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       load->dest.ssa.bit_size = 32;
  96       return nir_b2b1(b, &load->dest.ssa);
  97    } else {
  98       return &load->dest.ssa;
  99    }
 100 }
 101
 102 static void
 103 handle_constant_store(void *mem_ctx, struct var_info *info,
 104                       nir_deref_instr *deref, nir_const_value *val,
 105                       unsigned writemask,
 106                       glsl_type_size_align_func size_align)
 107 {
 108    assert(!nir_deref_instr_has_indirect(deref));
 109    const unsigned bit_size = glsl_get_bit_size(deref->type);
 110    const unsigned num_components = glsl_get_vector_elements(deref->type);
 111
 112    if (info->constant_data_size == 0) {
 113       unsigned var_size, var_align;
 114       size_align(info->var->type, &var_size, &var_align);
 115       info->constant_data_size = var_size;
 116       info->constant_data = rzalloc_size(mem_ctx, var_size);
 117    }
 118
 119    char *dst = (char *)info->constant_data +
 120                nir_deref_instr_get_const_offset(deref, size_align);
 121
 122    for (unsigned i = 0; i < num_components; i++) {
 123       if (!(writemask & (1 << i)))
 124          continue;
 125
 126       switch (bit_size) {
 127       case 1:
 128          /* Booleans are special-cased to be 32-bit */
 129          ((int32_t *)dst)[i] = -(int)val[i].b;
 130          break;
 131
 132       case 8:
 133          ((uint8_t *)dst)[i] = val[i].u8;
 134          break;
 135
 136       case 16:
 137          ((uint16_t *)dst)[i] = val[i].u16;
 138          break;
 139
 140       case 32:
 141          ((uint32_t *)dst)[i] = val[i].u32;
 142          break;
 143
 144       case 64:
 145          ((uint64_t *)dst)[i] = val[i].u64;
 146          break;
 147
 148       default:
 149          unreachable("Invalid bit size");
 150       }
 151    }
 152 }
 153
 154 /** Lower large constant variables to shader constant data
 155  *
 156  * This pass looks for large (type_size(var->type) > threshold) variables
 157  * which are statically constant and moves them into shader constant data.
 158  * This is especially useful when large tables are baked into the shader
 159  * source code because they can be moved into a UBO by the driver to reduce
 160  * register pressure and make indirect access cheaper.
 161  */
 162 bool
 163 nir_opt_large_constants(nir_shader *shader,
 164                         glsl_type_size_align_func size_align,
 165                         unsigned threshold)
 166 {
 167    /* Default to a natural alignment if none is provided */
 168    if (size_align == NULL)
 169       size_align = glsl_get_natural_size_align_bytes;
 170
 171    /* This only works with a single entrypoint */
 172    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 173
 174    /* This pass can only be run once */
 175    assert(shader->constant_data == NULL && shader->constant_data_size == 0);
 176
 177    unsigned num_locals = exec_list_length(&impl->locals);
 178    nir_index_vars(shader, impl, nir_var_function_temp);
 179
 180    if (num_locals == 0)
 181       return false;
 182
 183    struct var_info *var_infos = ralloc_array(NULL, struct var_info, num_locals);
 184    nir_foreach_variable(var, &impl->locals) {
 185       var_infos[var->index] = (struct var_info) {
 186          .var = var,
 187          .is_constant = true,
 188          .found_read = false,
 189       };
 190    }
 191
 192    nir_metadata_require(impl, nir_metadata_dominance);
 193
 194    /* First, walk through the shader and figure out what variables we can
 195     * lower to the constant blob.
 196     */
 197    nir_foreach_block(block, impl) {
 198       nir_foreach_instr(instr, block) {
 199          if (instr->type != nir_instr_type_intrinsic)
 200             continue;
 201
 202          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 203
 204          bool src_is_const = false;
 205          nir_deref_instr *src_deref = NULL, *dst_deref = NULL;
 206          unsigned writemask = 0;
 207          switch (intrin->intrinsic) {
 208          case nir_intrinsic_store_deref:
 209             dst_deref = nir_src_as_deref(intrin->src[0]);
 210             src_is_const = nir_src_is_const(intrin->src[1]);
 211             writemask = nir_intrinsic_write_mask(intrin);
 212             break;
 213
 214          case nir_intrinsic_load_deref:
 215             src_deref = nir_src_as_deref(intrin->src[0]);
 216             break;
 217
 218          case nir_intrinsic_copy_deref:
 219             assert(!"Lowering of copy_deref with large constants is prohibited");
 220             break;
 221
 222          default:
 223             continue;
 224          }
 225
 226          if (dst_deref && dst_deref->mode == nir_var_function_temp) {
 227             nir_variable *var = nir_deref_instr_get_variable(dst_deref);
 228             assert(var->data.mode == nir_var_function_temp);
 229
 230             struct var_info *info = &var_infos[var->index];
 231             if (!info->is_constant)
 232                continue;
 233
 234             if (!info->block)
 235                info->block = block;
 236
 237             /* We only consider variables constant if they only have constant
 238              * stores, all the stores come before any reads, and all stores
 239              * come from the same block.  We also can't handle indirect stores.
 240              */
 241             if (!src_is_const || info->found_read || block != info->block ||
 242                 nir_deref_instr_has_indirect(dst_deref)) {
 243                info->is_constant = false;
 244             } else {
 245                nir_const_value *val = nir_src_as_const_value(intrin->src[1]);
 246                handle_constant_store(var_infos, info, dst_deref, val, writemask,
 247                                      size_align);
 248             }
 249          }
 250
 251          if (src_deref && src_deref->mode == nir_var_function_temp) {
 252             nir_variable *var = nir_deref_instr_get_variable(src_deref);
 253             assert(var->data.mode == nir_var_function_temp);
 254
 255             /* We only consider variables constant if all the reads are
 256              * dominated by the block that writes to it.
 257              */
 258             struct var_info *info = &var_infos[var->index];
 259             if (!info->is_constant)
 260                continue;
 261
 262             if (!info->block || !nir_block_dominates(info->block, block))
 263                info->is_constant = false;
 264
 265             info->found_read = true;
 266          }
 267       }
 268    }
 269
 270    /* Allocate constant data space for each variable that just has constant
 271     * data.  We sort them by size and content so we can easily find
 272     * duplicates.
 273     */
 274    shader->constant_data_size = 0;
 275    qsort(var_infos, num_locals, sizeof(struct var_info), var_info_cmp);
 276    for (int i = 0; i < num_locals; i++) {
 277       struct var_info *info = &var_infos[i];
 278
 279       /* Fix up indices after we sorted. */
 280       info->var->index = i;
 281
 282       if (!info->is_constant)
 283          continue;
 284
 285       unsigned var_size, var_align;
 286       size_align(info->var->type, &var_size, &var_align);
 287       if (var_size <= threshold || !info->found_read) {
 288          /* Don't bother lowering small stuff or data that's never read */
 289          info->is_constant = false;
 290          continue;
 291       }
 292
 293       if (i > 0 && var_info_cmp(info, &var_infos[i - 1]) == 0) {
 294          info->var->data.location = var_infos[i - 1].var->data.location;
 295          info->duplicate = true;
 296       } else {
 297          info->var->data.location = ALIGN_POT(shader->constant_data_size, var_align);
 298          shader->constant_data_size = info->var->data.location + var_size;
 299       }
 300    }
 301
 302    if (shader->constant_data_size == 0) {
 303       ralloc_free(var_infos);
 304       return false;
 305    }
 306
 307    shader->constant_data = rzalloc_size(shader, shader->constant_data_size);
 308    for (int i = 0; i < num_locals; i++) {
 309       struct var_info *info = &var_infos[i];
 310       if (!info->duplicate && info->is_constant) {
 311          memcpy((char *)shader->constant_data + info->var->data.location,
 312                 info->constant_data, info->constant_data_size);
 313       }
 314    }
 315
 316    nir_builder b;
 317    nir_builder_init(&b, impl);
 318
 319    nir_foreach_block(block, impl) {
 320       nir_foreach_instr_safe(instr, block) {
 321          if (instr->type != nir_instr_type_intrinsic)
 322             continue;
 323
 324          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 325
 326          switch (intrin->intrinsic) {
 327          case nir_intrinsic_load_deref: {
 328             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 329             if (deref->mode != nir_var_function_temp)
 330                continue;
 331
 332             nir_variable *var = nir_deref_instr_get_variable(deref);
 333             struct var_info *info = &var_infos[var->index];
 334             if (info->is_constant) {
 335                b.cursor = nir_after_instr(&intrin->instr);
 336                nir_ssa_def *val = build_constant_load(&b, deref, size_align);
 337                nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 338                                         nir_src_for_ssa(val));
 339                nir_instr_remove(&intrin->instr);
 340                nir_deref_instr_remove_if_unused(deref);
 341             }
 342             break;
 343          }
 344
 345          case nir_intrinsic_store_deref: {
 346             nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 347             if (deref->mode != nir_var_function_temp)
 348                continue;
 349
 350             nir_variable *var = nir_deref_instr_get_variable(deref);
 351             struct var_info *info = &var_infos[var->index];
 352             if (info->is_constant) {
 353                nir_instr_remove(&intrin->instr);
 354                nir_deref_instr_remove_if_unused(deref);
 355             }
 356             break;
 357          }
 358          case nir_intrinsic_copy_deref:
 359          default:
 360             continue;
 361          }
 362       }
 363    }
 364
 365    /* Clean up the now unused variables */
 366    for (int i = 0; i < num_locals; i++) {
 367       struct var_info *info = &var_infos[i];
 368       if (info->is_constant)
 369          exec_node_remove(&info->var->node);
 370    }
 371
 372    ralloc_free(var_infos);
 373
 374    nir_metadata_preserve(impl, nir_metadata_block_index |
 375                                nir_metadata_dominance);
 376    return true;
 377 }