From 136bdb561b4ce05788e7b654c7e734cc35664b91 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9ment=20Foucault?= <foucault.clem@gmail.com>
Date: Sat, 12 Sep 2020 06:10:11 +0200
Subject: GPU: Add Image Load Store extension support

This wraps the functionality used to speedup EEVEE volumetrics.

This touches the rendering code of EEVEE as it should fix a mis-usage of
the GL barrier. The barrier changed type and location, removing an
unused barrier.
---
 source/blender/gpu/opengl/gl_state.cc | 95 +++++++++++++++++++++++++++++++++++
 1 file changed, 95 insertions(+)

(limited to 'source/blender/gpu/opengl/gl_state.cc')

diff --git a/source/blender/gpu/opengl/gl_state.cc b/source/blender/gpu/opengl/gl_state.cc
index 1678760e9cd..93753768928 100644
--- a/source/blender/gpu/opengl/gl_state.cc
+++ b/source/blender/gpu/opengl/gl_state.cc
@@ -76,6 +76,7 @@ void GLStateManager::apply_state(void)
   this->set_state(this->state);
   this->set_mutable_state(this->mutable_state);
   this->texture_bind_apply();
+  this->image_bind_apply();
   active_fb->apply_state();
 };
 
@@ -538,4 +539,98 @@ uint64_t GLStateManager::bound_texture_slots(void)
 
 /** \} */
 
+/* -------------------------------------------------------------------- */
+/** \name Image Binding (from image load store)
+ * \{ */
+
+void GLStateManager::image_bind(Texture *tex_, int unit)
+{
+  /* Minimum support is 8 image in the fragment shader. No image for other stages. */
+  BLI_assert(GPU_shader_image_load_store_support() && unit < 8);
+  GLTexture *tex = static_cast<GLTexture *>(tex_);
+  if (G.debug & G_DEBUG_GPU) {
+    tex->check_feedback_loop();
+  }
+  images_[unit] = tex->tex_id_;
+  formats_[unit] = to_gl_internal_format(tex->format_);
+  tex->is_bound_ = true;
+  dirty_image_binds_ |= 1ULL << unit;
+}
+
+void GLStateManager::image_unbind(Texture *tex_)
+{
+  GLTexture *tex = static_cast<GLTexture *>(tex_);
+  if (!tex->is_bound_) {
+    return;
+  }
+
+  GLuint tex_id = tex->tex_id_;
+  for (int i = 0; i < ARRAY_SIZE(images_); i++) {
+    if (images_[i] == tex_id) {
+      images_[i] = 0;
+      dirty_image_binds_ |= 1ULL << i;
+    }
+  }
+  tex->is_bound_ = false;
+}
+
+void GLStateManager::image_unbind_all(void)
+{
+  for (int i = 0; i < ARRAY_SIZE(images_); i++) {
+    if (images_[i] != 0) {
+      images_[i] = 0;
+      dirty_image_binds_ |= 1ULL << i;
+    }
+  }
+  this->image_bind_apply();
+}
+
+void GLStateManager::image_bind_apply(void)
+{
+  if (dirty_image_binds_ == 0) {
+    return;
+  }
+  uint32_t dirty_bind = dirty_image_binds_;
+  dirty_image_binds_ = 0;
+
+  int first = bitscan_forward_uint(dirty_bind);
+  int last = 32 - bitscan_reverse_uint(dirty_bind);
+  int count = last - first;
+
+  if (GLContext::multi_bind_support) {
+    glBindImageTextures(first, count, images_ + first);
+  }
+  else {
+    for (int unit = first; unit < last; unit++) {
+      if ((dirty_bind >> unit) & 1UL) {
+        glBindImageTexture(unit, images_[unit], 0, GL_TRUE, 0, GL_READ_WRITE, formats_[unit]);
+      }
+    }
+  }
+}
+
+uint8_t GLStateManager::bound_image_slots(void)
+{
+  uint8_t bound_slots = 0;
+  for (int i = 0; i < ARRAY_SIZE(images_); i++) {
+    if (images_[i] != 0) {
+      bound_slots |= 1ULL << i;
+    }
+  }
+  return bound_slots;
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Memory barrier
+ * \{ */
+
+void GLStateManager::issue_barrier(eGPUBarrier barrier_bits)
+{
+  glMemoryBarrier(to_gl(barrier_bits));
+}
+
+/** \} */
+
 }  // namespace blender::gpu
-- 
cgit v1.2.3