fluid-work IRC Logs-2013-06-09

Owned by Fluid IRC Log Bot
Last updated: Jun 09, 2013Version comment
3 min read

[04:24:50 CDT(-0500)] <thealphanerd> Bosmon: are you around?

[14:46:05 CDT(-0500)] <colinclark> thealphanerd: you around?

[14:46:11 CDT(-0500)] <thealphanerd> indeed

[14:46:14 CDT(-0500)] <thealphanerd> but only for a minute

[14:46:46 CDT(-0500)] <colinclark> ok

[14:46:53 CDT(-0500)] <colinclark> your problem should be pretty straightforward to resolve

[14:47:04 CDT(-0500)] <colinclark> ultimately, there are no pointers in JavaScript

[14:47:06 CDT(-0500)] <colinclark> just arrays of bytes

[14:47:15 CDT(-0500)] <colinclark> and arrays of indices into arrays of bytes (smile)

[14:47:42 CDT(-0500)] <colinclark> I'm trying to see whether you can specify to cwrap that the return type of a function should be a typed array

[14:47:52 CDT(-0500)] <colinclark> otherwise, you'll just have to read the values off the heap directly

[14:51:27 CDT(-0500)] <thealphanerd> which is what I am currently doing

[14:51:39 CDT(-0500)] <thealphanerd> kinda

[14:51:47 CDT(-0500)] <colinclark> you're not (smile)

[14:51:53 CDT(-0500)] <thealphanerd> oh wai

[14:52:05 CDT(-0500)] <thealphanerd> you mean have the webaudio function directly read off the heap?

[14:52:09 CDT(-0500)] <colinclark> you're reading values off the heap, one by one, and putting them into new array (smile)

[14:52:14 CDT(-0500)] <colinclark> sadly, you can't

[14:52:20 CDT(-0500)] <colinclark> but you can at least get rid of one iteration

[14:52:25 CDT(-0500)] <colinclark> so here's how typed arrays work

[14:52:33 CDT(-0500)] <colinclark> they're "views" on an ArrayBuffer

[14:52:42 CDT(-0500)] <thealphanerd> there use to be an emscripten function called array_copy

[14:52:53 CDT(-0500)] <thealphanerd> you can see this

[14:52:53 CDT(-0500)] <colinclark> You don't want to copy, that's the thing

[14:52:54 CDT(-0500)] <thealphanerd> https://github.com/kripken/emscripten/issues/87

[14:53:08 CDT(-0500)] <thealphanerd> ahhh because array_copy is what I am doing already

[14:53:22 CDT(-0500)] <colinclark> So, ArrayBuffers are just arrays of bytes

[14:53:26 CDT(-0500)] <colinclark> you can't do anything with them direclty

[14:53:31 CDT(-0500)] <colinclark> you need a "view" on them

[14:53:39 CDT(-0500)] <colinclark> which lets you index them by type

[14:54:02 CDT(-0500)] <colinclark> So, you need to know at which offset your data starts in the heap

[14:54:11 CDT(-0500)] <colinclark> and then you need to create a Float32Array view on it

[14:54:18 CDT(-0500)] <thealphanerd> oh ok

[14:54:20 CDT(-0500)] <thealphanerd> that is pretty easy

[14:54:36 CDT(-0500)] <thealphanerd> the function currently returns a pointer to the array

[14:54:37 CDT(-0500)] <colinclark> I assume that.ptr is just a index into the heap, is it?

[14:54:51 CDT(-0500)] <thealphanerd> yup

[14:55:07 CDT(-0500)] <colinclark> so it's super easy

[14:55:15 CDT(-0500)] <thealphanerd> so I should be able to use a Float32Array view over that pointer?

[14:55:21 CDT(-0500)] <colinclark> Check out the fourth constructor here: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Typed_arrays/Float32Array?redirectlocale=en-US&amp;redirectslug=JavaScript%2FTyped_arrays%2FFloat32Array

[14:55:43 CDT(-0500)] <thealphanerd> btw I gotta run in a few minutes

[14:55:51 CDT(-0500)] <colinclark> if that.ptr is just an index (in bytes) into the heap, then it's as easy as

[14:56:04 CDT(-0500)] <thealphanerd> that is exactly what it is

[14:56:10 CDT(-0500)] <thealphanerd> and this is exactly what I couldn't figure out last night

[14:56:21 CDT(-0500)] <colinclark> new Float32Array(heap.buffer, that.ptr, byteLength)

[14:56:42 CDT(-0500)] <colinclark> But I'm thinking cwrap should give you a way to say "this function returns a Float32Array"

[14:56:44 CDT(-0500)] <colinclark> who knows?

[14:57:09 CDT(-0500)] <thealphanerd> perhaps

[14:57:36 CDT(-0500)] <thealphanerd> but wait

[14:57:40 CDT(-0500)] <thealphanerd> it isn't a Float32Array

[14:57:42 CDT(-0500)] <thealphanerd> now that I think about it

[14:57:49 CDT(-0500)] <thealphanerd> its an array of pointers to floats

[14:58:13 CDT(-0500)] <colinclark> Presumably that's exactly what Emscripten takes care of, no?

[14:58:19 CDT(-0500)] <colinclark> Converting pointers to arrays?

[14:58:33 CDT(-0500)] <colinclark> again, there are no pointers in JavaScript (smile)

[14:58:35 CDT(-0500)] <colinclark> just arrays

[14:58:56 CDT(-0500)] <colinclark> Looking at the source code, it does look like cwrap will take "array" as a type argument

[14:58:58 CDT(-0500)] <thealphanerd> I gotta run

[14:59:02 CDT(-0500)] <colinclark> but I don't know if it will do what you want

[14:59:06 CDT(-0500)] <thealphanerd> but yeah these are the questions I figure they can aswer really easily

[14:59:09 CDT(-0500)] <colinclark> Good luck with this (smile)

[14:59:24 CDT(-0500)] <colinclark> I look forward to adding it to Flocking's benchmarks so we can see what asm.js will give us

[15:00:11 CDT(-0500)] <colinclark> (smile)

[15:00:39 CDT(-0500)] <thealphanerd> I have a feeling like using emscripten will prove less efficient for simple ugens

[15:00:50 CDT(-0500)] <thealphanerd> but scale way better for more complicated block diagrams

[15:01:20 CDT(-0500)] <thealphanerd> ok time to go

[15:01:21 CDT(-0500)] <thealphanerd> talk soon

[15:01:25 CDT(-0500)] <colinclark> see you