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    nir_intrinsic_instr *load =
  79       nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_constant);
  80    load->num_components = num_components;
  81    nir_intrinsic_set_base(load, var->data.location);
  82    nir_intrinsic_set_range(load, var_size);
  83    load->src[0] = nir_src_for_ssa(nir_build_deref_offset(b, deref, size_align));
  84    nir_ssa_dest_init(&load->instr, &load->dest,
  85                      num_components, bit_size, NULL);
  86    nir_builder_instr_insert(b, &load->instr);
  87
  88    if (load->dest.ssa.bit_size < 8) {
  89       /* Booleans are special-cased to be 32-bit
  90        *
  91        * Ideally, for drivers that can handle 32-bit booleans, we wouldn't
  92        * emit the i2b here.  However, at this point, the driver is likely to
  93        * still have 1-bit booleans so we need to at least convert bit sizes.
  94        * Unfortunately, we don't have a good way to annotate the load as
  95        * loading a known boolean value so the optimizer isn't going to be
  96        * able to get rid of the conversion.  Some day, we may solve that
  97        * problem but not today.
  98        */
  99       assert(glsl_type_is_boolean(deref->type));
 100       load->dest.ssa.bit_size = 32;
 101       return nir_i2b(b, &load->dest.ssa);
 102    } else {
 103       return &load->dest.ssa;
 104    }
 105 }
 106
 107 static void
 108 handle_constant_store(void *mem_ctx, struct var_info *info,
 109                       nir_deref_instr *deref, nir_const_value *val,
 110                       glsl_type_size_align_func size_align)
 111 {
 112    assert(!nir_deref_instr_has_indirect(deref));
 113    const unsigned bit_size = glsl_get_bit_size(deref->type);
 114    const unsigned num_components = glsl_get_vector_elements(deref->type);
 115
 116    if (info->constant_data_size == 0) {
 117       unsigned var_size, var_align;
 118       size_align(info->var->type, &var_size, &var_align);
 119       info->constant_data_size = var_size;
 120       info->constant_data = rzalloc_size(mem_ctx, var_size);
 121    }
 122
 123    char *dst = (char *)info->constant_data +
 124                nir_deref_instr_get_const_offset(deref, size_align);
 125
 126    switch (bit_size) {
 127    case 1:
 128       /* Booleans are special-cased to be 32-bit */
 129       for (unsigned i = 0; i < num_components; i++)
 130          ((int32_t *)dst)[i] = -(int)val[i].b;
 131       break;
 132
 133    case 8:
 134       for (unsigned i = 0; i < num_components; i++)
 135          ((uint8_t *)dst)[i] = val[i].u8;
 136       break;
 137
 138    case 16:
 139       for (unsigned i = 0; i < num_components; i++)
 140          ((uint16_t *)dst)[i] = val[i].u16;
 141       break;
 142
 143    case 32:
 144       for (unsigned i = 0; i < num_components; i++)
 145          ((uint32_t *)dst)[i] = val[i].u32;
 146       break;
 147
 148    case 64:
 149       for (unsigned i = 0; i < num_components; i++)
 150          ((uint64_t *)dst)[i] = val[i].u64;
 151       break;
 152
 153    default:
 154       unreachable("Invalid bit size");
 155    }
 156 }
 157
 158 /** Lower large constant variables to shader constant data
 159  *
 160  * This pass looks for large (type_size(var->type) > threshold) variables
 161  * which are statically constant and moves them into shader constant data.
 162  * This is especially useful when large tables are baked into the shader
 163  * source code because they can be moved into a UBO by the driver to reduce
 164  * register pressure and make indirect access cheaper.
 165  */
 166 bool
 167 nir_opt_large_constants(nir_shader *shader,
 168                         glsl_type_size_align_func size_align,
 169                         unsigned threshold)
 170 {
 171    /* Default to a natural alignment if none is provided */
 172    if (size_align == NULL)
 173       size_align = glsl_get_natural_size_align_bytes;
 174
 175    /* This only works with a single entrypoint */
 176    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 177
 178    /* This pass can only be run once */
 179    assert(shader->constant_data == NULL && shader->constant_data_size == 0);
 180
 181    /* The index parameter is unused for local variables so we'll use it for
 182     * indexing into our array of variable metadata.
 183     */
 184    unsigned num_locals = 0;
 185    nir_foreach_variable(var, &impl->locals)
 186       var->data.index = num_locals++;
 187
 188    if (num_locals == 0)
 189       return false;
 190
 191    struct var_info *var_infos = ralloc_array(NULL, struct var_info, num_locals);
 192    nir_foreach_variable(var, &impl->locals) {
 193       var_infos[var->data.index] = (struct var_info) {
 194          .var = var,
 195          .is_constant = true,
 196          .found_read = false,
 197       };
 198    }
 199
 200    nir_metadata_require(impl, nir_metadata_dominance);
 201
 202    /* First, walk through the shader and figure out what variables we can
 203     * lower to the constant blob.
 204     */
 205    nir_foreach_block(block, impl) {
 206       nir_foreach_instr(instr, block) {
 207          if (instr->type != nir_instr_type_intrinsic)
 208             continue;
 209
 210          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 211
 212          bool src_is_const = false;
 213          nir_deref_instr *src_deref = NULL, *dst_deref = NULL;
 214          switch (intrin->intrinsic) {
 215          case nir_intrinsic_store_deref:
 216             dst_deref = nir_src_as_deref(intrin->src[0]);
 217             src_is_const = nir_src_is_const(intrin->src[1]);
 218             break;
 219
 220          case nir_intrinsic_load_deref:
 221             src_deref = nir_src_as_deref(intrin->src[0]);
 222             break;
 223
 224          case nir_intrinsic_copy_deref:
 225             /* We always assume the src and therefore the dst are not
 226              * constants here. Copy and constant propagation passes should
 227              * have taken care of this in most cases anyway.
 228              */
 229             dst_deref = nir_src_as_deref(intrin->src[0]);
 230             src_deref = nir_src_as_deref(intrin->src[1]);
 231             src_is_const = false;
 232             break;
 233
 234          default:
 235             continue;
 236          }
 237
 238          if (dst_deref && dst_deref->mode == nir_var_function_temp) {
 239             nir_variable *var = nir_deref_instr_get_variable(dst_deref);
 240             assert(var->data.mode == nir_var_function_temp);
 241
 242             struct var_info *info = &var_infos[var->data.index];
 243             if (!info->is_constant)
 244                continue;
 245
 246             if (!info->block)
 247                info->block = block;
 248
 249             /* We only consider variables constant if they only have constant
 250              * stores, all the stores come before any reads, and all stores
 251              * come from the same block.  We also can't handle indirect stores.
 252              */
 253             if (!src_is_const || info->found_read || block != info->block ||
 254                 nir_deref_instr_has_indirect(dst_deref)) {
 255                info->is_constant = false;
 256             } else {
 257                nir_const_value *val = nir_src_as_const_value(intrin->src[1]);
 258                handle_constant_store(var_infos, info, dst_deref, val, size_align);
 259             }
 260          }
 261
 262          if (src_deref && src_deref->mode == nir_var_function_temp) {
 263             nir_variable *var = nir_deref_instr_get_variable(src_deref);
 264             assert(var->data.mode == nir_var_function_temp);
 265
 266             /* We only consider variables constant if all the reads are
 267              * dominated by the block that writes to it.
 268              */
 269             struct var_info *info = &var_infos[var->data.index];
 270             if (!info->is_constant)
 271                continue;
 272
 273             if (!info->block || !nir_block_dominates(info->block, block))
 274                info->is_constant = false;
 275
 276             info->found_read = true;
 277          }
 278       }
 279    }
 280
 281    /* Allocate constant data space for each variable that just has constant
 282     * data.  We sort them by size and content so we can easily find
 283     * duplicates.
 284     */
 285    shader->constant_data_size = 0;
 286    qsort(var_infos, num_locals, sizeof(struct var_info), var_info_cmp);
 287    for (int i = 0; i < num_locals; i++) {
 288       struct var_info *info = &var_infos[i];
 289
 290       /* Fix up indices after we sorted. */
 291       info->var->data.index = i;
 292
 293       if (!info->is_constant)
 294          continue;
 295
 296       unsigned var_size, var_align;
 297       size_align(info->var->type, &var_size, &var_align);
 298       if (var_size <= threshold || !info->found_read) {
 299          /* Don't bother lowering small stuff or data that's never read */
 300          info->is_constant = false;
 301          continue;
 302       }
 303
 304       if (i > 0 && var_info_cmp(info, &var_infos[i - 1]) == 0) {
 305          info->var->data.location = var_infos[i - 1].var->data.location;
 306          info->duplicate = true;
 307       } else {
 308          info->var->data.location = ALIGN_POT(shader->constant_data_size, var_align);
 309          shader->constant_data_size = info->var->data.location + var_size;
 310       }
 311    }
 312
 313    if (shader->constant_data_size == 0) {
 314       ralloc_free(var_infos);
 315       return false;
 316    }
 317
 318    shader->constant_data = rzalloc_size(shader, shader->constant_data_size);
 319    for (int i = 0; i < num_locals; i++) {
 320       struct var_info *info = &var_infos[i];
 321       if (!info->duplicate) {
 322          memcpy((char *)shader->constant_data + info->var->data.location,
 323                 info->constant_data, info->constant_data_size);
 324       }
 325    }
 326
 327    nir_builder b;
 328    nir_builder_init(&b, impl);
 329
 330    nir_foreach_block(block, impl) {
 331       nir_foreach_instr_safe(instr, block) {
 332          if (instr->type != nir_instr_type_intrinsic)
 333             continue;
 334
 335          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 336
 337          switch (intrin->intrinsic) {
 338          case nir_intrinsic_load_deref: {
 339             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 340             if (deref->mode != nir_var_function_temp)
 341                continue;
 342
 343             nir_variable *var = nir_deref_instr_get_variable(deref);
 344             struct var_info *info = &var_infos[var->data.index];
 345             if (info->is_constant) {
 346                b.cursor = nir_after_instr(&intrin->instr);
 347                nir_ssa_def *val = build_constant_load(&b, deref, size_align);
 348                nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 349                                         nir_src_for_ssa(val));
 350                nir_instr_remove(&intrin->instr);
 351                nir_deref_instr_remove_if_unused(deref);
 352             }
 353             break;
 354          }
 355
 356          case nir_intrinsic_store_deref: {
 357             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 358             if (deref->mode != nir_var_function_temp)
 359                continue;
 360
 361             nir_variable *var = nir_deref_instr_get_variable(deref);
 362             struct var_info *info = &var_infos[var->data.index];
 363             if (info->is_constant) {
 364                nir_instr_remove(&intrin->instr);
 365                nir_deref_instr_remove_if_unused(deref);
 366             }
 367             break;
 368          }
 369
 370          case nir_intrinsic_copy_deref: {
 371             nir_deref_instr *deref = nir_src_as_deref(intrin->src[1]);
 372             if (deref->mode != nir_var_function_temp)
 373                continue;
 374
 375             nir_variable *var = nir_deref_instr_get_variable(deref);
 376             struct var_info *info = &var_infos[var->data.index];
 377             if (info->is_constant) {
 378                b.cursor = nir_after_instr(&intrin->instr);
 379                nir_ssa_def *val = build_constant_load(&b, deref, size_align);
 380                nir_store_deref(&b, nir_src_as_deref(intrin->src[0]), val, ~0);
 381                nir_instr_remove(&intrin->instr);
 382                nir_deref_instr_remove_if_unused(deref);
 383             }
 384             break;
 385          }
 386
 387          default:
 388             continue;
 389          }
 390       }
 391    }
 392
 393    /* Clean up the now unused variables */
 394    for (int i = 0; i < num_locals; i++) {
 395       struct var_info *info = &var_infos[i];
 396       if (info->is_constant)
 397          exec_node_remove(&info->var->node);
 398    }
 399
 400    ralloc_free(var_infos);
 401
 402    nir_metadata_preserve(impl, nir_metadata_block_index |
 403                                nir_metadata_dominance);
 404    return true;
 405 }