[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Public WebGL] Support precompiled shaders as extensions

This is probably just a side note: The most expensive part of end-user compile-time should be target-architecture instruction scheduling and register allocation.  Both are needed for both SPIR-V and DX bytecode, and other higher level optimizations can be applied to both offline.  So, long term, there should not be a notable difference between SPIR-V and DX byte-code regarding end-user compile time.  The differences should be that SPIR-V is more portable, less platform specific, has a definition we control, and is larger, the latter of which we believe is a solvable problem to the extent it actually matters.


On 11/14/2016 5:17 PM, Maksims Mihejevs wrote:
So to summarise:

SPIR-V - does not solves compilation times.
Platform specific shader bytecode - is not the way due to many reasons explained above in the topic.

Now the questions:
1. If there is a work we can influence on making GPU compilation faster?
2. What can we do, to help WebGL community to engage with async shader compilation solutions, possibly glFenceSync?

Kind Regards,

On 14 November 2016 at 23:17, Florian Bösch <pyalot@gmail.com> wrote:
On Mon, Nov 14, 2016 at 11:37 PM, Florian Bösch <pyalot@gmail.com> wrote:
To be precise: No other platform than windows can implement WEBGL_dx_bytecode_shaders or any viable alternative to it. That's why it'd be a bad extension.

Also if and when a viable alternative does become... viable, WEBGL_dx_bytecode_shaders would be a bad idea because it's a platform specific format, and we'd all rather make an extension about something that doesn't require you to submit several different formats for no intrinsic reason other than hasty expediency.