OpenSubDiv: add support for an OpenGL evaluator

This evaluator is used in order to evaluate subdivision at render time, allowing for
faster renders of meshes with a subdivision surface modifier placed at the last
position in the modifier list.

When evaluating the subsurf modifier, we detect whether we can delegate evaluation
to the draw code. If so, the subdivision is first evaluated on the GPU using our own
custom evaluator (only the coarse data needs to be initially sent to the GPU), then,
buffers for the final `MeshBufferCache` are filled on the GPU using a set of
compute shaders. However, some buffers are still filled on the CPU side, if doing so
on the GPU is impractical (e.g. the line adjacency buffer used for x-ray, whose
logic is hardly GPU compatible).

This is done at the mesh buffer extraction level so that the result can be readily used
in the various OpenGL engines, without having to write custom geometry or tesselation
shaders.

We use our own subdivision evaluation shaders, instead of OpenSubDiv's vanilla one, in
order to control the data layout, and interpolation. For example, we store vertex colors
as compressed 16-bit integers, while OpenSubDiv's default evaluator only work for float
types.

In order to still access the modified geometry on the CPU side, for use in modifiers
or transform operators, a dedicated wrapper type is added `MESH_WRAPPER_TYPE_SUBD`.
Subdivision will be lazily evaluated via `BKE_object_get_evaluated_mesh` which will
create such a wrapper if possible. If the final subdivision surface is not needed on
the CPU side, `BKE_object_get_evaluated_mesh_no_subsurf` should be used.

Enabling or disabling GPU subdivision can be done through the user preferences (under
Viewport -> Subdivision).

See patch description for benchmarks.

Reviewed By: campbellbarton, jbakker, fclem, brecht, #eevee_viewport

Differential Revision: https://developer.blender.org/D12406

12 files changed, 105 insertions, 7 deletions

diff --git a/source/blender/gpu/GPU_context.h b/source/blender/gpu/GPU_context.h
index 5189fa1ae41..5e67441be27 100644
--- a/source/blender/gpu/GPU_context.h
+++ b/source/blender/gpu/GPU_context.h
@@ -40,6 +40,8 @@ typedef enum eGPUBackendType {
 void GPU_backend_init(eGPUBackendType backend);
 void GPU_backend_exit(void);
 
+eGPUBackendType GPU_backend_get_type(void);
+
 /** Opaque type hiding blender::gpu::Context. */
 typedef struct GPUContext GPUContext;
 
diff --git a/source/blender/gpu/GPU_index_buffer.h b/source/blender/gpu/GPU_index_buffer.h
index e4f1709173e..0f83e590597 100644
--- a/source/blender/gpu/GPU_index_buffer.h
+++ b/source/blender/gpu/GPU_index_buffer.h
@@ -53,6 +53,8 @@ void GPU_indexbuf_init_ex(GPUIndexBufBuilder *, GPUPrimType, uint index_len, uin
 void GPU_indexbuf_init(GPUIndexBufBuilder *, GPUPrimType, uint prim_len, uint vertex_len);
 GPUIndexBuf *GPU_indexbuf_build_on_device(uint index_len);
 
+void GPU_indexbuf_init_build_on_device(GPUIndexBuf *elem, uint index_len);
+
 /*
  * Thread safe.
  *
@@ -82,6 +84,16 @@ void GPU_indexbuf_build_in_place(GPUIndexBufBuilder *, GPUIndexBuf *);
 
 void GPU_indexbuf_bind_as_ssbo(GPUIndexBuf *elem, int binding);
 
+/* Upload data to the GPU (if not built on the device) and bind the buffer to its default target.
+ */
+void GPU_indexbuf_use(GPUIndexBuf *elem);
+
+/* Partially update the GPUIndexBuf which was already sent to the device, or built directly on the
+ * device. The data needs to be compatible with potential compression applied to the original
+ * indices when the index buffer was built, i.e., if the data was compressed to use shorts instead
+ * of ints, shorts should passed here. */
+void GPU_indexbuf_update_sub(GPUIndexBuf *elem, uint start, uint len, const void *data);
+
 /* Create a sub-range of an existing index-buffer. */
 GPUIndexBuf *GPU_indexbuf_create_subrange(GPUIndexBuf *elem_src, uint start, uint length);
 void GPU_indexbuf_create_subrange_in_place(GPUIndexBuf *elem,
diff --git a/source/blender/gpu/GPU_vertex_buffer.h b/source/blender/gpu/GPU_vertex_buffer.h
index 62a495abfb3..43a8e7fc4cb 100644
--- a/source/blender/gpu/GPU_vertex_buffer.h
+++ b/source/blender/gpu/GPU_vertex_buffer.h
@@ -91,6 +91,8 @@ void GPU_vertbuf_handle_ref_remove(GPUVertBuf *verts);
 
 void GPU_vertbuf_init_with_format_ex(GPUVertBuf *, const GPUVertFormat *, GPUUsageType);
 
+void GPU_vertbuf_init_build_on_device(GPUVertBuf *verts, GPUVertFormat *format, uint v_len);
+
 #define GPU_vertbuf_init_with_format(verts, format) \
   GPU_vertbuf_init_with_format_ex(verts, format, GPU_USAGE_STATIC)
 
@@ -172,6 +174,7 @@ const GPUVertFormat *GPU_vertbuf_get_format(const GPUVertBuf *verts);
 uint GPU_vertbuf_get_vertex_alloc(const GPUVertBuf *verts);
 uint GPU_vertbuf_get_vertex_len(const GPUVertBuf *verts);
 GPUVertBufStatus GPU_vertbuf_get_status(const GPUVertBuf *verts);
+void GPU_vertbuf_tag_dirty(GPUVertBuf *verts);
 
 /**
  * Should be rename to #GPU_vertbuf_data_upload.
@@ -179,12 +182,14 @@ GPUVertBufStatus GPU_vertbuf_get_status(const GPUVertBuf *verts);
 void GPU_vertbuf_use(GPUVertBuf *);
 void GPU_vertbuf_bind_as_ssbo(struct GPUVertBuf *verts, int binding);
 
+void GPU_vertbuf_wrap_handle(GPUVertBuf *verts, uint64_t handle);
+
 /**
  * XXX: do not use!
  * This is just a wrapper for the use of the Hair refine workaround.
  * To be used with #GPU_vertbuf_use().
  */
-void GPU_vertbuf_update_sub(GPUVertBuf *verts, uint start, uint len, void *data);
+void GPU_vertbuf_update_sub(GPUVertBuf *verts, uint start, uint len, const void *data);
 
 /* Metrics */
 uint GPU_vertbuf_get_memory_usage(void);
diff --git a/source/blender/gpu/intern/gpu_context.cc b/source/blender/gpu/intern/gpu_context.cc
index 5af15d1bc3d..98714269402 100644
--- a/source/blender/gpu/intern/gpu_context.cc
+++ b/source/blender/gpu/intern/gpu_context.cc
@@ -186,6 +186,15 @@ void GPU_backend_exit()
   g_backend = nullptr;
 }
 
+eGPUBackendType GPU_backend_get_type()
+{
+  if (g_backend && dynamic_cast<GLBackend *>(g_backend) != nullptr) {
+    return GPU_BACKEND_OPENGL;
+  }
+
+  return GPU_BACKEND_NONE;
+}
+
 GPUBackend *GPUBackend::get()
 {
   return g_backend;
diff --git a/source/blender/gpu/intern/gpu_index_buffer.cc b/source/blender/gpu/intern/gpu_index_buffer.cc
index 3472cc24a74..895b2a8461b 100644
--- a/source/blender/gpu/intern/gpu_index_buffer.cc
+++ b/source/blender/gpu/intern/gpu_index_buffer.cc
@@ -74,11 +74,16 @@ void GPU_indexbuf_init(GPUIndexBufBuilder *builder,
 GPUIndexBuf *GPU_indexbuf_build_on_device(uint index_len)
 {
   GPUIndexBuf *elem_ = GPU_indexbuf_calloc();
-  IndexBuf *elem = unwrap(elem_);
-  elem->init_build_on_device(index_len);
+  GPU_indexbuf_init_build_on_device(elem_, index_len);
   return elem_;
 }
 
+void GPU_indexbuf_init_build_on_device(GPUIndexBuf *elem, uint index_len)
+{
+  IndexBuf *elem_ = unwrap(elem);
+  elem_->init_build_on_device(index_len);
+}
+
 void GPU_indexbuf_join(GPUIndexBufBuilder *builder_to, const GPUIndexBufBuilder *builder_from)
 {
   BLI_assert(builder_to->data == builder_from->data);
@@ -410,9 +415,19 @@ int GPU_indexbuf_primitive_len(GPUPrimType prim_type)
   return indices_per_primitive(prim_type);
 }
 
+void GPU_indexbuf_use(GPUIndexBuf *elem)
+{
+  unwrap(elem)->upload_data();
+}
+
 void GPU_indexbuf_bind_as_ssbo(GPUIndexBuf *elem, int binding)
 {
   unwrap(elem)->bind_as_ssbo(binding);
 }
 
+void GPU_indexbuf_update_sub(GPUIndexBuf *elem, uint start, uint len, const void *data)
+{
+  unwrap(elem)->update_sub(start, len, data);
+}
+
 /** \} */
diff --git a/source/blender/gpu/intern/gpu_index_buffer_private.hh b/source/blender/gpu/intern/gpu_index_buffer_private.hh
index ed7dd830c8c..adc0145f867 100644
--- a/source/blender/gpu/intern/gpu_index_buffer_private.hh
+++ b/source/blender/gpu/intern/gpu_index_buffer_private.hh
@@ -92,11 +92,15 @@ class IndexBuf {
     return is_init_;
   };
 
+  virtual void upload_data(void) = 0;
+
   virtual void bind_as_ssbo(uint binding) = 0;
 
   virtual const uint32_t *read() const = 0;
   uint32_t *unmap(const uint32_t *mapped_memory) const;
 
+  virtual void update_sub(uint start, uint len, const void *data) = 0;
+
  private:
   inline void squeeze_indices_short(uint min_idx, uint max_idx);
   inline uint index_range(uint *r_min, uint *r_max);
diff --git a/source/blender/gpu/intern/gpu_vertex_buffer.cc b/source/blender/gpu/intern/gpu_vertex_buffer.cc
index 5ed9648387f..dba31f501f2 100644
--- a/source/blender/gpu/intern/gpu_vertex_buffer.cc
+++ b/source/blender/gpu/intern/gpu_vertex_buffer.cc
@@ -144,6 +144,12 @@ void GPU_vertbuf_init_with_format_ex(GPUVertBuf *verts_,
   unwrap(verts_)->init(format, usage);
 }
 
+void GPU_vertbuf_init_build_on_device(GPUVertBuf *verts, GPUVertFormat *format, uint v_len)
+{
+  GPU_vertbuf_init_with_format_ex(verts, format, GPU_USAGE_DEVICE_ONLY);
+  GPU_vertbuf_data_alloc(verts, v_len);
+}
+
 GPUVertBuf *GPU_vertbuf_duplicate(GPUVertBuf *verts_)
 {
   return wrap(unwrap(verts_)->duplicate());
@@ -313,6 +319,11 @@ GPUVertBufStatus GPU_vertbuf_get_status(const GPUVertBuf *verts)
   return unwrap(verts)->flag;
 }
 
+void GPU_vertbuf_tag_dirty(GPUVertBuf *verts)
+{
+  unwrap(verts)->flag |= GPU_VERTBUF_DATA_DIRTY;
+}
+
 uint GPU_vertbuf_get_memory_usage()
 {
   return VertBuf::memory_usage;
@@ -323,12 +334,17 @@ void GPU_vertbuf_use(GPUVertBuf *verts)
   unwrap(verts)->upload();
 }
 
+void GPU_vertbuf_wrap_handle(GPUVertBuf *verts, uint64_t handle)
+{
+  unwrap(verts)->wrap_handle(handle);
+}
+
 void GPU_vertbuf_bind_as_ssbo(struct GPUVertBuf *verts, int binding)
 {
   unwrap(verts)->bind_as_ssbo(binding);
 }
 
-void GPU_vertbuf_update_sub(GPUVertBuf *verts, uint start, uint len, void *data)
+void GPU_vertbuf_update_sub(GPUVertBuf *verts, uint start, uint len, const void *data)
 {
   unwrap(verts)->update_sub(start, len, data);
 }
diff --git a/source/blender/gpu/intern/gpu_vertex_buffer_private.hh b/source/blender/gpu/intern/gpu_vertex_buffer_private.hh
index 9531c2c1a5f..2f46295f45a 100644
--- a/source/blender/gpu/intern/gpu_vertex_buffer_private.hh
+++ b/source/blender/gpu/intern/gpu_vertex_buffer_private.hh
@@ -68,6 +68,8 @@ class VertBuf {
   void upload(void);
   virtual void bind_as_ssbo(uint binding) = 0;
 
+  virtual void wrap_handle(uint64_t handle) = 0;
+
   VertBuf *duplicate(void);
 
   /* Size of the data allocated. */
@@ -96,7 +98,7 @@ class VertBuf {
     }
   }
 
-  virtual void update_sub(uint start, uint len, void *data) = 0;
+  virtual void update_sub(uint start, uint len, const void *data) = 0;
   virtual const void *read() const = 0;
   virtual void *unmap(const void *mapped_data) const = 0;
 
diff --git a/source/blender/gpu/opengl/gl_index_buffer.cc b/source/blender/gpu/opengl/gl_index_buffer.cc
index e305f765ad9..82bab460ae3 100644
--- a/source/blender/gpu/opengl/gl_index_buffer.cc
+++ b/source/blender/gpu/opengl/gl_index_buffer.cc
@@ -81,4 +81,14 @@ bool GLIndexBuf::is_active() const
   return ibo_id_ == active_ibo_id;
 }
 
+void GLIndexBuf::upload_data()
+{
+  bind();
+}
+
+void GLIndexBuf::update_sub(uint start, uint len, const void *data)
+{
+  glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, start, len, data);
+}
+
 }  // namespace blender::gpu
diff --git a/source/blender/gpu/opengl/gl_index_buffer.hh b/source/blender/gpu/opengl/gl_index_buffer.hh
index 0dbdaa6d398..85d52447bc6 100644
--- a/source/blender/gpu/opengl/gl_index_buffer.hh
+++ b/source/blender/gpu/opengl/gl_index_buffer.hh
@@ -61,6 +61,10 @@ class GLIndexBuf : public IndexBuf {
     return (index_type_ == GPU_INDEX_U16) ? 0xFFFFu : 0xFFFFFFFFu;
   }
 
+  void upload_data(void) override;
+
+  void update_sub(uint start, uint len, const void *data) override;
+
  private:
   bool is_active() const;
 
diff --git a/source/blender/gpu/opengl/gl_vertex_buffer.cc b/source/blender/gpu/opengl/gl_vertex_buffer.cc
index ce16a491528..469ac2cf8d6 100644
--- a/source/blender/gpu/opengl/gl_vertex_buffer.cc
+++ b/source/blender/gpu/opengl/gl_vertex_buffer.cc
@@ -49,6 +49,10 @@ void GLVertBuf::resize_data()
 
 void GLVertBuf::release_data()
 {
+  if (is_wrapper_) {
+    return;
+  }
+
   if (vbo_id_ != 0) {
     GLContext::buf_free(vbo_id_);
     vbo_id_ = 0;
@@ -137,6 +141,16 @@ void *GLVertBuf::unmap(const void *mapped_data) const
   return result;
 }
 
+void GLVertBuf::wrap_handle(uint64_t handle)
+{
+  BLI_assert(vbo_id_ == 0);
+  BLI_assert(glIsBuffer(static_cast<uint>(handle)));
+  is_wrapper_ = true;
+  vbo_id_ = static_cast<uint>(handle);
+  /* We assume the data is already on the device, so no need to allocate or send it. */
+  flag = GPU_VERTBUF_DATA_UPLOADED;
+}
+
 bool GLVertBuf::is_active() const
 {
   if (!vbo_id_) {
@@ -147,7 +161,7 @@ bool GLVertBuf::is_active() const
   return vbo_id_ == active_vbo_id;
 }
 
-void GLVertBuf::update_sub(uint start, uint len, void *data)
+void GLVertBuf::update_sub(uint start, uint len, const void *data)
 {
   glBufferSubData(GL_ARRAY_BUFFER, start, len, data);
 }
diff --git a/source/blender/gpu/opengl/gl_vertex_buffer.hh b/source/blender/gpu/opengl/gl_vertex_buffer.hh
index 6c38a2225b3..27e4cc4f8e2 100644
--- a/source/blender/gpu/opengl/gl_vertex_buffer.hh
+++ b/source/blender/gpu/opengl/gl_vertex_buffer.hh
@@ -39,17 +39,22 @@ class GLVertBuf : public VertBuf {
  private:
   /** OpenGL buffer handle. Init on first upload. Immutable after that. */
   GLuint vbo_id_ = 0;
+  /** Defines whether the buffer handle is wrapped by this GLVertBuf, i.e. we do not own it and
+   * should not free it. */
+  bool is_wrapper_ = false;
   /** Size on the GPU. */
   size_t vbo_size_ = 0;
 
  public:
   void bind(void);
 
-  void update_sub(uint start, uint len, void *data) override;
+  void update_sub(uint start, uint len, const void *data) override;
 
   const void *read() const override;
   void *unmap(const void *mapped_data) const override;
 
+  void wrap_handle(uint64_t handle) override;
+
  protected:
   void acquire_data(void) override;
   void resize_data(void) override;