src/freedreno/ir3/ir3_nir_move_varying_inputs.c

   1 /*
   2  * Copyright © 2019 Red Hat
   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 "ir3_nir.h"
  25 #include "compiler/nir/nir_builder.h"
  26
  27 /**
  28  * This pass moves varying fetches (and the instructions they depend on
  29  * into the start block.
  30  *
  31  * We need to set the (ei) "end input" flag on the last varying fetch.
  32  * And we want to ensure that all threads execute the instruction that
  33  * sets (ei).  The easiest way to ensure this is to move all varying
  34  * fetches into the start block.  Which is something we used to get for
  35  * free by using lower_all_io_to_temps=true.
  36  *
  37  * This may come at the cost of additional register usage.  OTOH setting
  38  * the (ei) flag earlier probably frees up more VS to run.
  39  */
  40
  41
  42 typedef struct {
  43         nir_shader *shader;
  44         nir_block *start_block;
  45         struct nir_instr *cursor;
  46 } state;
  47
  48 static void move_instruction_to_start_block(state *state, nir_instr *instr);
  49
  50 static bool
  51 move_src(nir_src *src, void *state)
  52 {
  53         /* At this point we shouldn't have any non-ssa src: */
  54         debug_assert(src->is_ssa);
  55         move_instruction_to_start_block(state, src->ssa->parent_instr);
  56         return true;
  57 }
  58
  59 static void
  60 move_instruction_to_start_block(state *state, nir_instr *instr)
  61 {
  62         /* first move (recursively) all src's to ensure they appear before
  63          * load*_input that we are trying to move:
  64          */
  65         nir_foreach_src(instr, move_src, state);
  66
  67         /* and then move the instruction itself:
  68          */
  69         exec_node_remove(&instr->node);
  70
  71         if (state->cursor) {
  72                 exec_node_insert_after(&state->cursor->node, &instr->node);
  73         } else {
  74                 exec_list_push_head(&state->start_block->instr_list, &instr->node);
  75         }
  76
  77         state->cursor = instr;
  78         instr->block = state->start_block;
  79 }
  80
  81 static bool
  82 move_varying_inputs_block(state *state, nir_block *block)
  83 {
  84         bool progress = false;
  85
  86         nir_foreach_instr_safe (instr, block) {
  87                 if (instr->type != nir_instr_type_intrinsic)
  88                         continue;
  89
  90                 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
  91
  92                 switch (intr->intrinsic) {
  93                 case nir_intrinsic_load_interpolated_input:
  94                 case nir_intrinsic_load_input:
  95                         /* TODO any others to handle? */
  96                         break;
  97                 default:
  98                         continue;
  99                 }
 100
 101                 debug_assert(intr->dest.is_ssa);
 102
 103                 state->cursor = NULL;
 104                 move_instruction_to_start_block(state, instr);
 105
 106                 progress = true;
 107         }
 108
 109         return progress;
 110 }
 111
 112 bool
 113 ir3_nir_move_varying_inputs(nir_shader *shader)
 114 {
 115         bool progress = false;
 116
 117         debug_assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 118
 119         nir_foreach_function (function, shader) {
 120                 state state;
 121
 122                 if (!function->impl)
 123                         continue;
 124
 125                 state.shader = shader;
 126                 state.start_block = nir_start_block(function->impl);
 127
 128                 bool progress = false;
 129                 nir_foreach_block (block, function->impl) {
 130                         /* don't need to move anything that is already in the first block */
 131                         if (block == state.start_block)
 132                                 continue;
 133                         progress |= move_varying_inputs_block(&state, block);
 134                 }
 135
 136                 if (progress) {
 137                         nir_metadata_preserve(function->impl,
 138                                 nir_metadata_block_index | nir_metadata_dominance);
 139                 }
 140         }
 141
 142         return progress;
 143 }