Re: Double precision floats (IEEE-754 binary64)

Right - it would be an optional extension. Any Metal-based implementations would just not offer the extension.

Do you think you would be willing to open an issue in https://github.com/gpuweb/gpuweb/issues/new/choose advocating for FP64 and your use case? If there’s a bug, we can schedule it accordingly, and you can follow along with any spec updates regarding it.

Thanks,
Myles

> On Apr 27, 2022, at 2:22 PM, Rob Conde <rob.conde@ai-solutions.com> wrote:
> 
> I thought a big blocker there was that metal has no support for float64 - but maybe just an extension then?
> 
> Rob
> From: Kelsey Gilbert <kelsey.gilbert@mozilla.com>
> Sent: Wednesday, April 27, 2022 4:15 PM
> To: Samuele Decarli <sdecarli@esri.com>
> Cc: public-gpu@w3.org <public-gpu@w3.org>
> Subject: Re: Double precision floats (IEEE-754 binary64)
>  
> Float64 is likely eventually, yes. There are no concrete plans for them yet, though I expect it would be similar to what we are doing for float16.
> 
> However, we do have ieee floats in format, but not strictly speaking in accuracy/behavior, so be aware of that! 
> 
> On Wed, Apr 27, 2022 at 12:59 PM Samuele Decarli <sdecarli@esri.com> wrote:
> Hello,
> 
> I'm interested in knowing if there are plans to support double
> precision float (IEEE-754 binary64) types and operations in WGSL
> shaders, especially compute ones.
> 
> I see from the spec that there is a f32 (IEEE-754 binary32) type
> supported, as well as f16 (IEEE-754 binary16) (via extension). While
> doubles seem supported via AbstractFloat as compile-time/shader
> creation constants.
> Double precision floats are useful for scientific computation and GIS
> and are supported by some GPUs, particularly on desktops.
> Are there plans to add support for them to WebGPU (either in the
> required spec or via extensions)?
> 
> Best regards,
> Samuele