src/compiler/nir/nir_linking_helpers.c

   1 /*
   2  * Copyright © 2015 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
  24 #include "nir.h"
  25 #include "util/set.h"
  26 #include "util/hash_table.h"
  27
  28 /* This file contains various little helpers for doing simple linking in
  29  * NIR.  Eventually, we'll probably want a full-blown varying packing
  30  * implementation in here.  Right now, it just deletes unused things.
  31  */
  32
  33 /**
  34  * Returns the bits in the inputs_read, outputs_written, or
  35  * system_values_read bitfield corresponding to this variable.
  36  */
  37 static uint64_t
  38 get_variable_io_mask(nir_variable *var, gl_shader_stage stage)
  39 {
  40    if (var->data.location < 0)
  41       return 0;
  42
  43    unsigned location = var->data.patch ?
  44       var->data.location - VARYING_SLOT_PATCH0 : var->data.location;
  45
  46    assert(var->data.mode == nir_var_shader_in ||
  47           var->data.mode == nir_var_shader_out ||
  48           var->data.mode == nir_var_system_value);
  49    assert(var->data.location >= 0);
  50
  51    const struct glsl_type *type = var->type;
  52    if (nir_is_per_vertex_io(var, stage)) {
  53       assert(glsl_type_is_array(type));
  54       type = glsl_get_array_element(type);
  55    }
  56
  57    unsigned slots = glsl_count_attribute_slots(type, false);
  58    return ((1ull << slots) - 1) << location;
  59 }
  60
  61 static void
  62 tcs_add_output_reads(nir_shader *shader, uint64_t *read, uint64_t *patches_read)
  63 {
  64    nir_foreach_function(function, shader) {
  65       if (function->impl) {
  66          nir_foreach_block(block, function->impl) {
  67             nir_foreach_instr(instr, block) {
  68                if (instr->type != nir_instr_type_intrinsic)
  69                   continue;
  70
  71                nir_intrinsic_instr *intrin_instr =
  72                   nir_instr_as_intrinsic(instr);
  73                if (intrin_instr->intrinsic == nir_intrinsic_load_var &&
  74                    intrin_instr->variables[0]->var->data.mode ==
  75                    nir_var_shader_out) {
  76
  77                   nir_variable *var = intrin_instr->variables[0]->var;
  78                   if (var->data.patch) {
  79                      patches_read[var->data.location_frac] |=
  80                         get_variable_io_mask(intrin_instr->variables[0]->var,
  81                                              shader->info.stage);
  82                   } else {
  83                      read[var->data.location_frac] |=
  84                         get_variable_io_mask(intrin_instr->variables[0]->var,
  85                                              shader->info.stage);
  86                   }
  87                }
  88             }
  89          }
  90       }
  91    }
  92 }
  93
  94 static bool
  95 remove_unused_io_vars(nir_shader *shader, struct exec_list *var_list,
  96                       uint64_t *used_by_other_stage,
  97                       uint64_t *used_by_other_stage_patches)
  98 {
  99    bool progress = false;
 100    uint64_t *used;
 101
 102    nir_foreach_variable_safe(var, var_list) {
 103       if (var->data.patch)
 104          used = used_by_other_stage_patches;
 105       else
 106          used = used_by_other_stage;
 107
 108       if (var->data.location < VARYING_SLOT_VAR0 && var->data.location >= 0)
 109          continue;
 110
 111       if (var->data.always_active_io)
 112          continue;
 113
 114       uint64_t other_stage = used[var->data.location_frac];
 115
 116       if (!(other_stage & get_variable_io_mask(var, shader->info.stage))) {
 117          /* This one is invalid, make it a global variable instead */
 118          var->data.location = 0;
 119          var->data.mode = nir_var_global;
 120
 121          exec_node_remove(&var->node);
 122          exec_list_push_tail(&shader->globals, &var->node);
 123
 124          progress = true;
 125       }
 126    }
 127
 128    return progress;
 129 }
 130
 131 bool
 132 nir_remove_unused_varyings(nir_shader *producer, nir_shader *consumer)
 133 {
 134    assert(producer->info.stage != MESA_SHADER_FRAGMENT);
 135    assert(consumer->info.stage != MESA_SHADER_VERTEX);
 136
 137    uint64_t read[4] = { 0 }, written[4] = { 0 };
 138    uint64_t patches_read[4] = { 0 }, patches_written[4] = { 0 };
 139
 140    nir_foreach_variable(var, &producer->outputs) {
 141       if (var->data.patch) {
 142          patches_written[var->data.location_frac] |=
 143             get_variable_io_mask(var, producer->info.stage);
 144       } else {
 145          written[var->data.location_frac] |=
 146             get_variable_io_mask(var, producer->info.stage);
 147       }
 148    }
 149
 150    nir_foreach_variable(var, &consumer->inputs) {
 151       if (var->data.patch) {
 152          patches_read[var->data.location_frac] |=
 153             get_variable_io_mask(var, consumer->info.stage);
 154       } else {
 155          read[var->data.location_frac] |=
 156             get_variable_io_mask(var, consumer->info.stage);
 157       }
 158    }
 159
 160    /* Each TCS invocation can read data written by other TCS invocations,
 161     * so even if the outputs are not used by the TES we must also make
 162     * sure they are not read by the TCS before demoting them to globals.
 163     */
 164    if (producer->info.stage == MESA_SHADER_TESS_CTRL)
 165       tcs_add_output_reads(producer, read, patches_read);
 166
 167    bool progress = false;
 168    progress = remove_unused_io_vars(producer, &producer->outputs, read,
 169                                     patches_read);
 170
 171    progress = remove_unused_io_vars(consumer, &consumer->inputs, written,
 172                                     patches_written) || progress;
 173
 174    return progress;
 175 }