

ID: Q68559

6.00 6.00a 6.00ax 7.00 | 6.00 6.00a | 1.00 1.50

MS-DOS                 | OS/2       | WINDOWS

kbtool kbprb

The information in this article applies to:

The Microsoft C/C++ Compiler (CL.EXE) included with:

    - Microsoft C for MS-DOS, versions 6.0, 6.0a, and 6.0ax
    - Microsoft C for OS/2, versions 6.0 and 6.0a
    - Microsoft C/C++ for MS-DOS, version 7.0
    - Microsoft Visual C++ for Windows, versions 1.0 and 1.5

SYMPTOMS

The compiler optimizes expressions of the form "a+b == a" if a and b are floating-point variables (double or float type).

RESOLUTION

This is expected behavior for the compiler. The proper code is generated by compiling with the /Op option, which directs the compiler to use consistent floating-point calculations. If this is not a viable option, one of the following will work around the situation:

Enclose the expression "a+b" in parentheses.
Make the variables a and b integers, if applicable.
Compile with the quick compile (/qc) or fast compile (/f) options.

MORE INFORMATION

Sample code with floating-point expressions of the form "a+b == a" was compiled with optimization disabled (/Od) and the first few lines of the resulting assembly listing follows:

; double a,b; ; if (a+b == a) ;

    *** 00000b  9b d9 ee                fldz
    *** 00000e  9b dc 16 00 00          fcom    QWORD PTR _b
    *** 000013  9b dd d8                fstp    ST(0)
    *** (lines deleted)

; if ((a+b) == a) ;

    *** 000025  9b dd 06 00 00          fld     QWORD PTR _b
    *** 00002a  9b dc 06 00 00          fadd    QWORD PTR _a
    *** 00002f  9b dc 16 00 00          fcom    QWORD PTR _a
    *** 000034  9b dd d8                fstp    ST(0)
    *** (lines deleted)

Note that the first expression gets optimized to compare variable b to zero, rather than comparing a+b to a. In the second expression, a+b is correctly compared to a.

Compiling with the /Op option generates the correct code in both cases. The expression (a+b == a) might be used with floating point numbers to detect when b is negligibly small in relation to a.

Additional reference words: 1.00 1.50 6.00 6.00a 6.00ax 7.00 8.00 8.00c conditional KBCategory: kbtool kbprb KBSubcategory: CodeGen Keywords : kb16bitonly

Last Reviewed: July 18, 1997