diff options
Diffstat (limited to 'docs/opcodes/opcode-2d-cmp-kind.html')
| -rw-r--r-- | docs/opcodes/opcode-2d-cmp-kind.html | 121 |
1 files changed, 121 insertions, 0 deletions
diff --git a/docs/opcodes/opcode-2d-cmp-kind.html b/docs/opcodes/opcode-2d-cmp-kind.html new file mode 100644 index 000000000..431ccd40d --- /dev/null +++ b/docs/opcodes/opcode-2d-cmp-kind.html @@ -0,0 +1,121 @@ +<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"> + +<html> + +<head> +<title>cmp<kind></title> +<link rel=stylesheet href="opcode.css"> +</head> + +<body> + +<h1>cmp<kind></h1> + +<h2>Purpose</h2> + +<p> +Perform the indicated floating point or long comparison, storing 0 if the two +arguments are equal, 1 if the second argument is larger, or -1 if the first +argument is larger. The "bias" listed for the floating point operations +indicates how NaN comparisons are treated: "Gt bias" instructions return 1 for +NaN comparisons, and "lt bias" instructions return -1. +</p> +<p> +For example, to check to see if floating point a < b, then it is advisable to +use cmpg-float; a result of -1 indicates that the test was true, and the other +values indicate it was false either due to a valid comparison or because one +or the other values was NaN. +</p> + +<h2>Details</h2> + +<table class="instruc"> +<thead> +<tr> + <th>Op & Format</th> + <th>Mnemonic / Syntax</th> + <th>Arguments</th> +</tr> +</thead> +<tbody> +<tr> + <td>2d..31 23x</td> + <td>cmp<i>kind</i> vAA, vBB, vCC<br/> + 2d: cmpl-float <i>(lt bias)</i><br/> + 2e: cmpg-float <i>(gt bias)</i><br/> + 2f: cmpl-double <i>(lt bias)</i><br/> + 30: cmpg-double <i>(gt bias)</i><br/> + 31: cmp-long + </td> + <td><code>A:</code> destination register (8 bits)<br/> + <code>B:</code> first source register or pair<br/> + <code>C:</code> second source register or pair</td> +</tr> +</tbody> +</table> + +<h2>Constraints</h2> + +<ul> + <li> + A, B and C must be valid register indices in the current stack frame. + </li> + <li> + For the two -float variants, both vB and vC must be of type float. + </li> + <li> + For the two -double variants, both vB and vC must be the lower part of a + register pair holding a double value. + </li> + <li> + For the -long variant, both both vB and vC must be the lower part of a + register pair holding a long value. + </li> +</ul> + +<h2>Behavior</h2> + +<ul> + <li> + The values of registers vB and vC are compared. The result, which is stored + in vA, is one of the following three: + <ul> + <li> + If vB < vC, then vA'=-1. + </li> + <li> + If vB == vC, then vA'=0. + </li> + <li> + If vC > vC, then vA'=1. + </li> + </ul> + </li> + <li> + For the -float and -double variants, an addition "bias" specifies what + happens if one or both of the arguments are NaN: + <ul> + <li> + A "lt bias" results in vA'=-1. + </li> + <li> + A "gt bias" results in vA'=1. + </li> + </ul> + </li> + <li> + If v(A-1) is the lower half of a register pair, v(A-1)' becomes undefined. + </li> + <li> + If v(A+1) is the upper half of a register pair, v(A+1)' becomes undefined. + </li> +</ul> + +<h2>Exceptions</h2> + +<p> +None. +</p> + +</body> +</html> |