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         VkSamplerYcbcrRangeKHR 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              VkSamplerYcbcrRangeKHR 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 static const nir_const_value *
  83 ycbcr_model_to_rgb_matrix(VkSamplerYcbcrModelConversionKHR model)
  84 {
  85    switch (model) {
  86    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601: {
  87       static const nir_const_value bt601[3] = {
  88          { .f32 = {  1.402f,             1.0f,  0.0f,               0.0f } },
  89          { .f32 = { -0.714136286201022f, 1.0f, -0.344136286201022f, 0.0f } },
  90          { .f32 = {  0.0f,               1.0f,  1.772f,             0.0f } }
  91       };
  92
  93       return bt601;
  94    }
  95    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709: {
  96       static const nir_const_value bt709[3] = {
  97          { .f32 = {  1.5748031496063f,   1.0f,  0.0,                0.0f } },
  98          { .f32 = { -0.468125209181067f, 1.0f, -0.187327487470334f, 0.0f } },
  99          { .f32 = {  0.0f,               1.0f,  1.85563184264242f,  0.0f } }
 100       };
 101
 102       return bt709;
 103    }
 104    case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020: {
 105       static const nir_const_value bt2020[3] = {
 106          { .f32 = { 1.4746f,             1.0f,  0.0f,               0.0f } },
 107          { .f32 = { -0.571353126843658f, 1.0f, -0.164553126843658f, 0.0f } },
 108          { .f32 = { 0.0f,                1.0f,  1.8814f,            0.0f } }
 109       };
 110
 111       return bt2020;
 112    }
 113    default:
 114       unreachable("missing Ycbcr model");
 115       return NULL;
 116    }
 117 }
 118
 119 static nir_ssa_def *
 120 convert_ycbcr(struct ycbcr_state *state,
 121               nir_ssa_def *raw_channels,
 122               uint32_t *bpcs)
 123 {
 124    nir_builder *b = state->builder;
 125    struct anv_ycbcr_conversion *conversion = state->conversion;
 126
 127    nir_ssa_def *expanded_channels =
 128       nir_vec4(b,
 129                chroma_range(b, nir_channel(b, raw_channels, 0),
 130                             bpcs[0], conversion->ycbcr_range),
 131                y_range(b, nir_channel(b, raw_channels, 1),
 132                        bpcs[1], conversion->ycbcr_range),
 133                chroma_range(b, nir_channel(b, raw_channels, 2),
 134                             bpcs[2], conversion->ycbcr_range),
 135                nir_imm_float(b, 1.0f));
 136
 137    if (conversion->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY)
 138       return expanded_channels;
 139
 140    const nir_const_value *conversion_matrix =
 141       ycbcr_model_to_rgb_matrix(conversion->ycbcr_model);
 142
 143    nir_ssa_def *converted_channels[] = {
 144       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix[0])),
 145       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix[1])),
 146       nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix[2]))
 147    };
 148
 149    return nir_vec4(b,
 150                    converted_channels[0], converted_channels[1],
 151                    converted_channels[2], nir_imm_float(b, 1.0f));
 152 }
 153
 154 /* TODO: we should probably replace this with a push constant/uniform. */
 155 static nir_ssa_def *
 156 get_texture_size(struct ycbcr_state *state, nir_deref_instr *texture)
 157 {
 158    if (state->image_size)
 159       return state->image_size;
 160
 161    nir_builder *b = state->builder;
 162    const struct glsl_type *type = texture->type;
 163    nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
 164
 165    tex->op = nir_texop_txs;
 166    tex->sampler_dim = glsl_get_sampler_dim(type);
 167    tex->is_array = glsl_sampler_type_is_array(type);
 168    tex->is_shadow = glsl_sampler_type_is_shadow(type);
 169    tex->dest_type = nir_type_int;
 170
 171    tex->src[0].src_type = nir_tex_src_texture_deref;
 172    tex->src[0].src = nir_src_for_ssa(&texture->dest.ssa);
 173
 174    nir_ssa_dest_init(&tex->instr, &tex->dest,
 175                      nir_tex_instr_dest_size(tex), 32, NULL);
 176    nir_builder_instr_insert(b, &tex->instr);
 177
 178    state->image_size = nir_i2f32(b, &tex->dest.ssa);
 179
 180    return state->image_size;
 181 }
 182
 183 static nir_ssa_def *
 184 implicit_downsampled_coord(nir_builder *b,
 185                            nir_ssa_def *value,
 186                            nir_ssa_def *max_value,
 187                            int div_scale)
 188 {
 189    return nir_fadd(b,
 190                    value,
 191                    nir_fdiv(b,
 192                             nir_imm_float(b, 1.0f),
 193                             nir_fmul(b,
 194                                      nir_imm_float(b, div_scale),
 195                                      max_value)));
 196 }
 197
 198 static nir_ssa_def *
 199 implicit_downsampled_coords(struct ycbcr_state *state,
 200                             nir_ssa_def *old_coords,
 201                             const struct anv_format_plane *plane_format)
 202 {
 203    nir_builder *b = state->builder;
 204    struct anv_ycbcr_conversion *conversion = state->conversion;
 205    nir_ssa_def *image_size = get_texture_size(state, state->tex_deref);
 206    nir_ssa_def *comp[4] = { NULL, };
 207    int c;
 208
 209    for (c = 0; c < ARRAY_SIZE(conversion->chroma_offsets); c++) {
 210       if (plane_format->denominator_scales[c] > 1 &&
 211           conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN) {
 212          comp[c] = implicit_downsampled_coord(b,
 213                                               nir_channel(b, old_coords, c),
 214                                               nir_channel(b, image_size, c),
 215                                               plane_format->denominator_scales[c]);
 216       } else {
 217          comp[c] = nir_channel(b, old_coords, c);
 218       }
 219    }
 220
 221    /* Leave other coordinates untouched */
 222    for (; c < old_coords->num_components; c++)
 223       comp[c] = nir_channel(b, old_coords, c);
 224
 225    return nir_vec(b, comp, old_coords->num_components);
 226 }
 227
 228 static nir_ssa_def *
 229 create_plane_tex_instr_implicit(struct ycbcr_state *state,
 230                                 uint32_t plane)
 231 {
 232    nir_builder *b = state->builder;
 233    struct anv_ycbcr_conversion *conversion = state->conversion;
 234    const struct anv_format_plane *plane_format =
 235       &conversion->format->planes[plane];
 236    nir_tex_instr *old_tex = state->origin_tex;
 237    nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs + 1);
 238
 239    for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
 240       tex->src[i].src_type = old_tex->src[i].src_type;
 241
 242       switch (old_tex->src[i].src_type) {
 243       case nir_tex_src_coord:
 244          if (plane_format->has_chroma && conversion->chroma_reconstruction) {
 245             assert(old_tex->src[i].src.is_ssa);
 246             tex->src[i].src =
 247                nir_src_for_ssa(implicit_downsampled_coords(state,
 248                                                            old_tex->src[i].src.ssa,
 249                                                            plane_format));
 250             break;
 251          }
 252          /* fall through */
 253       default:
 254          nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
 255          break;
 256       }
 257    }
 258    tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
 259    tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
 260
 261    tex->sampler_dim = old_tex->sampler_dim;
 262    tex->dest_type = old_tex->dest_type;
 263
 264    tex->op = old_tex->op;
 265    tex->coord_components = old_tex->coord_components;
 266    tex->is_new_style_shadow = old_tex->is_new_style_shadow;
 267    tex->component = old_tex->component;
 268
 269    tex->texture_index = old_tex->texture_index;
 270    tex->texture_array_size = old_tex->texture_array_size;
 271    tex->sampler_index = old_tex->sampler_index;
 272
 273    nir_ssa_dest_init(&tex->instr, &tex->dest,
 274                      old_tex->dest.ssa.num_components,
 275                      nir_dest_bit_size(old_tex->dest), NULL);
 276    nir_builder_instr_insert(b, &tex->instr);
 277
 278    return &tex->dest.ssa;
 279 }
 280
 281 static unsigned
 282 channel_to_component(enum isl_channel_select channel)
 283 {
 284    switch (channel) {
 285    case ISL_CHANNEL_SELECT_RED:
 286       return 0;
 287    case ISL_CHANNEL_SELECT_GREEN:
 288       return 1;
 289    case ISL_CHANNEL_SELECT_BLUE:
 290       return 2;
 291    case ISL_CHANNEL_SELECT_ALPHA:
 292       return 3;
 293    default:
 294       unreachable("invalid channel");
 295       return 0;
 296    }
 297 }
 298
 299 static enum isl_channel_select
 300 swizzle_channel(struct isl_swizzle swizzle, unsigned channel)
 301 {
 302    switch (channel) {
 303    case 0:
 304       return swizzle.r;
 305    case 1:
 306       return swizzle.g;
 307    case 2:
 308       return swizzle.b;
 309    case 3:
 310       return swizzle.a;
 311    default:
 312       unreachable("invalid channel");
 313       return 0;
 314    }
 315 }
 316
 317 static bool
 318 try_lower_tex_ycbcr(struct anv_pipeline_layout *layout,
 319                     nir_builder *builder,
 320                     nir_tex_instr *tex)
 321 {
 322    int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
 323    assert(deref_src_idx >= 0);
 324    nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
 325
 326    nir_variable *var = nir_deref_instr_get_variable(deref);
 327    const struct anv_descriptor_set_layout *set_layout =
 328       layout->set[var->data.descriptor_set].layout;
 329    const struct anv_descriptor_set_binding_layout *binding =
 330       &set_layout->binding[var->data.binding];
 331
 332    /* For the following instructions, we don't apply any change and let the
 333     * instruction apply to the first plane.
 334     */
 335    if (tex->op == nir_texop_txs ||
 336        tex->op == nir_texop_query_levels ||
 337        tex->op == nir_texop_lod)
 338       return false;
 339
 340    if (binding->immutable_samplers == NULL)
 341       return false;
 342
 343    unsigned texture_index = tex->texture_index;
 344    if (deref->deref_type != nir_deref_type_var) {
 345       assert(deref->deref_type == nir_deref_type_array);
 346       nir_const_value *const_index = nir_src_as_const_value(deref->arr.index);
 347       if (!const_index)
 348          return false;
 349       size_t hw_binding_size =
 350          anv_descriptor_set_binding_layout_get_hw_size(binding);
 351       texture_index += MIN2(const_index->u32[0], hw_binding_size - 1);
 352    }
 353    const struct anv_sampler *sampler =
 354       binding->immutable_samplers[texture_index];
 355
 356    if (sampler->conversion == NULL)
 357       return false;
 358
 359    struct ycbcr_state state = {
 360       .builder = builder,
 361       .origin_tex = tex,
 362       .tex_deref = deref,
 363       .conversion = sampler->conversion,
 364    };
 365
 366    builder->cursor = nir_before_instr(&tex->instr);
 367
 368    const struct anv_format *format = state.conversion->format;
 369    const struct isl_format_layout *y_isl_layout = NULL;
 370    for (uint32_t p = 0; p < format->n_planes; p++) {
 371       if (!format->planes[p].has_chroma)
 372          y_isl_layout = isl_format_get_layout(format->planes[p].isl_format);
 373    }
 374    assert(y_isl_layout != NULL);
 375    uint8_t y_bpc = y_isl_layout->channels_array[0].bits;
 376
 377    /* |ycbcr_comp| holds components in the order : Cr-Y-Cb */
 378    nir_ssa_def *ycbcr_comp[5] = { NULL, NULL, NULL,
 379                                   /* Use extra 2 channels for following swizzle */
 380                                   nir_imm_float(builder, 1.0f),
 381                                   nir_imm_float(builder, 0.0f),
 382    };
 383    uint8_t ycbcr_bpcs[5];
 384    memset(ycbcr_bpcs, y_bpc, sizeof(ycbcr_bpcs));
 385
 386    /* Go through all the planes and gather the samples into a |ycbcr_comp|
 387     * while applying a swizzle required by the spec:
 388     *
 389     *    R, G, B should respectively map to Cr, Y, Cb
 390     */
 391    for (uint32_t p = 0; p < format->n_planes; p++) {
 392       const struct anv_format_plane *plane_format = &format->planes[p];
 393       nir_ssa_def *plane_sample = create_plane_tex_instr_implicit(&state, p);
 394
 395       for (uint32_t pc = 0; pc < 4; pc++) {
 396          enum isl_channel_select ycbcr_swizzle =
 397             swizzle_channel(plane_format->ycbcr_swizzle, pc);
 398          if (ycbcr_swizzle == ISL_CHANNEL_SELECT_ZERO)
 399             continue;
 400
 401          unsigned ycbcr_component = channel_to_component(ycbcr_swizzle);
 402          ycbcr_comp[ycbcr_component] = nir_channel(builder, plane_sample, pc);
 403
 404          /* Also compute the number of bits for each component. */
 405          const struct isl_format_layout *isl_layout =
 406             isl_format_get_layout(plane_format->isl_format);
 407          ycbcr_bpcs[ycbcr_component] = isl_layout->channels_array[pc].bits;
 408       }
 409    }
 410
 411    /* Now remaps components to the order specified by the conversion. */
 412    nir_ssa_def *swizzled_comp[4] = { NULL, };
 413    uint32_t swizzled_bpcs[4] = { 0, };
 414
 415    for (uint32_t i = 0; i < ARRAY_SIZE(state.conversion->mapping); i++) {
 416       /* Maps to components in |ycbcr_comp| */
 417       static const uint32_t swizzle_mapping[] = {
 418          [VK_COMPONENT_SWIZZLE_ZERO] = 4,
 419          [VK_COMPONENT_SWIZZLE_ONE]  = 3,
 420          [VK_COMPONENT_SWIZZLE_R]    = 0,
 421          [VK_COMPONENT_SWIZZLE_G]    = 1,
 422          [VK_COMPONENT_SWIZZLE_B]    = 2,
 423          [VK_COMPONENT_SWIZZLE_A]    = 3,
 424       };
 425       const VkComponentSwizzle m = state.conversion->mapping[i];
 426
 427       if (m == VK_COMPONENT_SWIZZLE_IDENTITY) {
 428          swizzled_comp[i] = ycbcr_comp[i];
 429          swizzled_bpcs[i] = ycbcr_bpcs[i];
 430       } else {
 431          swizzled_comp[i] = ycbcr_comp[swizzle_mapping[m]];
 432          swizzled_bpcs[i] = ycbcr_bpcs[swizzle_mapping[m]];
 433       }
 434    }
 435
 436    nir_ssa_def *result = nir_vec(builder, swizzled_comp, 4);
 437    if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY)
 438       result = convert_ycbcr(&state, result, swizzled_bpcs);
 439
 440    nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
 441    nir_instr_remove(&tex->instr);
 442
 443    return true;
 444 }
 445
 446 bool
 447 anv_nir_lower_ycbcr_textures(nir_shader *shader,
 448                              struct anv_pipeline_layout *layout)
 449 {
 450    bool progress = false;
 451
 452    nir_foreach_function(function, shader) {
 453       if (!function->impl)
 454          continue;
 455
 456       bool function_progress = false;
 457       nir_builder builder;
 458       nir_builder_init(&builder, function->impl);
 459
 460       nir_foreach_block(block, function->impl) {
 461          nir_foreach_instr_safe(instr, block) {
 462             if (instr->type != nir_instr_type_tex)
 463                continue;
 464
 465             nir_tex_instr *tex = nir_instr_as_tex(instr);
 466             function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
 467          }
 468       }
 469
 470       if (function_progress) {
 471          nir_metadata_preserve(function->impl,
 472                                nir_metadata_block_index |
 473                                nir_metadata_dominance);
 474       }
 475
 476       progress |= function_progress;
 477    }
 478
 479    return progress;
 480 }