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 create_plane_tex_instr_implicit(struct ycbcr_state *state,
 161                                 uint32_t plane)
 162 {
 163         nir_builder *b = state->builder;
 164         nir_tex_instr *old_tex = state->origin_tex;
 165         nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs+ 1);
 166
 167         for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
 168                 tex->src[i].src_type = old_tex->src[i].src_type;
 169                 nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
 170         }
 171
 172         tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
 173         tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
 174
 175         tex->sampler_dim = old_tex->sampler_dim;
 176         tex->dest_type = old_tex->dest_type;
 177         tex->is_array = old_tex->is_array;
 178
 179         tex->op = old_tex->op;
 180         tex->coord_components = old_tex->coord_components;
 181         tex->is_new_style_shadow = old_tex->is_new_style_shadow;
 182         tex->component = old_tex->component;
 183
 184         tex->texture_index = old_tex->texture_index;
 185         tex->texture_array_size = old_tex->texture_array_size;
 186         tex->sampler_index = old_tex->sampler_index;
 187
 188         nir_ssa_dest_init(&tex->instr, &tex->dest,
 189                           old_tex->dest.ssa.num_components,
 190                           nir_dest_bit_size(old_tex->dest), NULL);
 191         nir_builder_instr_insert(b, &tex->instr);
 192
 193         return &tex->dest.ssa;
 194 }
 195
 196 struct swizzle_info {
 197         unsigned plane[4];
 198         unsigned swizzle[4];
 199 };
 200
 201 static struct swizzle_info
 202 get_plane_swizzles(VkFormat format)
 203 {
 204         int planes = vk_format_get_plane_count(format);
 205         switch (planes) {
 206         case 3:
 207                 return (struct swizzle_info) {
 208                         {2, 0, 1, 0},
 209                         {0, 0, 0, 3}
 210                 };
 211         case 2:
 212                 return (struct swizzle_info) {
 213                         {1, 0, 1, 0},
 214                         {1, 0, 0, 3}
 215                 };
 216         case 1:
 217                 return (struct swizzle_info) {
 218                         {0, 0, 0, 0},
 219                         {0, 1, 2, 3}
 220                 };
 221         default:
 222                 unreachable("unhandled plane count for ycbcr swizzling");
 223         }
 224 }
 225
 226
 227 static nir_ssa_def *
 228 build_swizzled_components(nir_builder *builder,
 229                           VkFormat format,
 230                           VkComponentMapping mapping,
 231                           nir_ssa_def **plane_values)
 232 {
 233         struct swizzle_info plane_swizzle = get_plane_swizzles(format);
 234         enum vk_swizzle swizzles[4];
 235         nir_ssa_def *values[4];
 236
 237         vk_format_compose_swizzles(&mapping, (const unsigned char[4]){0,1,2,3}, swizzles);
 238
 239         nir_ssa_def *zero = nir_imm_float(builder, 0.0f);
 240         nir_ssa_def *one = nir_imm_float(builder, 1.0f);
 241
 242         for (unsigned i = 0; i < 4; ++i) {
 243                 switch(swizzles[i]) {
 244                 case VK_SWIZZLE_X:
 245                 case VK_SWIZZLE_Y:
 246                 case VK_SWIZZLE_Z:
 247                 case VK_SWIZZLE_W: {
 248                         unsigned channel = swizzles[i] - VK_SWIZZLE_X;
 249                         values[i] = nir_channel(builder,
 250                                                 plane_values[plane_swizzle.plane[channel]],
 251                                                 plane_swizzle.swizzle[channel]);
 252                         break;
 253                 }
 254                 case VK_SWIZZLE_0:
 255                         values[i] = zero;
 256                         break;
 257                 case VK_SWIZZLE_1:
 258                         values[i] = one;
 259                         break;
 260                 default:
 261                         unreachable("unhandled swizzle");
 262                 }
 263         }
 264         return nir_vec(builder, values, 4);
 265 }
 266
 267 static bool
 268 try_lower_tex_ycbcr(const struct radv_pipeline_layout *layout,
 269                     nir_builder *builder,
 270                     nir_tex_instr *tex)
 271 {
 272         int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
 273         assert(deref_src_idx >= 0);
 274         nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
 275
 276         nir_variable *var = nir_deref_instr_get_variable(deref);
 277         const struct radv_descriptor_set_layout *set_layout =
 278                 layout->set[var->data.descriptor_set].layout;
 279         const struct radv_descriptor_set_binding_layout *binding =
 280                 &set_layout->binding[var->data.binding];
 281         const struct radv_sampler_ycbcr_conversion *ycbcr_samplers =
 282                 radv_immutable_ycbcr_samplers(set_layout, var->data.binding);
 283
 284         if (!ycbcr_samplers)
 285                 return false;
 286
 287         /* For the following instructions, we don't apply any change and let the
 288          * instruction apply to the first plane.
 289          */
 290         if (tex->op == nir_texop_txs ||
 291             tex->op == nir_texop_query_levels ||
 292             tex->op == nir_texop_lod)
 293                 return false;
 294
 295         assert(tex->texture_index == 0);
 296         unsigned array_index = 0;
 297         if (deref->deref_type != nir_deref_type_var) {
 298                 assert(deref->deref_type == nir_deref_type_array);
 299                 if (!nir_src_is_const(deref->arr.index))
 300                         return false;
 301                 array_index = nir_src_as_uint(deref->arr.index);
 302                 array_index = MIN2(array_index, binding->array_size - 1);
 303         }
 304         const struct radv_sampler_ycbcr_conversion *ycbcr_sampler = ycbcr_samplers + array_index;
 305
 306         if (ycbcr_sampler->format == VK_FORMAT_UNDEFINED)
 307                 return false;
 308
 309         struct ycbcr_state state = {
 310                 .builder = builder,
 311                 .origin_tex = tex,
 312                 .tex_deref = deref,
 313                 .conversion = ycbcr_sampler,
 314         };
 315
 316         builder->cursor = nir_before_instr(&tex->instr);
 317
 318         VkFormat format = state.conversion->format;
 319         const int plane_count = vk_format_get_plane_count(format);
 320         nir_ssa_def *plane_values[3];
 321
 322         for (int p = 0; p < plane_count; ++p) {
 323                 plane_values[p] = create_plane_tex_instr_implicit(&state, p);
 324         }
 325
 326         nir_ssa_def *result = build_swizzled_components(builder, format, ycbcr_sampler->components, plane_values);
 327         if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) {
 328                 VkFormat first_format = vk_format_get_plane_format(format, 0);
 329                 result = convert_ycbcr(&state, result, vk_format_get_component_bits(first_format, VK_FORMAT_COLORSPACE_RGB, VK_SWIZZLE_X));
 330         }
 331
 332         nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
 333         nir_instr_remove(&tex->instr);
 334
 335         return true;
 336 }
 337
 338 bool
 339 radv_nir_lower_ycbcr_textures(nir_shader *shader,
 340                              const struct radv_pipeline_layout *layout)
 341 {
 342         bool progress = false;
 343
 344         nir_foreach_function(function, shader) {
 345                 if (!function->impl)
 346                         continue;
 347
 348                 bool function_progress = false;
 349                 nir_builder builder;
 350                 nir_builder_init(&builder, function->impl);
 351
 352                 nir_foreach_block(block, function->impl) {
 353                         nir_foreach_instr_safe(instr, block) {
 354                                 if (instr->type != nir_instr_type_tex)
 355                                         continue;
 356
 357                                 nir_tex_instr *tex = nir_instr_as_tex(instr);
 358                                 function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
 359                         }
 360                 }
 361
 362                 if (function_progress) {
 363                         nir_metadata_preserve(function->impl,
 364                                               nir_metadata_block_index |
 365                                               nir_metadata_dominance);
 366                 }
 367
 368                 progress |= function_progress;
 369         }
 370
 371         return progress;
 372 }