mesa/src/vulkan/runtime/vk_meta_draw_rects.c

/*
 * Copyright © 2022 Collabora Ltd
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "vk_meta_private.h"

#include "vk_command_buffer.h"
#include "vk_command_pool.h"
#include "vk_device.h"

#include "nir_builder.h"

const VkPipelineVertexInputStateCreateInfo vk_meta_draw_rects_vi_state = {
   .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
   .vertexBindingDescriptionCount = 1,
   .pVertexBindingDescriptions = &(const VkVertexInputBindingDescription) {
      .binding = 0,
      .stride = 4 * sizeof(uint32_t),
      .inputRate = VK_VERTEX_INPUT_RATE_VERTEX,
   },
   .vertexAttributeDescriptionCount = 1,
   .pVertexAttributeDescriptions = &(const VkVertexInputAttributeDescription) {
      .location = 0,
      .binding = 0,
      .format = VK_FORMAT_R32G32B32A32_UINT,
      .offset = 0,
   },
};

const VkPipelineInputAssemblyStateCreateInfo vk_meta_draw_rects_ia_state = {
   .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
   .topology = VK_PRIMITIVE_TOPOLOGY_META_RECT_LIST_MESA,
   .primitiveRestartEnable = VK_FALSE,
};

const VkPipelineViewportStateCreateInfo vk_meta_draw_rects_vs_state = {
   .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
   .viewportCount = 1,
   .scissorCount = 1,
};

nir_shader *
vk_meta_draw_rects_vs_nir(struct vk_meta_device *device, bool use_gs)
{
   nir_builder build = nir_builder_init_simple_shader(MESA_SHADER_VERTEX, NULL,
                                                      "vk-meta-draw-rects-vs");
   nir_builder *b = &build;

   nir_variable *in = nir_variable_create(b->shader, nir_var_shader_in,
                                          glsl_uvec4_type(), "vtx_in");
   in->data.location = VERT_ATTRIB_GENERIC0;

   nir_variable *pos =
      nir_variable_create(b->shader, nir_var_shader_out, glsl_vec4_type(),
                          use_gs ? "pos_out" : "gl_Position");
   pos->data.location = use_gs ? VARYING_SLOT_VAR0 : VARYING_SLOT_POS;

   nir_variable *layer =
      nir_variable_create(b->shader, nir_var_shader_out, glsl_int_type(),
                          use_gs ? "layer_out" : "gl_Layer");
   layer->data.location = use_gs ? VARYING_SLOT_VAR1 : VARYING_SLOT_LAYER;

   nir_def *vtx = nir_load_var(b, in);
   nir_store_var(b, pos, nir_vec4(b, nir_channel(b, vtx, 0),
                                     nir_channel(b, vtx, 1),
                                     nir_channel(b, vtx, 2),
                                     nir_imm_float(b, 1)),
                 0xf);

   nir_store_var(b, layer, nir_iadd(b, nir_load_instance_id(b),
                                       nir_channel(b, vtx, 3)),
                 0x1);

   return b->shader;
}

nir_shader *
vk_meta_draw_rects_gs_nir(struct vk_meta_device *device)
{
   nir_builder build =
      nir_builder_init_simple_shader(MESA_SHADER_GEOMETRY, NULL,
                                     "vk-meta-draw-rects-gs");
   nir_builder *b = &build;

   nir_variable *pos_in =
      nir_variable_create(b->shader, nir_var_shader_in,
                          glsl_array_type(glsl_vec4_type(), 3, 0), "pos_in");
   pos_in->data.location = VARYING_SLOT_VAR0;

   nir_variable *layer_in =
      nir_variable_create(b->shader, nir_var_shader_in,
                          glsl_array_type(glsl_int_type(), 3, 0), "layer_in");
   layer_in->data.location = VARYING_SLOT_VAR1;

   nir_variable *pos_out =
      nir_variable_create(b->shader, nir_var_shader_out,
                          glsl_vec4_type(), "gl_Position");
   pos_out->data.location = VARYING_SLOT_POS;

   nir_variable *layer_out =
      nir_variable_create(b->shader, nir_var_shader_out,
                          glsl_int_type(), "gl_Layer");
   layer_out->data.location = VARYING_SLOT_LAYER;

   for (unsigned i = 0; i < 3; i++) {
      nir_deref_instr *pos_in_deref =
         nir_build_deref_array_imm(b, nir_build_deref_var(b, pos_in), i);
      nir_deref_instr *layer_in_deref =
         nir_build_deref_array_imm(b, nir_build_deref_var(b, layer_in), i);

      nir_store_var(b, pos_out, nir_load_deref(b, pos_in_deref), 0xf);
      nir_store_var(b, layer_out, nir_load_deref(b, layer_in_deref), 1);
      nir_emit_vertex(b);
   }

   nir_end_primitive(b);

   struct shader_info *info = &build.shader->info;
   info->gs.input_primitive = MESA_PRIM_TRIANGLES;
   info->gs.output_primitive = MESA_PRIM_TRIANGLE_STRIP;
   info->gs.vertices_in = 3;
   info->gs.vertices_out = 3;
   info->gs.invocations = 1;
   info->gs.active_stream_mask = 1;

   return b->shader;
}

struct vertex {
   float x, y, z;
   uint32_t layer;
};

static void
setup_viewport_scissor(struct vk_command_buffer *cmd,
                       uint32_t rect_count,
                       const struct vk_meta_rect *rects,
                       float *x_scale, float *y_scale)
{
   const struct vk_device_dispatch_table *disp =
      &cmd->base.device->dispatch_table;
   VkCommandBuffer _cmd = vk_command_buffer_to_handle(cmd);

   assert(rects[0].x0 < rects[0].x1 && rects[0].y0 < rects[0].y1);
   uint32_t xbits = rects[0].x1 - 1, ybits = rects[0].y1 - 1;
   float zmin = rects[0].z, zmax = rects[0].z;
   for (uint32_t r = 1; r < rect_count; r++) {
      assert(rects[r].x0 < rects[r].x1 && rects[r].y0 < rects[r].y1);
      xbits |= rects[r].x1 - 1;
      ybits |= rects[r].y1 - 1;
      zmin = fminf(zmin, rects[r].z);
      zmax = fminf(zmax, rects[r].z);
   }

   /* Annoyingly, we don't actually know the render area.  We assume that all
    * our rects are inside the render area.  We further assume the maximum
    * image and/or viewport size is a power of two.  This means we can round
    * up to a power of two without going outside any maximums.  Using a power
    * of two will ensure we don't lose precision when scaling coordinates.
    */
   int xmax_log2 = 1 + util_logbase2(xbits);
   int ymax_log2 = 1 + util_logbase2(ybits);

   assert(xmax_log2 >= 0 && xmax_log2 <= 31);
   assert(ymax_log2 >= 0 && ymax_log2 <= 31);

   /* We don't care about precise bounds on Z, only that it's inside [0, 1] if
    * the implementation only supports [0, 1].
    */
   if (zmin >= 0.0f && zmax <= 1.0f) {
      zmin = 0.0f;
      zmax = 1.0f;
   }

   VkViewport viewport = {
      .x = 0,
      .y = 0,
      .width = ldexpf(1.0, xmax_log2),
      .height = ldexpf(1.0, ymax_log2),
      .minDepth = zmin,
      .maxDepth = zmax,
   };
   disp->CmdSetViewport(_cmd, 0, 1, &viewport);

   VkRect2D scissor = {
      .offset = { 0, 0 },
      .extent = { 1u << xmax_log2, 1u << ymax_log2 },
   };
   disp->CmdSetScissor(_cmd, 0, 1, &scissor);

   /* Scaling factors */
   *x_scale = ldexpf(2.0, -xmax_log2);
   *y_scale = ldexpf(2.0, -ymax_log2);
}

static const uint32_t rect_vb_size_B = 6 * 4 * sizeof(float);

static VkResult
create_vertex_buffer(struct vk_command_buffer *cmd,
                     struct vk_meta_device *meta,
                     float x_scale, float y_scale,
                     uint32_t rect_count,
                     const struct vk_meta_rect *rects,
                     VkBuffer *buffer_out)
{
   VkResult result;

   const VkBufferCreateInfo vtx_buffer_info = {
      .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
      .size = rect_count * rect_vb_size_B,
      .usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
      .queueFamilyIndexCount = 1,
      .pQueueFamilyIndices = &cmd->pool->queue_family_index,
   };

   result = vk_meta_create_buffer(cmd, meta, &vtx_buffer_info, buffer_out);
   if (unlikely(result != VK_SUCCESS))
      return result;

   void *map;
   result = meta->cmd_bind_map_buffer(cmd, meta, *buffer_out, &map);
   if (unlikely(result != VK_SUCCESS))
      return result;

   for (uint32_t r = 0; r < rect_count; r++) {
      float x0 = rects[r].x0 * x_scale - 1.0f;
      float y0 = rects[r].y0 * y_scale - 1.0f;
      float x1 = rects[r].x1 * x_scale - 1.0f;
      float y1 = rects[r].y1 * y_scale - 1.0f;
      float z = rects[r].z;
      uint32_t w = rects[r].layer;

      struct vertex rect_vb_data[6] = {
         { x0, y1, z, w },
         { x0, y0, z, w },
         { x1, y1, z, w },

         { x1, y0, z, w },
         { x1, y1, z, w },
         { x0, y0, z, w },
      };
      assert(sizeof(rect_vb_data) == rect_vb_size_B);
      memcpy((char *)map + r * rect_vb_size_B, rect_vb_data, rect_vb_size_B);
   }

   return VK_SUCCESS;
}

void
vk_meta_draw_volume(struct vk_command_buffer *cmd,
                    struct vk_meta_device *meta,
                    const struct vk_meta_rect *rect,
                    uint32_t layer_count)
{
   const struct vk_device_dispatch_table *disp =
      &cmd->base.device->dispatch_table;
   VkCommandBuffer _cmd = vk_command_buffer_to_handle(cmd);

   float x_scale, y_scale;
   setup_viewport_scissor(cmd, 1, rect, &x_scale, &y_scale);

   VkBuffer vtx_buffer;
   VkResult result = create_vertex_buffer(cmd, meta, x_scale, y_scale,
                                          1, rect, &vtx_buffer);
   if (unlikely(result != VK_SUCCESS)) {
      /* TODO: Report error */
      return;
   }

   const VkDeviceSize zero = 0;
   disp->CmdBindVertexBuffers(_cmd, 0, 1, &vtx_buffer, &zero);

   disp->CmdDraw(_cmd, 6, layer_count, 0, 0);
}

void
vk_meta_draw_rects(struct vk_command_buffer *cmd,
                   struct vk_meta_device *meta,
                   uint32_t rect_count,
                   const struct vk_meta_rect *rects)
{
   const struct vk_device_dispatch_table *disp =
      &cmd->base.device->dispatch_table;
   VkCommandBuffer _cmd = vk_command_buffer_to_handle(cmd);

   /* Two triangles with VK_FORMAT_R16G16_UINT */
   const uint32_t rect_vb_size_B = 6 * 3 * sizeof(float);
   const uint32_t rects_per_draw =
      meta->max_bind_map_buffer_size_B / rect_vb_size_B;

   if (rect_count == 0)
      return;

   float x_scale, y_scale;
   setup_viewport_scissor(cmd, rect_count, rects, &x_scale, &y_scale);

   uint32_t next_rect = 0;
   while (next_rect < rect_count) {
      const uint32_t count = MIN2(rects_per_draw, rect_count - next_rect);

      VkBuffer vtx_buffer;
      VkResult result = create_vertex_buffer(cmd, meta, x_scale, y_scale,
                                             count, &rects[next_rect],
                                             &vtx_buffer);
      if (unlikely(result != VK_SUCCESS)) {
         /* TODO: Report error */
         return;
      }

      const VkDeviceSize zero = 0;
      disp->CmdBindVertexBuffers(_cmd, 0, 1, &vtx_buffer, &zero);

      disp->CmdDraw(_cmd, 6 * count, 1, 0, 0);

      next_rect += count;
   }
   assert(next_rect == rect_count);
}