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 #include "nir/nir_vulkan.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    struct anv_ycbcr_conversion *conversion;
  36 };
  37
  38 /* TODO: we should probably replace this with a push constant/uniform. */
  39 static nir_ssa_def *
  40 get_texture_size(struct ycbcr_state *state, nir_deref_instr *texture)
  41 {
  42    if (state->image_size)
  43       return state->image_size;
  44
  45    nir_builder *b = state->builder;
  46    const struct glsl_type *type = texture->type;
  47    nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
  48
  49    tex->op = nir_texop_txs;
  50    tex->sampler_dim = glsl_get_sampler_dim(type);
  51    tex->is_array = glsl_sampler_type_is_array(type);
  52    tex->is_shadow = glsl_sampler_type_is_shadow(type);
  53    tex->dest_type = nir_type_int;
  54
  55    tex->src[0].src_type = nir_tex_src_texture_deref;
  56    tex->src[0].src = nir_src_for_ssa(&texture->dest.ssa);
  57
  58    nir_ssa_dest_init(&tex->instr, &tex->dest,
  59                      nir_tex_instr_dest_size(tex), 32, NULL);
  60    nir_builder_instr_insert(b, &tex->instr);
  61
  62    state->image_size = nir_i2f32(b, &tex->dest.ssa);
  63
  64    return state->image_size;
  65 }
  66
  67 static nir_ssa_def *
  68 implicit_downsampled_coord(nir_builder *b,
  69                            nir_ssa_def *value,
  70                            nir_ssa_def *max_value,
  71                            int div_scale)
  72 {
  73    return nir_fadd(b,
  74                    value,
  75                    nir_fdiv(b,
  76                             nir_imm_float(b, 1.0f),
  77                             nir_fmul(b,
  78                                      nir_imm_float(b, div_scale),
  79                                      max_value)));
  80 }
  81
  82 static nir_ssa_def *
  83 implicit_downsampled_coords(struct ycbcr_state *state,
  84                             nir_ssa_def *old_coords,
  85                             const struct anv_format_plane *plane_format)
  86 {
  87    nir_builder *b = state->builder;
  88    struct anv_ycbcr_conversion *conversion = state->conversion;
  89    nir_ssa_def *image_size = get_texture_size(state, state->tex_deref);
  90    nir_ssa_def *comp[4] = { NULL, };
  91    int c;
  92
  93    for (c = 0; c < ARRAY_SIZE(conversion->chroma_offsets); c++) {
  94       if (plane_format->denominator_scales[c] > 1 &&
  95           conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN) {
  96          comp[c] = implicit_downsampled_coord(b,
  97                                               nir_channel(b, old_coords, c),
  98                                               nir_channel(b, image_size, c),
  99                                               plane_format->denominator_scales[c]);
 100       } else {
 101          comp[c] = nir_channel(b, old_coords, c);
 102       }
 103    }
 104
 105    /* Leave other coordinates untouched */
 106    for (; c < old_coords->num_components; c++)
 107       comp[c] = nir_channel(b, old_coords, c);
 108
 109    return nir_vec(b, comp, old_coords->num_components);
 110 }
 111
 112 static nir_ssa_def *
 113 create_plane_tex_instr_implicit(struct ycbcr_state *state,
 114                                 uint32_t plane)
 115 {
 116    nir_builder *b = state->builder;
 117    struct anv_ycbcr_conversion *conversion = state->conversion;
 118    const struct anv_format_plane *plane_format =
 119       &conversion->format->planes[plane];
 120    nir_tex_instr *old_tex = state->origin_tex;
 121    nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs + 1);
 122
 123    for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
 124       tex->src[i].src_type = old_tex->src[i].src_type;
 125
 126       switch (old_tex->src[i].src_type) {
 127       case nir_tex_src_coord:
 128          if (plane_format->has_chroma && conversion->chroma_reconstruction) {
 129             assert(old_tex->src[i].src.is_ssa);
 130             tex->src[i].src =
 131                nir_src_for_ssa(implicit_downsampled_coords(state,
 132                                                            old_tex->src[i].src.ssa,
 133                                                            plane_format));
 134             break;
 135          }
 136          /* fall through */
 137       default:
 138          nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
 139          break;
 140       }
 141    }
 142    tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
 143    tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
 144
 145    tex->sampler_dim = old_tex->sampler_dim;
 146    tex->dest_type = old_tex->dest_type;
 147
 148    tex->op = old_tex->op;
 149    tex->coord_components = old_tex->coord_components;
 150    tex->is_new_style_shadow = old_tex->is_new_style_shadow;
 151    tex->component = old_tex->component;
 152
 153    tex->texture_index = old_tex->texture_index;
 154    tex->sampler_index = old_tex->sampler_index;
 155    tex->is_array = old_tex->is_array;
 156
 157    nir_ssa_dest_init(&tex->instr, &tex->dest,
 158                      old_tex->dest.ssa.num_components,
 159                      nir_dest_bit_size(old_tex->dest), NULL);
 160    nir_builder_instr_insert(b, &tex->instr);
 161
 162    return &tex->dest.ssa;
 163 }
 164
 165 static unsigned
 166 channel_to_component(enum isl_channel_select channel)
 167 {
 168    switch (channel) {
 169    case ISL_CHANNEL_SELECT_RED:
 170       return 0;
 171    case ISL_CHANNEL_SELECT_GREEN:
 172       return 1;
 173    case ISL_CHANNEL_SELECT_BLUE:
 174       return 2;
 175    case ISL_CHANNEL_SELECT_ALPHA:
 176       return 3;
 177    default:
 178       unreachable("invalid channel");
 179       return 0;
 180    }
 181 }
 182
 183 static enum isl_channel_select
 184 swizzle_channel(struct isl_swizzle swizzle, unsigned channel)
 185 {
 186    switch (channel) {
 187    case 0:
 188       return swizzle.r;
 189    case 1:
 190       return swizzle.g;
 191    case 2:
 192       return swizzle.b;
 193    case 3:
 194       return swizzle.a;
 195    default:
 196       unreachable("invalid channel");
 197       return 0;
 198    }
 199 }
 200
 201 static bool
 202 try_lower_tex_ycbcr(const struct anv_pipeline_layout *layout,
 203                     nir_builder *builder,
 204                     nir_tex_instr *tex)
 205 {
 206    int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
 207    assert(deref_src_idx >= 0);
 208    nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
 209
 210    nir_variable *var = nir_deref_instr_get_variable(deref);
 211    const struct anv_descriptor_set_layout *set_layout =
 212       layout->set[var->data.descriptor_set].layout;
 213    const struct anv_descriptor_set_binding_layout *binding =
 214       &set_layout->binding[var->data.binding];
 215
 216    /* For the following instructions, we don't apply any change and let the
 217     * instruction apply to the first plane.
 218     */
 219    if (tex->op == nir_texop_txs ||
 220        tex->op == nir_texop_query_levels ||
 221        tex->op == nir_texop_lod)
 222       return false;
 223
 224    if (binding->immutable_samplers == NULL)
 225       return false;
 226
 227    assert(tex->texture_index == 0);
 228    unsigned array_index = 0;
 229    if (deref->deref_type != nir_deref_type_var) {
 230       assert(deref->deref_type == nir_deref_type_array);
 231       if (!nir_src_is_const(deref->arr.index))
 232          return false;
 233       array_index = nir_src_as_uint(deref->arr.index);
 234       array_index = MIN2(array_index, binding->array_size - 1);
 235    }
 236    const struct anv_sampler *sampler = binding->immutable_samplers[array_index];
 237
 238    if (sampler->conversion == NULL)
 239       return false;
 240
 241    struct ycbcr_state state = {
 242       .builder = builder,
 243       .origin_tex = tex,
 244       .tex_deref = deref,
 245       .conversion = sampler->conversion,
 246    };
 247
 248    builder->cursor = nir_before_instr(&tex->instr);
 249
 250    const struct anv_format *format = state.conversion->format;
 251    const struct isl_format_layout *y_isl_layout = NULL;
 252    for (uint32_t p = 0; p < format->n_planes; p++) {
 253       if (!format->planes[p].has_chroma)
 254          y_isl_layout = isl_format_get_layout(format->planes[p].isl_format);
 255    }
 256    assert(y_isl_layout != NULL);
 257    uint8_t y_bpc = y_isl_layout->channels_array[0].bits;
 258
 259    /* |ycbcr_comp| holds components in the order : Cr-Y-Cb */
 260    nir_ssa_def *zero = nir_imm_float(builder, 0.0f);
 261    nir_ssa_def *one = nir_imm_float(builder, 1.0f);
 262    /* Use extra 2 channels for following swizzle */
 263    nir_ssa_def *ycbcr_comp[5] = { zero, zero, zero, one, zero };
 264
 265    uint8_t ycbcr_bpcs[5];
 266    memset(ycbcr_bpcs, y_bpc, sizeof(ycbcr_bpcs));
 267
 268    /* Go through all the planes and gather the samples into a |ycbcr_comp|
 269     * while applying a swizzle required by the spec:
 270     *
 271     *    R, G, B should respectively map to Cr, Y, Cb
 272     */
 273    for (uint32_t p = 0; p < format->n_planes; p++) {
 274       const struct anv_format_plane *plane_format = &format->planes[p];
 275       nir_ssa_def *plane_sample = create_plane_tex_instr_implicit(&state, p);
 276
 277       for (uint32_t pc = 0; pc < 4; pc++) {
 278          enum isl_channel_select ycbcr_swizzle =
 279             swizzle_channel(plane_format->ycbcr_swizzle, pc);
 280          if (ycbcr_swizzle == ISL_CHANNEL_SELECT_ZERO)
 281             continue;
 282
 283          unsigned ycbcr_component = channel_to_component(ycbcr_swizzle);
 284          ycbcr_comp[ycbcr_component] = nir_channel(builder, plane_sample, pc);
 285
 286          /* Also compute the number of bits for each component. */
 287          const struct isl_format_layout *isl_layout =
 288             isl_format_get_layout(plane_format->isl_format);
 289          ycbcr_bpcs[ycbcr_component] = isl_layout->channels_array[pc].bits;
 290       }
 291    }
 292
 293    /* Now remaps components to the order specified by the conversion. */
 294    nir_ssa_def *swizzled_comp[4] = { NULL, };
 295    uint32_t swizzled_bpcs[4] = { 0, };
 296
 297    for (uint32_t i = 0; i < ARRAY_SIZE(state.conversion->mapping); i++) {
 298       /* Maps to components in |ycbcr_comp| */
 299       static const uint32_t swizzle_mapping[] = {
 300          [VK_COMPONENT_SWIZZLE_ZERO] = 4,
 301          [VK_COMPONENT_SWIZZLE_ONE]  = 3,
 302          [VK_COMPONENT_SWIZZLE_R]    = 0,
 303          [VK_COMPONENT_SWIZZLE_G]    = 1,
 304          [VK_COMPONENT_SWIZZLE_B]    = 2,
 305          [VK_COMPONENT_SWIZZLE_A]    = 3,
 306       };
 307       const VkComponentSwizzle m = state.conversion->mapping[i];
 308
 309       if (m == VK_COMPONENT_SWIZZLE_IDENTITY) {
 310          swizzled_comp[i] = ycbcr_comp[i];
 311          swizzled_bpcs[i] = ycbcr_bpcs[i];
 312       } else {
 313          swizzled_comp[i] = ycbcr_comp[swizzle_mapping[m]];
 314          swizzled_bpcs[i] = ycbcr_bpcs[swizzle_mapping[m]];
 315       }
 316    }
 317
 318    nir_ssa_def *result = nir_vec(builder, swizzled_comp, 4);
 319    if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) {
 320       result = nir_convert_ycbcr_to_rgb(builder,
 321                                         state.conversion->ycbcr_model,
 322                                         state.conversion->ycbcr_range,
 323                                         result,
 324                                         swizzled_bpcs);
 325    }
 326
 327    nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
 328    nir_instr_remove(&tex->instr);
 329
 330    return true;
 331 }
 332
 333 bool
 334 anv_nir_lower_ycbcr_textures(nir_shader *shader,
 335                              const struct anv_pipeline_layout *layout)
 336 {
 337    bool progress = false;
 338
 339    nir_foreach_function(function, shader) {
 340       if (!function->impl)
 341          continue;
 342
 343       bool function_progress = false;
 344       nir_builder builder;
 345       nir_builder_init(&builder, function->impl);
 346
 347       nir_foreach_block(block, function->impl) {
 348          nir_foreach_instr_safe(instr, block) {
 349             if (instr->type != nir_instr_type_tex)
 350                continue;
 351
 352             nir_tex_instr *tex = nir_instr_as_tex(instr);
 353             function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
 354          }
 355       }
 356
 357       if (function_progress) {
 358          nir_metadata_preserve(function->impl,
 359                                nir_metadata_block_index |
 360                                nir_metadata_dominance);
 361       }
 362
 363       progress |= function_progress;
 364    }
 365
 366    return progress;
 367 }