GPU: Compute Pipeline.

With the compute pipeline calculation can be offloaded to the GPU.
This patch only adds the framework for compute. So no changes for users at
this moment.

NOTE: As this is an OpenGL4.3 feature it must always have a fallback.

Use `GPU_compute_shader_support` to check if compute pipeline can be used.
Check `gpu_shader_compute*` test cases for usage.

This patch also adds support for shader storage buffer objects and device only
vertex/index buffers.

An alternative that had been discussed was adding this to the `GPUBatch`, this
was eventually not chosen as it would lead to more code when used as part of a
shading group. The idea is that we add an `eDRWCommandType` in the near
future.

Reviewed By: fclem

Differential Revision: https://developer.blender.org/D10913

13 files changed, 246 insertions, 14 deletions

diff --git a/source/blender/gpu/opengl/gl_backend.cc b/source/blender/gpu/opengl/gl_backend.cc
index 31b6549fc3b..fb03a2c2d2a 100644
--- a/source/blender/gpu/opengl/gl_backend.cc
+++ b/source/blender/gpu/opengl/gl_backend.cc
@@ -437,6 +437,8 @@ 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;
+  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);
   glGetIntegerv(GL_MAX_CUBE_MAP_TEXTURE_SIZE, &GLContext::max_cubemap_size);
diff --git a/source/blender/gpu/opengl/gl_backend.hh b/source/blender/gpu/opengl/gl_backend.hh
index 231e5811b45..e9dcdffced0 100644
--- a/source/blender/gpu/opengl/gl_backend.hh
+++ b/source/blender/gpu/opengl/gl_backend.hh
@@ -28,6 +28,7 @@
 #include "BLI_vector.hh"
 
 #include "gl_batch.hh"
+#include "gl_compute.hh"
 #include "gl_context.hh"
 #include "gl_drawlist.hh"
 #include "gl_framebuffer.hh"
@@ -126,6 +127,12 @@ class GLBackend : public GPUBackend {
     return shared_orphan_list_;
   };
 
+  void compute_dispatch(int groups_x_len, int groups_y_len, int groups_z_len) override
+  {
+    GLContext::get()->state_manager_active_get()->apply_state();
+    GLCompute::dispatch(groups_x_len, groups_y_len, groups_z_len);
+  }
+
  private:
   static void platform_init(void);
   static void platform_exit(void);
diff --git a/source/blender/gpu/opengl/gl_compute.cc b/source/blender/gpu/opengl/gl_compute.cc
new file mode 100644
index 00000000000..fa8317dde4a
--- /dev/null
+++ b/source/blender/gpu/opengl/gl_compute.cc
@@ -0,0 +1,35 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+/** \file
+ * \ingroup gpu
+ */
+
+#include "gl_compute.hh"
+
+#include "gl_debug.hh"
+
+#include "glew-mx.h"
+
+namespace blender::gpu {
+
+void GLCompute::dispatch(int group_x_len, int group_y_len, int group_z_len)
+{
+  glDispatchCompute(group_x_len, group_y_len, group_z_len);
+  debug::check_gl_error("Dispatch Compute");
+}
+
+}  // namespace blender::gpu
diff --git a/source/blender/gpu/opengl/gl_compute.hh b/source/blender/gpu/opengl/gl_compute.hh
new file mode 100644
index 00000000000..2fd918ddd10
--- /dev/null
+++ b/source/blender/gpu/opengl/gl_compute.hh
@@ -0,0 +1,30 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+/** \file
+ * \ingroup gpu
+ */
+
+#pragma once
+
+namespace blender::gpu {
+
+class GLCompute {
+ public:
+  static void dispatch(int group_x_len, int group_y_len, int group_z_len);
+};
+
+}  // namespace blender::gpu
diff --git a/source/blender/gpu/opengl/gl_index_buffer.cc b/source/blender/gpu/opengl/gl_index_buffer.cc
index e2c18c5d0b9..e305f765ad9 100644
--- a/source/blender/gpu/opengl/gl_index_buffer.cc
+++ b/source/blender/gpu/opengl/gl_index_buffer.cc
@@ -40,17 +40,14 @@ void GLIndexBuf::bind()
     return;
   }
 
-  if (ibo_id_ == 0) {
+  const bool allocate_on_device = ibo_id_ == 0;
+  if (allocate_on_device) {
     glGenBuffers(1, &ibo_id_);
-
-    if (data_ == nullptr) {
-      debug::raise_gl_error("Trying to use Index Buffer but the buffer contains no data");
-    }
   }
 
   glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, ibo_id_);
 
-  if (data_ != nullptr) {
+  if (data_ != nullptr || allocate_on_device) {
     size_t size = this->size_get();
     /* Sends data to GPU. */
     glBufferData(GL_ELEMENT_ARRAY_BUFFER, size, data_, GL_STATIC_DRAW);
@@ -59,4 +56,29 @@ void GLIndexBuf::bind()
   }
 }
 
+void GLIndexBuf::bind_as_ssbo(uint binding)
+{
+  bind();
+  BLI_assert(ibo_id_ != 0);
+  glBindBufferBase(GL_SHADER_STORAGE_BUFFER, binding, ibo_id_);
+}
+
+const uint32_t *GLIndexBuf::read() const
+{
+  BLI_assert(is_active());
+  void *data = glMapBuffer(GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY);
+  uint32_t *result = static_cast<uint32_t *>(data);
+  return result;
+}
+
+bool GLIndexBuf::is_active() const
+{
+  if (!ibo_id_) {
+    return false;
+  }
+  int active_ibo_id = 0;
+  glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &active_ibo_id);
+  return ibo_id_ == active_ibo_id;
+}
+
 }  // namespace blender::gpu
diff --git a/source/blender/gpu/opengl/gl_index_buffer.hh b/source/blender/gpu/opengl/gl_index_buffer.hh
index b84934bb77f..0dbdaa6d398 100644
--- a/source/blender/gpu/opengl/gl_index_buffer.hh
+++ b/source/blender/gpu/opengl/gl_index_buffer.hh
@@ -34,6 +34,7 @@ namespace blender::gpu {
 class GLIndexBuf : public IndexBuf {
   friend class GLBatch;
   friend class GLDrawList;
+  friend class GLShader; /* For compute shaders. */
 
  private:
   GLuint ibo_id_ = 0;
@@ -42,6 +43,9 @@ class GLIndexBuf : public IndexBuf {
   ~GLIndexBuf();
 
   void bind(void);
+  void bind_as_ssbo(uint binding) override;
+
+  const uint32_t *read() const override;
 
   void *offset_ptr(uint additional_vertex_offset) const
   {
@@ -57,6 +61,9 @@ class GLIndexBuf : public IndexBuf {
     return (index_type_ == GPU_INDEX_U16) ? 0xFFFFu : 0xFFFFFFFFu;
   }
 
+ private:
+  bool is_active() const;
+
   MEM_CXX_CLASS_ALLOC_FUNCS("GLIndexBuf")
 };
 
diff --git a/source/blender/gpu/opengl/gl_shader.cc b/source/blender/gpu/opengl/gl_shader.cc
index dd08a67517e..e77347d99eb 100644
--- a/source/blender/gpu/opengl/gl_shader.cc
+++ b/source/blender/gpu/opengl/gl_shader.cc
@@ -26,6 +26,7 @@
 #include "BLI_string.h"
 #include "BLI_vector.hh"
 
+#include "GPU_capabilities.h"
 #include "GPU_platform.h"
 
 #include "gl_backend.hh"
@@ -63,6 +64,7 @@ GLShader::~GLShader()
   glDeleteShader(vert_shader_);
   glDeleteShader(geom_shader_);
   glDeleteShader(frag_shader_);
+  glDeleteShader(compute_shader_);
   glDeleteProgram(shader_program_);
 }
 
@@ -72,7 +74,7 @@ GLShader::~GLShader()
 /** \name Shader stage creation
  * \{ */
 
-char *GLShader::glsl_patch_get()
+static char *glsl_patch_default_get()
 {
   /** Used for shader patching. Init once. */
   static char patch[512] = "\0";
@@ -111,6 +113,30 @@ char *GLShader::glsl_patch_get()
   return patch;
 }
 
+static char *glsl_patch_compute_get()
+{
+  /** Used for shader patching. Init once. */
+  static char patch[512] = "\0";
+  if (patch[0] != '\0') {
+    return patch;
+  }
+
+  size_t slen = 0;
+  /* Version need to go first. */
+  STR_CONCAT(patch, slen, "#version 430\n");
+  STR_CONCAT(patch, slen, "#extension GL_ARB_compute_shader :enable\n");
+  BLI_assert(slen < sizeof(patch));
+  return patch;
+}
+
+char *GLShader::glsl_patch_get(GLenum gl_stage)
+{
+  if (gl_stage == GL_COMPUTE_SHADER) {
+    return glsl_patch_compute_get();
+  }
+  return glsl_patch_default_get();
+}
+
 /* Create, compile and attach the shader stage to the shader program. */
 GLuint GLShader::create_shader_stage(GLenum gl_stage, MutableSpan<const char *> sources)
 {
@@ -121,7 +147,7 @@ GLuint GLShader::create_shader_stage(GLenum gl_stage, MutableSpan<const char *>
   }
 
   /* Patch the shader code using the first source slot. */
-  sources[0] = glsl_patch_get();
+  sources[0] = glsl_patch_get(gl_stage);
 
   glShaderSource(shader, sources.size(), sources.data(), nullptr);
   glCompileShader(shader);
@@ -142,6 +168,9 @@ GLuint GLShader::create_shader_stage(GLenum gl_stage, MutableSpan<const char *>
         case GL_FRAGMENT_SHADER:
           this->print_log(sources, log, "FragShader", !status);
           break;
+        case GL_COMPUTE_SHADER:
+          this->print_log(sources, log, "ComputeShader", !status);
+          break;
       }
     }
   }
@@ -172,6 +201,11 @@ void GLShader::fragment_shader_from_glsl(MutableSpan<const char *> sources)
   frag_shader_ = this->create_shader_stage(GL_FRAGMENT_SHADER, sources);
 }
 
+void GLShader::compute_shader_from_glsl(MutableSpan<const char *> sources)
+{
+  compute_shader_ = this->create_shader_stage(GL_COMPUTE_SHADER, sources);
+}
+
 bool GLShader::finalize()
 {
   if (compilation_failed_) {
diff --git a/source/blender/gpu/opengl/gl_shader.hh b/source/blender/gpu/opengl/gl_shader.hh
index 152eb2f068a..48aaaf2283d 100644
--- a/source/blender/gpu/opengl/gl_shader.hh
+++ b/source/blender/gpu/opengl/gl_shader.hh
@@ -43,6 +43,7 @@ class GLShader : public Shader {
   GLuint vert_shader_ = 0;
   GLuint geom_shader_ = 0;
   GLuint frag_shader_ = 0;
+  GLuint compute_shader_ = 0;
   /** True if any shader failed to compile. */
   bool compilation_failed_ = false;
 
@@ -56,6 +57,7 @@ class GLShader : public Shader {
   void vertex_shader_from_glsl(MutableSpan<const char *> sources) override;
   void geometry_shader_from_glsl(MutableSpan<const char *> sources) override;
   void fragment_shader_from_glsl(MutableSpan<const char *> sources) override;
+  void compute_shader_from_glsl(MutableSpan<const char *> sources) override;
   bool finalize(void) override;
 
   void transform_feedback_names_set(Span<const char *> name_list,
@@ -75,7 +77,7 @@ class GLShader : public Shader {
   int program_handle_get(void) const override;
 
  private:
-  char *glsl_patch_get(void);
+  char *glsl_patch_get(GLenum gl_stage);
 
   GLuint create_shader_stage(GLenum gl_stage, MutableSpan<const char *> sources);
 
diff --git a/source/blender/gpu/opengl/gl_shader_interface.cc b/source/blender/gpu/opengl/gl_shader_interface.cc
index 5870c645bf4..9cf072b2e8a 100644
--- a/source/blender/gpu/opengl/gl_shader_interface.cc
+++ b/source/blender/gpu/opengl/gl_shader_interface.cc
@@ -29,6 +29,8 @@
 
 #include "gl_shader_interface.hh"
 
+#include "GPU_capabilities.h"
+
 namespace blender::gpu {
 
 /* -------------------------------------------------------------------- */
@@ -125,6 +127,18 @@ static inline int image_binding(int32_t program,
       return -1;
   }
 }
+
+static inline int ssbo_binding(int32_t program, uint32_t ssbo_index)
+{
+  GLint binding = -1;
+  GLenum property = GL_BUFFER_BINDING;
+  GLint values_written = 0;
+  glGetProgramResourceiv(
+      program, GL_SHADER_STORAGE_BLOCK, ssbo_index, 1, &property, 1, &values_written, &binding);
+
+  return binding;
+}
+
 /** \} */
 
 /* -------------------------------------------------------------------- */
@@ -149,6 +163,13 @@ GLShaderInterface::GLShaderInterface(GLuint program)
   glGetProgramiv(program, GL_ACTIVE_UNIFORMS, &active_uniform_len);
   uniform_len = active_uniform_len;
 
+  GLint max_ssbo_name_len = 0, ssbo_len = 0;
+  if (GPU_shader_storage_buffer_objects_support()) {
+    glGetProgramInterfaceiv(program, GL_SHADER_STORAGE_BLOCK, GL_ACTIVE_RESOURCES, &ssbo_len);
+    glGetProgramInterfaceiv(
+        program, GL_SHADER_STORAGE_BLOCK, GL_MAX_NAME_LENGTH, &max_ssbo_name_len);
+  }
+
   BLI_assert(ubo_len <= 16 && "enabled_ubo_mask_ is uint16_t");
 
   /* Work around driver bug with Intel HD 4600 on Windows 7/8, where
@@ -162,6 +183,9 @@ GLShaderInterface::GLShaderInterface(GLuint program)
   if (uniform_len > 0 && max_uniform_name_len == 0) {
     max_uniform_name_len = 256;
   }
+  if (ssbo_len > 0 && max_ssbo_name_len == 0) {
+    max_ssbo_name_len = 256;
+  }
 
   /* GL_ACTIVE_UNIFORMS lied to us! Remove the UBO uniforms from the total before
    * allocating the uniform array. */
@@ -186,11 +210,12 @@ GLShaderInterface::GLShaderInterface(GLuint program)
   }
   MEM_freeN(ubo_uni_ids);
 
-  int input_tot_len = attr_len + ubo_len + uniform_len;
+  int input_tot_len = attr_len + ubo_len + uniform_len + ssbo_len;
   inputs_ = (ShaderInput *)MEM_callocN(sizeof(ShaderInput) * input_tot_len, __func__);
 
   const uint32_t name_buffer_len = attr_len * max_attr_name_len + ubo_len * max_ubo_name_len +
-                                   uniform_len * max_uniform_name_len;
+                                   uniform_len * max_uniform_name_len +
+                                   ssbo_len * max_ssbo_name_len;
   name_buffer_ = (char *)MEM_mallocN(name_buffer_len, "name_buffer");
   uint32_t name_buffer_offset = 0;
 
@@ -257,6 +282,22 @@ GLShaderInterface::GLShaderInterface(GLuint program)
     }
   }
 
+  /* SSBOs */
+  for (int i = 0; i < ssbo_len; i++) {
+    char *name = name_buffer_ + name_buffer_offset;
+    GLsizei remaining_buffer = name_buffer_len - name_buffer_offset;
+    GLsizei name_len = 0;
+    glGetProgramResourceName(
+        program, GL_SHADER_STORAGE_BLOCK, i, remaining_buffer, &name_len, name);
+
+    const GLint binding = ssbo_binding(program, i);
+
+    ShaderInput *input = &inputs_[attr_len_ + ubo_len_ + uniform_len_ + ssbo_len_++];
+    input->binding = input->location = binding;
+
+    name_buffer_offset += this->set_input_name(input, name, name_len);
+  }
+
   /* Builtin Uniforms */
   for (int32_t u_int = 0; u_int < GPU_NUM_UNIFORMS; u_int++) {
     GPUUniformBuiltin u = static_cast<GPUUniformBuiltin>(u_int);
diff --git a/source/blender/gpu/opengl/gl_state.hh b/source/blender/gpu/opengl/gl_state.hh
index 651c3c22afa..3b4b40b1d10 100644
--- a/source/blender/gpu/opengl/gl_state.hh
+++ b/source/blender/gpu/opengl/gl_state.hh
@@ -121,6 +121,9 @@ static inline GLbitfield to_gl(eGPUBarrier barrier_bits)
   if (barrier_bits & GPU_BARRIER_TEXTURE_FETCH) {
     barrier |= GL_TEXTURE_FETCH_BARRIER_BIT;
   }
+  if (barrier_bits & GPU_BARRIER_SHADER_STORAGE) {
+    barrier |= GL_SHADER_STORAGE_BARRIER_BIT;
+  }
   return barrier;
 }
 
diff --git a/source/blender/gpu/opengl/gl_texture.cc b/source/blender/gpu/opengl/gl_texture.cc
index b65686165d9..e2478a9976c 100644
--- a/source/blender/gpu/opengl/gl_texture.cc
+++ b/source/blender/gpu/opengl/gl_texture.cc
@@ -368,7 +368,7 @@ void GLTexture::copy_to(Texture *dst_)
 void *GLTexture::read(int mip, eGPUDataFormat type)
 {
   BLI_assert(!(format_flag_ & GPU_FORMAT_COMPRESSED));
-  BLI_assert(mip <= mipmaps_);
+  BLI_assert(mip <= mipmaps_ || mip == 0);
   BLI_assert(validate_data_format(format_, type));
 
   /* NOTE: mip_size_get() won't override any dimension that is equal to 0. */
diff --git a/source/blender/gpu/opengl/gl_vertex_buffer.cc b/source/blender/gpu/opengl/gl_vertex_buffer.cc
index a56d5269fde..ce16a491528 100644
--- a/source/blender/gpu/opengl/gl_vertex_buffer.cc
+++ b/source/blender/gpu/opengl/gl_vertex_buffer.cc
@@ -29,6 +29,10 @@ namespace blender::gpu {
 
 void GLVertBuf::acquire_data()
 {
+  if (usage_ == GPU_USAGE_DEVICE_ONLY) {
+    return;
+  }
+
   /* Discard previous data if any. */
   MEM_SAFE_FREE(data);
   data = (uchar *)MEM_mallocN(sizeof(uchar) * this->size_alloc_get(), __func__);
@@ -36,6 +40,10 @@ void GLVertBuf::acquire_data()
 
 void GLVertBuf::resize_data()
 {
+  if (usage_ == GPU_USAGE_DEVICE_ONLY) {
+    return;
+  }
+
   data = (uchar *)MEM_reallocN(data, sizeof(uchar) * this->size_alloc_get());
 }
 
@@ -94,8 +102,10 @@ void GLVertBuf::bind()
     vbo_size_ = this->size_used_get();
     /* Orphan the vbo to avoid sync then upload data. */
     glBufferData(GL_ARRAY_BUFFER, vbo_size_, nullptr, to_gl(usage_));
-    glBufferSubData(GL_ARRAY_BUFFER, 0, vbo_size_, data);
-
+    /* Do not transfer data from host to device when buffer is device only. */
+    if (usage_ != GPU_USAGE_DEVICE_ONLY) {
+      glBufferSubData(GL_ARRAY_BUFFER, 0, vbo_size_, data);
+    }
     memory_usage += vbo_size_;
 
     if (usage_ == GPU_USAGE_STATIC) {
@@ -106,6 +116,37 @@ void GLVertBuf::bind()
   }
 }
 
+void GLVertBuf::bind_as_ssbo(uint binding)
+{
+  bind();
+  BLI_assert(vbo_id_ != 0);
+  glBindBufferBase(GL_SHADER_STORAGE_BUFFER, binding, vbo_id_);
+}
+
+const void *GLVertBuf::read() const
+{
+  BLI_assert(is_active());
+  void *result = glMapBuffer(GL_ARRAY_BUFFER, GL_READ_ONLY);
+  return result;
+}
+
+void *GLVertBuf::unmap(const void *mapped_data) const
+{
+  void *result = MEM_mallocN(vbo_size_, __func__);
+  memcpy(result, mapped_data, vbo_size_);
+  return result;
+}
+
+bool GLVertBuf::is_active() const
+{
+  if (!vbo_id_) {
+    return false;
+  }
+  int active_vbo_id = 0;
+  glGetIntegerv(GL_ARRAY_BUFFER_BINDING, &active_vbo_id);
+  return vbo_id_ == active_vbo_id;
+}
+
 void GLVertBuf::update_sub(uint start, uint len, 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 e2bf6cd00e8..6c38a2225b3 100644
--- a/source/blender/gpu/opengl/gl_vertex_buffer.hh
+++ b/source/blender/gpu/opengl/gl_vertex_buffer.hh
@@ -47,12 +47,19 @@ class GLVertBuf : public VertBuf {
 
   void update_sub(uint start, uint len, void *data) override;
 
+  const void *read() const override;
+  void *unmap(const void *mapped_data) const override;
+
  protected:
   void acquire_data(void) override;
   void resize_data(void) override;
   void release_data(void) override;
   void upload_data(void) override;
   void duplicate_data(VertBuf *dst) override;
+  void bind_as_ssbo(uint binding) override;
+
+ private:
+  bool is_active() const;
 
   MEM_CXX_CLASS_ALLOC_FUNCS("GLVertBuf");
 };
@@ -65,6 +72,7 @@ static inline GLenum to_gl(GPUUsageType type)
     case GPU_USAGE_DYNAMIC:
       return GL_DYNAMIC_DRAW;
     case GPU_USAGE_STATIC:
+    case GPU_USAGE_DEVICE_ONLY:
       return GL_STATIC_DRAW;
     default:
       BLI_assert(0);