[Bf-blender-cvs] [8f9599d17e8] master: DrawManager: Use Compute Shader to Update Hair.
Jeroen Bakker
noreply at git.blender.org
Wed May 26 17:03:58 CEST 2021
Commit: 8f9599d17e80254928d2d72081a4c7e0dee64038
Author: Jeroen Bakker
Date: Wed May 26 17:02:32 2021 +0200
Branches: master
https://developer.blender.org/rB8f9599d17e80254928d2d72081a4c7e0dee64038
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
===================================================================
M source/blender/draw/CMakeLists.txt
M source/blender/draw/intern/DRW_render.h
M source/blender/draw/intern/draw_cache_impl_hair.c
M source/blender/draw/intern/draw_hair.c
M source/blender/draw/intern/draw_manager.h
M source/blender/draw/intern/draw_manager_data.c
M source/blender/draw/intern/draw_manager_exec.c
M source/blender/draw/intern/shaders/common_hair_lib.glsl
A source/blender/draw/intern/shaders/common_hair_refine_comp.glsl
M source/blender/draw/intern/shaders/common_hair_refine_vert.glsl
M source/blender/gpu/GPU_capabilities.h
M source/blender/gpu/intern/gpu_capabilities.cc
M source/blender/gpu/intern/gpu_capabilities_private.hh
M source/blender/gpu/opengl/gl_backend.cc
===================================================================
diff --git a/source/blender/draw/CMakeLists.txt b/source/blender/draw/CMakeLists.txt
index 045adf4b380..95c0f5d300c 100644
--- a/source/blender/draw/CMakeLists.txt
+++ b/source/blender/draw/CMakeLists.txt
@@ -321,6 +321,7 @@ data_to_c_simple(intern/shaders/common_globals_lib.glsl SRC)
data_to_c_simple(intern/shaders/common_pointcloud_lib.glsl SRC)
data_to_c_simple(intern/shaders/common_hair_lib.glsl SRC)
data_to_c_simple(intern/shaders/common_hair_refine_vert.glsl SRC)
+data_to_c_simple(intern/shaders/common_hair_refine_comp.glsl SRC)
data_to_c_simple(intern/shaders/common_math_lib.glsl SRC)
data_to_c_simple(intern/shaders/common_math_geom_lib.glsl SRC)
data_to_c_simple(intern/shaders/common_view_lib.glsl SRC)
diff --git a/source/blender/draw/intern/DRW_render.h b/source/blender/draw/intern/DRW_render.h
index 2545cfa65dc..5071658fd82 100644
--- a/source/blender/draw/intern/DRW_render.h
+++ b/source/blender/draw/intern/DRW_render.h
@@ -438,6 +438,10 @@ void DRW_shgroup_call_range(
void DRW_shgroup_call_instance_range(
DRWShadingGroup *shgroup, Object *ob, struct GPUBatch *geom, uint i_sta, uint i_ct);
+void DRW_shgroup_call_compute(DRWShadingGroup *shgroup,
+ int groups_x_len,
+ int groups_y_len,
+ int groups_z_len);
void DRW_shgroup_call_procedural_points(DRWShadingGroup *sh, Object *ob, uint point_count);
void DRW_shgroup_call_procedural_lines(DRWShadingGroup *sh, Object *ob, uint line_count);
void DRW_shgroup_call_procedural_triangles(DRWShadingGroup *sh, Object *ob, uint tri_count);
@@ -575,6 +579,9 @@ void DRW_shgroup_uniform_vec4_array_copy(DRWShadingGroup *shgroup,
const char *name,
const float (*value)[4],
int arraysize);
+void DRW_shgroup_vertex_buffer(DRWShadingGroup *shgroup,
+ const char *name,
+ struct GPUVertBuf *vertex_buffer);
bool DRW_shgroup_is_empty(DRWShadingGroup *shgroup);
diff --git a/source/blender/draw/intern/draw_cache_impl_hair.c b/source/blender/draw/intern/draw_cache_impl_hair.c
index fd28ac00186..6424b21666d 100644
--- a/source/blender/draw/intern/draw_cache_impl_hair.c
+++ b/source/blender/draw/intern/draw_cache_impl_hair.c
@@ -243,7 +243,8 @@ static void hair_batch_cache_ensure_procedural_final_points(ParticleHairCache *c
GPUVertFormat format = {0};
GPU_vertformat_attr_add(&format, "pos", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
- cache->final[subdiv].proc_buf = GPU_vertbuf_create_with_format(&format);
+ cache->final[subdiv].proc_buf = GPU_vertbuf_create_with_format_ex(&format,
+ GPU_USAGE_DEVICE_ONLY);
/* Create a destination buffer for the transform feedback. Sized appropriately */
/* Those are points! not line segments. */
diff --git a/source/blender/draw/intern/draw_hair.c b/source/blender/draw/intern/draw_hair.c
index bca227a24e2..258777b34fb 100644
--- a/source/blender/draw/intern/draw_hair.c
+++ b/source/blender/draw/intern/draw_hair.c
@@ -36,15 +36,28 @@
#include "BKE_duplilist.h"
#include "GPU_batch.h"
+#include "GPU_capabilities.h"
+#include "GPU_compute.h"
#include "GPU_shader.h"
+#include "GPU_texture.h"
#include "GPU_vertex_buffer.h"
#include "draw_hair_private.h"
#ifndef __APPLE__
# define USE_TRANSFORM_FEEDBACK
+# define USE_COMPUTE_SHADERS
#endif
+BLI_INLINE bool drw_hair_use_compute_shaders(void)
+{
+#ifdef USE_COMPUTE_SHADERS
+ return GPU_compute_shader_support();
+#else
+ return false;
+#endif
+}
+
typedef enum ParticleRefineShader {
PART_REFINE_CATMULL_ROM = 0,
PART_REFINE_MAX_SHADER,
@@ -71,6 +84,7 @@ static DRWPass *g_tf_pass; /* XXX can be a problem with multiple DRWManager in t
extern char datatoc_common_hair_lib_glsl[];
extern char datatoc_common_hair_refine_vert_glsl[];
+extern char datatoc_common_hair_refine_comp_glsl[];
extern char datatoc_gpu_shader_3D_smooth_color_frag_glsl[];
static GPUShader *hair_refine_shader_get(ParticleRefineShader sh)
@@ -79,15 +93,26 @@ static GPUShader *hair_refine_shader_get(ParticleRefineShader sh)
return g_refine_shaders[sh];
}
- char *vert_with_lib = BLI_string_joinN(datatoc_common_hair_lib_glsl,
- datatoc_common_hair_refine_vert_glsl);
+#ifdef USE_COMPUTE_SHADERS
+ const bool do_compute = drw_hair_use_compute_shaders();
+ if (do_compute) {
+ g_refine_shaders[sh] = GPU_shader_create_compute(datatoc_common_hair_refine_comp_glsl,
+ datatoc_common_hair_lib_glsl,
+ "#define HAIR_PHASE_SUBDIV\n",
+ __func__);
+ return g_refine_shaders[sh];
+ }
+#endif
#ifdef USE_TRANSFORM_FEEDBACK
+ char *shader_src = BLI_string_joinN(datatoc_common_hair_lib_glsl,
+ datatoc_common_hair_refine_vert_glsl);
const char *var_names[1] = {"finalColor"};
g_refine_shaders[sh] = DRW_shader_create_with_transform_feedback(
- vert_with_lib, NULL, "#define HAIR_PHASE_SUBDIV\n", GPU_SHADER_TFB_POINTS, var_names, 1);
+ shader_src, NULL, "#define HAIR_PHASE_SUBDIV\n", GPU_SHADER_TFB_POINTS, var_names, 1);
+
#else
- g_refine_shaders[sh] = DRW_shader_create(vert_with_lib,
+ g_refine_shaders[sh] = DRW_shader_create(shader_src,
NULL,
datatoc_gpu_shader_3D_smooth_color_frag_glsl,
"#define blender_srgb_to_framebuffer_space(a) a\n"
@@ -95,14 +120,14 @@ static GPUShader *hair_refine_shader_get(ParticleRefineShader sh)
"#define TF_WORKAROUND\n");
#endif
- MEM_freeN(vert_with_lib);
+ MEM_freeN(shader_src);
return g_refine_shaders[sh];
}
void DRW_hair_init(void)
{
-#ifdef USE_TRANSFORM_FEEDBACK
+#if defined(USE_TRANSFORM_FEEDBACK) || defined(USE_COMPUTE_SHADERS)
g_tf_pass = DRW_pass_create("Update Hair Pass", 0);
#else
g_tf_pass = DRW_pass_create("Update Hair Pass", DRW_STATE_WRITE_COLOR);
@@ -125,6 +150,67 @@ void DRW_hair_init(void)
}
}
+static void drw_hair_particle_cache_shgrp_attach_resources(DRWShadingGroup *shgrp,
+ ParticleHairCache *cache,
+ const int subdiv)
+{
+ DRW_shgroup_uniform_texture(shgrp, "hairPointBuffer", cache->point_tex);
+ DRW_shgroup_uniform_texture(shgrp, "hairStrandBuffer", cache->strand_tex);
+ DRW_shgroup_uniform_texture(shgrp, "hairStrandSegBuffer", cache->strand_seg_tex);
+ DRW_shgroup_uniform_int(shgrp, "hairStrandsRes", &cache->final[subdiv].strands_res, 1);
+}
+
+static void drw_hair_particle_cache_update_compute(ParticleHairCache *cache, const int subdiv)
+{
+ const int strands_len = cache->strands_len;
+ const int final_points_len = cache->final[subdiv].strands_res * strands_len;
+ if (final_points_len > 0) {
+ GPUShader *shader = hair_refine_shader_get(PART_REFINE_CATMULL_ROM);
+ DRWShadingGroup *shgrp = DRW_shgroup_create(shader, g_tf_pass);
+ drw_hair_particle_cache_shgrp_attach_resources(shgrp, cache, subdiv);
+ DRW_shgroup_vertex_buffer(shgrp, "hairPointOutputBuffer", cache->final[subdiv].proc_buf);
+
+ const int max_strands_per_call = GPU_max_work_group_count(0);
+ int strands_start = 0;
+ while (strands_start < strands_len) {
+ int batch_strands_len = MIN2(strands_len - strands_start, max_strands_per_call);
+ DRWShadingGroup *subgroup = DRW_shgroup_create_sub(shgrp);
+ DRW_shgroup_uniform_int_copy(subgroup, "hairStrandOffset", strands_start);
+ DRW_shgroup_call_compute(subgroup, batch_strands_len, cache->final[subdiv].strands_res, 1);
+ strands_start += batch_strands_len;
+ }
+ }
+}
+
+static void drw_hair_particle_cache_update_transform_feedback(ParticleHairCache *cache,
+ const int subdiv)
+{
+ const int final_points_len = cache->final[subdiv].strands_res * cache->strands_len;
+ if (final_points_len > 0) {
+ GPUShader *tf_shader = hair_refine_shader_get(PART_REFINE_CATMULL_ROM);
+
+#ifdef USE_TRANSFORM_FEEDBACK
+ DRWShadingGroup *tf_shgrp = DRW_shgroup_transform_feedback_create(
+ tf_shader, g_tf_pass, cache->final[subdiv].proc_buf);
+#else
+ DRWShadingGroup *tf_shgrp = DRW_shgroup_create(tf_shader, g_tf_pass);
+
+ ParticleRefineCall *pr_call = MEM_mallocN(sizeof(*pr_call), __func__);
+ pr_call->next = g_tf_calls;
+ pr_call->vbo = cache->final[subdiv].proc_buf;
+ pr_call->shgrp = tf_shgrp;
+ pr_call->vert_len = final_points_len;
+ g_tf_calls = pr_call;
+ DRW_shgroup_uniform_int(tf_shgrp, "targetHeight", &g_tf_target_height, 1);
+ DRW_shgroup_uniform_int(tf_shgrp, "targetWidth", &g_tf_target_width, 1);
+ DRW_shgroup_uniform_int(tf_shgrp, "idOffset", &g_tf_id_offset, 1);
+#endif
+
+ drw_hair_particle_cache_shgrp_attach_resources(tf_shgrp, cache, subdiv);
+ DRW_shgroup_call_procedural_points(tf_shgrp, NULL, final_points_len);
+ }
+}
+
static ParticleHairCache *drw_hair_particle_cache_get(
Object *object, ParticleSystem *psys, ModifierData *md, int subdiv, int thickness_res)
{
@@ -140,32 +226,11 @@ static ParticleHairCache *drw_hair_particle_cache_get(
}
if (update) {
- int final_points_len = cache->final[subdiv].strands_res * cache->strands_len;
- if (final_points_len > 0) {
- GPUShader *tf_shader = hair_refine_shader_get(PART_REFINE_CATMULL_ROM);
-
-#ifdef USE_TRANSFORM_FEEDBACK
- DRWShadingGroup *tf_shgrp = DRW_shgroup_transform_feedback_create(
- tf_shader, g_tf_pass, cache->final[subdiv].proc_buf);
-#else
- DRWShadingGroup *tf_shgrp = DRW_shgroup_create(tf_shader, g_tf_pass);
-
- ParticleRefineCall *pr_call = MEM_mallocN(sizeof(*pr_call), __func__);
- pr_call->next = g_tf_calls;
- pr_call->vbo = cache->final[subdiv].proc_buf;
- pr_call->shgrp
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list