From 8f9599d17e80254928d2d72081a4c7e0dee64038 Mon Sep 17 00:00:00 2001
From: Jeroen Bakker <jbakker>
Date: Wed, 26 May 2021 17:02:32 +0200
Subject: DrawManager: Use Compute Shader to Update Hair.

This patch will use compute shaders to create the VBO for hair.
The previous implementation uses tranform feedback.

Timings master (transform feedback with GPU_USAGE_STATIC between 0.000069s and 0.000362s
Timings transform feedback with GPU_USAGE_DEVICE_ONLY. between 0.000057s and 0.000122s
Timings compute shader between 0.000032 and 0.000092s

Future improvements:
* Generate hair Index buffer using compute shaders: currently done single threaded on CPU, easy to add as compute shader.

Reviewed By: fclem

Differential Revision: https://developer.blender.org/D11057
---
 source/blender/gpu/GPU_capabilities.h                 |  2 ++
 source/blender/gpu/intern/gpu_capabilities.cc         | 10 ++++++++++
 source/blender/gpu/intern/gpu_capabilities_private.hh |  2 ++
 source/blender/gpu/opengl/gl_backend.cc               |  8 ++++++++
 4 files changed, 22 insertions(+)

(limited to 'source/blender/gpu')

diff --git a/source/blender/gpu/GPU_capabilities.h b/source/blender/gpu/GPU_capabilities.h
index 45c656b49be..0c054d4f264 100644
--- a/source/blender/gpu/GPU_capabilities.h
+++ b/source/blender/gpu/GPU_capabilities.h
@@ -37,6 +37,8 @@ int GPU_max_textures(void);
 int GPU_max_textures_vert(void);
 int GPU_max_textures_geom(void);
 int GPU_max_textures_frag(void);
+int GPU_max_work_group_count(int index);
+int GPU_max_work_group_size(int index);
 int GPU_max_uniforms_vert(void);
 int GPU_max_uniforms_frag(void);
 int GPU_max_batch_indices(void);
diff --git a/source/blender/gpu/intern/gpu_capabilities.cc b/source/blender/gpu/intern/gpu_capabilities.cc
index bedc9ad3092..c6e9dc210cb 100644
--- a/source/blender/gpu/intern/gpu_capabilities.cc
+++ b/source/blender/gpu/intern/gpu_capabilities.cc
@@ -82,6 +82,16 @@ int GPU_max_textures(void)
   return GCaps.max_textures;
 }
 
+int GPU_max_work_group_count(int index)
+{
+  return GCaps.max_work_group_count[index];
+}
+
+int GPU_max_work_group_size(int index)
+{
+  return GCaps.max_work_group_size[index];
+}
+
 int GPU_max_uniforms_vert(void)
 {
   return GCaps.max_uniforms_vert;
diff --git a/source/blender/gpu/intern/gpu_capabilities_private.hh b/source/blender/gpu/intern/gpu_capabilities_private.hh
index ee7ef1e69e6..95cf7fd335d 100644
--- a/source/blender/gpu/intern/gpu_capabilities_private.hh
+++ b/source/blender/gpu/intern/gpu_capabilities_private.hh
@@ -41,6 +41,8 @@ struct GPUCapabilities {
   int max_textures_vert = 0;
   int max_textures_geom = 0;
   int max_textures_frag = 0;
+  int max_work_group_count[3] = {0, 0, 0};
+  int max_work_group_size[3] = {0, 0, 0};
   int max_uniforms_vert = 0;
   int max_uniforms_frag = 0;
   int max_batch_indices = 0;
diff --git a/source/blender/gpu/opengl/gl_backend.cc b/source/blender/gpu/opengl/gl_backend.cc
index fb03a2c2d2a..d85f9f7684d 100644
--- a/source/blender/gpu/opengl/gl_backend.cc
+++ b/source/blender/gpu/opengl/gl_backend.cc
@@ -438,6 +438,14 @@ void GLBackend::capabilities_init()
   GCaps.mem_stats_support = GLEW_NVX_gpu_memory_info || GLEW_ATI_meminfo;
   GCaps.shader_image_load_store_support = GLEW_ARB_shader_image_load_store;
   GCaps.compute_shader_support = GLEW_ARB_compute_shader;
+  if (GCaps.compute_shader_support) {
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, &GCaps.max_work_group_count[0]);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 1, &GCaps.max_work_group_count[1]);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 2, &GCaps.max_work_group_count[2]);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 0, &GCaps.max_work_group_size[0]);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 1, &GCaps.max_work_group_size[1]);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 2, &GCaps.max_work_group_size[2]);
+  }
   GCaps.shader_storage_buffer_objects_support = GLEW_ARB_shader_storage_buffer_object;
   /* GL specific capabilities. */
   glGetIntegerv(GL_MAX_3D_TEXTURE_SIZE, &GLContext::max_texture_3d_size);
-- 
cgit v1.2.3