src/compiler/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       /* we handle modifiers in a separate pass */
  63       if (instr->src[i].abs || instr->src[i].negate)
  64          return false;
  65    }
  66
  67    return instr->op == nir_op_vec2 ||
  68           instr->op == nir_op_vec3 ||
  69           instr->op == nir_op_vec4;
  70 }
  71
  72 static bool
  73 is_swizzleless_move(nir_alu_instr *instr)
  74 {
  75    if (is_move(instr)) {
  76       for (unsigned i = 0; i < 4; i++) {
  77          if (!((instr->dest.write_mask >> i) & 1))
  78             break;
  79          if (instr->src[0].swizzle[i] != i)
  80             return false;
  81       }
  82       return true;
  83    } else if (is_vec(instr)) {
  84       nir_ssa_def *def = NULL;
  85       for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
  86          if (instr->src[i].swizzle[0] != i)
  87             return false;
  88
  89          if (def == NULL) {
  90             def = instr->src[i].src.ssa;
  91          } else if (instr->src[i].src.ssa != def) {
  92             return false;
  93          }
  94       }
  95       return true;
  96    } else {
  97       return false;
  98    }
  99 }
 100
 101 static bool
 102 copy_prop_src(nir_src *src, nir_instr *parent_instr, nir_if *parent_if,
 103               unsigned num_components)
 104 {
 105    if (!src->is_ssa) {
 106       if (src->reg.indirect)
 107          return copy_prop_src(src->reg.indirect, parent_instr, parent_if, 1);
 108       return false;
 109    }
 110
 111    nir_instr *src_instr = src->ssa->parent_instr;
 112    nir_ssa_def *copy_def;
 113    if (src_instr->type == nir_instr_type_alu) {
 114       nir_alu_instr *alu_instr = nir_instr_as_alu(src_instr);
 115       if (!is_swizzleless_move(alu_instr))
 116          return false;
 117
 118       if (alu_instr->src[0].src.ssa->num_components != num_components)
 119          return false;
 120
 121       copy_def= alu_instr->src[0].src.ssa;
 122    } else {
 123       return false;
 124    }
 125
 126    if (parent_instr) {
 127       nir_instr_rewrite_src(parent_instr, src, nir_src_for_ssa(copy_def));
 128    } else {
 129       assert(src == &parent_if->condition);
 130       nir_if_rewrite_condition(parent_if, nir_src_for_ssa(copy_def));
 131    }
 132
 133    return true;
 134 }
 135
 136 static bool
 137 copy_prop_alu_src(nir_alu_instr *parent_alu_instr, unsigned index)
 138 {
 139    nir_alu_src *src = &parent_alu_instr->src[index];
 140    if (!src->src.is_ssa) {
 141       if (src->src.reg.indirect)
 142          return copy_prop_src(src->src.reg.indirect, &parent_alu_instr->instr,
 143                               NULL, 1);
 144       return false;
 145    }
 146
 147    nir_instr *src_instr =  src->src.ssa->parent_instr;
 148    if (src_instr->type != nir_instr_type_alu)
 149       return false;
 150
 151    nir_alu_instr *alu_instr = nir_instr_as_alu(src_instr);
 152    if (!is_move(alu_instr) && !is_vec(alu_instr))
 153       return false;
 154
 155    nir_ssa_def *def;
 156    unsigned new_swizzle[4] = {0, 0, 0, 0};
 157
 158    if (alu_instr->op == nir_op_fmov ||
 159        alu_instr->op == nir_op_imov) {
 160       for (unsigned i = 0; i < 4; i++)
 161          new_swizzle[i] = alu_instr->src[0].swizzle[src->swizzle[i]];
 162       def = alu_instr->src[0].src.ssa;
 163    } else {
 164       def = NULL;
 165
 166       for (unsigned i = 0; i < 4; i++) {
 167          if (!nir_alu_instr_channel_used(parent_alu_instr, index, i))
 168             continue;
 169
 170          nir_ssa_def *new_def = alu_instr->src[src->swizzle[i]].src.ssa;
 171          if (def == NULL)
 172             def = new_def;
 173          else {
 174             if (def != new_def)
 175                return false;
 176          }
 177          new_swizzle[i] = alu_instr->src[src->swizzle[i]].swizzle[0];
 178       }
 179    }
 180
 181    for (unsigned i = 0; i < 4; i++)
 182       src->swizzle[i] = new_swizzle[i];
 183
 184    nir_instr_rewrite_src(&parent_alu_instr->instr, &src->src,
 185                          nir_src_for_ssa(def));
 186
 187    return true;
 188 }
 189
 190 static bool
 191 copy_prop_dest(nir_dest *dest, nir_instr *instr)
 192 {
 193    if (!dest->is_ssa && dest->reg.indirect)
 194       return copy_prop_src(dest->reg.indirect, instr, NULL, 1);
 195
 196    return false;
 197 }
 198
 199 static bool
 200 copy_prop_instr(nir_instr *instr)
 201 {
 202    bool progress = false;
 203    switch (instr->type) {
 204    case nir_instr_type_alu: {
 205       nir_alu_instr *alu_instr = nir_instr_as_alu(instr);
 206
 207       for (unsigned i = 0; i < nir_op_infos[alu_instr->op].num_inputs; i++)
 208          while (copy_prop_alu_src(alu_instr, i))
 209             progress = true;
 210
 211       while (copy_prop_dest(&alu_instr->dest.dest, instr))
 212          progress = true;
 213
 214       return progress;
 215    }
 216
 217    case nir_instr_type_deref: {
 218       nir_deref_instr *deref = nir_instr_as_deref(instr);
 219
 220       if (deref->deref_type != nir_deref_type_var) {
 221          assert(deref->dest.is_ssa);
 222          const unsigned comps = deref->dest.ssa.num_components;
 223          while (copy_prop_src(&deref->parent, instr, NULL, comps))
 224             progress = true;
 225       }
 226
 227       if (deref->deref_type == nir_deref_type_array ||
 228           deref->deref_type == nir_deref_type_ptr_as_array) {
 229          while (copy_prop_src(&deref->arr.index, instr, NULL, 1))
 230             progress = true;
 231       }
 232
 233       return progress;
 234    }
 235
 236    case nir_instr_type_tex: {
 237       nir_tex_instr *tex = nir_instr_as_tex(instr);
 238       for (unsigned i = 0; i < tex->num_srcs; i++) {
 239          unsigned num_components = nir_tex_instr_src_size(tex, i);
 240          while (copy_prop_src(&tex->src[i].src, instr, NULL, num_components))
 241             progress = true;
 242       }
 243
 244       while (copy_prop_dest(&tex->dest, instr))
 245          progress = true;
 246
 247       return progress;
 248    }
 249
 250    case nir_instr_type_intrinsic: {
 251       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 252       for (unsigned i = 0;
 253            i < nir_intrinsic_infos[intrin->intrinsic].num_srcs; i++) {
 254          unsigned num_components = nir_intrinsic_src_components(intrin, i);
 255
 256          while (copy_prop_src(&intrin->src[i], instr, NULL, num_components))
 257             progress = true;
 258       }
 259
 260       if (nir_intrinsic_infos[intrin->intrinsic].has_dest) {
 261          while (copy_prop_dest(&intrin->dest, instr))
 262             progress = true;
 263       }
 264
 265       return progress;
 266    }
 267
 268    case nir_instr_type_phi: {
 269       nir_phi_instr *phi = nir_instr_as_phi(instr);
 270       assert(phi->dest.is_ssa);
 271       unsigned num_components = phi->dest.ssa.num_components;
 272       nir_foreach_phi_src(src, phi) {
 273          while (copy_prop_src(&src->src, instr, NULL, num_components))
 274             progress = true;
 275       }
 276
 277       return progress;
 278    }
 279
 280    default:
 281       return false;
 282    }
 283 }
 284
 285 static bool
 286 copy_prop_if(nir_if *if_stmt)
 287 {
 288    return copy_prop_src(&if_stmt->condition, NULL, if_stmt, 1);
 289 }
 290
 291 static bool
 292 nir_copy_prop_impl(nir_function_impl *impl)
 293 {
 294    bool progress = false;
 295
 296    nir_foreach_block(block, impl) {
 297       nir_foreach_instr(instr, block) {
 298          if (copy_prop_instr(instr))
 299             progress = true;
 300       }
 301
 302       nir_if *if_stmt = nir_block_get_following_if(block);
 303       if (if_stmt && copy_prop_if(if_stmt))
 304          progress = true;
 305       }
 306
 307    if (progress) {
 308       nir_metadata_preserve(impl, nir_metadata_block_index |
 309                                   nir_metadata_dominance);
 310    }
 311
 312    return progress;
 313 }
 314
 315 bool
 316 nir_copy_prop(nir_shader *shader)
 317 {
 318    bool progress = false;
 319
 320    nir_foreach_function(function, shader) {
 321       if (function->impl && nir_copy_prop_impl(function->impl))
 322          progress = true;
 323    }
 324
 325    return progress;
 326 }