Miguel Pozo pragma37
  • Joined on 2018-04-26
Miguel Pozo commented on issue blender/blender#120100 2024-04-26 17:48:06 +02:00
EEVEE-Next: Reduce longer compilation time

I’ve been taking a look at ways to improve shader compilation times.

As a recap, there are 3 main issues on the user side (mainly on Nvidia):

  1. Engine startup. The first time EEVEE-Next is…
Miguel Pozo created pull request blender/blender#121137 2024-04-26 17:33:06 +02:00
WIP: Draw: Add default material fallback detection
Miguel Pozo pushed to pull-is-default-material at pragma37/blender 2024-04-26 17:20:57 +02:00
182368ffad Add default material fallback detection
c87e8790bb Fix: GPv3: Redraw editors showing grease pencil keys when new on is added
03cb588529 Refactor: Cycles: Replace fixed Tangent input with custom input
62151ffaeb Fix: GPv3: Thickness modifier conversion of "uniform thickness"
7e3aa5a7a9 Vulkan: Add to_string functions for dynamic rendering
Compare 10 commits »
Miguel Pozo created branch pull-is-default-material in pragma37/blender 2024-04-26 17:20:56 +02:00
Miguel Pozo created pull request blender/blender#121093 2024-04-25 20:02:29 +02:00
WIP: GPU: GL_ARB/KHR_parallel_shader_compile support
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-25 19:21:50 +02:00
4ee65d5654 Undo OpenGL optimization step
Miguel Pozo pushed to main at pragma37/test-parallel-shader-compilation 2024-04-25 18:46:02 +02:00
27e4bbcf0c Initial commit
Miguel Pozo created branch main in pragma37/test-parallel-shader-compilation 2024-04-25 18:46:02 +02:00
Miguel Pozo created repository pragma37/test-parallel-shader-compilation 2024-04-25 18:44:40 +02:00
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-25 15:31:39 +02:00
c3a7fe9fec Avoid GPU_shader_deferred_compilation_is_ready checks
ca0c6beeac Set compiler threads from context initialization
Compare 2 commits »
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-24 21:30:24 +02:00
022e27d031 GPU_material_compile_deferred
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-24 21:21:50 +02:00
ffbc2b730b Workaround: Allow deferred material linking result check
759e7f268d Multithreaded EEVEE-Next static shader compilation
889bc2de75 Defer linking result check
Compare 3 commits »
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-24 18:50:31 +02:00
c5b9d09fc3 Shader::interface getter
7656d13347 Initial test
25128aa93b Enable parallel shader compilation
Compare 3 commits »
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-24 18:09:17 +02:00
bc322ef14e Shader::interface getter
Compare 2 commits »
Miguel Pozo pushed to test-parallel-compilation at pragma37/blender 2024-04-23 21:10:06 +02:00
490bccc438 Initial test
65c04f0f15 Enable parallel shader compilation
37da07e51a Geometry Nodes: support volume grid sockets in bake node
7b3e063704 GPv3: Overlay for Weight Paint mode
7a8e4995c8 Fix #120966: Remove `sample_clamp` for EEVEE raytracing presents
Compare 10 commits »
Miguel Pozo created branch test-parallel-compilation in pragma37/blender 2024-04-23 21:10:05 +02:00
Miguel Pozo pushed to pull-eevee-jittered-shoft-shadows at pragma37/blender 2024-04-23 17:32:31 +02:00
e44482833f Scale projection bounds by jitter overblur
942d7fedd4 Fix area point_on_light_shape
d30e6af62f Cleanup
Compare 3 commits »
Miguel Pozo commented on pull request blender/blender#119753 2024-04-22 16:14:05 +02:00
EEVEE-Next: Jittered Soft Shadows

When moving the view or an object the shadow is not centered at the origin of the light and not using the full spread. Which lead me to think that the random sampling is not fully disabled in…

Miguel Pozo commented on pull request blender/blender#119753 2024-04-22 16:00:50 +02:00
EEVEE-Next: Jittered Soft Shadows

Done, but I can't say I like it.

Miguel Pozo commented on pull request blender/blender#119753 2024-04-22 15:59:00 +02:00
EEVEE-Next: Jittered Soft Shadows

It's invert if we assume we only use unit quaternion. The same happens with the rotate function, it assumes a unit quaternion.

I think it's a fair assumption (at least in the GPU code), but…