DRWManager: New implementation.

This is a new implementation of the draw manager using modern
rendering practices and GPU driven culling.

This only ports features that are not considered deprecated or to be
removed.

The old DRW API is kept working along side this new one, and does not
interfeer with it. However this needed some more hacking inside the
draw_view_lib.glsl. At least the create info are well separated.

The reviewer might start by looking at `draw_pass_test.cc` to see the
API in usage.

Important files are `draw_pass.hh`, `draw_command.hh`,
`draw_command_shared.hh`.

In a nutshell (for a developper used to old DRW API):
- `DRWShadingGroups` are replaced by `Pass<T>::Sub`.
- Contrary to DRWShadingGroups, all commands recorded inside a pass or
   sub-pass (even binds / push_constant / uniforms) will be executed in order.
- All memory is managed per object (except for Sub-Pass which are managed
   by their parent pass) and not from draw manager pools. So passes "can"
   potentially be recorded once and submitted multiple time (but this is
   not really encouraged for now). The only implicit link is between resource
   lifetime and `ResourceHandles`
- Sub passes can be any level deep.
- IMPORTANT: All state propagate from sub pass to subpass. There is no
   state stack concept anymore. Ensure the correct render state is set before
   drawing anything using `Pass::state_set()`.
- The drawcalls now needs a `ResourceHandle` instead of an `Object *`.
   This is to remove any implicit dependency between `Pass` and `Manager`.
   This was a huge problem in old implementation since the manager did not
   know what to pull from the object. Now it is explicitly requested by the
   engine.
- The pases need to be submitted to a `draw::Manager` instance which can
   be retrieved using `DRW_manager_get()` (for now).

Internally:
- All object data are stored in contiguous storage buffers. Removing a lot
   of complexity in the pass submission.
- Draw calls are sorted and visibility tested on GPU. Making more modern
   culling and better instancing usage possible in the future.
- Unit Tests have been added for regression testing and avoid most API
   breakage.
- `draw::View` now contains culling data for all objects in the scene
   allowing caching for multiple views.
- Bounding box and sphere final setup is moved to GPU.
- Some global resources locations have been hardcoded to reduce complexity.

What is missing:
- ~~Workaround for lack of gl_BaseInstanceARB.~~ Done
- ~~Object Uniform Attributes.~~ Done (Not in this patch)
- Workaround for hardware supporting a maximum of 8 SSBO.

Reviewed By: jbakker

Differential Revision: https://developer.blender.org/D15817

1 files changed, 600 insertions, 0 deletions

diff --git a/source/blender/draw/intern/draw_command.cc b/source/blender/draw/intern/draw_command.cc
new file mode 100644
index 00000000000..7d5ea5c2048
--- /dev/null
+++ b/source/blender/draw/intern/draw_command.cc
@@ -0,0 +1,600 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later
+ * Copyright 2022 Blender Foundation. */
+
+/** \file
+ * \ingroup draw
+ */
+
+#include "GPU_batch.h"
+#include "GPU_capabilities.h"
+#include "GPU_compute.h"
+#include "GPU_debug.h"
+
+#include "draw_command.hh"
+#include "draw_shader.h"
+#include "draw_view.hh"
+
+#include <bitset>
+#include <sstream>
+
+namespace blender::draw::command {
+
+/* -------------------------------------------------------------------- */
+/** \name Commands Execution
+ * \{ */
+
+void ShaderBind::execute(RecordingState &state) const
+{
+  if (assign_if_different(state.shader, shader)) {
+    GPU_shader_bind(shader);
+  }
+}
+
+void ResourceBind::execute() const
+{
+  if (slot == -1) {
+    return;
+  }
+  switch (type) {
+    case ResourceBind::Type::Sampler:
+      GPU_texture_bind_ex(is_reference ? *texture_ref : texture, sampler, slot, false);
+      break;
+    case ResourceBind::Type::Image:
+      GPU_texture_image_bind(is_reference ? *texture_ref : texture, slot);
+      break;
+    case ResourceBind::Type::UniformBuf:
+      GPU_uniformbuf_bind(is_reference ? *uniform_buf_ref : uniform_buf, slot);
+      break;
+    case ResourceBind::Type::StorageBuf:
+      GPU_storagebuf_bind(is_reference ? *storage_buf_ref : storage_buf, slot);
+      break;
+  }
+}
+
+void PushConstant::execute(RecordingState &state) const
+{
+  if (location == -1) {
+    return;
+  }
+  switch (type) {
+    case PushConstant::Type::IntValue:
+      GPU_shader_uniform_vector_int(state.shader, location, comp_len, array_len, int4_value);
+      break;
+    case PushConstant::Type::IntReference:
+      GPU_shader_uniform_vector_int(state.shader, location, comp_len, array_len, int_ref);
+      break;
+    case PushConstant::Type::FloatValue:
+      GPU_shader_uniform_vector(state.shader, location, comp_len, array_len, float4_value);
+      break;
+    case PushConstant::Type::FloatReference:
+      GPU_shader_uniform_vector(state.shader, location, comp_len, array_len, float_ref);
+      break;
+  }
+}
+
+void Draw::execute(RecordingState &state) const
+{
+  state.front_facing_set(handle.has_inverted_handedness());
+
+  if (GPU_shader_draw_parameters_support() == false) {
+    GPU_batch_resource_id_buf_set(batch, state.resource_id_buf);
+  }
+
+  GPU_batch_set_shader(batch, state.shader);
+  GPU_batch_draw_advanced(batch, vertex_first, vertex_len, 0, instance_len);
+}
+
+void DrawMulti::execute(RecordingState &state) const
+{
+  DrawMultiBuf::DrawCommandBuf &indirect_buf = multi_draw_buf->command_buf_;
+  DrawMultiBuf::DrawGroupBuf &groups = multi_draw_buf->group_buf_;
+
+  uint group_index = this->group_first;
+  while (group_index != (uint)-1) {
+    const DrawGroup &group = groups[group_index];
+
+    if (group.vertex_len > 0) {
+      if (GPU_shader_draw_parameters_support() == false) {
+        GPU_batch_resource_id_buf_set(group.gpu_batch, state.resource_id_buf);
+      }
+
+      GPU_batch_set_shader(group.gpu_batch, state.shader);
+
+      constexpr intptr_t stride = sizeof(DrawCommand);
+      /* We have 2 indirect command reserved per draw group. */
+      intptr_t offset = stride * group_index * 2;
+
+      /* Draw negatively scaled geometry first. */
+      if (group.len - group.front_facing_len > 0) {
+        state.front_facing_set(true);
+        GPU_batch_draw_indirect(group.gpu_batch, indirect_buf, offset);
+      }
+
+      if (group.front_facing_len > 0) {
+        state.front_facing_set(false);
+        GPU_batch_draw_indirect(group.gpu_batch, indirect_buf, offset + stride);
+      }
+    }
+
+    group_index = group.next;
+  }
+}
+
+void DrawIndirect::execute(RecordingState &state) const
+{
+  state.front_facing_set(handle.has_inverted_handedness());
+
+  GPU_batch_draw_indirect(batch, *indirect_buf, 0);
+}
+
+void Dispatch::execute(RecordingState &state) const
+{
+  if (is_reference) {
+    GPU_compute_dispatch(state.shader, size_ref->x, size_ref->y, size_ref->z);
+  }
+  else {
+    GPU_compute_dispatch(state.shader, size.x, size.y, size.z);
+  }
+}
+
+void DispatchIndirect::execute(RecordingState &state) const
+{
+  GPU_compute_dispatch_indirect(state.shader, *indirect_buf);
+}
+
+void Barrier::execute() const
+{
+  GPU_memory_barrier(type);
+}
+
+void Clear::execute() const
+{
+  GPUFrameBuffer *fb = GPU_framebuffer_active_get();
+  GPU_framebuffer_clear(fb, (eGPUFrameBufferBits)clear_channels, color, depth, stencil);
+}
+
+void StateSet::execute(RecordingState &recording_state) const
+{
+  /**
+   * Does not support locked state for the moment and never should.
+   * Better implement a less hacky selection!
+   */
+  BLI_assert(DST.state_lock == 0);
+
+  if (!assign_if_different(recording_state.pipeline_state, new_state)) {
+    return;
+  }
+
+  /* Keep old API working. Keep the state tracking in sync. */
+  /* TODO(fclem): Move at the end of a pass. */
+  DST.state = new_state;
+
+  GPU_state_set(to_write_mask(new_state),
+                to_blend(new_state),
+                to_face_cull_test(new_state),
+                to_depth_test(new_state),
+                to_stencil_test(new_state),
+                to_stencil_op(new_state),
+                to_provoking_vertex(new_state));
+
+  if (new_state & DRW_STATE_SHADOW_OFFSET) {
+    GPU_shadow_offset(true);
+  }
+  else {
+    GPU_shadow_offset(false);
+  }
+
+  /* TODO: this should be part of shader state. */
+  if (new_state & DRW_STATE_CLIP_PLANES) {
+    GPU_clip_distances(recording_state.view_clip_plane_count);
+  }
+  else {
+    GPU_clip_distances(0);
+  }
+
+  if (new_state & DRW_STATE_IN_FRONT_SELECT) {
+    /* XXX `GPU_depth_range` is not a perfect solution
+     * since very distant geometries can still be occluded.
+     * Also the depth test precision of these geometries is impaired.
+     * However, it solves the selection for the vast majority of cases. */
+    GPU_depth_range(0.0f, 0.01f);
+  }
+  else {
+    GPU_depth_range(0.0f, 1.0f);
+  }
+
+  if (new_state & DRW_STATE_PROGRAM_POINT_SIZE) {
+    GPU_program_point_size(true);
+  }
+  else {
+    GPU_program_point_size(false);
+  }
+}
+
+void StencilSet::execute() const
+{
+  GPU_stencil_write_mask_set(write_mask);
+  GPU_stencil_compare_mask_set(compare_mask);
+  GPU_stencil_reference_set(reference);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Commands Serialization for debugging
+ * \{ */
+
+std::string ShaderBind::serialize() const
+{
+  return std::string(".shader_bind(") + GPU_shader_get_name(shader) + ")";
+}
+
+std::string ResourceBind::serialize() const
+{
+  switch (type) {
+    case Type::Sampler:
+      return std::string(".bind_texture") + (is_reference ? "_ref" : "") + "(" +
+             std::to_string(slot) +
+             (sampler != GPU_SAMPLER_MAX ? ", sampler=" + std::to_string(sampler) : "") + ")";
+    case Type::Image:
+      return std::string(".bind_image") + (is_reference ? "_ref" : "") + "(" +
+             std::to_string(slot) + ")";
+    case Type::UniformBuf:
+      return std::string(".bind_uniform_buf") + (is_reference ? "_ref" : "") + "(" +
+             std::to_string(slot) + ")";
+    case Type::StorageBuf:
+      return std::string(".bind_storage_buf") + (is_reference ? "_ref" : "") + "(" +
+             std::to_string(slot) + ")";
+    default:
+      BLI_assert_unreachable();
+      return "";
+  }
+}
+
+std::string PushConstant::serialize() const
+{
+  std::stringstream ss;
+  for (int i = 0; i < array_len; i++) {
+    switch (comp_len) {
+      case 1:
+        switch (type) {
+          case Type::IntValue:
+            ss << int1_value;
+            break;
+          case Type::IntReference:
+            ss << int_ref[i];
+            break;
+          case Type::FloatValue:
+            ss << float1_value;
+            break;
+          case Type::FloatReference:
+            ss << float_ref[i];
+            break;
+        }
+        break;
+      case 2:
+        switch (type) {
+          case Type::IntValue:
+            ss << int2_value;
+            break;
+          case Type::IntReference:
+            ss << int2_ref[i];
+            break;
+          case Type::FloatValue:
+            ss << float2_value;
+            break;
+          case Type::FloatReference:
+            ss << float2_ref[i];
+            break;
+        }
+        break;
+      case 3:
+        switch (type) {
+          case Type::IntValue:
+            ss << int3_value;
+            break;
+          case Type::IntReference:
+            ss << int3_ref[i];
+            break;
+          case Type::FloatValue:
+            ss << float3_value;
+            break;
+          case Type::FloatReference:
+            ss << float3_ref[i];
+            break;
+        }
+        break;
+      case 4:
+        switch (type) {
+          case Type::IntValue:
+            ss << int4_value;
+            break;
+          case Type::IntReference:
+            ss << int4_ref[i];
+            break;
+          case Type::FloatValue:
+            ss << float4_value;
+            break;
+          case Type::FloatReference:
+            ss << float4_ref[i];
+            break;
+        }
+        break;
+      case 16:
+        switch (type) {
+          case Type::IntValue:
+          case Type::IntReference:
+            BLI_assert_unreachable();
+            break;
+          case Type::FloatValue:
+            ss << *reinterpret_cast<const float4x4 *>(&float4_value);
+            break;
+          case Type::FloatReference:
+            ss << *float4x4_ref;
+            break;
+        }
+        break;
+    }
+    if (i < array_len - 1) {
+      ss << ", ";
+    }
+  }
+
+  return std::string(".push_constant(") + std::to_string(location) + ", data=" + ss.str() + ")";
+}
+
+std::string Draw::serialize() const
+{
+  std::string inst_len = (instance_len == (uint)-1) ? "from_batch" : std::to_string(instance_len);
+  std::string vert_len = (vertex_len == (uint)-1) ? "from_batch" : std::to_string(vertex_len);
+  std::string vert_first = (vertex_first == (uint)-1) ? "from_batch" :
+                                                        std::to_string(vertex_first);
+  return std::string(".draw(inst_len=") + inst_len + ", vert_len=" + vert_len +
+         ", vert_first=" + vert_first + ", res_id=" + std::to_string(handle.resource_index()) +
+         ")";
+}
+
+std::string DrawMulti::serialize(std::string line_prefix) const
+{
+  DrawMultiBuf::DrawGroupBuf &groups = multi_draw_buf->group_buf_;
+
+  MutableSpan<DrawPrototype> prototypes(multi_draw_buf->prototype_buf_.data(),
+                                        multi_draw_buf->prototype_count_);
+
+  /* This emulates the GPU sorting but without the unstable draw order. */
+  std::sort(
+      prototypes.begin(), prototypes.end(), [](const DrawPrototype &a, const DrawPrototype &b) {
+        return (a.group_id < b.group_id) ||
+               (a.group_id == b.group_id && a.resource_handle > b.resource_handle);
+      });
+
+  /* Compute prefix sum to have correct offsets. */
+  uint prefix_sum = 0u;
+  for (DrawGroup &group : groups) {
+    group.start = prefix_sum;
+    prefix_sum += group.front_proto_len + group.back_proto_len;
+  }
+
+  std::stringstream ss;
+
+  uint group_len = 0;
+  uint group_index = this->group_first;
+  while (group_index != (uint)-1) {
+    const DrawGroup &grp = groups[group_index];
+
+    ss << std::endl << line_prefix << "  .group(id=" << group_index << ", len=" << grp.len << ")";
+
+    intptr_t offset = grp.start;
+
+    if (grp.back_proto_len > 0) {
+      for (DrawPrototype &proto : prototypes.slice({offset, grp.back_proto_len})) {
+        BLI_assert(proto.group_id == group_index);
+        ResourceHandle handle(proto.resource_handle);
+        BLI_assert(handle.has_inverted_handedness());
+        ss << std::endl
+           << line_prefix << "    .proto(instance_len=" << std::to_string(proto.instance_len)
+           << ", resource_id=" << std::to_string(handle.resource_index()) << ", back_face)";
+      }
+      offset += grp.back_proto_len;
+    }
+
+    if (grp.front_proto_len > 0) {
+      for (DrawPrototype &proto : prototypes.slice({offset, grp.front_proto_len})) {
+        BLI_assert(proto.group_id == group_index);
+        ResourceHandle handle(proto.resource_handle);
+        BLI_assert(!handle.has_inverted_handedness());
+        ss << std::endl
+           << line_prefix << "    .proto(instance_len=" << std::to_string(proto.instance_len)
+           << ", resource_id=" << std::to_string(handle.resource_index()) << ", front_face)";
+      }
+    }
+
+    group_index = grp.next;
+    group_len++;
+  }
+
+  ss << std::endl;
+
+  return line_prefix + ".draw_multi(" + std::to_string(group_len) + ")" + ss.str();
+}
+
+std::string DrawIndirect::serialize() const
+{
+  return std::string(".draw_indirect()");
+}
+
+std::string Dispatch::serialize() const
+{
+  int3 sz = is_reference ? *size_ref : size;
+  return std::string(".dispatch") + (is_reference ? "_ref" : "") + "(" + std::to_string(sz.x) +
+         ", " + std::to_string(sz.y) + ", " + std::to_string(sz.z) + ")";
+}
+
+std::string DispatchIndirect::serialize() const
+{
+  return std::string(".dispatch_indirect()");
+}
+
+std::string Barrier::serialize() const
+{
+  /* TOOD(fclem): Better serialization... */
+  return std::string(".barrier(") + std::to_string(type) + ")";
+}
+
+std::string Clear::serialize() const
+{
+  std::stringstream ss;
+  if (eGPUFrameBufferBits(clear_channels) & GPU_COLOR_BIT) {
+    ss << "color=" << color;
+    if (eGPUFrameBufferBits(clear_channels) & (GPU_DEPTH_BIT | GPU_STENCIL_BIT)) {
+      ss << ", ";
+    }
+  }
+  if (eGPUFrameBufferBits(clear_channels) & GPU_DEPTH_BIT) {
+    ss << "depth=" << depth;
+    if (eGPUFrameBufferBits(clear_channels) & GPU_STENCIL_BIT) {
+      ss << ", ";
+    }
+  }
+  if (eGPUFrameBufferBits(clear_channels) & GPU_STENCIL_BIT) {
+    ss << "stencil=0b" << std::bitset<8>(stencil) << ")";
+  }
+  return std::string(".clear(") + ss.str() + ")";
+}
+
+std::string StateSet::serialize() const
+{
+  /* TOOD(fclem): Better serialization... */
+  return std::string(".state_set(") + std::to_string(new_state) + ")";
+}
+
+std::string StencilSet::serialize() const
+{
+  std::stringstream ss;
+  ss << ".stencil_set(write_mask=0b" << std::bitset<8>(write_mask) << ", compare_mask=0b"
+     << std::bitset<8>(compare_mask) << ", reference=0b" << std::bitset<8>(reference);
+  return ss.str();
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Commands buffers binding / command / resource ID generation
+ * \{ */
+
+void DrawCommandBuf::bind(RecordingState &state,
+                          Vector<Header, 0> &headers,
+                          Vector<Undetermined, 0> &commands)
+{
+  UNUSED_VARS(headers, commands);
+
+  resource_id_count_ = 0;
+
+  for (const Header &header : headers) {
+    if (header.type != Type::Draw) {
+      continue;
+    }
+
+    Draw &cmd = commands[header.index].draw;
+
+    int batch_vert_len, batch_vert_first, batch_base_index, batch_inst_len;
+    /* Now that GPUBatches are guaranteed to be finished, extract their parameters. */
+    GPU_batch_draw_parameter_get(
+        cmd.batch, &batch_vert_len, &batch_vert_first, &batch_base_index, &batch_inst_len);
+    /* Instancing attributes are not supported using the new pipeline since we use the base
+     * instance to set the correct resource_id. Workaround is a storage_buf + gl_InstanceID. */
+    BLI_assert(batch_inst_len == 1);
+
+    if (cmd.vertex_len == (uint)-1) {
+      cmd.vertex_len = batch_vert_len;
+    }
+
+    if (cmd.handle.raw > 0) {
+      /* Save correct offset to start of resource_id buffer region for this draw. */
+      uint instance_first = resource_id_count_;
+      resource_id_count_ += cmd.instance_len;
+      /* Ensure the buffer is big enough. */
+      resource_id_buf_.get_or_resize(resource_id_count_ - 1);
+
+      /* Copy the resource id for all instances. */
+      uint index = cmd.handle.resource_index();
+      for (int i = instance_first; i < (instance_first + cmd.instance_len); i++) {
+        resource_id_buf_[i] = index;
+      }
+    }
+  }
+
+  resource_id_buf_.push_update();
+
+  if (GPU_shader_draw_parameters_support() == false) {
+    state.resource_id_buf = resource_id_buf_;
+  }
+  else {
+    GPU_storagebuf_bind(resource_id_buf_, DRW_RESOURCE_ID_SLOT);
+  }
+}
+
+void DrawMultiBuf::bind(RecordingState &state,
+                        Vector<Header, 0> &headers,
+                        Vector<Undetermined, 0> &commands,
+                        VisibilityBuf &visibility_buf)
+{
+  UNUSED_VARS(headers, commands);
+
+  GPU_debug_group_begin("DrawMultiBuf.bind");
+
+  resource_id_count_ = 0u;
+  for (DrawGroup &group : MutableSpan<DrawGroup>(group_buf_.data(), group_count_)) {
+    /* Compute prefix sum of all instance of previous group. */
+    group.start = resource_id_count_;
+    resource_id_count_ += group.len;
+
+    int batch_inst_len;
+    /* Now that GPUBatches are guaranteed to be finished, extract their parameters. */
+    GPU_batch_draw_parameter_get(group.gpu_batch,
+                                 &group.vertex_len,
+                                 &group.vertex_first,
+                                 &group.base_index,
+                                 &batch_inst_len);
+
+    /* Instancing attributes are not supported using the new pipeline since we use the base
+     * instance to set the correct resource_id. Workaround is a storage_buf + gl_InstanceID. */
+    BLI_assert(batch_inst_len == 1);
+    UNUSED_VARS_NDEBUG(batch_inst_len);
+
+    /* Now that we got the batch infos, we can set the counters to 0. */
+    group.total_counter = group.front_facing_counter = group.back_facing_counter = 0;
+  }
+
+  group_buf_.push_update();
+  prototype_buf_.push_update();
+  /* Allocate enough for the expansion pass. */
+  resource_id_buf_.get_or_resize(resource_id_count_);
+  /* Two command per group. */
+  command_buf_.get_or_resize(group_count_ * 2);
+
+  if (prototype_count_ > 0) {
+    GPUShader *shader = DRW_shader_draw_command_generate_get();
+    GPU_shader_bind(shader);
+    GPU_shader_uniform_1i(shader, "prototype_len", prototype_count_);
+    GPU_storagebuf_bind(group_buf_, GPU_shader_get_ssbo(shader, "group_buf"));
+    GPU_storagebuf_bind(visibility_buf, GPU_shader_get_ssbo(shader, "visibility_buf"));
+    GPU_storagebuf_bind(prototype_buf_, GPU_shader_get_ssbo(shader, "prototype_buf"));
+    GPU_storagebuf_bind(command_buf_, GPU_shader_get_ssbo(shader, "command_buf"));
+    GPU_storagebuf_bind(resource_id_buf_, DRW_RESOURCE_ID_SLOT);
+    GPU_compute_dispatch(shader, divide_ceil_u(prototype_count_, DRW_COMMAND_GROUP_SIZE), 1, 1);
+    if (GPU_shader_draw_parameters_support() == false) {
+      GPU_memory_barrier(GPU_BARRIER_VERTEX_ATTRIB_ARRAY);
+      state.resource_id_buf = resource_id_buf_;
+    }
+    else {
+      GPU_memory_barrier(GPU_BARRIER_SHADER_STORAGE);
+    }
+  }
+
+  GPU_debug_group_end();
+}
+
+/** \} */
+
+};  // namespace blender::draw::command