Home | History | Annotate | Download | only in opcodes 
      1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
      2 
      3 <html>
      4 
      5 <head>
      6 <title>cmp&lt;kind&gt;</title>
      7 <link rel=stylesheet href="opcode.css">
      8 </head>
      9 
     10 <body>
     11 
     12 <h1>cmp&lt;kind&gt;</h1>
     13 
     14 <h2>Purpose</h2>
     15 
     16 <p>
     17 Perform the indicated floating point or long comparison, storing 0 if the two
     18 arguments are equal, 1 if the second argument is larger, or -1 if the first
     19 argument is larger. The "bias" listed for the floating point operations
     20 indicates how NaN comparisons are treated: "Gt bias" instructions return 1 for
     21 NaN comparisons, and "lt bias" instructions return -1.
     22 </p>
     23 <p>
     24 For example, to check to see if floating point a < b, then it is advisable to
     25 use cmpg-float; a result of -1 indicates that the test was true, and the other
     26 values indicate it was false either due to a valid comparison or because one
     27 or the other values was NaN. 
     28 </p>
     29 
     30 <h2>Details</h2>
     31 
     32 <table class="instruc">
     33 <thead>
     34 <tr>
     35   <th>Op &amp; Format</th>
     36   <th>Mnemonic / Syntax</th>
     37   <th>Arguments</th>
     38 </tr>
     39 </thead>
     40 <tbody>
     41 <tr>
     42   <td>2d..31 23x</td>
     43   <td>cmp<i>kind</i> vAA, vBB, vCC<br/>
     44     2d: cmpl-float <i>(lt bias)</i><br/>
     45     2e: cmpg-float <i>(gt bias)</i><br/>
     46     2f: cmpl-double <i>(lt bias)</i><br/>
     47     30: cmpg-double <i>(gt bias)</i><br/>
     48     31: cmp-long
     49   </td>
     50   <td><code>A:</code> destination register (8 bits)<br/>
     51     <code>B:</code> first source register or pair<br/>
     52     <code>C:</code> second source register or pair</td>
     53 </tr>
     54 </tbody>
     55 </table>
     56 
     57 <h2>Constraints</h2>
     58 
     59 <ul>
     60   <li>
     61     A, B and C must be valid register indices in the current stack frame.
     62   </li>
     63   <li>
     64     For the two -float variants, both vB and vC must be of type float.
     65   </li>
     66   <li>
     67     For the two -double variants, both vB and vC must be the lower part of a
     68     register pair holding a double value.
     69   </li>
     70   <li>
     71     For the -long variant, both both vB and vC must be the lower part of a
     72     register pair holding a long value.
     73   </li>
     74 </ul> 
     75 
     76 <h2>Behavior</h2>
     77 
     78 <ul>
     79   <li>
     80     The values of registers vB and vC are compared. The result, which is stored
     81     in vA, is one of the following three:
     82     <ul>
     83       <li>
     84         If vB < vC, then vA'=-1.
     85       
     86       
  • 
     87         If vB == vC, then vA'=0.
     88       </li>
     89       <li>
     90         If vC > vC, then vA'=1.
     91       </li>
     92     </ul>
     93   </li>
     94   <li>
     95     For the -float and -double variants, an addition "bias" specifies what
     96     happens if one or both of the arguments are NaN:
     97     <ul>
     98       <li>
     99         A "lt bias" results in vA'=-1.
    100       </li>
    101       <li>
    102         A "gt bias" results in vA'=1.
    103       </li>
    104     </ul>
    105   </li>
    106   <li>
    107     If v(A-1) is the lower half of a register pair, v(A-1)' becomes undefined.
    108   </li>
    109   <li>
    110     If v(A+1) is the upper half of a register pair, v(A+1)' becomes undefined.
    111   </li>
    112 </ul> 
    113     
    114 <h2>Exceptions</h2>
    115 
    116 <p>
    117 None.
    118 </p>
    119 
    120 </body>
    121 </html>
    122