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 = exec_list_length(&impl->locals);
 180    nir_index_vars(shader, impl, nir_var_function_temp);
 181
 182    if (num_locals == 0) {
 183       nir_shader_preserve_all_metadata(shader);
 184       return false;
 185    }
 186
 187    struct var_info *var_infos = ralloc_array(NULL, struct var_info, num_locals);
 188    nir_foreach_variable(var, &impl->locals) {
 189       var_infos[var->index] = (struct var_info) {
 190          .var = var,
 191          .is_constant = true,
 192          .found_read = false,
 193       };
 194    }
 195
 196    nir_metadata_require(impl, nir_metadata_dominance);
 197
 198    /* First, walk through the shader and figure out what variables we can
 199     * lower to the constant blob.
 200     */
 201    nir_foreach_block(block, impl) {
 202       nir_foreach_instr(instr, block) {
 203          if (instr->type != nir_instr_type_intrinsic)
 204             continue;
 205
 206          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 207
 208          bool src_is_const = false;
 209          nir_deref_instr *src_deref = NULL, *dst_deref = NULL;
 210          unsigned writemask = 0;
 211          switch (intrin->intrinsic) {
 212          case nir_intrinsic_store_deref:
 213             dst_deref = nir_src_as_deref(intrin->src[0]);
 214             src_is_const = nir_src_is_const(intrin->src[1]);
 215             writemask = nir_intrinsic_write_mask(intrin);
 216             break;
 217
 218          case nir_intrinsic_load_deref:
 219             src_deref = nir_src_as_deref(intrin->src[0]);
 220             break;
 221
 222          case nir_intrinsic_copy_deref:
 223             assert(!"Lowering of copy_deref with large constants is prohibited");
 224             break;
 225
 226          default:
 227             continue;
 228          }
 229
 230          if (dst_deref && dst_deref->mode == nir_var_function_temp) {
 231             nir_variable *var = nir_deref_instr_get_variable(dst_deref);
 232             assert(var->data.mode == nir_var_function_temp);
 233
 234             struct var_info *info = &var_infos[var->index];
 235             if (!info->is_constant)
 236                continue;
 237
 238             if (!info->block)
 239                info->block = block;
 240
 241             /* We only consider variables constant if they only have constant
 242              * stores, all the stores come before any reads, and all stores
 243              * come from the same block.  We also can't handle indirect stores.
 244              */
 245             if (!src_is_const || info->found_read || block != info->block ||
 246                 nir_deref_instr_has_indirect(dst_deref)) {
 247                info->is_constant = false;
 248             } else {
 249                nir_const_value *val = nir_src_as_const_value(intrin->src[1]);
 250                handle_constant_store(var_infos, info, dst_deref, val, writemask,
 251                                      size_align);
 252             }
 253          }
 254
 255          if (src_deref && src_deref->mode == nir_var_function_temp) {
 256             nir_variable *var = nir_deref_instr_get_variable(src_deref);
 257             assert(var->data.mode == nir_var_function_temp);
 258
 259             /* We only consider variables constant if all the reads are
 260              * dominated by the block that writes to it.
 261              */
 262             struct var_info *info = &var_infos[var->index];
 263             if (!info->is_constant)
 264                continue;
 265
 266             if (!info->block || !nir_block_dominates(info->block, block))
 267                info->is_constant = false;
 268
 269             info->found_read = true;
 270          }
 271       }
 272    }
 273
 274    /* Allocate constant data space for each variable that just has constant
 275     * data.  We sort them by size and content so we can easily find
 276     * duplicates.
 277     */
 278    shader->constant_data_size = 0;
 279    qsort(var_infos, num_locals, sizeof(struct var_info), var_info_cmp);
 280    for (int i = 0; i < num_locals; i++) {
 281       struct var_info *info = &var_infos[i];
 282
 283       /* Fix up indices after we sorted. */
 284       info->var->index = i;
 285
 286       if (!info->is_constant)
 287          continue;
 288
 289       unsigned var_size, var_align;
 290       size_align(info->var->type, &var_size, &var_align);
 291       if (var_size <= threshold || !info->found_read) {
 292          /* Don't bother lowering small stuff or data that's never read */
 293          info->is_constant = false;
 294          continue;
 295       }
 296
 297       if (i > 0 && var_info_cmp(info, &var_infos[i - 1]) == 0) {
 298          info->var->data.location = var_infos[i - 1].var->data.location;
 299          info->duplicate = true;
 300       } else {
 301          info->var->data.location = ALIGN_POT(shader->constant_data_size, var_align);
 302          shader->constant_data_size = info->var->data.location + var_size;
 303       }
 304    }
 305
 306    if (shader->constant_data_size == 0) {
 307       nir_shader_preserve_all_metadata(shader);
 308       ralloc_free(var_infos);
 309       return false;
 310    }
 311
 312    shader->constant_data = rzalloc_size(shader, shader->constant_data_size);
 313    for (int i = 0; i < num_locals; i++) {
 314       struct var_info *info = &var_infos[i];
 315       if (!info->duplicate && info->is_constant) {
 316          memcpy((char *)shader->constant_data + info->var->data.location,
 317                 info->constant_data, info->constant_data_size);
 318       }
 319    }
 320
 321    nir_builder b;
 322    nir_builder_init(&b, impl);
 323
 324    nir_foreach_block(block, impl) {
 325       nir_foreach_instr_safe(instr, block) {
 326          if (instr->type != nir_instr_type_intrinsic)
 327             continue;
 328
 329          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 330
 331          switch (intrin->intrinsic) {
 332          case nir_intrinsic_load_deref: {
 333             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 334             if (deref->mode != nir_var_function_temp)
 335                continue;
 336
 337             nir_variable *var = nir_deref_instr_get_variable(deref);
 338             struct var_info *info = &var_infos[var->index];
 339             if (info->is_constant) {
 340                b.cursor = nir_after_instr(&intrin->instr);
 341                nir_ssa_def *val = build_constant_load(&b, deref, size_align);
 342                nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 343                                         nir_src_for_ssa(val));
 344                nir_instr_remove(&intrin->instr);
 345                nir_deref_instr_remove_if_unused(deref);
 346             }
 347             break;
 348          }
 349
 350          case nir_intrinsic_store_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             struct var_info *info = &var_infos[var->index];
 357             if (info->is_constant) {
 358                nir_instr_remove(&intrin->instr);
 359                nir_deref_instr_remove_if_unused(deref);
 360             }
 361             break;
 362          }
 363          case nir_intrinsic_copy_deref:
 364          default:
 365             continue;
 366          }
 367       }
 368    }
 369
 370    /* Clean up the now unused variables */
 371    for (int i = 0; i < num_locals; i++) {
 372       struct var_info *info = &var_infos[i];
 373       if (info->is_constant)
 374          exec_node_remove(&info->var->node);
 375    }
 376
 377    ralloc_free(var_infos);
 378
 379    nir_metadata_preserve(impl, nir_metadata_block_index |
 380                                nir_metadata_dominance);
 381    return true;
 382 }