!fjjkgHFcwtkREywzfk:matrix.org

Naga Shader Translator

79 Members
On the way to producing SPIR-V!6 Servers

Load older messages


SenderMessageTime
16 Apr 2021
@kvark:matrix.orgkvarkwe don't support varying matrices right now, only scalars and vectors21:49:10
17 Apr 2021
@jimb:mozilla.orgjimb Hmm. Emacs's flycheck-mode runs cargo check in the background, and that sometimes seems to confuse clippy, such that clippy says all is well, but then cargo clean; cargo clippy has lots to say 01:48:35
@jimb:mozilla.orgjimb So, panic! is frowned upon, but unreachable! is permitted? 02:11:20
@kvark:matrix.orgkvarkcorrect04:10:52
@napokue:matrix.orgnapokueI am trying to make a Github Action that is able to verify our HLSL shaders. However, I see the dxc compiler requires a TargetProfile to be specified. This TargetProfile is specific to a shader stage and shader model version. As far as I could could there are about 50 different TargetProfiles. Does someone maybe know if there is there any way to determine what TargetProfile to use with what HLSL shader?14:51:51
@kvark:matrix.orgkvarknapokue: you'd want a naming scheme, similar to how GLSL is tested. So anything.Vertex.hlsl is for the vs_6_0 profile15:01:09
@napokue:matrix.orgnapokue Is that best practice? So if I have a vertex shader that uses shadermodel 6, then it would be named like: XXX_vs_6_0.hlsl? 15:02:43
@kvark:matrix.orgkvarkWe currently do not include the profile versions in the snapshot names. Maybe we will, but today if you follow the other backends, you'd need xxx.Vertex.hlsl, and check shader model 6_0 always15:04:53
@napokue:matrix.orgnapokueSure! So I can (for now) assume that we always compile against only shadermodel 6.0?15:05:54
@kvark:matrix.orgkvarkYeah15:07:03
@kvark:matrix.orgkvark The problem of profile selection is common across backends: there are multiple MSl versions, GLSL versions, even SPIRV versions 15:07:51
@napokue:matrix.orgnapokueYeah true, I never really worked in that detail with shaders. Do we actually some day want to support multiple versions?15:14:58
@napokue:matrix.orgnapokueProblem with HLSL is that with shadermodel 6, it will only work on dx12 as far as I can understand from their documentation.15:16:00
@kvark:matrix.orgkvarkWe support multiple versions, just not in the snapshots today15:39:43
@kvark:matrix.orgkvark
In reply to @napokue:matrix.org
Problem with HLSL is that with shadermodel 6, it will only work on dx12 as far as I can understand from their documentation.
Yeah, that should be our focus
15:40:04
@napokue:matrix.orgnapokueQuick newbie question, but do all our makefile commands use linux commands? I guess for windows I have to use other syntax?15:40:53
@kvark:matrix.orgkvark I haven't tried on Windows 15:41:35
@napokue:matrix.orgnapokueNone of the makefile commands works at my Windows system15:41:52
@napokue:matrix.orgnapokue * None of the makefile commands works at my Windows system15:42:35
@napokue:matrix.orgnapokuemakefile targets*15:42:52
@kvark:matrix.orgkvark Makefile in general is available on Windows, via "choco install make", iirc 15:50:28
@kvark:matrix.orgkvark Dunno if WSL makes it easier 15:50:36
@napokue:matrix.orgnapokueYes I have make installed via chocolately, but I don't have WSL installed. Will do some more research15:54:40
@kvark:matrix.orgkvarkGfx CI invokes make, check it out15:58:11
@napokue:matrix.orgnapokueThanks, was already doing that16:01:17
@jimb:mozilla.orgjimb kvark: The WGSL spec says, "The Entry point IO types include the following: ... Structures containing only built-in variables and user-defined IO. The structure must not contain a nested structure." Is Naga IR also intended to have that restriction? 16:04:42
@kvark:matrix.orgkvarkNot today, no. I assumed that nesting is permitted. Nesting may be useful for return types.16:08:07
@kvark:matrix.orgkvarkCurrently, not all backends work with neating, so if we want to restrict the IR on this front, it's ok to do16:09:11
@ashley_:matrix.orgAshley (expenses on gh)Are there any plans for adding an optimiser to Naga? I don't think it's a huge priority at all but I was checking out the code for spirv-opt and it's pretty impenetrable c++, so long term it'd be nice to have an alternative17:59:53
@kvark:matrix.orgkvark Not currently, no 18:01:48

There are no newer messages yet.


Back to Room List