DrawManager: Use Compute Shader to Update Hair.

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

Timings before: between 0.000069s and 0.000362s.
Timings after:  between 0.000032s and 0.000092s.

Speedup isn't noticeable by end-users. The patch is used to test
the new compute shader pipeline and integrate it with the draw
manager. Allowing EEVEE, Workbench and other draw engines to
use compute shaders with the introduction of `DRW_shgroup_call_compute`
and `DRW_shgroup_vertex_buffer`.

Future improvements are possible by generating the index buffer
of hair directly on the GPU.

NOTE: that compute shaders aren't supported by Apple and still use
the transform feedback workaround.

Reviewed By: fclem

Differential Revision: https://developer.blender.org/D11057

1 files changed, 153 insertions, 47 deletions

diff --git a/source/blender/draw/intern/draw_hair.c b/source/blender/draw/intern/draw_hair.c
index bca227a24e2..bf3b10bccd3 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,
@@ -71,38 +84,89 @@ 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)
+/* TODO(jbakker): move shader creation to `draw_shaders` and add test cases. */
+/* TODO(jbakker): replace defines with `constexpr` to check compilation on all OSs. Currently the
+ * APPLE codepath does not compile on other platforms and vice versa. */
+#ifdef USE_COMPUTE_SHADERS
+static GPUShader *hair_refine_shader_compute_create(ParticleRefineShader UNUSED(refinement))
 {
-  if (g_refine_shaders[sh]) {
-    return g_refine_shaders[sh];
-  }
-
-  char *vert_with_lib = BLI_string_joinN(datatoc_common_hair_lib_glsl,
-                                         datatoc_common_hair_refine_vert_glsl);
+  GPUShader *sh = NULL;
+  sh = GPU_shader_create_compute(datatoc_common_hair_refine_comp_glsl,
+                                 datatoc_common_hair_lib_glsl,
+                                 "#define HAIR_PHASE_SUBDIV\n",
+                                 __func__);
+  return sh;
+}
+#endif
 
 #ifdef USE_TRANSFORM_FEEDBACK
+static GPUShader *hair_refine_shader_transform_feedback_create(
+    ParticleRefineShader UNUSED(refinement))
+{
+  GPUShader *sh = NULL;
+
+  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);
-#else
-  g_refine_shaders[sh] = DRW_shader_create(vert_with_lib,
-                                           NULL,
-                                           datatoc_gpu_shader_3D_smooth_color_frag_glsl,
-                                           "#define blender_srgb_to_framebuffer_space(a) a\n"
-                                           "#define HAIR_PHASE_SUBDIV\n"
-                                           "#define TF_WORKAROUND\n");
+  sh = DRW_shader_create_with_transform_feedback(
+      shader_src, NULL, "#define HAIR_PHASE_SUBDIV\n", GPU_SHADER_TFB_POINTS, var_names, 1);
+  MEM_freeN(shader_src);
+
+  return sh;
+}
 #endif
 
-  MEM_freeN(vert_with_lib);
+static GPUShader *hair_refine_shader_transform_feedback_workaround_create(
+    ParticleRefineShader UNUSED(refinement))
+{
+  GPUShader *sh = NULL;
+
+  char *shader_src = BLI_string_joinN(datatoc_common_hair_lib_glsl,
+                                      datatoc_common_hair_refine_vert_glsl);
+  sh = DRW_shader_create(shader_src,
+                         NULL,
+                         datatoc_gpu_shader_3D_smooth_color_frag_glsl,
+                         "#define blender_srgb_to_framebuffer_space(a) a\n"
+                         "#define HAIR_PHASE_SUBDIV\n"
+                         "#define TF_WORKAROUND\n");
+  MEM_freeN(shader_src);
+
+  return sh;
+}
+
+static GPUShader *hair_refine_shader_get(ParticleRefineShader refinement)
+{
+  if (g_refine_shaders[refinement]) {
+    return g_refine_shaders[refinement];
+  }
+
+#ifdef USE_COMPUTE_SHADERS
+  if (drw_hair_use_compute_shaders()) {
+    g_refine_shaders[refinement] = hair_refine_shader_compute_create(refinement);
+    if (g_refine_shaders[refinement]) {
+      return g_refine_shaders[refinement];
+    }
+  }
+#endif
+
+#ifdef USE_TRANSFORM_FEEDBACK
+  g_refine_shaders[refinement] = hair_refine_shader_transform_feedback_create(refinement);
+  if (g_refine_shaders[refinement]) {
+    return g_refine_shaders[refinement];
+  }
+#endif
 
-  return g_refine_shaders[sh];
+  g_refine_shaders[refinement] = hair_refine_shader_transform_feedback_workaround_create(
+      refinement);
+  return g_refine_shaders[refinement];
 }
 
 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 +189,67 @@ 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 +265,11 @@ 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 = 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_shgroup_uniform_texture(tf_shgrp, "hairPointBuffer", cache->point_tex);
-      DRW_shgroup_uniform_texture(tf_shgrp, "hairStrandBuffer", cache->strand_tex);
-      DRW_shgroup_uniform_texture(tf_shgrp, "hairStrandSegBuffer", cache->strand_seg_tex);
-      DRW_shgroup_uniform_int(tf_shgrp, "hairStrandsRes", &cache->final[subdiv].strands_res, 1);
-      DRW_shgroup_call_procedural_points(tf_shgrp, NULL, final_points_len);
+    if (drw_hair_use_compute_shaders()) {
+      drw_hair_particle_cache_update_compute(cache, subdiv);
+    }
+    else {
+      drw_hair_particle_cache_update_transform_feedback(cache, subdiv);
     }
   }
   return cache;
@@ -367,9 +471,11 @@ void DRW_hair_update(void)
   MEM_freeN(data);
   GPU_framebuffer_free(fb);
 #else
-  /* TODO(fclem): replace by compute shader. */
-  /* Just render using transform feedback. */
+  /* Just render the pass when using compute shaders or transform feedback. */
   DRW_draw_pass(g_tf_pass);
+  if (drw_hair_use_compute_shaders()) {
+    GPU_memory_barrier(GPU_BARRIER_SHADER_STORAGE);
+  }
 #endif
 }