src/amd/vulkan/radv_nir_lower_ycbcr_textures.c

   1 /*
   2  * Copyright © 2017 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "radv_private.h"
  25 #include "radv_shader.h"
  26 #include "vk_format.h"
  27 #include "nir/nir.h"
  28 #include "nir/nir_builder.h"
  29
  30 struct ycbcr_state {
  31         nir_builder *builder;
  32         nir_ssa_def *image_size;
  33         nir_tex_instr *origin_tex;
  34         nir_deref_instr *tex_deref;
  35         const struct radv_sampler_ycbcr_conversion *conversion;
  36 };
  37
  38 static nir_ssa_def *
  39 y_range(nir_builder *b,
  40         nir_ssa_def *y_channel,
  41         int bpc,
  42         VkSamplerYcbcrRange range)
  43 {
  44         switch (range) {
  45         case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
  46                 return y_channel;
  47         case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
  48                 return nir_fmul(b,
  49                                 nir_fadd(b,
  50                                          nir_fmul(b, y_channel,
  51                                                   nir_imm_float(b, pow(2, bpc) - 1)),
  52                                          nir_imm_float(b, -16.0f * pow(2, bpc - 8))),
  53                                 nir_frcp(b, nir_imm_float(b, 219.0f * pow(2, bpc - 8))));
  54         default:
  55                 unreachable("missing Ycbcr range");
  56                 return NULL;
  57         }
  58 }
  59
  60 static nir_ssa_def *
  61 chroma_range(nir_builder *b,
  62              nir_ssa_def *chroma_channel,
  63              int bpc,
  64              VkSamplerYcbcrRange range)
  65 {
  66         switch (range) {
  67         case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
  68                 return nir_fadd(b, chroma_channel,
  69                                 nir_imm_float(b, -pow(2, bpc - 1) / (pow(2, bpc) - 1.0f)));
  70         case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
  71                 return nir_fmul(b,
  72                                 nir_fadd(b,
  73                                          nir_fmul(b, chroma_channel,
  74                                                   nir_imm_float(b, pow(2, bpc) - 1)),
  75                                          nir_imm_float(b, -128.0f * pow(2, bpc - 8))),
  76                                 nir_frcp(b, nir_imm_float(b, 224.0f * pow(2, bpc - 8))));
  77         default:
  78                 unreachable("missing Ycbcr range");
  79                 return NULL;
  80         }
  81 }
  82
  83 typedef struct nir_const_value_3_4 {
  84         nir_const_value v[3][4];
  85 } nir_const_value_3_4;
  86
  87 static const nir_const_value_3_4 *
  88 ycbcr_model_to_rgb_matrix(VkSamplerYcbcrModelConversion model)
  89 {
  90         switch (model) {
  91         case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601: {
  92                 static const nir_const_value_3_4 bt601 = { {
  93                         { { .f32 =  1.402f             }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
  94                         { { .f32 = -0.714136286201022f }, { .f32 = 1.0f }, { .f32 = -0.344136286201022f }, { .f32 = 0.0f } },
  95                         { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.772f             }, { .f32 = 0.0f } },
  96                 } };
  97
  98                 return &bt601;
  99         }
 100         case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709: {
 101                 static const nir_const_value_3_4 bt709 = { {
 102                         { { .f32 =  1.5748031496063f   }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
 103                         { { .f32 = -0.468125209181067f }, { .f32 = 1.0f }, { .f32 = -0.187327487470334f }, { .f32 = 0.0f } },
 104                         { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.85563184264242f  }, { .f32 = 0.0f } },
 105                 } };
 106
 107                 return &bt709;
 108         }
 109         case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020: {
 110                 static const nir_const_value_3_4 bt2020 = { {
 111                         { { .f32 =  1.4746f            }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
 112                         { { .f32 = -0.571353126843658f }, { .f32 = 1.0f }, { .f32 = -0.164553126843658f }, { .f32 = 0.0f } },
 113                         { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.8814f            }, { .f32 = 0.0f } },
 114                 } };
 115
 116                 return &bt2020;
 117         }
 118         default:
 119                 unreachable("missing Ycbcr model");
 120                 return NULL;
 121         }
 122 }
 123
 124 static nir_ssa_def *
 125 convert_ycbcr(struct ycbcr_state *state,
 126               nir_ssa_def *raw_channels,
 127               uint8_t bits)
 128 {
 129         nir_builder *b = state->builder;
 130         const struct radv_sampler_ycbcr_conversion *conversion = state->conversion;
 131
 132         nir_ssa_def *expanded_channels =
 133                 nir_vec4(b,
 134                          chroma_range(b, nir_channel(b, raw_channels, 0),
 135                                       bits, conversion->ycbcr_range),
 136                          y_range(b, nir_channel(b, raw_channels, 1),
 137                                  bits, conversion->ycbcr_range),
 138                          chroma_range(b, nir_channel(b, raw_channels, 2),
 139                                       bits, conversion->ycbcr_range),
 140                          nir_imm_float(b, 1.0f));
 141
 142         if (conversion->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY)
 143                 return expanded_channels;
 144
 145         const nir_const_value_3_4 *conversion_matrix =
 146                 ycbcr_model_to_rgb_matrix(conversion->ycbcr_model);
 147
 148         nir_ssa_def *converted_channels[] = {
 149                 nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[0])),
 150                 nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[1])),
 151                 nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[2]))
 152         };
 153
 154         return nir_vec4(b,
 155                         converted_channels[0], converted_channels[1],
 156                         converted_channels[2], nir_imm_float(b, 1.0f));
 157 }
 158
 159 static nir_ssa_def *
 160 get_texture_size(struct ycbcr_state *state, nir_deref_instr *texture)
 161 {
 162         nir_builder *b = state->builder;
 163         const struct glsl_type *type = texture->type;
 164         nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
 165
 166         tex->op = nir_texop_txs;
 167         tex->sampler_dim = glsl_get_sampler_dim(type);
 168         tex->is_array = glsl_sampler_type_is_array(type);
 169         tex->is_shadow = glsl_sampler_type_is_shadow(type);
 170         tex->dest_type = nir_type_int;
 171
 172         tex->src[0].src_type = nir_tex_src_texture_deref;
 173         tex->src[0].src = nir_src_for_ssa(&texture->dest.ssa);
 174
 175         nir_ssa_dest_init(&tex->instr, &tex->dest,
 176                           nir_tex_instr_dest_size(tex), 32, NULL);
 177         nir_builder_instr_insert(b, &tex->instr);
 178
 179         return nir_i2f32(b, &tex->dest.ssa);
 180 }
 181
 182 static nir_ssa_def *
 183 implicit_downsampled_coord(nir_builder *b,
 184                            nir_ssa_def *value,
 185                            nir_ssa_def *max_value,
 186                            int div_scale)
 187 {
 188         return nir_fadd(b,
 189                         value,
 190                         nir_fdiv(b,
 191                                  nir_imm_float(b, 1.0f),
 192                                  nir_fmul(b,
 193                                           nir_imm_float(b, div_scale),
 194                                           max_value)));
 195 }
 196
 197 static nir_ssa_def *
 198 implicit_downsampled_coords(struct ycbcr_state *state,
 199                             nir_ssa_def *old_coords)
 200 {
 201         nir_builder *b = state->builder;
 202         const struct radv_sampler_ycbcr_conversion *conversion = state->conversion;
 203         nir_ssa_def *image_size = NULL;
 204         nir_ssa_def *comp[4] = { NULL, };
 205         const struct vk_format_description *fmt_desc = vk_format_description(state->conversion->format);
 206         const unsigned divisors[2] = {fmt_desc->width_divisor, fmt_desc->height_divisor};
 207
 208         for (int c = 0; c < old_coords->num_components; c++) {
 209                 if (c < ARRAY_SIZE(divisors) && divisors[c] > 1 &&
 210                     conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN) {
 211                         if (!image_size)
 212                                 image_size = get_texture_size(state, state->tex_deref);
 213
 214                         comp[c] = implicit_downsampled_coord(b,
 215                                                              nir_channel(b, old_coords, c),
 216                                                              nir_channel(b, image_size, c),
 217                                                              divisors[c]);
 218                 } else {
 219                         comp[c] = nir_channel(b, old_coords, c);
 220                 }
 221         }
 222
 223         return nir_vec(b, comp, old_coords->num_components);
 224 }
 225
 226 static nir_ssa_def *
 227 create_plane_tex_instr_implicit(struct ycbcr_state *state,
 228                                 uint32_t plane)
 229 {
 230         nir_builder *b = state->builder;
 231         nir_tex_instr *old_tex = state->origin_tex;
 232         nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs+ 1);
 233         for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
 234                 tex->src[i].src_type = old_tex->src[i].src_type;
 235
 236                 switch (old_tex->src[i].src_type) {
 237                 case nir_tex_src_coord:
 238                         if (plane && true/*state->conversion->chroma_reconstruction*/) {
 239                                 assert(old_tex->src[i].src.is_ssa);
 240                                 tex->src[i].src =
 241                                         nir_src_for_ssa(implicit_downsampled_coords(state,
 242                                                                                     old_tex->src[i].src.ssa));
 243                                 break;
 244                         }
 245                 /* fall through */
 246                 default:
 247                         nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
 248                         break;
 249                 }
 250         }
 251
 252         tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
 253         tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
 254
 255         tex->sampler_dim = old_tex->sampler_dim;
 256         tex->dest_type = old_tex->dest_type;
 257         tex->is_array = old_tex->is_array;
 258
 259         tex->op = old_tex->op;
 260         tex->coord_components = old_tex->coord_components;
 261         tex->is_new_style_shadow = old_tex->is_new_style_shadow;
 262         tex->component = old_tex->component;
 263
 264         tex->texture_index = old_tex->texture_index;
 265         tex->texture_array_size = old_tex->texture_array_size;
 266         tex->sampler_index = old_tex->sampler_index;
 267
 268         nir_ssa_dest_init(&tex->instr, &tex->dest,
 269                           old_tex->dest.ssa.num_components,
 270                           nir_dest_bit_size(old_tex->dest), NULL);
 271         nir_builder_instr_insert(b, &tex->instr);
 272
 273         return &tex->dest.ssa;
 274 }
 275
 276 struct swizzle_info {
 277         unsigned plane[4];
 278         unsigned swizzle[4];
 279 };
 280
 281 static struct swizzle_info
 282 get_plane_swizzles(VkFormat format)
 283 {
 284         int planes = vk_format_get_plane_count(format);
 285         switch (planes) {
 286         case 3:
 287                 return (struct swizzle_info) {
 288                         {2, 0, 1, 0},
 289                         {0, 0, 0, 3}
 290                 };
 291         case 2:
 292                 return (struct swizzle_info) {
 293                         {1, 0, 1, 0},
 294                         {1, 0, 0, 3}
 295                 };
 296         case 1:
 297                 return (struct swizzle_info) {
 298                         {0, 0, 0, 0},
 299                         {0, 1, 2, 3}
 300                 };
 301         default:
 302                 unreachable("unhandled plane count for ycbcr swizzling");
 303         }
 304 }
 305
 306
 307 static nir_ssa_def *
 308 build_swizzled_components(nir_builder *builder,
 309                           VkFormat format,
 310                           VkComponentMapping mapping,
 311                           nir_ssa_def **plane_values)
 312 {
 313         struct swizzle_info plane_swizzle = get_plane_swizzles(format);
 314         enum vk_swizzle swizzles[4];
 315         nir_ssa_def *values[4];
 316
 317         vk_format_compose_swizzles(&mapping, (const unsigned char[4]){0,1,2,3}, swizzles);
 318
 319         nir_ssa_def *zero = nir_imm_float(builder, 0.0f);
 320         nir_ssa_def *one = nir_imm_float(builder, 1.0f);
 321
 322         for (unsigned i = 0; i < 4; ++i) {
 323                 switch(swizzles[i]) {
 324                 case VK_SWIZZLE_X:
 325                 case VK_SWIZZLE_Y:
 326                 case VK_SWIZZLE_Z:
 327                 case VK_SWIZZLE_W: {
 328                         unsigned channel = swizzles[i] - VK_SWIZZLE_X;
 329                         values[i] = nir_channel(builder,
 330                                                 plane_values[plane_swizzle.plane[channel]],
 331                                                 plane_swizzle.swizzle[channel]);
 332                         break;
 333                 }
 334                 case VK_SWIZZLE_0:
 335                         values[i] = zero;
 336                         break;
 337                 case VK_SWIZZLE_1:
 338                         values[i] = one;
 339                         break;
 340                 default:
 341                         unreachable("unhandled swizzle");
 342                 }
 343         }
 344         return nir_vec(builder, values, 4);
 345 }
 346
 347 static bool
 348 try_lower_tex_ycbcr(const struct radv_pipeline_layout *layout,
 349                     nir_builder *builder,
 350                     nir_tex_instr *tex)
 351 {
 352         int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
 353         assert(deref_src_idx >= 0);
 354         nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
 355
 356         nir_variable *var = nir_deref_instr_get_variable(deref);
 357         const struct radv_descriptor_set_layout *set_layout =
 358                 layout->set[var->data.descriptor_set].layout;
 359         const struct radv_descriptor_set_binding_layout *binding =
 360                 &set_layout->binding[var->data.binding];
 361         const struct radv_sampler_ycbcr_conversion *ycbcr_samplers =
 362                 radv_immutable_ycbcr_samplers(set_layout, var->data.binding);
 363
 364         if (!ycbcr_samplers)
 365                 return false;
 366
 367         /* For the following instructions, we don't apply any change and let the
 368          * instruction apply to the first plane.
 369          */
 370         if (tex->op == nir_texop_txs ||
 371             tex->op == nir_texop_query_levels ||
 372             tex->op == nir_texop_lod)
 373                 return false;
 374
 375         assert(tex->texture_index == 0);
 376         unsigned array_index = 0;
 377         if (deref->deref_type != nir_deref_type_var) {
 378                 assert(deref->deref_type == nir_deref_type_array);
 379                 if (!nir_src_is_const(deref->arr.index))
 380                         return false;
 381                 array_index = nir_src_as_uint(deref->arr.index);
 382                 array_index = MIN2(array_index, binding->array_size - 1);
 383         }
 384         const struct radv_sampler_ycbcr_conversion *ycbcr_sampler = ycbcr_samplers + array_index;
 385
 386         if (ycbcr_sampler->format == VK_FORMAT_UNDEFINED)
 387                 return false;
 388
 389         struct ycbcr_state state = {
 390                 .builder = builder,
 391                 .origin_tex = tex,
 392                 .tex_deref = deref,
 393                 .conversion = ycbcr_sampler,
 394         };
 395
 396         builder->cursor = nir_before_instr(&tex->instr);
 397
 398         VkFormat format = state.conversion->format;
 399         const int plane_count = vk_format_get_plane_count(format);
 400         nir_ssa_def *plane_values[3];
 401
 402         for (int p = 0; p < plane_count; ++p) {
 403                 plane_values[p] = create_plane_tex_instr_implicit(&state, p);
 404         }
 405
 406         nir_ssa_def *result = build_swizzled_components(builder, format, ycbcr_sampler->components, plane_values);
 407         if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) {
 408                 VkFormat first_format = vk_format_get_plane_format(format, 0);
 409                 result = convert_ycbcr(&state, result, vk_format_get_component_bits(first_format, VK_FORMAT_COLORSPACE_RGB, VK_SWIZZLE_X));
 410         }
 411
 412         nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
 413         nir_instr_remove(&tex->instr);
 414
 415         return true;
 416 }
 417
 418 bool
 419 radv_nir_lower_ycbcr_textures(nir_shader *shader,
 420                              const struct radv_pipeline_layout *layout)
 421 {
 422         bool progress = false;
 423
 424         nir_foreach_function(function, shader) {
 425                 if (!function->impl)
 426                         continue;
 427
 428                 bool function_progress = false;
 429                 nir_builder builder;
 430                 nir_builder_init(&builder, function->impl);
 431
 432                 nir_foreach_block(block, function->impl) {
 433                         nir_foreach_instr_safe(instr, block) {
 434                                 if (instr->type != nir_instr_type_tex)
 435                                         continue;
 436
 437                                 nir_tex_instr *tex = nir_instr_as_tex(instr);
 438                                 function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
 439                         }
 440                 }
 441
 442                 if (function_progress) {
 443                         nir_metadata_preserve(function->impl,
 444                                               nir_metadata_block_index |
 445                                               nir_metadata_dominance);
 446                 }
 447
 448                 progress |= function_progress;
 449         }
 450
 451         return progress;
 452 }