integer container for float

basiliscos has asked for the wisdom of the Perl Monks concerning the following question:

Hello Monks,

The business values I operated are floats; meanwhile the module I use to them in shared memory uses integers, that also guarantees the atomicity of writing int-values in shared memory.

Is there any way to pack (not convert) float to integer, without loosing precision? Way# 1

my $packed_integer = int($float * 1_000_000);
my $unpacked_float = $packed_integer / 1_000_000;
[download]

Obviously, I loose precision here. In C I can do to manipulation with pointers, to get the result I like:

assert(sizeof(int) >= sizeof(float));
 
float f = ...;
int *i_ptr = (int*)&f;
int i = *int;
[download]

Then I've found can do something like that via pack/upack build-ins in Perl:

my $v = 1/3;
my $v2 = unpack("F", pack("j", unpack("j", pack("F", $v))));
print ("v = $v\nv = $v2\n");

# v = 0.333333333333333
# v = 0.333333333333333
[download]

Seems fine.

Is there any hidden problem with pack/unpack here?

How to emulate C's assert in perl, i.e. how to get sizeof:

die("float cannot be packed into integer") 
  unless ($int_size >= $float_size);
[download]

May be there is already some module that does that? If there is no, does it worth to make a small module with pack + assert as above?

Thanks!

WBR, basiliscos.

Comment on integer container for float Select or Download Code

Replies are listed 'Best First'.
Re: integer container for float by BrowserUk (Patriarch) on Mar 28, 2016 at 11:38 UTC
You can check that a native int will hold a native double: `use Config; die "Int/double size mismatch" unless $Config{ ivsize } == $Config{ nv +size };` [download] And if you get passed that test, then the following should always work: `$packedDoubleAsInt = unpack 'Q', pack 'd', 3.14159265358979;; print $packedDoubleAsInt;; 4614256656552045841 $unpacked = unpack 'd', pack 'Q', $packedDoubleAsInt;; print $unpacked;; 3.14159265358979 $unpacked = unpack 'd', pack 'Q', 4614256656552045841;; print $unpacked;; 3.14159265358979` [download] With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday' Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error. "Science is about questioning the status quo. Questioning authority". I knew I was on the right track :) In the absence of evidence, opinion is indistinguishable from prejudice.	[reply] [d/l] [select]
Re: integer container for float by syphilis (Archbishop) on Mar 28, 2016 at 12:44 UTC
BrowserUk has nicely covered the case where $Config{ivsize} == 8. Similarly it's possible to pack/unpack single precision floats when $Config{ivsize} == 4. Obviously, you can't use the "Q" template because, according to the docs, that applies only to ivsize of 8. But you can use the "j" (or "J") templates with the "f" template: `#!perl -l use strict; use warnings; use Config; my $val = 3.625; if($Config{ivsize} == 4) { my $to_int = unpack 'j', pack 'f', $val;; print $to_int; # 1080557568 my $to_float = unpack 'f', pack 'j', $to_int;; print $to_float; # 3.625 } else { print "Can't pack C floats to IV/UV"; }` [download] Cheers, Rob	[reply] [d/l]
Re^2: integer container for float by BrowserUk (Patriarch) on Mar 28, 2016 at 14:08 UTC
Indeed. He could even use 2 x 32-bit ints as an intermediary for a double: `($lo, $hi) = unpack 'NN', pack 'd', 3.14159265358979;; print $lo, $hi;; 288179284 4213246272 print unpack 'd', pack 'NN', 288179284, 4213246272;; 3.14159265358979` [download] Or even 4 x 16-bit ints: `( $_0, $_1, $_2, $_3 ) = unpack 'n', pack 'd', exp(1);; print unpack 'd', pack 'n', $_0, $_1, $_2, $_3;; 2.71828182845905` [download] With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday' Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error. "Science is about questioning the status quo. Questioning authority". I knew I was on the right track :) In the absence of evidence, opinion is indistinguishable from prejudice.	[reply] [d/l] [select]
Re: integer container for float by Anonymous Monk on Mar 28, 2016 at 16:22 UTC
Just a side note which probably won't help you any: there's a problem with your C code, too. In modern C, one cannot access an object via pointer of incompatible type^#1. This is called type punning and it violates aliasing constraints. You can still compile such code with `-fno-strict-aliasing`, but this is a poor practice. Going via union is to be preferred. #1 excepting character type—one may inspect the stored representation of any object through `(unsigned char *).`	[reply] [d/l] [select]