src/glsl/nir/nir_opt_copy_propagate.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  *    Connor Abbott (cwabbott0@gmail.com)
  25  *
  26  */
  27
  28 #include "nir.h"
  29 #include <main/imports.h>
  30
  31 /**
  32  * SSA-based copy propagation
  33  */
  34
  35 static bool is_move(nir_alu_instr *instr)
  36 {
  37    if (instr->op != nir_op_fmov &&
  38        instr->op != nir_op_imov)
  39       return false;
  40
  41    if (instr->dest.saturate)
  42       return false;
  43
  44    /* we handle modifiers in a separate pass */
  45
  46    if (instr->src[0].abs || instr->src[0].negate)
  47       return false;
  48
  49    if (!instr->src[0].src.is_ssa)
  50       return false;
  51
  52    return true;
  53
  54 }
  55
  56 static bool is_vec(nir_alu_instr *instr)
  57 {
  58    for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
  59       if (!instr->src[i].src.is_ssa)
  60          return false;
  61
  62    return instr->op == nir_op_vec2 ||
  63           instr->op == nir_op_vec3 ||
  64           instr->op == nir_op_vec4;
  65 }
  66
  67 static bool
  68 is_swizzleless_move(nir_alu_instr *instr)
  69 {
  70    if (is_move(instr)) {
  71       for (unsigned i = 0; i < 4; i++) {
  72          if (!((instr->dest.write_mask >> i) & 1))
  73             break;
  74          if (instr->src[0].swizzle[i] != i)
  75             return false;
  76       }
  77       return true;
  78    } else if (is_vec(instr)) {
  79       nir_ssa_def *def = NULL;
  80       for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
  81          if (instr->src[i].swizzle[0] != i)
  82             return false;
  83
  84          if (def == NULL) {
  85             def = instr->src[i].src.ssa;
  86          } else if (instr->src[i].src.ssa != def) {
  87             return false;
  88          }
  89       }
  90       return true;
  91    } else {
  92       return false;
  93    }
  94 }
  95
  96 typedef struct {
  97    nir_ssa_def *def;
  98    bool found;
  99 } search_def_state;
 100
 101 static bool
 102 search_def(nir_src *src, void *_state)
 103 {
 104    search_def_state *state = (search_def_state *) _state;
 105
 106    if (src->is_ssa && src->ssa == state->def)
 107       state->found = true;
 108
 109    return true;
 110 }
 111
 112 static void
 113 rewrite_src_instr(nir_src *src, nir_ssa_def *new_def, nir_instr *parent_instr)
 114 {
 115    nir_ssa_def *old_def = src->ssa;
 116
 117    src->ssa = new_def;
 118
 119    /*
 120     * The instruction could still use the old definition in one of its other
 121     * sources, so only remove the instruction from the uses if there are no
 122     * more uses left.
 123     */
 124
 125    search_def_state search_state;
 126    search_state.def = old_def;
 127    search_state.found = false;
 128    nir_foreach_src(parent_instr, search_def, &search_state);
 129    if (!search_state.found) {
 130       struct set_entry *entry = _mesa_set_search(old_def->uses, parent_instr);
 131       assert(entry);
 132       _mesa_set_remove(old_def->uses, entry);
 133    }
 134
 135    _mesa_set_add(new_def->uses, parent_instr);
 136 }
 137
 138 static void
 139 rewrite_src_if(nir_if *if_stmt, nir_ssa_def *new_def)
 140 {
 141    nir_ssa_def *old_def = if_stmt->condition.ssa;
 142
 143    if_stmt->condition.ssa = new_def;
 144
 145    struct set_entry *entry = _mesa_set_search(old_def->if_uses, if_stmt);
 146    assert(entry);
 147    _mesa_set_remove(old_def->if_uses, entry);
 148
 149    _mesa_set_add(new_def->if_uses, if_stmt);
 150 }
 151
 152 static bool
 153 copy_prop_src(nir_src *src, nir_instr *parent_instr, nir_if *parent_if)
 154 {
 155    if (!src->is_ssa) {
 156       if (src->reg.indirect)
 157          return copy_prop_src(src, parent_instr, parent_if);
 158       return false;
 159    }
 160
 161    nir_instr *src_instr = src->ssa->parent_instr;
 162    if (src_instr->type != nir_instr_type_alu)
 163       return false;
 164
 165    nir_alu_instr *alu_instr = nir_instr_as_alu(src_instr);
 166    if (!is_swizzleless_move(alu_instr))
 167       return false;
 168
 169    /* Don't let copy propagation land us with a phi that has more
 170     * components in its source than it has in its destination.  That badly
 171     * messes up out-of-ssa.
 172     */
 173    if (parent_instr && parent_instr->type == nir_instr_type_phi) {
 174       nir_phi_instr *phi = nir_instr_as_phi(parent_instr);
 175       assert(phi->dest.is_ssa);
 176       if (phi->dest.ssa.num_components !=
 177           alu_instr->src[0].src.ssa->num_components)
 178          return false;
 179    }
 180
 181    if (parent_instr)
 182       rewrite_src_instr(src, alu_instr->src[0].src.ssa, parent_instr);
 183    else
 184       rewrite_src_if(parent_if, alu_instr->src[0].src.ssa);
 185
 186    return true;
 187 }
 188
 189 static bool
 190 copy_prop_alu_src(nir_alu_instr *parent_alu_instr, unsigned index)
 191 {
 192    nir_alu_src *src = &parent_alu_instr->src[index];
 193    if (!src->src.is_ssa) {
 194       if (src->src.reg.indirect)
 195          return copy_prop_src(src->src.reg.indirect, &parent_alu_instr->instr,
 196                               NULL);
 197       return false;
 198    }
 199
 200    nir_instr *src_instr =  src->src.ssa->parent_instr;
 201    if (src_instr->type != nir_instr_type_alu)
 202       return false;
 203
 204    nir_alu_instr *alu_instr = nir_instr_as_alu(src_instr);
 205    if (!is_move(alu_instr) && !is_vec(alu_instr))
 206       return false;
 207
 208    nir_ssa_def *def;
 209    unsigned new_swizzle[4] = {0, 0, 0, 0};
 210
 211    if (alu_instr->op == nir_op_fmov ||
 212        alu_instr->op == nir_op_imov) {
 213       for (unsigned i = 0; i < 4; i++)
 214          new_swizzle[i] = alu_instr->src[0].swizzle[src->swizzle[i]];
 215       def = alu_instr->src[0].src.ssa;
 216    } else {
 217       def = NULL;
 218
 219       for (unsigned i = 0; i < 4; i++) {
 220          if (!nir_alu_instr_channel_used(parent_alu_instr, index, i))
 221             continue;
 222
 223          nir_ssa_def *new_def = alu_instr->src[src->swizzle[i]].src.ssa;
 224          if (def == NULL)
 225             def = new_def;
 226          else {
 227             if (def != new_def)
 228                return false;
 229          }
 230          new_swizzle[i] = alu_instr->src[src->swizzle[i]].swizzle[0];
 231       }
 232    }
 233
 234    for (unsigned i = 0; i < 4; i++)
 235       src->swizzle[i] = new_swizzle[i];
 236
 237    rewrite_src_instr(&src->src, def, &parent_alu_instr->instr);
 238
 239    return true;
 240 }
 241
 242 typedef struct {
 243    nir_instr *parent_instr;
 244    bool progress;
 245 } copy_prop_state;
 246
 247 static bool
 248 copy_prop_src_cb(nir_src *src, void *_state)
 249 {
 250    copy_prop_state *state = (copy_prop_state *) _state;
 251    while (copy_prop_src(src, state->parent_instr, NULL))
 252       state->progress = true;
 253
 254    return true;
 255 }
 256
 257 static bool
 258 copy_prop_instr(nir_instr *instr)
 259 {
 260    if (instr->type == nir_instr_type_alu) {
 261       nir_alu_instr *alu_instr = nir_instr_as_alu(instr);
 262       bool progress = false;
 263
 264       for (unsigned i = 0; i < nir_op_infos[alu_instr->op].num_inputs; i++)
 265          while (copy_prop_alu_src(alu_instr, i))
 266             progress = true;
 267
 268       if (!alu_instr->dest.dest.is_ssa && alu_instr->dest.dest.reg.indirect)
 269          while (copy_prop_src(alu_instr->dest.dest.reg.indirect, instr, NULL))
 270             progress = true;
 271
 272       return progress;
 273    }
 274
 275    copy_prop_state state;
 276    state.parent_instr = instr;
 277    state.progress = false;
 278    nir_foreach_src(instr, copy_prop_src_cb, &state);
 279
 280    return state.progress;
 281 }
 282
 283 static bool
 284 copy_prop_if(nir_if *if_stmt)
 285 {
 286    return copy_prop_src(&if_stmt->condition, NULL, if_stmt);
 287 }
 288
 289 static bool
 290 copy_prop_block(nir_block *block, void *_state)
 291 {
 292    bool *progress = (bool *) _state;
 293
 294    nir_foreach_instr(block, instr) {
 295       if (copy_prop_instr(instr))
 296          *progress = true;
 297    }
 298
 299    if (block->cf_node.node.next != NULL && /* check that we aren't the end node */
 300        !nir_cf_node_is_last(&block->cf_node) &&
 301        nir_cf_node_next(&block->cf_node)->type == nir_cf_node_if) {
 302       nir_if *if_stmt = nir_cf_node_as_if(nir_cf_node_next(&block->cf_node));
 303       if (copy_prop_if(if_stmt))
 304          *progress = true;
 305    }
 306
 307    return true;
 308 }
 309
 310 bool
 311 nir_copy_prop_impl(nir_function_impl *impl)
 312 {
 313    bool progress = false;
 314
 315    nir_foreach_block(impl, copy_prop_block, &progress);
 316    return progress;
 317 }
 318
 319 bool
 320 nir_copy_prop(nir_shader *shader)
 321 {
 322    bool progress = false;
 323
 324    nir_foreach_overload(shader, overload) {
 325       if (overload->impl && nir_copy_prop_impl(overload->impl))
 326          progress = true;
 327    }
 328
 329    return progress;
 330 }