[Issue 21526] New: x87 not rounding to precision on assignment on some platforms

[d-bugmail at puremagic.com]

https://issues.dlang.org/show_bug.cgi?id=21526

          Issue ID: 21526
           Summary: x87 not rounding to precision on assignment on some
                    platforms
           Product: D
           Version: D2
          Hardware: x86
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P1
         Component: dmd
          Assignee: nobody at puremagic.com
          Reporter: bugzilla at digitalmars.com

For the code:

 float f(float a, float b)
 {
    float c = a + b;
    float d = c - b;
    return d;
 }

When compiling with -m32 -O, on Windows c is stored to memory rounding it to
float precision. This is correct.

        push    EAX
        push    EAX
        fld     float ptr 010h[ESP]
        fadd    float ptr 0Ch[ESP]
        fstp    float ptr [ESP] // store
        fld     float ptr [ESP] // reload
        fsub    float ptr 0Ch[ESP]
        fstp    float ptr 4[ESP]
        fld     float ptr 4[ESP]
        add     ESP,8
        ret     8

On Linux, it does not, which is incorrect.

        fld     float ptr 8[ESP]
        fadd    float ptr 4[ESP]
        fsub    float ptr 4[ESP]
        ret     8

--