Lesson 8 – Geometry shaders
Environment mapping
PV227 – GPU Rendering
Jiˇrí Chmelík, Jan ˇCejka
Fakulta informatiky Masarykovy univerzity
7. 11. 2016
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 1 / 13
Intermezzo – Environment mapping
Reflections: Environment mapping
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 2 / 13
Environment mapping
Source: Wikipedia
−−−−−−−−−−→
ReflectedRay =
−−−−−−−−→
CameraRay − 2 ·
−→
N · dot(
−→
N ,
−−−−−−−−→
CameraRay)
In GLSL: ReflectedRay = reflect(CameraRay, N)
Assumes N is normalized
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 3 / 13
Task: Implement environment mapping
Task 1: Implement environment mapping in
reflection_fragment.glsl
Mix the environment reflection with the color of the object
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 4 / 13
Updating the cube map
When the surrounding changes, the cube map with the
environment should be updated.
Six faces of the cube map means:
six cameras,
six framebuffer objects
six times traversing the scene
Already implemented in the code.
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 5 / 13
Layered rendering
Renders into multiple textures at the same time
Good for cube maps, stereo rendering etc.
Different from attachments of FBOs
Attachments: Primitives are rasterized at the same places
Layers: Each layer has different primitives
Renders triangles into layered textures:
cube maps (6 layers)
2D texture arrays
3D textures, 1D texture arrays, cube map arrays
Use glFramebufferTexture to attach a layered texture into a
framebuffer
All textures at all attachments must be layered
Another usage of geometry shaders
New output variable in geometry shaders: gl_Layer
Specifies the index of the layer into which the primitive is sent
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 6 / 13
Updating the cube map – layered rendering
Updating all faces simultaneously means:
six cameras available at the same time
one framebuffer object with all faces
traversing the scene once
special vertex and geometry shaders
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 7 / 13
Task: Implement layered rendering
Task 2: Implement layered rendering in
texture_to_cube_geometry.glsl and compare the rendering speed
Generate 6 triangles (18 vertices), one for each face
Some vertex data do not change, the are computed in VS
Pass them through geometry shader without change
Some vertex data (gl_Position and gl_Layer) are different for each
face.
Compute their values in geometry shader
Optional task: Implement the same for the skybox in
skybox_to_cube_geometry.glsl
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 8 / 13
Instanced geometry shader
Problem: the geometry shader processes 18 vertices sequentially,
not in parallel
Possible solution: Instanced geometry shaders
Similar to instancing
Geometry shader is run multiple times per each input primitive
In GS: Instances = Incovations
Defined in geometry shader:
layout (triangles, invocations = 6) in;
layout (triangle_strip, max_vertices = 3) out;
Special variable gl_InvocationID:
Only in geometry shader
Similar to gl_InstanceID
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 9 / 13
Task: Implement instanced geometry shaders
Task 3: Implement instanced geometry shaders in
texture_to_cube_invocations_geometry.glsl and compare the
rendering speed
Generate 1 triangle (3 vertices), 6 incovations, one invocation for
each face
Optional task: Implement the same for the skybox in
skybox_to_cube_invocations_geometry.glsl
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 10 / 13
Parallelize even more
Use instancning, i.e. render each object six times.
Everything is computed in vertex shader, all vertices in parallel
Geometry shader only copies the data of each vertex and sets
gl_Layer
Task 4: Implement this in texture_to_cube_instancing_vertex.glsl
and texture_to_cube_instancing_geometry.glsl and compare the
rendering speed
Option: Implement the same for the skybox in
skybox_to_cube_instancing_vertex.glsl and
skybox_to_cube_instancing_geometry.glsl
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 11 / 13
Parallelize even more, skip geometry shaders
Modern graphics card may set gl_Layer also in vertex shaders,
thus skipping the geometry shader completely
We need OpenGL extension
GL_ARB_shader_viewport_layer_array, unfortunately, it is not
available on computers in B311
Optional task 5: Implement this in
texture_to_cube_instancing_no_gs_vertex.glsl and in
skybox_to_cube_instancing_no_gs_vertex.glsl and compare the
rendering speed
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 12 / 13
Geometry Shaders Today
not for tessellation, surpassed by tessellation shaders
probably not for culling (not necessary)
expanding a point to a quad (particle systems), compete with
instancing
expanding a line to a quad (grass, hair), in combination with
tessellation shaders
transform feedback (outputs vertices back into VBOs)
layered rendering, instanced geometry shaders
not enough parallel, compete with instancing, not necessary when
gl_Layer is set in vertex shaders
PV227 – GPU Rendering (FI MUNI) Lesson 8 – Geometry shaders 7. 11. 2016 13 / 13