Difference between revisions of "System Function DR"

From NARS2000
Jump to navigationJump to search
Line 183: Line 183:
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;⎕ct←0 ⋄ ⎕pp←99<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;6403 ⎕dr ¯64↑1<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;6403 ⎕dr ¯64↑1<br />
¯0<br />
¯0<br />
Line 188: Line 189:
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;1 ⎕dr QNaN<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;1 ⎕dr QNaN<br />
FFF8000000000000<br />
FFF8000000000000<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;⎕ct←0 ⋄ QNaN=1 2 3 ∞<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;QNaN=1 2 3 ∞<br />
1 1 1 1<br />
1 1 1 1<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;1 ⎕dr '000fffffffffffff'</apll> &mdash; The largest positive denormal<apll><br />
2.2250738585072009E¯308<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;1 ⎕dr '0000000000000001'</apll> &mdash; The smallest positive denormal<apll><br />
4.9406564584124654E¯324<br />

Revision as of 15:52, 9 October 2008

Monadic Function

Z←⎕DR R returns the data representation of R.
R is an arbitrary array.
Z is a numeric scalar which represents the datatype of R.

The datatypes are encoded with a unique index in the low-order two digits, and the bits per element in the remaining digits:

  •  100: Boolean, one bit per element
  • 1601: Character, 16 bits per element
  • 6402: Integer, 64 bits per element
  • 6403: Floating Point, 64 bits per element (double precision)
  • 6404: Arithmetic Progression Array, 64 bits each for the offset and multiplier
  • 3208: Heterogeneous, 32 bits per element (each is a pointer)
  • 3209: Nested, 32 bits per element (each is a pointer)

For example,

      ⎕DR 1 0 1
      ⎕dr ⌈/⍬
      ⎕dr 2 64⍴1
— Note this is really an APA, not Boolean
      ⎕dr 2 64⍴1 1
— This one is Boolean because Reshape produces APAs for integer singleton right arguments only

Dyadic Function

Z←L ⎕DR R converts R to the datatype indicated by L.
R is an arbitrary array.
L is an integer scalar or one-element vector datatype (see the table above), or a special value (see below).
Z is R where each of the values are converted to the datatype indicated by L.

If the conversion is from a narrower datatype to a wider datatype, there must be exactly enough columns in the right argument to match a multiple of the size of the wider datatype. For example, when converting from character (16-bit) to integer (64-bit), the last column of the right argument must be a multiple of 4 (= 64/16); otherwise, a LENGTH ERROR is signalled.

      6402 ⎕dr 'NARS2000' — Eight 16-bit characters (=128 bits) convert to two 64-bit integers
23362775258562638 13511005043687474

If the conversion is from a wider datatype to a narrower datatype, the number of values in the result is a multiple of the ratio of the wider datatype to the narrower datatype. For example, when converting from integer (64-bit) to character (16-bit), the last column in the result is the product of last column of the right argument and 4 (= 64/16).

For example,

      ⍴⎕←1601 ⎕dr 23362775258562638 13511005043687474

Keep in mind how Arithmetic Progression Arrays are created and represented as they can fool you. For example, you might try the following to convert from Boolean to integer:

      6402 ⎕dr 2 64⍴1
1 0 2 64

However, this doesn't produce the expected result because this particular right argument is an APA, not a Boolean vector. That is, under certain circumstances, the reshape primitive creates an APA. In this case, the APA is an array with an offset of 1, a multiplier of 0 and a shape of 2 64.

On the other hand, this expression

      6402 ⎕dr 2 64⍴1 1

produces the expected result because the right argument is now Boolean.

Special Values

There are several special values you may use as a left argument to ⎕dr:

  • The value 0 displays the datatype of the right argument as a text string so you don't need to remember the datatype numbers.

    For example,

          ⎕dr 'a'
          0 ⎕dr 'a'
    Character (1601): 16 bits per element
          0 ⎕dr ⍳12
    Arithmetic Progression Array (6404): 64 bit offset + 64 bit multiplier

  • The value 1 converts between character and floating point. This argument makes it easy to see the representation of floating point numbers in order to help understand precision and other floating point issues.

    For example,

          1 ⎕pp←99
          1 ⎕dr 1.1
          1 ⎕dr '3fd',13⍴'5'
          1 ⎕dr ¯∞ ∞
          1 ⎕dr '7fe',13⍴'f'
    — The largest positive number
          1 ⎕dr '001',13⍴'0'
    — The smallest positive number
          1 ⎕dr '801',13⍴'0'
    — The largest negative number
          1 ⎕dr 'ffe',13⍴'f'
    — The smallest negative number

  • The value 2 converts between character and 64-bit integer. This argument makes it easy to see the representation of integers.

    For example,

          2 ⎕dr ¯1
          2 ⎕dr '7',15⍴'f'
    — The largest positive integer
          2 ⎕dr '8',15⍴'0'
    — The smallest negative integer
          2 ⎕dr 9223372036854775807 ¯9223372036854775808

A Word of Caution

This system function allows you to create special numbers we don't currently support in that no other primitive generates these numbers. Examples of such special numbers include Quiet NaNs, Signaling NaNs, Negative Zero, and Denormals. If the system doesn't behave as you expect when using these special numbers, don't be surprised.

For example,

      ⎕ct←0 ⋄ ⎕pp←99
      6403 ⎕dr ¯64↑1
      QNaN←6403 ⎕dr ¯64↑13⍴1
— A Quiet NaN (Not a Number)
      1 ⎕dr QNaN
      QNaN=1 2 3 ∞
1 1 1 1
      1 ⎕dr '000fffffffffffff'
— The largest positive denormal
      1 ⎕dr '0000000000000001'
— The smallest positive denormal