kernel4.comp raw

   1  // SPDX-License-Identifier: Apache-2.0 OR MIT OR Unlicense
   2  
   3  // This is "kernel 4" in a 4-kernel pipeline. It renders the commands
   4  // in the per-tile command list to an image.
   5  
   6  // Right now, this kernel stores the image in a buffer, but a better
   7  // plan is to use a texture. This is because of limited support.
   8  
   9  #version 450
  10  #extension GL_GOOGLE_include_directive : enable
  11  #ifdef ENABLE_IMAGE_INDICES
  12  #extension GL_EXT_nonuniform_qualifier : enable
  13  #endif
  14  
  15  #include "mem.h"
  16  #include "setup.h"
  17  
  18  #define CHUNK_X 2
  19  #define CHUNK_Y 4
  20  #define CHUNK CHUNK_X * CHUNK_Y
  21  #define CHUNK_DX (TILE_WIDTH_PX / CHUNK_X)
  22  #define CHUNK_DY (TILE_HEIGHT_PX / CHUNK_Y)
  23  layout(local_size_x = CHUNK_DX, local_size_y = CHUNK_DY) in;
  24  
  25  layout(set = 0, binding = 1) restrict readonly buffer ConfigBuf {
  26      Config conf;
  27  };
  28  
  29  layout(rgba8, set = 0, binding = 2) uniform restrict writeonly image2D image;
  30  
  31  #ifdef ENABLE_IMAGE_INDICES
  32  layout(rgba8, set = 0, binding = 3) uniform restrict readonly image2D images[];
  33  #else
  34  layout(rgba8, set = 0, binding = 3) uniform restrict readonly image2D images[1];
  35  #endif
  36  
  37  #include "ptcl.h"
  38  #include "tile.h"
  39  
  40  mediump vec3 tosRGB(mediump vec3 rgb) {
  41      bvec3 cutoff = greaterThanEqual(rgb, vec3(0.0031308));
  42      mediump vec3 below = vec3(12.92)*rgb;
  43      mediump vec3 above = vec3(1.055)*pow(rgb, vec3(0.41666)) - vec3(0.055);
  44      return mix(below, above, cutoff);
  45  }
  46  
  47  mediump vec3 fromsRGB(mediump vec3 srgb) {
  48      // Formula from EXT_sRGB.
  49      bvec3 cutoff = greaterThanEqual(srgb, vec3(0.04045));
  50      mediump vec3 below = srgb/vec3(12.92);
  51      mediump vec3 above = pow((srgb + vec3(0.055))/vec3(1.055), vec3(2.4));
  52      return mix(below, above, cutoff);
  53  }
  54  
  55  // unpacksRGB unpacks a color in the sRGB color space to a vec4 in the linear color
  56  // space.
  57  mediump vec4 unpacksRGB(uint srgba) {
  58      mediump vec4 color = unpackUnorm4x8(srgba).wzyx;
  59      return vec4(fromsRGB(color.rgb), color.a);
  60  }
  61  
  62  // packsRGB packs a color in the linear color space into its 8-bit sRGB equivalent.
  63  uint packsRGB(mediump vec4 rgba) {
  64      rgba = vec4(tosRGB(rgba.rgb), rgba.a);
  65      return packUnorm4x8(rgba.wzyx);
  66  }
  67  
  68  uvec2 chunk_offset(uint i) {
  69      return uvec2(i % CHUNK_X * CHUNK_DX, i / CHUNK_X * CHUNK_DY);
  70  }
  71  
  72  mediump vec4[CHUNK] fillImage(uvec2 xy, CmdImage cmd_img) {
  73      mediump vec4 rgba[CHUNK];
  74      for (uint i = 0; i < CHUNK; i++) {
  75          ivec2 uv = ivec2(xy + chunk_offset(i)) + cmd_img.offset;
  76          mediump vec4 fg_rgba;
  77  #ifdef ENABLE_IMAGE_INDICES
  78          fg_rgba = imageLoad(images[cmd_img.index], uv);
  79  #else
  80          fg_rgba = imageLoad(images[0], uv);
  81  #endif
  82          fg_rgba.rgb = fromsRGB(fg_rgba.rgb);
  83          rgba[i] = fg_rgba;
  84      }
  85      return rgba;
  86  }
  87  
  88  void main() {
  89      uint tile_ix = gl_WorkGroupID.y * conf.width_in_tiles + gl_WorkGroupID.x;
  90      Alloc cmd_alloc = slice_mem(conf.ptcl_alloc, tile_ix * PTCL_INITIAL_ALLOC, PTCL_INITIAL_ALLOC);
  91      CmdRef cmd_ref = CmdRef(cmd_alloc.offset);
  92  
  93      // Read scrach space allocation, written first in the command list.
  94      Alloc scratch_alloc = alloc_read(cmd_alloc, cmd_ref.offset);
  95      cmd_ref.offset += Alloc_size;
  96  
  97      uvec2 xy_uint = uvec2(gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_WorkGroupID.x, gl_LocalInvocationID.y + TILE_HEIGHT_PX * gl_WorkGroupID.y);
  98      vec2 xy = vec2(xy_uint);
  99      mediump vec4 rgba[CHUNK];
 100      for (uint i = 0; i < CHUNK; i++) {
 101          rgba[i] = vec4(0.0);
 102          // TODO: remove this debug image support when the actual image method is plumbed.
 103  #ifdef DEBUG_IMAGES
 104  #ifdef ENABLE_IMAGE_INDICES
 105          if (xy_uint.x < 1024 && xy_uint.y < 1024) {
 106              rgba[i] = imageLoad(images[gl_WorkGroupID.x / 64], ivec2(xy_uint + chunk_offset(i))/4);
 107          }
 108  #else
 109          if (xy_uint.x < 1024 && xy_uint.y < 1024) {
 110              rgb[i] = imageLoad(images[0], ivec2(xy_uint + chunk_offset(i))/4).rgb;
 111          }
 112  #endif
 113  #endif
 114      }
 115  
 116      mediump float area[CHUNK];
 117      uint clip_depth = 0;
 118      bool mem_ok = mem_error == NO_ERROR;
 119      while (mem_ok) {
 120          uint tag = Cmd_tag(cmd_alloc, cmd_ref).tag;
 121          if (tag == Cmd_End) {
 122              break;
 123          }
 124          switch (tag) {
 125          case Cmd_Stroke:
 126              // Calculate distance field from all the line segments in this tile.
 127              CmdStroke stroke = Cmd_Stroke_read(cmd_alloc, cmd_ref);
 128              mediump float df[CHUNK];
 129              for (uint k = 0; k < CHUNK; k++) df[k] = 1e9;
 130              TileSegRef tile_seg_ref = TileSegRef(stroke.tile_ref);
 131              do {
 132                  TileSeg seg = TileSeg_read(new_alloc(tile_seg_ref.offset, TileSeg_size, mem_ok), tile_seg_ref);
 133                  vec2 line_vec = seg.vector;
 134                  for (uint k = 0; k < CHUNK; k++) {
 135                      vec2 dpos = xy + vec2(0.5, 0.5) - seg.origin;
 136                      dpos += vec2(chunk_offset(k));
 137                      float t = clamp(dot(line_vec, dpos) / dot(line_vec, line_vec), 0.0, 1.0);
 138                      df[k] = min(df[k], length(line_vec * t - dpos));
 139                  }
 140                  tile_seg_ref = seg.next;
 141              } while (tile_seg_ref.offset != 0);
 142              for (uint k = 0; k < CHUNK; k++) {
 143                  area[k] = clamp(stroke.half_width + 0.5 - df[k], 0.0, 1.0);
 144              }
 145              cmd_ref.offset += 4 + CmdStroke_size;
 146              break;
 147          case Cmd_Fill:
 148              CmdFill fill = Cmd_Fill_read(cmd_alloc, cmd_ref);
 149              for (uint k = 0; k < CHUNK; k++) area[k] = float(fill.backdrop);
 150              tile_seg_ref = TileSegRef(fill.tile_ref);
 151              // Calculate coverage based on backdrop + coverage of each line segment
 152              do {
 153                  TileSeg seg = TileSeg_read(new_alloc(tile_seg_ref.offset, TileSeg_size, mem_ok), tile_seg_ref);
 154                  for (uint k = 0; k < CHUNK; k++) {
 155                      vec2 my_xy = xy + vec2(chunk_offset(k));
 156                      vec2 start = seg.origin - my_xy;
 157                      vec2 end = start + seg.vector;
 158                      vec2 window = clamp(vec2(start.y, end.y), 0.0, 1.0);
 159                      if (window.x != window.y) {
 160                          vec2 t = (window - start.y) / seg.vector.y;
 161                          vec2 xs = vec2(mix(start.x, end.x, t.x), mix(start.x, end.x, t.y));
 162                          float xmin = min(min(xs.x, xs.y), 1.0) - 1e-6;
 163                          float xmax = max(xs.x, xs.y);
 164                          float b = min(xmax, 1.0);
 165                          float c = max(b, 0.0);
 166                          float d = max(xmin, 0.0);
 167                          float a = (b + 0.5 * (d * d - c * c) - xmin) / (xmax - xmin);
 168                          area[k] += a * (window.x - window.y);
 169                      }
 170                      area[k] += sign(seg.vector.x) * clamp(my_xy.y - seg.y_edge + 1.0, 0.0, 1.0);
 171                  }
 172                  tile_seg_ref = seg.next;
 173              } while (tile_seg_ref.offset != 0);
 174              for (uint k = 0; k < CHUNK; k++) {
 175                  area[k] = min(abs(area[k]), 1.0);
 176              }
 177              cmd_ref.offset += 4 + CmdFill_size;
 178              break;
 179          case Cmd_Solid:
 180              for (uint k = 0; k < CHUNK; k++) {
 181                  area[k] = 1.0;
 182              }
 183              cmd_ref.offset += 4;
 184              break;
 185          case Cmd_Alpha:
 186              CmdAlpha alpha = Cmd_Alpha_read(cmd_alloc, cmd_ref);
 187              for (uint k = 0; k < CHUNK; k++) {
 188                  area[k] = alpha.alpha;
 189              }
 190              cmd_ref.offset += 4 + CmdAlpha_size;
 191              break;
 192          case Cmd_Color:
 193              CmdColor color = Cmd_Color_read(cmd_alloc, cmd_ref);
 194              mediump vec4 fg = unpacksRGB(color.rgba_color);
 195              for (uint k = 0; k < CHUNK; k++) {
 196                  mediump vec4 fg_k = fg * area[k];
 197                  rgba[k] = rgba[k] * (1.0 - fg_k.a) + fg_k;
 198              }
 199              cmd_ref.offset += 4 + CmdColor_size;
 200              break;
 201          case Cmd_Image:
 202              CmdImage fill_img = Cmd_Image_read(cmd_alloc, cmd_ref);
 203              mediump vec4 img[CHUNK] = fillImage(xy_uint, fill_img);
 204              for (uint k = 0; k < CHUNK; k++) {
 205                  mediump vec4 fg_k = img[k] * area[k];
 206                  rgba[k] = rgba[k] * (1.0 - fg_k.a) + fg_k;
 207              }
 208              cmd_ref.offset += 4 + CmdImage_size;
 209              break;
 210          case Cmd_BeginClip:
 211              uint base_ix = (scratch_alloc.offset >> 2) + CLIP_STATE_SIZE * (clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
 212                  gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y);
 213              for (uint k = 0; k < CHUNK; k++) {
 214                  uvec2 offset = chunk_offset(k);
 215                  uint srgb = packsRGB(vec4(rgba[k]));
 216                  mediump float alpha = clamp(abs(area[k]), 0.0, 1.0);
 217                  write_mem(scratch_alloc, base_ix + 0 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX), srgb);
 218                  write_mem(scratch_alloc, base_ix + 1 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX), floatBitsToUint(alpha));
 219                  rgba[k] = vec4(0.0);
 220              }
 221              clip_depth++;
 222              cmd_ref.offset += 4;
 223              break;
 224          case Cmd_EndClip:
 225              clip_depth--;
 226              base_ix = (scratch_alloc.offset >> 2) + CLIP_STATE_SIZE * (clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
 227                  gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y);
 228              for (uint k = 0; k < CHUNK; k++) {
 229                  uvec2 offset = chunk_offset(k);
 230                  uint srgb = read_mem(scratch_alloc, base_ix + 0 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX));
 231                  uint alpha = read_mem(scratch_alloc, base_ix + 1 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX));
 232                  mediump vec4 bg = unpacksRGB(srgb);
 233                  mediump vec4 fg = rgba[k] * area[k] * uintBitsToFloat(alpha);
 234                  rgba[k] = bg * (1.0 - fg.a) + fg;
 235              }
 236              cmd_ref.offset += 4;
 237              break;
 238          case Cmd_Jump:
 239              cmd_ref = CmdRef(Cmd_Jump_read(cmd_alloc, cmd_ref).new_ref);
 240              cmd_alloc.offset = cmd_ref.offset;
 241              break;
 242          }
 243      }
 244  
 245      for (uint i = 0; i < CHUNK; i++) {
 246          imageStore(image, ivec2(xy_uint + chunk_offset(i)), vec4(tosRGB(rgba[i].rgb), rgba[i].a));
 247      }
 248  }
 249