src/compiler/nir/nir_opt_large_constants.c

   1 /*
   2  * Copyright © 2018 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 "nir.h"
  25 #include "nir_builder.h"
  26 #include "nir_deref.h"
  27
  28 struct var_info {
  29    nir_variable *var;
  30
  31    bool is_constant;
  32    bool found_read;
  33    bool duplicate;
  34
  35    /* Block that has all the variable stores.  All the blocks with reads
  36     * should be dominated by this block.
  37     */
  38    nir_block *block;
  39
  40    /* If is_constant, hold the collected constant data for this var. */
  41    uint32_t constant_data_size;
  42    void *constant_data;
  43 };
  44
  45 static int
  46 var_info_cmp(const void *_a, const void *_b)
  47 {
  48    const struct var_info *a = _a;
  49    const struct var_info *b = _b;
  50    uint32_t a_size = a->constant_data_size;
  51    uint32_t b_size = b->constant_data_size;
  52
  53    if (a_size < b_size) {
  54       return -1;
  55    } else if (a_size > b_size) {
  56       return 1;
  57    } else if (a_size == 0) {
  58       /* Don't call memcmp with invalid pointers. */
  59       return 0;
  60    } else {
  61       return memcmp(a->constant_data, b->constant_data, a_size);
  62    }
  63 }
  64
  65 static nir_ssa_def *
  66 build_constant_load(nir_builder *b, nir_deref_instr *deref,
  67                     glsl_type_size_align_func size_align)
  68 {
  69    nir_variable *var = nir_deref_instr_get_variable(deref);
  70
  71    const unsigned bit_size = glsl_get_bit_size(deref->type);
  72    const unsigned num_components = glsl_get_vector_elements(deref->type);
  73
  74    UNUSED unsigned var_size, var_align;
  75    size_align(var->type, &var_size, &var_align);
  76    assert(var->data.location % var_align == 0);
  77
  78    UNUSED unsigned deref_size, deref_align;
  79    size_align(deref->type, &deref_size, &deref_align);
  80
  81    nir_intrinsic_instr *load =
  82       nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_constant);
  83    load->num_components = num_components;
  84    nir_intrinsic_set_base(load, var->data.location);
  85    nir_intrinsic_set_range(load, var_size);
  86    nir_intrinsic_set_align(load, deref_align, 0);
  87    load->src[0] = nir_src_for_ssa(nir_build_deref_offset(b, deref, size_align));
  88    nir_ssa_dest_init(&load->instr, &load->dest,
  89                      num_components, bit_size, NULL);
  90    nir_builder_instr_insert(b, &load->instr);
  91
  92    if (load->dest.ssa.bit_size < 8) {
  93       /* Booleans are special-cased to be 32-bit */
  94       assert(glsl_type_is_boolean(deref->type));
  95       assert(deref_size == num_components * 4);
  96       load->dest.ssa.bit_size = 32;
  97       return nir_b2b1(b, &load->dest.ssa);
  98    } else {
  99       assert(deref_size == num_components * bit_size / 8);
 100       return &load->dest.ssa;
 101    }
 102 }
 103
 104 static void
 105 handle_constant_store(void *mem_ctx, struct var_info *info,
 106                       nir_deref_instr *deref, nir_const_value *val,
 107                       unsigned writemask,
 108                       glsl_type_size_align_func size_align)
 109 {
 110    assert(!nir_deref_instr_has_indirect(deref));
 111    const unsigned bit_size = glsl_get_bit_size(deref->type);
 112    const unsigned num_components = glsl_get_vector_elements(deref->type);
 113
 114    if (info->constant_data_size == 0) {
 115       unsigned var_size, var_align;
 116       size_align(info->var->type, &var_size, &var_align);
 117       info->constant_data_size = var_size;
 118       info->constant_data = rzalloc_size(mem_ctx, var_size);
 119    }
 120
 121    char *dst = (char *)info->constant_data +
 122                nir_deref_instr_get_const_offset(deref, size_align);
 123
 124    for (unsigned i = 0; i < num_components; i++) {
 125       if (!(writemask & (1 << i)))
 126          continue;
 127
 128       switch (bit_size) {
 129       case 1:
 130          /* Booleans are special-cased to be 32-bit */
 131          ((int32_t *)dst)[i] = -(int)val[i].b;
 132          break;
 133
 134       case 8:
 135          ((uint8_t *)dst)[i] = val[i].u8;
 136          break;
 137
 138       case 16:
 139          ((uint16_t *)dst)[i] = val[i].u16;
 140          break;
 141
 142       case 32:
 143          ((uint32_t *)dst)[i] = val[i].u32;
 144          break;
 145
 146       case 64:
 147          ((uint64_t *)dst)[i] = val[i].u64;
 148          break;
 149
 150       default:
 151          unreachable("Invalid bit size");
 152       }
 153    }
 154 }
 155
 156 /** Lower large constant variables to shader constant data
 157  *
 158  * This pass looks for large (type_size(var->type) > threshold) variables
 159  * which are statically constant and moves them into shader constant data.
 160  * This is especially useful when large tables are baked into the shader
 161  * source code because they can be moved into a UBO by the driver to reduce
 162  * register pressure and make indirect access cheaper.
 163  */
 164 bool
 165 nir_opt_large_constants(nir_shader *shader,
 166                         glsl_type_size_align_func size_align,
 167                         unsigned threshold)
 168 {
 169    /* Default to a natural alignment if none is provided */
 170    if (size_align == NULL)
 171       size_align = glsl_get_natural_size_align_bytes;
 172
 173    /* This only works with a single entrypoint */
 174    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 175
 176    /* This pass can only be run once */
 177    assert(shader->constant_data == NULL && shader->constant_data_size == 0);
 178
 179    unsigned num_locals = nir_function_impl_index_vars(impl);
 180
 181    if (num_locals == 0) {
 182       nir_shader_preserve_all_metadata(shader);
 183       return false;
 184    }
 185
 186    struct var_info *var_infos = ralloc_array(NULL, struct var_info, num_locals);
 187    nir_foreach_function_temp_variable(var, impl) {
 188       var_infos[var->index] = (struct var_info) {
 189          .var = var,
 190          .is_constant = true,
 191          .found_read = false,
 192       };
 193    }
 194
 195    nir_metadata_require(impl, nir_metadata_dominance);
 196
 197    /* First, walk through the shader and figure out what variables we can
 198     * lower to the constant blob.
 199     */
 200    nir_foreach_block(block, impl) {
 201       nir_foreach_instr(instr, block) {
 202          if (instr->type == nir_instr_type_deref) {
 203             /* If we ever see a complex use of a deref_var, we have to assume
 204              * that variable is non-constant because we can't guarantee we
 205              * will find all of the writers of that variable.
 206              */
 207             nir_deref_instr *deref = nir_instr_as_deref(instr);
 208             if (deref->deref_type == nir_deref_type_var &&
 209                 deref->mode == nir_var_function_temp &&
 210                 nir_deref_instr_has_complex_use(deref))
 211                var_infos[deref->var->index].is_constant = false;
 212             continue;
 213          }
 214
 215          if (instr->type != nir_instr_type_intrinsic)
 216             continue;
 217
 218          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 219
 220          bool src_is_const = false;
 221          nir_deref_instr *src_deref = NULL, *dst_deref = NULL;
 222          unsigned writemask = 0;
 223          switch (intrin->intrinsic) {
 224          case nir_intrinsic_store_deref:
 225             dst_deref = nir_src_as_deref(intrin->src[0]);
 226             src_is_const = nir_src_is_const(intrin->src[1]);
 227             writemask = nir_intrinsic_write_mask(intrin);
 228             break;
 229
 230          case nir_intrinsic_load_deref:
 231             src_deref = nir_src_as_deref(intrin->src[0]);
 232             break;
 233
 234          case nir_intrinsic_copy_deref:
 235             assert(!"Lowering of copy_deref with large constants is prohibited");
 236             break;
 237
 238          default:
 239             continue;
 240          }
 241
 242          if (dst_deref && dst_deref->mode == nir_var_function_temp) {
 243             nir_variable *var = nir_deref_instr_get_variable(dst_deref);
 244             if (var == NULL)
 245                continue;
 246
 247             assert(var->data.mode == nir_var_function_temp);
 248
 249             struct var_info *info = &var_infos[var->index];
 250             if (!info->is_constant)
 251                continue;
 252
 253             if (!info->block)
 254                info->block = block;
 255
 256             /* We only consider variables constant if they only have constant
 257              * stores, all the stores come before any reads, and all stores
 258              * come from the same block.  We also can't handle indirect stores.
 259              */
 260             if (!src_is_const || info->found_read || block != info->block ||
 261                 nir_deref_instr_has_indirect(dst_deref)) {
 262                info->is_constant = false;
 263             } else {
 264                nir_const_value *val = nir_src_as_const_value(intrin->src[1]);
 265                handle_constant_store(var_infos, info, dst_deref, val, writemask,
 266                                      size_align);
 267             }
 268          }
 269
 270          if (src_deref && src_deref->mode == nir_var_function_temp) {
 271             nir_variable *var = nir_deref_instr_get_variable(src_deref);
 272             if (var == NULL)
 273                continue;
 274
 275             assert(var->data.mode == nir_var_function_temp);
 276
 277             /* We only consider variables constant if all the reads are
 278              * dominated by the block that writes to it.
 279              */
 280             struct var_info *info = &var_infos[var->index];
 281             if (!info->is_constant)
 282                continue;
 283
 284             if (!info->block || !nir_block_dominates(info->block, block))
 285                info->is_constant = false;
 286
 287             info->found_read = true;
 288          }
 289       }
 290    }
 291
 292    /* Allocate constant data space for each variable that just has constant
 293     * data.  We sort them by size and content so we can easily find
 294     * duplicates.
 295     */
 296    shader->constant_data_size = 0;
 297    qsort(var_infos, num_locals, sizeof(struct var_info), var_info_cmp);
 298    for (int i = 0; i < num_locals; i++) {
 299       struct var_info *info = &var_infos[i];
 300
 301       /* Fix up indices after we sorted. */
 302       info->var->index = i;
 303
 304       if (!info->is_constant)
 305          continue;
 306
 307       unsigned var_size, var_align;
 308       size_align(info->var->type, &var_size, &var_align);
 309       if (var_size <= threshold || !info->found_read) {
 310          /* Don't bother lowering small stuff or data that's never read */
 311          info->is_constant = false;
 312          continue;
 313       }
 314
 315       if (i > 0 && var_info_cmp(info, &var_infos[i - 1]) == 0) {
 316          info->var->data.location = var_infos[i - 1].var->data.location;
 317          info->duplicate = true;
 318       } else {
 319          info->var->data.location = ALIGN_POT(shader->constant_data_size, var_align);
 320          shader->constant_data_size = info->var->data.location + var_size;
 321       }
 322    }
 323
 324    if (shader->constant_data_size == 0) {
 325       nir_shader_preserve_all_metadata(shader);
 326       ralloc_free(var_infos);
 327       return false;
 328    }
 329
 330    shader->constant_data = rzalloc_size(shader, shader->constant_data_size);
 331    for (int i = 0; i < num_locals; i++) {
 332       struct var_info *info = &var_infos[i];
 333       if (!info->duplicate && info->is_constant) {
 334          memcpy((char *)shader->constant_data + info->var->data.location,
 335                 info->constant_data, info->constant_data_size);
 336       }
 337    }
 338
 339    nir_builder b;
 340    nir_builder_init(&b, impl);
 341
 342    nir_foreach_block(block, impl) {
 343       nir_foreach_instr_safe(instr, block) {
 344          if (instr->type != nir_instr_type_intrinsic)
 345             continue;
 346
 347          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 348
 349          switch (intrin->intrinsic) {
 350          case nir_intrinsic_load_deref: {
 351             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 352             if (deref->mode != nir_var_function_temp)
 353                continue;
 354
 355             nir_variable *var = nir_deref_instr_get_variable(deref);
 356             if (var == NULL)
 357                continue;
 358
 359             struct var_info *info = &var_infos[var->index];
 360             if (info->is_constant) {
 361                b.cursor = nir_after_instr(&intrin->instr);
 362                nir_ssa_def *val = build_constant_load(&b, deref, size_align);
 363                nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 364                                         nir_src_for_ssa(val));
 365                nir_instr_remove(&intrin->instr);
 366                nir_deref_instr_remove_if_unused(deref);
 367             }
 368             break;
 369          }
 370
 371          case nir_intrinsic_store_deref: {
 372             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 373             if (deref->mode != nir_var_function_temp)
 374                continue;
 375
 376             nir_variable *var = nir_deref_instr_get_variable(deref);
 377             if (var == NULL)
 378                continue;
 379
 380             struct var_info *info = &var_infos[var->index];
 381             if (info->is_constant) {
 382                nir_instr_remove(&intrin->instr);
 383                nir_deref_instr_remove_if_unused(deref);
 384             }
 385             break;
 386          }
 387          case nir_intrinsic_copy_deref:
 388          default:
 389             continue;
 390          }
 391       }
 392    }
 393
 394    /* Clean up the now unused variables */
 395    for (int i = 0; i < num_locals; i++) {
 396       struct var_info *info = &var_infos[i];
 397       if (info->is_constant)
 398          exec_node_remove(&info->var->node);
 399    }
 400
 401    ralloc_free(var_infos);
 402
 403    nir_metadata_preserve(impl, nir_metadata_block_index |
 404                                nir_metadata_dominance);
 405    return true;
 406 }