src/glsl/nir/nir_lower_to_source_mods.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
  30 /*
  31  * This pass lowers the neg, abs, and sat operations to source modifiers on
  32  * ALU operations to make things nicer for the backend.  It's just much
  33  * easier to not have them when we're doing optimizations.
  34  */
  35
  36 static bool
  37 nir_lower_to_source_mods_block(nir_block *block, void *state)
  38 {
  39    nir_foreach_instr(block, instr) {
  40       if (instr->type != nir_instr_type_alu)
  41          continue;
  42
  43       nir_alu_instr *alu = nir_instr_as_alu(instr);
  44
  45       for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) {
  46          if (!alu->src[i].src.is_ssa)
  47             continue;
  48
  49          if (alu->src[i].src.ssa->parent_instr->type != nir_instr_type_alu)
  50             continue;
  51
  52          nir_alu_instr *parent = nir_instr_as_alu(alu->src[i].src.ssa->parent_instr);
  53
  54          if (parent->dest.saturate)
  55             continue;
  56
  57          switch (nir_op_infos[alu->op].input_types[i]) {
  58          case nir_type_float:
  59             if (parent->op != nir_op_fmov)
  60                continue;
  61             break;
  62          case nir_type_int:
  63             if (parent->op != nir_op_imov)
  64                continue;
  65             break;
  66          default:
  67             continue;
  68          }
  69
  70          nir_instr_rewrite_src(instr, &alu->src[i].src, parent->src[0].src);
  71          if (alu->src[i].abs) {
  72             /* abs trumps both neg and abs, do nothing */
  73          } else {
  74             alu->src[i].negate = (alu->src[i].negate != parent->src[0].negate);
  75             alu->src[i].abs |= parent->src[0].abs;
  76          }
  77
  78          for (int j = 0; j < 4; ++j) {
  79             if (!nir_alu_instr_channel_used(alu, i, j))
  80                continue;
  81             alu->src[i].swizzle[j] = parent->src[0].swizzle[alu->src[i].swizzle[j]];
  82          }
  83
  84          if (parent->dest.dest.ssa.uses->entries == 0 &&
  85              parent->dest.dest.ssa.if_uses->entries == 0)
  86             nir_instr_remove(&parent->instr);
  87       }
  88
  89       switch (alu->op) {
  90       case nir_op_fsat:
  91          alu->op = nir_op_fmov;
  92          alu->dest.saturate = true;
  93          break;
  94       case nir_op_ineg:
  95          alu->op = nir_op_imov;
  96          alu->src[0].negate = !alu->src[0].negate;
  97          break;
  98       case nir_op_fneg:
  99          alu->op = nir_op_fmov;
 100          alu->src[0].negate = !alu->src[0].negate;
 101          break;
 102       case nir_op_iabs:
 103          alu->op = nir_op_imov;
 104          alu->src[0].abs = true;
 105          alu->src[0].negate = false;
 106          break;
 107       case nir_op_fabs:
 108          alu->op = nir_op_fmov;
 109          alu->src[0].abs = true;
 110          alu->src[0].negate = false;
 111          break;
 112       default:
 113          break;
 114       }
 115
 116       /* We've covered sources.  Now we're going to try and saturate the
 117        * destination if we can.
 118        */
 119
 120       if (!alu->dest.dest.is_ssa)
 121          continue;
 122
 123       /* We can only saturate float destinations */
 124       if (nir_op_infos[alu->op].output_type != nir_type_float)
 125          continue;
 126
 127       if (alu->dest.dest.ssa.if_uses->entries != 0)
 128          continue;
 129
 130       bool all_children_are_sat = true;
 131       struct set_entry *entry;
 132       set_foreach(alu->dest.dest.ssa.uses, entry) {
 133          const nir_instr *child = entry->key;
 134          if (child->type != nir_instr_type_alu) {
 135             all_children_are_sat = false;
 136             continue;
 137          }
 138
 139          nir_alu_instr *child_alu = nir_instr_as_alu(child);
 140          if (child_alu->src[0].negate || child_alu->src[0].abs) {
 141             all_children_are_sat = false;
 142             continue;
 143          }
 144
 145          if (child_alu->op != nir_op_fsat &&
 146              !(child_alu->op == nir_op_fmov && child_alu->dest.saturate)) {
 147             all_children_are_sat = false;
 148             continue;
 149          }
 150       }
 151
 152       if (!all_children_are_sat)
 153          continue;
 154
 155       alu->dest.saturate = true;
 156
 157       set_foreach(alu->dest.dest.ssa.uses, entry) {
 158          nir_alu_instr *child_alu = nir_instr_as_alu((nir_instr *)entry->key);
 159          child_alu->op = nir_op_fmov;
 160          child_alu->dest.saturate = false;
 161          /* We could propagate the dest of our instruction to the
 162           * destinations of the uses here.  However, one quick round of
 163           * copy propagation will clean that all up and then we don't have
 164           * the complexity.
 165           */
 166       }
 167    }
 168
 169    return true;
 170 }
 171
 172 static void
 173 nir_lower_to_source_mods_impl(nir_function_impl *impl)
 174 {
 175    nir_foreach_block(impl, nir_lower_to_source_mods_block, NULL);
 176 }
 177
 178 void
 179 nir_lower_to_source_mods(nir_shader *shader)
 180 {
 181    nir_foreach_overload(shader, overload) {
 182       if (overload->impl)
 183          nir_lower_to_source_mods_impl(overload->impl);
 184    }
 185 }