Inside Macintosh: PowerPC Numerics / Part 3 - Numerics in PowerPC Assembly Language

Chapter 11 - Introduction to Assembly-Language Numerics

Numerics Example Using PowerPC Assembly Language

Listing 11-1 is a code example that shows when the PowerPC assembly-language numeric features might be useful. The instructions used in this example are described in the Motorola PowerPC 601 RISC Microprocessor User's Manual. This example evaluates the polynomial

$x3+2x2+-5$

It illustrates the evaluation of a polynomial

$c$₀xⁿ+c₁x^n+-1+...+c_n

using Horner's recurrence

$--\; c$₀

$--\; (r\times x)+c$_j for j = 1ton

On entry, general-purpose register GPR0 contains the degree n (<256) of the polynomial, and floating-point register F1 points to a function argument x. The coefficient table consists of $n+1$ double-format coefficients, starting with $c$₀ . In this particular polynomial, $n\; =\; 3,\; c$₀ = 1, c₁= 2, c₂ = 0, and c₃ .

Listing 11-1 Polynomial evaluation

r0:   equ   0                 # general-purpose register 0
r5:   equ   5                 # general-purpose register 5
f0:   equ   0                 # floating-point register 0
f1:   equ   1                 # floating-point register 1
f2:   equ   2                 # floating-point register 2
CTR:  equ   9                 # Count Register for loops

extern   polyeval{DS}         # export the routine descriptor
extern   .polyeval            # export the entry point
# put the code in a program control section
csect    polyeval{PR}
 
#high-level languages prepend a period to function names
.polyeval:
   lwz      r0,0(r5)          # r0 = degree
   lfd      f0,4(r5)          # f0 = leading coefficient, c0
   addic    r5,r5,4           # r5 = address of leading coeff. &c0
   mtspr    CTR,r0            # CTR = r0
loop:
   lfdu     f2,8(r5)          # f2 = next coefficient 
                              # update r5 = r5 + 8
   fmadd    f0,f0,f1,f2       # f0 = f0 * f1 + f2; ...
                              # res = res * x + c[j]
   bdnz     loop              # CTR = CTR - 1, branch if CTR  0
   fmr      f1,f0             # f1 = f0
   blr                        # return through the Link Register
   nop
 
 
#
# Set up the table of contents.  It must include at least the 
# exported routines. It may also contain global data or pointers 
# to data.
#
polyeval_TOC:  tc    polyeval{tc}, polyeval{PR}
 
 
#
# Build a transition vector for all exported routines so they can 
# be accessed through an inter-TOC call.
#
 csect   polyeval{DS}      # it's in a separate control section
 dc.l    .polyeval         # contains the entry point
 dc.l    0                 # loader will fill in correct TOC 
                           # pointer
 dc.l    0                 # save space for environment pointer