src/intel/vulkan/anv_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 "anv_nir.h"
  25 #include "anv_private.h"
  26 #include "nir/nir.h"
  27 #include "nir/nir_builder.h"
  28
  29 struct ycbcr_state {
  30    nir_builder *builder;
  31    nir_ssa_def *image_size;
  32    nir_tex_instr *origin_tex;
  33    nir_deref_instr *tex_deref;
  34    struct anv_ycbcr_conversion *conversion;
  35 };
  36
  37 static nir_ssa_def *
  38 y_range(nir_builder *b,
  39         nir_ssa_def *y_channel,
  40         int bpc,
  41         VkSamplerYcbcrRange range)
  42 {
  43    switch (range) {
  44    case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
  45       return y_channel;
  46    case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
  47       return nir_fmul(b,
  48                       nir_fadd(b,
  49                                nir_fmul(b, y_channel,
  50                                         nir_imm_float(b, pow(2, bpc) - 1)),
  51                                nir_imm_float(b, -16.0f * pow(2, bpc - 8))),
  52                       nir_frcp(b, nir_imm_float(b, 219.0f * pow(2, bpc - 8))));
  53    default:
  54       unreachable("missing Ycbcr range");
  55       return NULL;
  56    }
  57 }
  58
  59 static nir_ssa_def *
  60 chroma_range(nir_builder *b,
  61              nir_ssa_def *chroma_channel,
  62              int bpc,
  63              VkSamplerYcbcrRange range)
  64 {
  65    switch (range) {
  66    case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
  67       return nir_fadd(b, chroma_channel,
  68                       nir_imm_float(b, -pow(2, bpc - 1) / (pow(2, bpc) - 1.0f)));
  69    case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
  70       return nir_fmul(b,
  71                       nir_fadd(b,
  72                                nir_fmul(b, chroma_channel,
  73                                         nir_imm_float(b, pow(2, bpc) - 1)),
  74                                nir_imm_float(b, -128.0f * pow(2, bpc - 8))),
  75                       nir_frcp(b, nir_imm_float(b, 224.0f * pow(2, bpc - 8))));
  76    default:
  77       unreachable("missing Ycbcr range");
  78       return NULL;
  79    }
  80 }
  81
  82 typedef struct nir_const_value_3_4 {
  83    nir_const_value v[3][4];
  84 } nir_const_value_3_4;
  85
  86 static const nir_const_value_3_4 *
  87 ycbcr_model_to_rgb_matrix(VkSamplerYcbcrModelConversion model)
  88 {
  89    switch (model) {
  90    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601: {
  91       static const nir_const_value_3_4 bt601 = { {
  92          { { .f32 =  1.402f             }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
  93          { { .f32 = -0.714136286201022f }, { .f32 = 1.0f }, { .f32 = -0.344136286201022f }, { .f32 = 0.0f } },
  94          { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.772f             }, { .f32 = 0.0f } },
  95       } };
  96
  97       return &bt601;
  98    }
  99    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709: {
 100       static const nir_const_value_3_4 bt709 = { {
 101          { { .f32 =  1.5748031496063f   }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
 102          { { .f32 = -0.468125209181067f }, { .f32 = 1.0f }, { .f32 = -0.187327487470334f }, { .f32 = 0.0f } },
 103          { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.85563184264242f  }, { .f32 = 0.0f } },
 104       } };
 105
 106       return &bt709;
 107    }
 108    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020: {
 109       static const nir_const_value_3_4 bt2020 = { {
 110          { { .f32 =  1.4746f            }, { .f32 = 1.0f }, { .f32 =  0.0f               }, { .f32 = 0.0f } },
 111          { { .f32 = -0.571353126843658f }, { .f32 = 1.0f }, { .f32 = -0.164553126843658f }, { .f32 = 0.0f } },
 112          { { .f32 =  0.0f               }, { .f32 = 1.0f }, { .f32 =  1.8814f            }, { .f32 = 0.0f } },
 113       } };
 114
 115       return &bt2020;
 116    }
 117    default:
 118       unreachable("missing Ycbcr model");
 119       return NULL;
 120    }
 121 }
 122
 123 static nir_ssa_def *
 124 convert_ycbcr(struct ycbcr_state *state,
 125               nir_ssa_def *raw_channels,
 126               uint32_t *bpcs)
 127 {
 128    nir_builder *b = state->builder;
 129    struct anv_ycbcr_conversion *conversion = state->conversion;
 130
 131    nir_ssa_def *expanded_channels =
 132       nir_vec4(b,
 133                chroma_range(b, nir_channel(b, raw_channels, 0),
 134                             bpcs[0], conversion->ycbcr_range),
 135                y_range(b, nir_channel(b, raw_channels, 1),
 136                        bpcs[1], conversion->ycbcr_range),
 137                chroma_range(b, nir_channel(b, raw_channels, 2),
 138                             bpcs[2], conversion->ycbcr_range),
 139                nir_imm_float(b, 1.0f));
 140
 141    if (conversion->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY)
 142       return expanded_channels;
 143
 144    const nir_const_value_3_4 *conversion_matrix =
 145       ycbcr_model_to_rgb_matrix(conversion->ycbcr_model);
 146
 147    nir_ssa_def *converted_channels[] = {
 148       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[0])),
 149       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[1])),
 150       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[2]))
 151    };
 152
 153    return nir_vec4(b,
 154                    converted_channels[0], converted_channels[1],
 155                    converted_channels[2], nir_imm_float(b, 1.0f));
 156 }
 157
 158 /* TODO: we should probably replace this with a push constant/uniform. */
 159 static nir_ssa_def *
 160 get_texture_size(struct ycbcr_state *state, nir_deref_instr *texture)
 161 {
 162    if (state->image_size)
 163       return state->image_size;
 164
 165    nir_builder *b = state->builder;
 166    const struct glsl_type *type = texture->type;
 167    nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
 168
 169    tex->op = nir_texop_txs;
 170    tex->sampler_dim = glsl_get_sampler_dim(type);
 171    tex->is_array = glsl_sampler_type_is_array(type);
 172    tex->is_shadow = glsl_sampler_type_is_shadow(type);
 173    tex->dest_type = nir_type_int;
 174
 175    tex->src[0].src_type = nir_tex_src_texture_deref;
 176    tex->src[0].src = nir_src_for_ssa(&texture->dest.ssa);
 177
 178    nir_ssa_dest_init(&tex->instr, &tex->dest,
 179                      nir_tex_instr_dest_size(tex), 32, NULL);
 180    nir_builder_instr_insert(b, &tex->instr);
 181
 182    state->image_size = nir_i2f32(b, &tex->dest.ssa);
 183
 184    return state->image_size;
 185 }
 186
 187 static nir_ssa_def *
 188 implicit_downsampled_coord(nir_builder *b,
 189                            nir_ssa_def *value,
 190                            nir_ssa_def *max_value,
 191                            int div_scale)
 192 {
 193    return nir_fadd(b,
 194                    value,
 195                    nir_fdiv(b,
 196                             nir_imm_float(b, 1.0f),
 197                             nir_fmul(b,
 198                                      nir_imm_float(b, div_scale),
 199                                      max_value)));
 200 }
 201
 202 static nir_ssa_def *
 203 implicit_downsampled_coords(struct ycbcr_state *state,
 204                             nir_ssa_def *old_coords,
 205                             const struct anv_format_plane *plane_format)
 206 {
 207    nir_builder *b = state->builder;
 208    struct anv_ycbcr_conversion *conversion = state->conversion;
 209    nir_ssa_def *image_size = get_texture_size(state, state->tex_deref);
 210    nir_ssa_def *comp[4] = { NULL, };
 211    int c;
 212
 213    for (c = 0; c < ARRAY_SIZE(conversion->chroma_offsets); c++) {
 214       if (plane_format->denominator_scales[c] > 1 &&
 215           conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN) {
 216          comp[c] = implicit_downsampled_coord(b,
 217                                               nir_channel(b, old_coords, c),
 218                                               nir_channel(b, image_size, c),
 219                                               plane_format->denominator_scales[c]);
 220       } else {
 221          comp[c] = nir_channel(b, old_coords, c);
 222       }
 223    }
 224
 225    /* Leave other coordinates untouched */
 226    for (; c < old_coords->num_components; c++)
 227       comp[c] = nir_channel(b, old_coords, c);
 228
 229    return nir_vec(b, comp, old_coords->num_components);
 230 }
 231
 232 static nir_ssa_def *
 233 create_plane_tex_instr_implicit(struct ycbcr_state *state,
 234                                 uint32_t plane)
 235 {
 236    nir_builder *b = state->builder;
 237    struct anv_ycbcr_conversion *conversion = state->conversion;
 238    const struct anv_format_plane *plane_format =
 239       &conversion->format->planes[plane];
 240    nir_tex_instr *old_tex = state->origin_tex;
 241    nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs + 1);
 242
 243    for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
 244       tex->src[i].src_type = old_tex->src[i].src_type;
 245
 246       switch (old_tex->src[i].src_type) {
 247       case nir_tex_src_coord:
 248          if (plane_format->has_chroma && conversion->chroma_reconstruction) {
 249             assert(old_tex->src[i].src.is_ssa);
 250             tex->src[i].src =
 251                nir_src_for_ssa(implicit_downsampled_coords(state,
 252                                                            old_tex->src[i].src.ssa,
 253                                                            plane_format));
 254             break;
 255          }
 256          /* fall through */
 257       default:
 258          nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
 259          break;
 260       }
 261    }
 262    tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
 263    tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
 264
 265    tex->sampler_dim = old_tex->sampler_dim;
 266    tex->dest_type = old_tex->dest_type;
 267
 268    tex->op = old_tex->op;
 269    tex->coord_components = old_tex->coord_components;
 270    tex->is_new_style_shadow = old_tex->is_new_style_shadow;
 271    tex->component = old_tex->component;
 272
 273    tex->texture_index = old_tex->texture_index;
 274    tex->texture_array_size = old_tex->texture_array_size;
 275    tex->sampler_index = old_tex->sampler_index;
 276    tex->is_array = old_tex->is_array;
 277
 278    nir_ssa_dest_init(&tex->instr, &tex->dest,
 279                      old_tex->dest.ssa.num_components,
 280                      nir_dest_bit_size(old_tex->dest), NULL);
 281    nir_builder_instr_insert(b, &tex->instr);
 282
 283    return &tex->dest.ssa;
 284 }
 285
 286 static unsigned
 287 channel_to_component(enum isl_channel_select channel)
 288 {
 289    switch (channel) {
 290    case ISL_CHANNEL_SELECT_RED:
 291       return 0;
 292    case ISL_CHANNEL_SELECT_GREEN:
 293       return 1;
 294    case ISL_CHANNEL_SELECT_BLUE:
 295       return 2;
 296    case ISL_CHANNEL_SELECT_ALPHA:
 297       return 3;
 298    default:
 299       unreachable("invalid channel");
 300       return 0;
 301    }
 302 }
 303
 304 static enum isl_channel_select
 305 swizzle_channel(struct isl_swizzle swizzle, unsigned channel)
 306 {
 307    switch (channel) {
 308    case 0:
 309       return swizzle.r;
 310    case 1:
 311       return swizzle.g;
 312    case 2:
 313       return swizzle.b;
 314    case 3:
 315       return swizzle.a;
 316    default:
 317       unreachable("invalid channel");
 318       return 0;
 319    }
 320 }
 321
 322 static bool
 323 try_lower_tex_ycbcr(struct anv_pipeline_layout *layout,
 324                     nir_builder *builder,
 325                     nir_tex_instr *tex)
 326 {
 327    int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
 328    assert(deref_src_idx >= 0);
 329    nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
 330
 331    nir_variable *var = nir_deref_instr_get_variable(deref);
 332    const struct anv_descriptor_set_layout *set_layout =
 333       layout->set[var->data.descriptor_set].layout;
 334    const struct anv_descriptor_set_binding_layout *binding =
 335       &set_layout->binding[var->data.binding];
 336
 337    /* For the following instructions, we don't apply any change and let the
 338     * instruction apply to the first plane.
 339     */
 340    if (tex->op == nir_texop_txs ||
 341        tex->op == nir_texop_query_levels ||
 342        tex->op == nir_texop_lod)
 343       return false;
 344
 345    if (binding->immutable_samplers == NULL)
 346       return false;
 347
 348    assert(tex->texture_index == 0);
 349    unsigned array_index = 0;
 350    if (deref->deref_type != nir_deref_type_var) {
 351       assert(deref->deref_type == nir_deref_type_array);
 352       if (!nir_src_is_const(deref->arr.index))
 353          return false;
 354       array_index = nir_src_as_uint(deref->arr.index);
 355       array_index = MIN2(array_index, binding->array_size - 1);
 356    }
 357    const struct anv_sampler *sampler = binding->immutable_samplers[array_index];
 358
 359    if (sampler->conversion == NULL)
 360       return false;
 361
 362    struct ycbcr_state state = {
 363       .builder = builder,
 364       .origin_tex = tex,
 365       .tex_deref = deref,
 366       .conversion = sampler->conversion,
 367    };
 368
 369    builder->cursor = nir_before_instr(&tex->instr);
 370
 371    const struct anv_format *format = state.conversion->format;
 372    const struct isl_format_layout *y_isl_layout = NULL;
 373    for (uint32_t p = 0; p < format->n_planes; p++) {
 374       if (!format->planes[p].has_chroma)
 375          y_isl_layout = isl_format_get_layout(format->planes[p].isl_format);
 376    }
 377    assert(y_isl_layout != NULL);
 378    uint8_t y_bpc = y_isl_layout->channels_array[0].bits;
 379
 380    /* |ycbcr_comp| holds components in the order : Cr-Y-Cb */
 381    nir_ssa_def *zero = nir_imm_float(builder, 0.0f);
 382    nir_ssa_def *one = nir_imm_float(builder, 1.0f);
 383    /* Use extra 2 channels for following swizzle */
 384    nir_ssa_def *ycbcr_comp[5] = { zero, zero, zero, one, zero };
 385
 386    uint8_t ycbcr_bpcs[5];
 387    memset(ycbcr_bpcs, y_bpc, sizeof(ycbcr_bpcs));
 388
 389    /* Go through all the planes and gather the samples into a |ycbcr_comp|
 390     * while applying a swizzle required by the spec:
 391     *
 392     *    R, G, B should respectively map to Cr, Y, Cb
 393     */
 394    for (uint32_t p = 0; p < format->n_planes; p++) {
 395       const struct anv_format_plane *plane_format = &format->planes[p];
 396       nir_ssa_def *plane_sample = create_plane_tex_instr_implicit(&state, p);
 397
 398       for (uint32_t pc = 0; pc < 4; pc++) {
 399          enum isl_channel_select ycbcr_swizzle =
 400             swizzle_channel(plane_format->ycbcr_swizzle, pc);
 401          if (ycbcr_swizzle == ISL_CHANNEL_SELECT_ZERO)
 402             continue;
 403
 404          unsigned ycbcr_component = channel_to_component(ycbcr_swizzle);
 405          ycbcr_comp[ycbcr_component] = nir_channel(builder, plane_sample, pc);
 406
 407          /* Also compute the number of bits for each component. */
 408          const struct isl_format_layout *isl_layout =
 409             isl_format_get_layout(plane_format->isl_format);
 410          ycbcr_bpcs[ycbcr_component] = isl_layout->channels_array[pc].bits;
 411       }
 412    }
 413
 414    /* Now remaps components to the order specified by the conversion. */
 415    nir_ssa_def *swizzled_comp[4] = { NULL, };
 416    uint32_t swizzled_bpcs[4] = { 0, };
 417
 418    for (uint32_t i = 0; i < ARRAY_SIZE(state.conversion->mapping); i++) {
 419       /* Maps to components in |ycbcr_comp| */
 420       static const uint32_t swizzle_mapping[] = {
 421          [VK_COMPONENT_SWIZZLE_ZERO] = 4,
 422          [VK_COMPONENT_SWIZZLE_ONE]  = 3,
 423          [VK_COMPONENT_SWIZZLE_R]    = 0,
 424          [VK_COMPONENT_SWIZZLE_G]    = 1,
 425          [VK_COMPONENT_SWIZZLE_B]    = 2,
 426          [VK_COMPONENT_SWIZZLE_A]    = 3,
 427       };
 428       const VkComponentSwizzle m = state.conversion->mapping[i];
 429
 430       if (m == VK_COMPONENT_SWIZZLE_IDENTITY) {
 431          swizzled_comp[i] = ycbcr_comp[i];
 432          swizzled_bpcs[i] = ycbcr_bpcs[i];
 433       } else {
 434          swizzled_comp[i] = ycbcr_comp[swizzle_mapping[m]];
 435          swizzled_bpcs[i] = ycbcr_bpcs[swizzle_mapping[m]];
 436       }
 437    }
 438
 439    nir_ssa_def *result = nir_vec(builder, swizzled_comp, 4);
 440    if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY)
 441       result = convert_ycbcr(&state, result, swizzled_bpcs);
 442
 443    nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
 444    nir_instr_remove(&tex->instr);
 445
 446    return true;
 447 }
 448
 449 bool
 450 anv_nir_lower_ycbcr_textures(nir_shader *shader,
 451                              struct anv_pipeline_layout *layout)
 452 {
 453    bool progress = false;
 454
 455    nir_foreach_function(function, shader) {
 456       if (!function->impl)
 457          continue;
 458
 459       bool function_progress = false;
 460       nir_builder builder;
 461       nir_builder_init(&builder, function->impl);
 462
 463       nir_foreach_block(block, function->impl) {
 464          nir_foreach_instr_safe(instr, block) {
 465             if (instr->type != nir_instr_type_tex)
 466                continue;
 467
 468             nir_tex_instr *tex = nir_instr_as_tex(instr);
 469             function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
 470          }
 471       }
 472
 473       if (function_progress) {
 474          nir_metadata_preserve(function->impl,
 475                                nir_metadata_block_index |
 476                                nir_metadata_dominance);
 477       }
 478
 479       progress |= function_progress;
 480    }
 481
 482    return progress;
 483 }