DRW: Add compute_ref calls, barriers calls, and vertex_buffer_ref

- Compute ref let the size of dispatch be modified just before drawing.
- Barrier call makes it possible to chain multiple compute passes in one pass.
- DRW_shgroup_vertex_buffer_ref is the analog of DRW_shgroup_uniform_block_ref.

4 files changed, 82 insertions, 1 deletions

diff --git a/source/blender/draw/intern/DRW_render.h b/source/blender/draw/intern/DRW_render.h
index 8c56d21746d..29a321df556 100644
--- a/source/blender/draw/intern/DRW_render.h
+++ b/source/blender/draw/intern/DRW_render.h
@@ -468,6 +468,10 @@ void DRW_shgroup_call_compute(DRWShadingGroup *shgroup,
                               int groups_x_len,
                               int groups_y_len,
                               int groups_z_len);
+/**
+ * \warning this keeps the ref to groups_ref until it actually dispatch.
+ */
+void DRW_shgroup_call_compute_ref(DRWShadingGroup *shgroup, int groups_ref[3]);
 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);
@@ -534,6 +538,11 @@ void DRW_shgroup_stencil_set(DRWShadingGroup *shgroup,
 void DRW_shgroup_stencil_mask(DRWShadingGroup *shgroup, uint mask);
 
 /**
+ * Issue a barrier command.
+ */
+void DRW_shgroup_barrier(DRWShadingGroup *shgroup, eGPUBarrier type);
+
+/**
  * Issue a clear command.
  */
 void DRW_shgroup_clear_framebuffer(DRWShadingGroup *shgroup,
@@ -627,6 +636,9 @@ void DRW_shgroup_uniform_vec4_array_copy(DRWShadingGroup *shgroup,
 void DRW_shgroup_vertex_buffer(DRWShadingGroup *shgroup,
                                const char *name,
                                struct GPUVertBuf *vertex_buffer);
+void DRW_shgroup_vertex_buffer_ref(DRWShadingGroup *shgroup,
+                                   const char *name,
+                                   struct GPUVertBuf **vertex_buffer);
 
 bool DRW_shgroup_is_empty(DRWShadingGroup *shgroup);
 
diff --git a/source/blender/draw/intern/draw_manager.h b/source/blender/draw/intern/draw_manager.h
index 38cd3ecb9b6..73fd157426c 100644
--- a/source/blender/draw/intern/draw_manager.h
+++ b/source/blender/draw/intern/draw_manager.h
@@ -205,8 +205,10 @@ typedef enum {
 
   /* Compute Commands. */
   DRW_CMD_COMPUTE = 8,
+  DRW_CMD_COMPUTE_REF = 9,
 
   /* Other Commands */
+  DRW_CMD_BARRIER = 11,
   DRW_CMD_CLEAR = 12,
   DRW_CMD_DRWSTATE = 13,
   DRW_CMD_STENCIL = 14,
@@ -249,6 +251,14 @@ typedef struct DRWCommandCompute {
   int groups_z_len;
 } DRWCommandCompute;
 
+typedef struct DRWCommandComputeRef {
+  int *groups_ref;
+} DRWCommandComputeRef;
+
+typedef struct DRWCommandBarrier {
+  eGPUBarrier type;
+} DRWCommandBarrier;
+
 typedef struct DRWCommandDrawProcedural {
   GPUBatch *batch;
   DRWResourceHandle handle;
@@ -286,6 +296,8 @@ typedef union DRWCommand {
   DRWCommandDrawInstanceRange instance_range;
   DRWCommandDrawProcedural procedural;
   DRWCommandCompute compute;
+  DRWCommandComputeRef compute_ref;
+  DRWCommandBarrier barrier;
   DRWCommandSetMutableState state;
   DRWCommandSetStencil stencil;
   DRWCommandSetSelectID select_id;
@@ -314,6 +326,7 @@ typedef enum {
   DRW_UNIFORM_BLOCK_REF,
   DRW_UNIFORM_TFEEDBACK_TARGET,
   DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE,
+  DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE_REF,
   /** Per drawcall uniforms/UBO */
   DRW_UNIFORM_BLOCK_OBMATS,
   DRW_UNIFORM_BLOCK_OBINFOS,
@@ -345,6 +358,11 @@ struct DRWUniform {
       GPUUniformBuf *block;
       GPUUniformBuf **block_ref;
     };
+    /* DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE */
+    union {
+      GPUVertBuf *vertbuf;
+      GPUVertBuf **vertbuf_ref;
+    };
     /* DRW_UNIFORM_FLOAT_COPY */
     float fvalue[4];
     /* DRW_UNIFORM_INT_COPY */
diff --git a/source/blender/draw/intern/draw_manager_data.c b/source/blender/draw/intern/draw_manager_data.c
index f07fc94f5d1..9091ae632d0 100644
--- a/source/blender/draw/intern/draw_manager_data.c
+++ b/source/blender/draw/intern/draw_manager_data.c
@@ -460,6 +460,19 @@ void DRW_shgroup_vertex_buffer(DRWShadingGroup *shgroup,
       shgroup, location, DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE, vertex_buffer, 0, 0, 1);
 }
 
+void DRW_shgroup_vertex_buffer_ref(DRWShadingGroup *shgroup,
+                                   const char *name,
+                                   GPUVertBuf **vertex_buffer)
+{
+  int location = GPU_shader_get_ssbo(shgroup->shader, name);
+  if (location == -1) {
+    BLI_assert_msg(0, "Unable to locate binding of shader storage buffer objects.");
+    return;
+  }
+  drw_shgroup_uniform_create_ex(
+      shgroup, location, DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE_REF, vertex_buffer, 0, 0, 1);
+}
+
 /** \} */
 
 /* -------------------------------------------------------------------- */
@@ -730,6 +743,18 @@ static void drw_command_compute(DRWShadingGroup *shgroup,
   cmd->groups_z_len = groups_z_len;
 }
 
+static void drw_command_compute_ref(DRWShadingGroup *shgroup, int groups_ref[3])
+{
+  DRWCommandComputeRef *cmd = drw_command_create(shgroup, DRW_CMD_COMPUTE_REF);
+  cmd->groups_ref = groups_ref;
+}
+
+static void drw_command_barrier(DRWShadingGroup *shgroup, eGPUBarrier type)
+{
+  DRWCommandBarrier *cmd = drw_command_create(shgroup, DRW_CMD_BARRIER);
+  cmd->type = type;
+}
+
 static void drw_command_draw_procedural(DRWShadingGroup *shgroup,
                                         GPUBatch *batch,
                                         DRWResourceHandle handle,
@@ -855,6 +880,20 @@ void DRW_shgroup_call_compute(DRWShadingGroup *shgroup,
   drw_command_compute(shgroup, groups_x_len, groups_y_len, groups_z_len);
 }
 
+void DRW_shgroup_call_compute_ref(DRWShadingGroup *shgroup, int groups_ref[3])
+{
+  BLI_assert(GPU_compute_shader_support());
+
+  drw_command_compute_ref(shgroup, groups_ref);
+}
+
+void DRW_shgroup_barrier(DRWShadingGroup *shgroup, eGPUBarrier type)
+{
+  BLI_assert(GPU_compute_shader_support());
+
+  drw_command_barrier(shgroup, type);
+}
+
 static void drw_shgroup_call_procedural_add_ex(DRWShadingGroup *shgroup,
                                                GPUBatch *geom,
                                                Object *ob,
diff --git a/source/blender/draw/intern/draw_manager_exec.c b/source/blender/draw/intern/draw_manager_exec.c
index 2095a8483d6..a579403975f 100644
--- a/source/blender/draw/intern/draw_manager_exec.c
+++ b/source/blender/draw/intern/draw_manager_exec.c
@@ -662,8 +662,11 @@ static void draw_update_uniforms(DRWShadingGroup *shgroup,
           *use_tfeedback = GPU_shader_transform_feedback_enable(shgroup->shader,
                                                                 ((GPUVertBuf *)uni->pvalue));
           break;
+        case DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE_REF:
+          GPU_vertbuf_bind_as_ssbo(*uni->vertbuf_ref, uni->location);
+          break;
         case DRW_UNIFORM_VERTEX_BUFFER_AS_STORAGE:
-          GPU_vertbuf_bind_as_ssbo((GPUVertBuf *)uni->pvalue, uni->location);
+          GPU_vertbuf_bind_as_ssbo(uni->vertbuf, uni->location);
           break;
           /* Legacy/Fallback support. */
         case DRW_UNIFORM_BASE_INSTANCE:
@@ -1049,6 +1052,15 @@ static void draw_shgroup(DRWShadingGroup *shgroup, DRWState pass_state)
                                cmd->compute.groups_y_len,
                                cmd->compute.groups_z_len);
           break;
+        case DRW_CMD_COMPUTE_REF:
+          GPU_compute_dispatch(shgroup->shader,
+                               cmd->compute_ref.groups_ref[0],
+                               cmd->compute_ref.groups_ref[1],
+                               cmd->compute_ref.groups_ref[2]);
+          break;
+        case DRW_CMD_BARRIER:
+          GPU_memory_barrier(cmd->barrier.type);
+          break;
       }
     }