src/compiler/nir/nir_lower_bool_to_bitsize.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
  27 static bool
  28 assert_ssa_def_is_not_1bit(nir_ssa_def *def, UNUSED void *unused)
  29 {
  30    assert(def->bit_size > 1);
  31    return true;
  32 }
  33
  34 static bool
  35 rewrite_1bit_ssa_def_to_32bit(nir_ssa_def *def, void *_progress)
  36 {
  37    bool *progress = _progress;
  38    if (def->bit_size == 1) {
  39       def->bit_size = 32;
  40       *progress = true;
  41    }
  42    return true;
  43 }
  44
  45 static uint32_t
  46 get_bool_convert_opcode(uint32_t dst_bit_size)
  47 {
  48    switch (dst_bit_size) {
  49    case 32: return nir_op_i2i32;
  50    case 16: return nir_op_i2i16;
  51    case 8:  return nir_op_i2i8;
  52    default:
  53       unreachable("invalid boolean bit-size");
  54    }
  55 }
  56
  57 static void
  58 make_sources_canonical(nir_builder *b, nir_alu_instr *alu, uint32_t start_idx)
  59 {
  60    /* TODO: for now we take the bit-size of the first source as the canonical
  61     * form but we could try to be smarter.
  62     */
  63    const nir_op_info *op_info = &nir_op_infos[alu->op];
  64    uint32_t bit_size = nir_src_bit_size(alu->src[start_idx].src);
  65    for (uint32_t i = start_idx + 1; i < op_info->num_inputs; i++) {
  66       if (nir_src_bit_size(alu->src[i].src) != bit_size) {
  67          b->cursor = nir_before_instr(&alu->instr);
  68          nir_op convert_op = get_bool_convert_opcode(bit_size);
  69          nir_ssa_def *new_src =
  70             nir_build_alu(b, convert_op, alu->src[i].src.ssa, NULL, NULL, NULL);
  71          /* Retain the write mask and swizzle of the original instruction so
  72           * that we don’t unnecessarily create a vectorized instruction.
  73           */
  74          nir_alu_instr *conv_instr =
  75             nir_instr_as_alu(nir_builder_last_instr(b));
  76          conv_instr->dest.write_mask = alu->dest.write_mask;
  77          conv_instr->dest.dest.ssa.num_components =
  78             alu->dest.dest.ssa.num_components;
  79          memcpy(conv_instr->src[0].swizzle,
  80                 alu->src[i].swizzle,
  81                 sizeof(conv_instr->src[0].swizzle));
  82          nir_instr_rewrite_src(&alu->instr,
  83                                &alu->src[i].src, nir_src_for_ssa(new_src));
  84          /* The swizzle will have been handled by the conversion instruction
  85           * so we can reset it back to the default
  86           */
  87          for (unsigned j = 0; j < NIR_MAX_VEC_COMPONENTS; j++)
  88             alu->src[i].swizzle[j] = j;
  89       }
  90    }
  91 }
  92
  93 static bool
  94 lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
  95 {
  96    const nir_op_info *op_info = &nir_op_infos[alu->op];
  97
  98    /* For operations that can take multiple boolean sources we need to ensure
  99     * that all booleans have the same bit-size
 100     */
 101    switch (alu->op) {
 102    case nir_op_mov:
 103    case nir_op_vec2:
 104    case nir_op_vec3:
 105    case nir_op_vec4:
 106    case nir_op_inot:
 107    case nir_op_iand:
 108    case nir_op_ior:
 109    case nir_op_ixor:
 110       if (nir_dest_bit_size(alu->dest.dest) > 1)
 111          break; /* Not a boolean instruction */
 112       /* Fallthrough */
 113
 114    case nir_op_ball_fequal2:
 115    case nir_op_ball_fequal3:
 116    case nir_op_ball_fequal4:
 117    case nir_op_bany_fnequal2:
 118    case nir_op_bany_fnequal3:
 119    case nir_op_bany_fnequal4:
 120    case nir_op_ball_iequal2:
 121    case nir_op_ball_iequal3:
 122    case nir_op_ball_iequal4:
 123    case nir_op_bany_inequal2:
 124    case nir_op_bany_inequal3:
 125    case nir_op_bany_inequal4:
 126    case nir_op_ieq:
 127    case nir_op_ine:
 128       make_sources_canonical(b, alu, 0);
 129       break;
 130
 131    case nir_op_bcsel:
 132       /* bcsel may be choosing between boolean sources too */
 133       if (nir_dest_bit_size(alu->dest.dest) == 1)
 134          make_sources_canonical(b, alu, 1);
 135       break;
 136
 137    default:
 138       break;
 139    }
 140
 141    /* Now that we have a canonical boolean bit-size, go on and rewrite the
 142     * instruction to match the canonical bit-size.
 143     */
 144    uint32_t bit_size = nir_src_bit_size(alu->src[0].src);
 145    assert(bit_size > 1);
 146
 147    nir_op opcode = alu->op;
 148    switch (opcode) {
 149    case nir_op_mov:
 150    case nir_op_vec2:
 151    case nir_op_vec3:
 152    case nir_op_vec4:
 153    case nir_op_inot:
 154    case nir_op_iand:
 155    case nir_op_ior:
 156    case nir_op_ixor:
 157       /* Nothing to do here, we do not specialize these opcodes by bit-size */
 158       break;
 159
 160    case nir_op_f2b1:
 161       opcode = bit_size == 8 ? nir_op_f2b8 :
 162                                bit_size == 16 ? nir_op_f2b16 : nir_op_f2b32;
 163       break;
 164
 165    case nir_op_i2b1:
 166       opcode = bit_size == 8 ? nir_op_i2b8 :
 167                                bit_size == 16 ? nir_op_i2b16 : nir_op_i2b32;
 168       break;
 169
 170    case nir_op_b2b1:
 171       /* Since the canonical bit size is the size of the src, it's a no-op */
 172       opcode = nir_op_mov;
 173       break;
 174
 175    case nir_op_b2b32:
 176       /* For up-converting booleans, sign-extend */
 177       opcode = nir_op_i2i32;
 178       break;
 179
 180    case nir_op_flt:
 181       opcode = bit_size == 8 ? nir_op_flt8 :
 182                                bit_size == 16 ? nir_op_flt16 : nir_op_flt32;
 183       break;
 184
 185    case nir_op_fge:
 186       opcode = bit_size == 8 ? nir_op_fge8 :
 187                                bit_size == 16 ? nir_op_fge16 : nir_op_fge32;
 188       break;
 189
 190    case nir_op_feq:
 191       opcode = bit_size == 8 ? nir_op_feq8 :
 192                                bit_size == 16 ? nir_op_feq16 : nir_op_feq32;
 193       break;
 194
 195    case nir_op_fne:
 196       opcode = bit_size == 8 ? nir_op_fne8 :
 197                                bit_size == 16 ? nir_op_fne16 : nir_op_fne32;
 198       break;
 199
 200    case nir_op_ilt:
 201       opcode = bit_size == 8 ? nir_op_ilt8 :
 202                                bit_size == 16 ? nir_op_ilt16 : nir_op_ilt32;
 203       break;
 204
 205    case nir_op_ige:
 206       opcode = bit_size == 8 ? nir_op_ige8 :
 207                                bit_size == 16 ? nir_op_ige16 : nir_op_ige32;
 208       break;
 209
 210    case nir_op_ieq:
 211       opcode = bit_size == 8 ? nir_op_ieq8 :
 212                                bit_size == 16 ? nir_op_ieq16 : nir_op_ieq32;
 213       break;
 214
 215    case nir_op_ine:
 216       opcode = bit_size == 8 ? nir_op_ine8 :
 217                                bit_size == 16 ? nir_op_ine16 : nir_op_ine32;
 218       break;
 219
 220    case nir_op_ult:
 221       opcode = bit_size == 8 ? nir_op_ult8 :
 222                                bit_size == 16 ? nir_op_ult16 : nir_op_ult32;
 223       break;
 224
 225    case nir_op_uge:
 226       opcode = bit_size == 8 ? nir_op_uge8 :
 227                                bit_size == 16 ? nir_op_uge16 : nir_op_uge32;
 228       break;
 229
 230    case nir_op_ball_fequal2:
 231       opcode = bit_size == 8 ? nir_op_b8all_fequal2 :
 232                                bit_size == 16 ? nir_op_b16all_fequal2 :
 233                                                 nir_op_b32all_fequal2;
 234       break;
 235
 236    case nir_op_ball_fequal3:
 237       opcode = bit_size == 8 ? nir_op_b8all_fequal3 :
 238                                bit_size == 16 ? nir_op_b16all_fequal3 :
 239                                                 nir_op_b32all_fequal3;
 240       break;
 241
 242    case nir_op_ball_fequal4:
 243       opcode = bit_size == 8 ? nir_op_b8all_fequal4 :
 244                                bit_size == 16 ? nir_op_b16all_fequal4 :
 245                                                 nir_op_b32all_fequal4;
 246       break;
 247
 248    case nir_op_bany_fnequal2:
 249       opcode = bit_size == 8 ? nir_op_b8any_fnequal2 :
 250                                bit_size == 16 ? nir_op_b16any_fnequal2 :
 251                                                 nir_op_b32any_fnequal2;
 252       break;
 253
 254    case nir_op_bany_fnequal3:
 255       opcode = bit_size == 8 ? nir_op_b8any_fnequal3 :
 256                                bit_size == 16 ? nir_op_b16any_fnequal3 :
 257                                                 nir_op_b32any_fnequal3;
 258       break;
 259
 260    case nir_op_bany_fnequal4:
 261       opcode = bit_size == 8 ? nir_op_b8any_fnequal4 :
 262                                bit_size == 16 ? nir_op_b16any_fnequal4 :
 263                                                 nir_op_b32any_fnequal4;
 264       break;
 265
 266    case nir_op_ball_iequal2:
 267       opcode = bit_size == 8 ? nir_op_b8all_iequal2 :
 268                                bit_size == 16 ? nir_op_b16all_iequal2 :
 269                                                 nir_op_b32all_iequal2;
 270       break;
 271
 272    case nir_op_ball_iequal3:
 273       opcode = bit_size == 8 ? nir_op_b8all_iequal3 :
 274                                bit_size == 16 ? nir_op_b16all_iequal3 :
 275                                                 nir_op_b32all_iequal3;
 276       break;
 277
 278    case nir_op_ball_iequal4:
 279       opcode = bit_size == 8 ? nir_op_b8all_iequal4 :
 280                                bit_size == 16 ? nir_op_b16all_iequal4 :
 281                                                 nir_op_b32all_iequal4;
 282       break;
 283
 284    case nir_op_bany_inequal2:
 285       opcode = bit_size == 8 ? nir_op_b8any_inequal2 :
 286                                bit_size == 16 ? nir_op_b16any_inequal2 :
 287                                                 nir_op_b32any_inequal2;
 288       break;
 289
 290    case nir_op_bany_inequal3:
 291       opcode = bit_size == 8 ? nir_op_b8any_inequal3 :
 292                                bit_size == 16 ? nir_op_b16any_inequal3 :
 293                                                 nir_op_b32any_inequal3;
 294       break;
 295
 296    case nir_op_bany_inequal4:
 297       opcode = bit_size == 8 ? nir_op_b8any_inequal4 :
 298                                bit_size == 16 ? nir_op_b16any_inequal4 :
 299                                                 nir_op_b32any_inequal4;
 300       break;
 301
 302    case nir_op_bcsel:
 303       opcode = bit_size == 8 ? nir_op_b8csel :
 304                                bit_size == 16 ? nir_op_b16csel : nir_op_b32csel;
 305
 306       /* The destination of the selection may have a different bit-size from
 307        * the bcsel condition.
 308        */
 309       bit_size = nir_src_bit_size(alu->src[1].src);
 310       break;
 311
 312    default:
 313       assert(alu->dest.dest.ssa.bit_size > 1);
 314       for (unsigned i = 0; i < op_info->num_inputs; i++)
 315          assert(alu->src[i].src.ssa->bit_size > 1);
 316       return false;
 317    }
 318
 319    alu->op = opcode;
 320
 321    if (alu->dest.dest.ssa.bit_size == 1)
 322       alu->dest.dest.ssa.bit_size = bit_size;
 323
 324    return true;
 325 }
 326
 327 static bool
 328 lower_load_const_instr(nir_load_const_instr *load)
 329 {
 330    bool progress = false;
 331
 332    if (load->def.bit_size > 1)
 333       return progress;
 334
 335    /* TODO: It is not clear if there is any case in which we can ever hit
 336     * this path, so for now we just provide a 32-bit default.
 337     *
 338     * TODO2: after some changed on nir_const_value and other on upstream, we
 339     * removed the initialization of a general value like this:
 340     *   nir_const_value value = load->value
 341     *
 342     * to initialize per value component. Need to confirm if that is correct,
 343     * but look at the TOO before.
 344     */
 345    for (unsigned i = 0; i < load->def.num_components; i++) {
 346       load->value[i].u32 = load->value[i].b ? NIR_TRUE : NIR_FALSE;
 347       load->def.bit_size = 32;
 348       progress = true;
 349    }
 350
 351    return progress;
 352 }
 353
 354 static bool
 355 lower_phi_instr(nir_builder *b, nir_phi_instr *phi)
 356 {
 357    if (nir_dest_bit_size(phi->dest) != 1)
 358       return false;
 359
 360    /* Ensure all phi sources have a canonical bit-size. We choose the
 361     * bit-size of the first phi source as the canonical form.
 362     *
 363     * TODO: maybe we can be smarter about how we choose the canonical form.
 364     */
 365    uint32_t dst_bit_size = 0;
 366    nir_foreach_phi_src(phi_src, phi) {
 367       uint32_t src_bit_size = nir_src_bit_size(phi_src->src);
 368       if (dst_bit_size == 0) {
 369          dst_bit_size = src_bit_size;
 370       } else if (src_bit_size != dst_bit_size) {
 371          assert(phi_src->src.is_ssa);
 372          b->cursor = nir_before_src(&phi_src->src, false);
 373          nir_op convert_op = get_bool_convert_opcode(dst_bit_size);
 374          nir_ssa_def *new_src =
 375             nir_build_alu(b, convert_op, phi_src->src.ssa, NULL, NULL, NULL);
 376          nir_instr_rewrite_src(&phi->instr, &phi_src->src,
 377                                nir_src_for_ssa(new_src));
 378       }
 379    }
 380
 381    phi->dest.ssa.bit_size = dst_bit_size;
 382
 383    return true;
 384 }
 385
 386 static bool
 387 nir_lower_bool_to_bitsize_impl(nir_builder *b, nir_function_impl *impl)
 388 {
 389    bool progress = false;
 390
 391    nir_foreach_block(block, impl) {
 392       nir_foreach_instr_safe(instr, block) {
 393          switch (instr->type) {
 394          case nir_instr_type_alu:
 395             progress |= lower_alu_instr(b, nir_instr_as_alu(instr));
 396             break;
 397
 398          case nir_instr_type_load_const:
 399             progress |= lower_load_const_instr(nir_instr_as_load_const(instr));
 400             break;
 401
 402          case nir_instr_type_phi:
 403             progress |= lower_phi_instr(b, nir_instr_as_phi(instr));
 404             break;
 405
 406          case nir_instr_type_ssa_undef:
 407          case nir_instr_type_intrinsic:
 408          case nir_instr_type_tex:
 409             nir_foreach_ssa_def(instr, rewrite_1bit_ssa_def_to_32bit,
 410                                 &progress);
 411             break;
 412
 413          default:
 414             nir_foreach_ssa_def(instr, assert_ssa_def_is_not_1bit, NULL);
 415          }
 416       }
 417    }
 418
 419    if (progress) {
 420       nir_metadata_preserve(impl, nir_metadata_block_index |
 421                                   nir_metadata_dominance);
 422    }
 423
 424    return progress;
 425 }
 426
 427 bool
 428 nir_lower_bool_to_bitsize(nir_shader *shader)
 429 {
 430    nir_builder b;
 431    bool progress = false;
 432
 433    nir_foreach_function(function, shader) {
 434       if (function->impl) {
 435          nir_builder_init(&b, function->impl);
 436          progress = nir_lower_bool_to_bitsize_impl(&b, function->impl) || progress;
 437       }
 438    }
 439
 440    return progress;
 441 }