Open chirsz-ever opened 2 years ago
OK. But this indeed can only be done under a new option. We can't regress code that has already taken this into account.
Note that #2936 is asking for the same transformation for GLSL as is done for HLSL. For fixup-clipspace, should we do for HLSL what is being requested for GLSL?
The strategy of SPIRV-Cross's fixup_clipspace
option is:
For GLSL targets, enabling this will convert a shader which assumes
[0, w]
depth range (Vulkan / D3D / Metal) into[-w, w]
range. For MSL and HLSL targets, enabling this will convert a shader in[-w, w]
depth range (OpenGL) to[0, w]
depth range.
IMO, this is because D3D and Metal have NDCs with [0, 1]
z-range and OpenGL has a [-1, 1]
z-range.
An HLSL shader always assumes that the NDC z-range is [0, 1]
, so we don't need to worry about it when convert it to SPIR-V for Vulkan.
But there are also SPIR-V for OpenGL and glslang support it. In this case maybe we need add a line to transform Z-coordinate form [0, 1]
to [-1. 1]
:
gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;
I think this kind of situations is very rare.
My conclusion is, the --fixup-clipspace
option should not be allowed or do the [0, w]
to [-w, w]
transformation if the source code format is HLSL.
OpenGL uses the NDC with z of [-1, 1], and Vulkan, DirectX, Metal use the NDC with z of [0, 1].
When we want fix a vertex shader originally written for OpenGL, we can add an extra line:
Or generate SPIR-V code of this meaning.
SPIRV-Cross has an option for it.
Like #2936, I want use the same source code of shaders for OpenGL and Vulkan.