| The Android Open Source Project | 89c1feb | 2008-12-17 18:03:55 -0800 | [diff] [blame] | 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<kind></title> |
| 7 | <link rel=stylesheet href="opcode.css"> |
| 8 | </head> |
| 9 | |
| 10 | <body> |
| 11 | |
| 12 | <h1>cmp<kind></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 & 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 | </li> |
| 86 | <li> |
| 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> |