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                       unsigned writemask,
 111                       glsl_type_size_align_func size_align)
 112 {
 113    assert(!nir_deref_instr_has_indirect(deref));
 114    const unsigned bit_size = glsl_get_bit_size(deref->type);
 115    const unsigned num_components = glsl_get_vector_elements(deref->type);
 116
 117    if (info->constant_data_size == 0) {
 118       unsigned var_size, var_align;
 119       size_align(info->var->type, &var_size, &var_align);
 120       info->constant_data_size = var_size;
 121       info->constant_data = rzalloc_size(mem_ctx, var_size);
 122    }
 123
 124    char *dst = (char *)info->constant_data +
 125                nir_deref_instr_get_const_offset(deref, size_align);
 126
 127    for (unsigned i = 0; i < num_components; i++) {
 128       if (!(writemask & (1 << i)))
 129          continue;
 130
 131       switch (bit_size) {
 132       case 1:
 133          /* Booleans are special-cased to be 32-bit */
 134          ((int32_t *)dst)[i] = -(int)val[i].b;
 135          break;
 136
 137       case 8:
 138          ((uint8_t *)dst)[i] = val[i].u8;
 139          break;
 140
 141       case 16:
 142          ((uint16_t *)dst)[i] = val[i].u16;
 143          break;
 144
 145       case 32:
 146          ((uint32_t *)dst)[i] = val[i].u32;
 147          break;
 148
 149       case 64:
 150          ((uint64_t *)dst)[i] = val[i].u64;
 151          break;
 152
 153       default:
 154          unreachable("Invalid bit size");
 155       }
 156    }
 157 }
 158
 159 /** Lower large constant variables to shader constant data
 160  *
 161  * This pass looks for large (type_size(var->type) > threshold) variables
 162  * which are statically constant and moves them into shader constant data.
 163  * This is especially useful when large tables are baked into the shader
 164  * source code because they can be moved into a UBO by the driver to reduce
 165  * register pressure and make indirect access cheaper.
 166  */
 167 bool
 168 nir_opt_large_constants(nir_shader *shader,
 169                         glsl_type_size_align_func size_align,
 170                         unsigned threshold)
 171 {
 172    /* Default to a natural alignment if none is provided */
 173    if (size_align == NULL)
 174       size_align = glsl_get_natural_size_align_bytes;
 175
 176    /* This only works with a single entrypoint */
 177    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 178
 179    /* This pass can only be run once */
 180    assert(shader->constant_data == NULL && shader->constant_data_size == 0);
 181
 182    unsigned num_locals = exec_list_length(&impl->locals);
 183    nir_index_vars(shader, impl, nir_var_function_temp);
 184
 185    if (num_locals == 0)
 186       return false;
 187
 188    struct var_info *var_infos = ralloc_array(NULL, struct var_info, num_locals);
 189    nir_foreach_variable(var, &impl->locals) {
 190       var_infos[var->index] = (struct var_info) {
 191          .var = var,
 192          .is_constant = true,
 193          .found_read = false,
 194       };
 195    }
 196
 197    nir_metadata_require(impl, nir_metadata_dominance);
 198
 199    /* First, walk through the shader and figure out what variables we can
 200     * lower to the constant blob.
 201     */
 202    nir_foreach_block(block, impl) {
 203       nir_foreach_instr(instr, block) {
 204          if (instr->type != nir_instr_type_intrinsic)
 205             continue;
 206
 207          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 208
 209          bool src_is_const = false;
 210          nir_deref_instr *src_deref = NULL, *dst_deref = NULL;
 211          unsigned writemask = 0;
 212          switch (intrin->intrinsic) {
 213          case nir_intrinsic_store_deref:
 214             dst_deref = nir_src_as_deref(intrin->src[0]);
 215             src_is_const = nir_src_is_const(intrin->src[1]);
 216             writemask = nir_intrinsic_write_mask(intrin);
 217             break;
 218
 219          case nir_intrinsic_load_deref:
 220             src_deref = nir_src_as_deref(intrin->src[0]);
 221             break;
 222
 223          case nir_intrinsic_copy_deref:
 224             assert(!"Lowering of copy_deref with large constants is prohibited");
 225             break;
 226
 227          default:
 228             continue;
 229          }
 230
 231          if (dst_deref && dst_deref->mode == nir_var_function_temp) {
 232             nir_variable *var = nir_deref_instr_get_variable(dst_deref);
 233             assert(var->data.mode == nir_var_function_temp);
 234
 235             struct var_info *info = &var_infos[var->index];
 236             if (!info->is_constant)
 237                continue;
 238
 239             if (!info->block)
 240                info->block = block;
 241
 242             /* We only consider variables constant if they only have constant
 243              * stores, all the stores come before any reads, and all stores
 244              * come from the same block.  We also can't handle indirect stores.
 245              */
 246             if (!src_is_const || info->found_read || block != info->block ||
 247                 nir_deref_instr_has_indirect(dst_deref)) {
 248                info->is_constant = false;
 249             } else {
 250                nir_const_value *val = nir_src_as_const_value(intrin->src[1]);
 251                handle_constant_store(var_infos, info, dst_deref, val, writemask,
 252                                      size_align);
 253             }
 254          }
 255
 256          if (src_deref && src_deref->mode == nir_var_function_temp) {
 257             nir_variable *var = nir_deref_instr_get_variable(src_deref);
 258             assert(var->data.mode == nir_var_function_temp);
 259
 260             /* We only consider variables constant if all the reads are
 261              * dominated by the block that writes to it.
 262              */
 263             struct var_info *info = &var_infos[var->index];
 264             if (!info->is_constant)
 265                continue;
 266
 267             if (!info->block || !nir_block_dominates(info->block, block))
 268                info->is_constant = false;
 269
 270             info->found_read = true;
 271          }
 272       }
 273    }
 274
 275    /* Allocate constant data space for each variable that just has constant
 276     * data.  We sort them by size and content so we can easily find
 277     * duplicates.
 278     */
 279    shader->constant_data_size = 0;
 280    qsort(var_infos, num_locals, sizeof(struct var_info), var_info_cmp);
 281    for (int i = 0; i < num_locals; i++) {
 282       struct var_info *info = &var_infos[i];
 283
 284       /* Fix up indices after we sorted. */
 285       info->var->index = i;
 286
 287       if (!info->is_constant)
 288          continue;
 289
 290       unsigned var_size, var_align;
 291       size_align(info->var->type, &var_size, &var_align);
 292       if (var_size <= threshold || !info->found_read) {
 293          /* Don't bother lowering small stuff or data that's never read */
 294          info->is_constant = false;
 295          continue;
 296       }
 297
 298       if (i > 0 && var_info_cmp(info, &var_infos[i - 1]) == 0) {
 299          info->var->data.location = var_infos[i - 1].var->data.location;
 300          info->duplicate = true;
 301       } else {
 302          info->var->data.location = ALIGN_POT(shader->constant_data_size, var_align);
 303          shader->constant_data_size = info->var->data.location + var_size;
 304       }
 305    }
 306
 307    if (shader->constant_data_size == 0) {
 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 }