🚀 go-pugleaf

Our moderator muses: "Interestingly enough, it [computing a function
involving four calls to cos()] still appears to be quicker [in FORTRAN]
than cosc.c [the C version]".

The answer is that in FORTRAN, cos() is an intrinsic function, and the
compiler uses the direct floating point opcode.  In C, cos() is an
external function, and the compiler generates a call to an external named
"cos".

Compare the code generated for "sin(x) + cos(x)" in FORTRAN and C.  In
FORTRAN (a5@(-0x7ffc) is x):

        fcoss   a5@(-0x7ffc),fp0
        fsins   a5@(-0x7ffc),fp1
        faddx   fp1,fp0

In C (a6@(-12) and a6@(-16) are x):

        movl    a6@(-12),sp@-
        movl    a6@(-16),sp@-
        jbsr    _sin
        addqw   #8,sp
        movl    d0,a6@(-24)
        movl    d1,a6@(-20)
        movl    a6@(-12),sp@-
        movl    a6@(-16),sp@-
        jbsr    _cos
        addqw   #8,sp
        movl    d1,sp@-
        movl    d0,sp@-
        fmoved  sp@+,fp0
        faddd   a6@(-24),fp0

Does anyone know of an optimizer which will convert calls to the standard
math routines to their 68881 instruction counterparts?

Chuck Musciano
Advanced Technology Department
Harris Corporation
(407) 727-6131
ARPA: chuck@trantor.harris-atd.com