src/glsl/nir/nir_opt_constant_folding.c

   1 /*
   2  * Copyright © 2014 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  * Authors:
  24  *    Jason Ekstrand (jason@jlekstrand.net)
  25  *
  26  */
  27
  28 #include "nir.h"
  29 #include <math.h>
  30
  31 /*
  32  * Implements SSA-based constant folding.
  33  */
  34
  35 struct constant_fold_state {
  36    void *mem_ctx;
  37    nir_function_impl *impl;
  38    bool progress;
  39 };
  40
  41 #define SRC_COMP(T, IDX, CMP) src[IDX]->value.T[instr->src[IDX].swizzle[CMP]]
  42 #define SRC(T, IDX) SRC_COMP(T, IDX, i)
  43 #define DEST_COMP(T, CMP) dest->value.T[CMP]
  44 #define DEST(T) DEST_COMP(T, i)
  45
  46 #define FOLD_PER_COMP(EXPR) \
  47    for (unsigned i = 0; i < instr->dest.dest.ssa.num_components; i++) { \
  48       EXPR; \
  49    } \
  50
  51 static bool
  52 constant_fold_alu_instr(nir_alu_instr *instr, void *mem_ctx)
  53 {
  54    nir_load_const_instr *src[4], *dest;
  55
  56    if (!instr->dest.dest.is_ssa)
  57       return false;
  58
  59    for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
  60       if (!instr->src[i].src.is_ssa)
  61          return false;
  62
  63       if (instr->src[i].src.ssa->parent_instr->type != nir_instr_type_load_const)
  64          return false;
  65
  66       /* We shouldn't have any source modifiers in the optimization loop. */
  67       assert(!instr->src[i].abs && !instr->src[i].negate);
  68
  69       src[i] = nir_instr_as_load_const(instr->src[i].src.ssa->parent_instr);
  70    }
  71
  72    /* We shouldn't have any saturate modifiers in the optimization loop. */
  73    assert(!instr->dest.saturate);
  74
  75    dest = nir_load_const_instr_create(mem_ctx,
  76                                       instr->dest.dest.ssa.num_components);
  77
  78    switch (instr->op) {
  79    case nir_op_ineg:
  80       FOLD_PER_COMP(DEST(i) = -SRC(i, 0));
  81       break;
  82    case nir_op_fneg:
  83       FOLD_PER_COMP(DEST(f) = -SRC(f, 0));
  84       break;
  85    case nir_op_inot:
  86       FOLD_PER_COMP(DEST(i) = ~SRC(i, 0));
  87       break;
  88    case nir_op_fnot:
  89       FOLD_PER_COMP(DEST(f) = (SRC(f, 0) == 0.0f) ? 1.0f : 0.0f);
  90       break;
  91    case nir_op_frcp:
  92       FOLD_PER_COMP(DEST(f) = 1.0f / SRC(f, 0));
  93       break;
  94    case nir_op_frsq:
  95       FOLD_PER_COMP(DEST(f) = 1.0f / sqrt(SRC(f, 0)));
  96       break;
  97    case nir_op_fsqrt:
  98       FOLD_PER_COMP(DEST(f) = sqrtf(SRC(f, 0)));
  99       break;
 100    case nir_op_fexp:
 101       FOLD_PER_COMP(DEST(f) = expf(SRC(f, 0)));
 102       break;
 103    case nir_op_flog:
 104       FOLD_PER_COMP(DEST(f) = logf(SRC(f, 0)));
 105       break;
 106    case nir_op_fexp2:
 107       FOLD_PER_COMP(DEST(f) = exp2f(SRC(f, 0)));
 108       break;
 109    case nir_op_flog2:
 110       FOLD_PER_COMP(DEST(f) = log2f(SRC(f, 0)));
 111       break;
 112    case nir_op_f2i:
 113       FOLD_PER_COMP(DEST(i) = SRC(f, 0));
 114       break;
 115    case nir_op_f2u:
 116       FOLD_PER_COMP(DEST(u) = SRC(f, 0));
 117       break;
 118    case nir_op_i2f:
 119       FOLD_PER_COMP(DEST(f) = SRC(i, 0));
 120       break;
 121    case nir_op_f2b:
 122       FOLD_PER_COMP(DEST(u) = (SRC(i, 0) == 0.0f) ? NIR_FALSE : NIR_TRUE);
 123       break;
 124    case nir_op_b2f:
 125       FOLD_PER_COMP(DEST(f) = SRC(u, 0) ? 1.0f : 0.0f);
 126       break;
 127    case nir_op_i2b:
 128       FOLD_PER_COMP(DEST(u) = SRC(i, 0) ? NIR_TRUE : NIR_FALSE);
 129       break;
 130    case nir_op_u2f:
 131       FOLD_PER_COMP(DEST(f) = SRC(u, 0));
 132       break;
 133    case nir_op_bany2:
 134       DEST_COMP(u, 0) = (SRC_COMP(u, 0, 0) || SRC_COMP(u, 0, 1)) ?
 135                         NIR_TRUE : NIR_FALSE;
 136       break;
 137    case nir_op_fadd:
 138       FOLD_PER_COMP(DEST(f) = SRC(f, 0) + SRC(f, 1));
 139       break;
 140    case nir_op_iadd:
 141       FOLD_PER_COMP(DEST(i) = SRC(i, 0) + SRC(i, 1));
 142       break;
 143    case nir_op_fsub:
 144       FOLD_PER_COMP(DEST(f) = SRC(f, 0) - SRC(f, 1));
 145       break;
 146    case nir_op_isub:
 147       FOLD_PER_COMP(DEST(i) = SRC(i, 0) - SRC(i, 1));
 148       break;
 149    case nir_op_fmul:
 150       FOLD_PER_COMP(DEST(f) = SRC(f, 0) * SRC(f, 1));
 151       break;
 152    case nir_op_imul:
 153       FOLD_PER_COMP(DEST(i) = SRC(i, 0) * SRC(i, 1));
 154       break;
 155    case nir_op_fdiv:
 156       FOLD_PER_COMP(DEST(f) = SRC(f, 0) / SRC(f, 1));
 157       break;
 158    case nir_op_idiv:
 159       FOLD_PER_COMP(DEST(i) = SRC(i, 0) / SRC(i, 1));
 160       break;
 161    case nir_op_udiv:
 162       FOLD_PER_COMP(DEST(u) = SRC(u, 0) / SRC(u, 1));
 163       break;
 164    case nir_op_flt:
 165       FOLD_PER_COMP(DEST(u) = (SRC(f, 0) < SRC(f, 1)) ? NIR_TRUE : NIR_FALSE);
 166       break;
 167    case nir_op_fge:
 168       FOLD_PER_COMP(DEST(u) = (SRC(f, 0) >= SRC(f, 1)) ? NIR_TRUE : NIR_FALSE);
 169       break;
 170    case nir_op_feq:
 171       FOLD_PER_COMP(DEST(u) = (SRC(f, 0) == SRC(f, 1)) ? NIR_TRUE : NIR_FALSE);
 172       break;
 173    case nir_op_fne:
 174       FOLD_PER_COMP(DEST(u) = (SRC(f, 0) != SRC(f, 1)) ? NIR_TRUE : NIR_FALSE);
 175       break;
 176    case nir_op_ilt:
 177       FOLD_PER_COMP(DEST(u) = (SRC(i, 0) < SRC(i, 1)) ? NIR_TRUE : NIR_FALSE);
 178       break;
 179    case nir_op_ige:
 180       FOLD_PER_COMP(DEST(u) = (SRC(i, 0) >= SRC(i, 1)) ? NIR_TRUE : NIR_FALSE);
 181       break;
 182    case nir_op_ieq:
 183       FOLD_PER_COMP(DEST(u) = (SRC(i, 0) == SRC(i, 1)) ? NIR_TRUE : NIR_FALSE);
 184       break;
 185    case nir_op_ine:
 186       FOLD_PER_COMP(DEST(u) = (SRC(i, 0) != SRC(i, 1)) ? NIR_TRUE : NIR_FALSE);
 187       break;
 188    case nir_op_ult:
 189       FOLD_PER_COMP(DEST(u) = (SRC(u, 0) < SRC(u, 1)) ? NIR_TRUE : NIR_FALSE);
 190       break;
 191    case nir_op_uge:
 192       FOLD_PER_COMP(DEST(u) = (SRC(u, 0) >= SRC(u, 1)) ? NIR_TRUE : NIR_FALSE);
 193       break;
 194    case nir_op_ishl:
 195       FOLD_PER_COMP(DEST(i) = SRC(i, 0) << SRC(i, 1));
 196       break;
 197    case nir_op_ishr:
 198       FOLD_PER_COMP(DEST(i) = SRC(i, 0) >> SRC(i, 1));
 199       break;
 200    case nir_op_ushr:
 201       FOLD_PER_COMP(DEST(u) = SRC(u, 0) >> SRC(u, 1));
 202       break;
 203    case nir_op_iand:
 204       FOLD_PER_COMP(DEST(i) = SRC(i, 0) & SRC(i, 1));
 205       break;
 206    case nir_op_ior:
 207       FOLD_PER_COMP(DEST(i) = SRC(i, 0) | SRC(i, 1));
 208       break;
 209    case nir_op_ixor:
 210       FOLD_PER_COMP(DEST(i) = SRC(i, 0) ^ SRC(i, 1));
 211       break;
 212    default:
 213       ralloc_free(dest);
 214       return false;
 215    }
 216
 217    nir_instr_insert_before(&instr->instr, &dest->instr);
 218
 219    nir_src new_src = {
 220       .is_ssa = true,
 221       .ssa = &dest->def,
 222    };
 223
 224    nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, new_src, mem_ctx);
 225
 226    nir_instr_remove(&instr->instr);
 227    ralloc_free(instr);
 228
 229    return true;
 230 }
 231
 232 static bool
 233 constant_fold_deref(nir_instr *instr, nir_deref_var *deref)
 234 {
 235    bool progress = false;
 236
 237    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 238       if (tail->deref_type != nir_deref_type_array)
 239          continue;
 240
 241       nir_deref_array *arr = nir_deref_as_array(tail);
 242
 243       if (arr->deref_array_type == nir_deref_array_type_indirect &&
 244           arr->indirect.is_ssa &&
 245           arr->indirect.ssa->parent_instr->type == nir_instr_type_load_const) {
 246          nir_load_const_instr *indirect =
 247             nir_instr_as_load_const(arr->indirect.ssa->parent_instr);
 248
 249          arr->base_offset += indirect->value.u[0];
 250
 251          nir_src empty = {
 252             .is_ssa = true,
 253             .ssa = NULL,
 254          };
 255
 256          nir_instr_rewrite_src(instr, &arr->indirect, empty);
 257
 258          arr->deref_array_type = nir_deref_array_type_direct;
 259
 260          progress = true;
 261       }
 262    }
 263
 264    return progress;
 265 }
 266
 267 static bool
 268 constant_fold_intrinsic_instr(nir_intrinsic_instr *instr)
 269 {
 270    bool progress = false;
 271
 272    unsigned num_vars = nir_intrinsic_infos[instr->intrinsic].num_variables;
 273    for (unsigned i = 0; i < num_vars; i++) {
 274       progress |= constant_fold_deref(&instr->instr, instr->variables[i]);
 275    }
 276
 277    return progress;
 278 }
 279
 280 static bool
 281 constant_fold_tex_instr(nir_tex_instr *instr)
 282 {
 283    if (instr->sampler)
 284       return constant_fold_deref(&instr->instr, instr->sampler);
 285    else
 286       return false;
 287 }
 288
 289 static bool
 290 constant_fold_block(nir_block *block, void *void_state)
 291 {
 292    struct constant_fold_state *state = void_state;
 293
 294    nir_foreach_instr_safe(block, instr) {
 295       switch (instr->type) {
 296       case nir_instr_type_alu:
 297          state->progress |= constant_fold_alu_instr(nir_instr_as_alu(instr),
 298                                                     state->mem_ctx);
 299          break;
 300       case nir_instr_type_intrinsic:
 301          state->progress |=
 302             constant_fold_intrinsic_instr(nir_instr_as_intrinsic(instr));
 303          break;
 304       case nir_instr_type_tex:
 305          state->progress |= constant_fold_tex_instr(nir_instr_as_tex(instr));
 306          break;
 307       default:
 308          /* Don't know how to constant fold */
 309          break;
 310       }
 311    }
 312
 313    return true;
 314 }
 315
 316 static bool
 317 nir_opt_constant_folding_impl(nir_function_impl *impl)
 318 {
 319    struct constant_fold_state state;
 320
 321    state.mem_ctx = ralloc_parent(impl);
 322    state.impl = impl;
 323    state.progress = false;
 324
 325    nir_foreach_block(impl, constant_fold_block, &state);
 326
 327    if (state.progress)
 328       nir_metadata_preserve(impl, nir_metadata_block_index |
 329                                   nir_metadata_dominance);
 330
 331    return state.progress;
 332 }
 333
 334 bool
 335 nir_opt_constant_folding(nir_shader *shader)
 336 {
 337    bool progress = false;
 338
 339    nir_foreach_overload(shader, overload) {
 340       if (overload->impl)
 341          progress |= nir_opt_constant_folding_impl(overload->impl);
 342    }
 343
 344    return progress;
 345 }