src/compiler/nir/nir_lower_io_to_temporaries.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 /*
  25  * Implements a pass that lowers output and/or input variables to a
  26  * temporary plus an output variable with a single copy at each exit
  27  * point of the shader and/or an input variable with a single copy
  28  * at the entrance point of the shader.  This way the output variable
  29  * is only ever written once and/or input is only read once, and there
  30  * are no indirect outut/input accesses.
  31  */
  32
  33 #include "nir.h"
  34
  35 struct lower_io_state {
  36    nir_shader *shader;
  37    nir_function_impl *entrypoint;
  38    struct exec_list old_outputs;
  39    struct exec_list old_inputs;
  40 };
  41
  42 static void
  43 emit_copies(nir_cursor cursor, nir_shader *shader, struct exec_list *new_vars,
  44           struct exec_list *old_vars)
  45 {
  46    assert(exec_list_length(new_vars) == exec_list_length(old_vars));
  47
  48    foreach_two_lists(new_node, new_vars, old_node, old_vars) {
  49       nir_variable *newv = exec_node_data(nir_variable, new_node, node);
  50       nir_variable *temp = exec_node_data(nir_variable, old_node, node);
  51
  52       /* No need to copy the contents of a non-fb_fetch_output output variable
  53        * to the temporary allocated for it, since its initial value is
  54        * undefined.
  55        */
  56       if (temp->data.mode == nir_var_shader_out &&
  57           !temp->data.fb_fetch_output)
  58          continue;
  59
  60       /* Can't copy the contents of the temporary back to a read-only
  61        * interface variable.  The value of the temporary won't have been
  62        * modified by the shader anyway.
  63        */
  64       if (newv->data.read_only)
  65          continue;
  66
  67       nir_intrinsic_instr *copy =
  68          nir_intrinsic_instr_create(shader, nir_intrinsic_copy_var);
  69       copy->variables[0] = nir_deref_var_create(copy, newv);
  70       copy->variables[1] = nir_deref_var_create(copy, temp);
  71
  72       nir_instr_insert(cursor, &copy->instr);
  73    }
  74 }
  75
  76 static void
  77 emit_output_copies_impl(struct lower_io_state *state, nir_function_impl *impl)
  78 {
  79    if (state->shader->stage == MESA_SHADER_GEOMETRY) {
  80       /* For geometry shaders, we have to emit the output copies right
  81        * before each EmitVertex call.
  82        */
  83       nir_foreach_block(block, impl) {
  84          nir_foreach_instr(instr, block) {
  85             if (instr->type != nir_instr_type_intrinsic)
  86                continue;
  87
  88             nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
  89             if (intrin->intrinsic == nir_intrinsic_emit_vertex) {
  90                nir_cursor cursor = nir_before_instr(&intrin->instr);
  91                emit_copies(cursor, state->shader, &state->shader->outputs,
  92                            &state->old_outputs);
  93             }
  94          }
  95       }
  96    } else if (impl == state->entrypoint) {
  97       nir_cursor cursor = nir_before_block(nir_start_block(impl));
  98       emit_copies(cursor, state->shader, &state->old_outputs,
  99                   &state->shader->outputs);
 100
 101       /* For all other shader types, we need to do the copies right before
 102        * the jumps to the end block.
 103        */
 104       struct set_entry *block_entry;
 105       set_foreach(impl->end_block->predecessors, block_entry) {
 106          struct nir_block *block = (void *)block_entry->key;
 107          nir_cursor cursor = nir_after_block_before_jump(block);
 108          emit_copies(cursor, state->shader, &state->shader->outputs,
 109                      &state->old_outputs);
 110       }
 111    }
 112 }
 113
 114 static void
 115 emit_input_copies_impl(struct lower_io_state *state, nir_function_impl *impl)
 116 {
 117    if (impl == state->entrypoint) {
 118       nir_cursor cursor = nir_before_block(nir_start_block(impl));
 119       emit_copies(cursor, state->shader, &state->old_inputs,
 120                   &state->shader->inputs);
 121    }
 122 }
 123
 124 static nir_variable *
 125 create_shadow_temp(struct lower_io_state *state, nir_variable *var)
 126 {
 127    nir_variable *nvar = ralloc(state->shader, nir_variable);
 128    memcpy(nvar, var, sizeof *nvar);
 129
 130    /* The original is now the temporary */
 131    nir_variable *temp = var;
 132
 133    /* Reparent the name to the new variable */
 134    ralloc_steal(nvar, nvar->name);
 135
 136    /* Reparent the constant initializer (if any) */
 137    ralloc_steal(nvar, nvar->constant_initializer);
 138
 139    /* Give the original a new name with @<mode>-temp appended */
 140    const char *mode = (temp->data.mode == nir_var_shader_in) ? "in" : "out";
 141    temp->name = ralloc_asprintf(var, "%s@%s-temp", mode, nvar->name);
 142    temp->data.mode = nir_var_global;
 143    temp->data.read_only = false;
 144    temp->data.fb_fetch_output = false;
 145    temp->constant_initializer = NULL;
 146
 147    return nvar;
 148 }
 149
 150 void
 151 nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint,
 152                             bool outputs, bool inputs)
 153 {
 154    struct lower_io_state state;
 155
 156    if (shader->stage == MESA_SHADER_TESS_CTRL)
 157       return;
 158
 159    state.shader = shader;
 160    state.entrypoint = entrypoint->impl;
 161
 162    if (inputs)
 163       exec_list_move_nodes_to(&shader->inputs, &state.old_inputs);
 164    else
 165       exec_list_make_empty(&state.old_inputs);
 166
 167    if (outputs)
 168       exec_list_move_nodes_to(&shader->outputs, &state.old_outputs);
 169    else
 170       exec_list_make_empty(&state.old_outputs);
 171
 172    /* Walk over all of the outputs turn each output into a temporary and
 173     * make a new variable for the actual output.
 174     */
 175    nir_foreach_variable(var, &state.old_outputs) {
 176       nir_variable *output = create_shadow_temp(&state, var);
 177       exec_list_push_tail(&shader->outputs, &output->node);
 178    }
 179
 180    /* and same for inputs: */
 181    nir_foreach_variable(var, &state.old_inputs) {
 182       nir_variable *input = create_shadow_temp(&state, var);
 183       exec_list_push_tail(&shader->inputs, &input->node);
 184    }
 185
 186    nir_foreach_function(function, shader) {
 187       if (function->impl == NULL)
 188          continue;
 189
 190       if (inputs)
 191          emit_input_copies_impl(&state, function->impl);
 192
 193       if (outputs)
 194          emit_output_copies_impl(&state, function->impl);
 195
 196       nir_metadata_preserve(function->impl, nir_metadata_block_index |
 197                                             nir_metadata_dominance);
 198    }
 199
 200    exec_list_append(&shader->globals, &state.old_inputs);
 201    exec_list_append(&shader->globals, &state.old_outputs);
 202 }