src/mesa/drivers/dri/i965/brw_nir.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
  24 #include "brw_nir.h"
  25 #include "brw_shader.h"
  26 #include "glsl/glsl_parser_extras.h"
  27 #include "glsl/nir/glsl_to_nir.h"
  28 #include "program/prog_to_nir.h"
  29
  30 static void
  31 nir_optimize(nir_shader *nir, bool is_scalar)
  32 {
  33    bool progress;
  34    do {
  35       progress = false;
  36       nir_lower_vars_to_ssa(nir);
  37       nir_validate_shader(nir);
  38
  39       if (is_scalar) {
  40          nir_lower_alu_to_scalar(nir);
  41          nir_validate_shader(nir);
  42       }
  43
  44       progress |= nir_copy_prop(nir);
  45       nir_validate_shader(nir);
  46
  47       if (is_scalar) {
  48          nir_lower_phis_to_scalar(nir);
  49          nir_validate_shader(nir);
  50       }
  51
  52       progress |= nir_copy_prop(nir);
  53       nir_validate_shader(nir);
  54       progress |= nir_opt_dce(nir);
  55       nir_validate_shader(nir);
  56       progress |= nir_opt_cse(nir);
  57       nir_validate_shader(nir);
  58       progress |= nir_opt_peephole_select(nir);
  59       nir_validate_shader(nir);
  60       progress |= nir_opt_algebraic(nir);
  61       nir_validate_shader(nir);
  62       progress |= nir_opt_constant_folding(nir);
  63       nir_validate_shader(nir);
  64       progress |= nir_opt_remove_phis(nir);
  65       nir_validate_shader(nir);
  66    } while (progress);
  67 }
  68
  69 nir_shader *
  70 brw_create_nir(struct brw_context *brw,
  71                const struct gl_shader_program *shader_prog,
  72                const struct gl_program *prog,
  73                gl_shader_stage stage,
  74                bool is_scalar)
  75 {
  76    struct gl_context *ctx = &brw->ctx;
  77    const nir_shader_compiler_options *options =
  78       ctx->Const.ShaderCompilerOptions[stage].NirOptions;
  79    struct gl_shader *shader = shader_prog ? shader_prog->_LinkedShaders[stage] : NULL;
  80    bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
  81    nir_shader *nir;
  82
  83    /* First, lower the GLSL IR or Mesa IR to NIR */
  84    if (shader_prog) {
  85       nir = glsl_to_nir(shader, options);
  86    } else {
  87       nir = prog_to_nir(prog, options);
  88       nir_convert_to_ssa(nir); /* turn registers into SSA */
  89    }
  90    nir_validate_shader(nir);
  91
  92    nir_lower_global_vars_to_local(nir);
  93    nir_validate_shader(nir);
  94
  95    nir_lower_tex_projector(nir);
  96    nir_validate_shader(nir);
  97
  98    nir_normalize_cubemap_coords(nir);
  99    nir_validate_shader(nir);
 100
 101    nir_split_var_copies(nir);
 102    nir_validate_shader(nir);
 103
 104    nir_optimize(nir, is_scalar);
 105
 106    /* Lower a bunch of stuff */
 107    nir_lower_var_copies(nir);
 108    nir_validate_shader(nir);
 109
 110    /* Get rid of split copies */
 111    nir_optimize(nir, is_scalar);
 112
 113    if (is_scalar) {
 114       nir_assign_var_locations(&nir->uniforms,
 115                                &nir->num_uniforms,
 116                                type_size_scalar);
 117       nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_scalar);
 118       nir_assign_var_locations(&nir->outputs, &nir->num_outputs, type_size_scalar);
 119       nir_lower_io(nir, type_size_scalar);
 120    } else {
 121       nir_assign_var_locations(&nir->uniforms,
 122                                &nir->num_uniforms,
 123                                type_size_vec4);
 124
 125       nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_vec4);
 126
 127       foreach_list_typed(nir_variable, var, node, &nir->outputs)
 128          var->data.driver_location = var->data.location;
 129
 130       nir_lower_io(nir, type_size_vec4);
 131    }
 132
 133    nir_validate_shader(nir);
 134
 135    nir_remove_dead_variables(nir);
 136    nir_validate_shader(nir);
 137
 138    if (shader_prog) {
 139       nir_lower_samplers(nir, shader_prog, stage);
 140       nir_validate_shader(nir);
 141    }
 142
 143    nir_lower_system_values(nir);
 144    nir_validate_shader(nir);
 145
 146    nir_lower_atomics(nir);
 147    nir_validate_shader(nir);
 148
 149    nir_optimize(nir, is_scalar);
 150
 151    if (brw->gen >= 6) {
 152       /* Try and fuse multiply-adds */
 153       nir_opt_peephole_ffma(nir);
 154       nir_validate_shader(nir);
 155    }
 156
 157    nir_opt_algebraic_late(nir);
 158    nir_validate_shader(nir);
 159
 160    nir_lower_locals_to_regs(nir);
 161    nir_validate_shader(nir);
 162
 163    nir_lower_to_source_mods(nir);
 164    nir_validate_shader(nir);
 165    nir_copy_prop(nir);
 166    nir_validate_shader(nir);
 167    nir_opt_dce(nir);
 168    nir_validate_shader(nir);
 169
 170    if (unlikely(debug_enabled)) {
 171       /* Re-index SSA defs so we print more sensible numbers. */
 172       nir_foreach_overload(nir, overload) {
 173          if (overload->impl)
 174             nir_index_ssa_defs(overload->impl);
 175       }
 176
 177       fprintf(stderr, "NIR (SSA form) for %s shader:\n",
 178               _mesa_shader_stage_to_string(stage));
 179       nir_print_shader(nir, stderr);
 180    }
 181
 182    nir_convert_from_ssa(nir, is_scalar);
 183    nir_validate_shader(nir);
 184
 185    if (!is_scalar) {
 186       nir_lower_vec_to_movs(nir);
 187       nir_validate_shader(nir);
 188    }
 189
 190    /* This is the last pass we run before we start emitting stuff.  It
 191     * determines when we need to insert boolean resolves on Gen <= 5.  We
 192     * run it last because it stashes data in instr->pass_flags and we don't
 193     * want that to be squashed by other NIR passes.
 194     */
 195    if (brw->gen <= 5)
 196       brw_nir_analyze_boolean_resolves(nir);
 197
 198    nir_sweep(nir);
 199
 200    if (unlikely(debug_enabled)) {
 201       fprintf(stderr, "NIR (final form) for %s shader:\n",
 202               _mesa_shader_stage_to_string(stage));
 203       nir_print_shader(nir, stderr);
 204    }
 205
 206    return nir;
 207 }
 208
 209 enum brw_reg_type
 210 brw_type_for_nir_type(nir_alu_type type)
 211 {
 212    switch (type) {
 213    case nir_type_unsigned:
 214       return BRW_REGISTER_TYPE_UD;
 215    case nir_type_bool:
 216    case nir_type_int:
 217       return BRW_REGISTER_TYPE_D;
 218    case nir_type_float:
 219       return BRW_REGISTER_TYPE_F;
 220    default:
 221       unreachable("unknown type");
 222    }
 223
 224    return BRW_REGISTER_TYPE_F;
 225 }
 226
 227 /* Returns the glsl_base_type corresponding to a nir_alu_type.
 228  * This is used by both brw_vec4_nir and brw_fs_nir.
 229  */
 230 enum glsl_base_type
 231 brw_glsl_base_type_for_nir_type(nir_alu_type type)
 232 {
 233    switch (type) {
 234    case nir_type_float:
 235       return GLSL_TYPE_FLOAT;
 236
 237    case nir_type_int:
 238       return GLSL_TYPE_INT;
 239
 240    case nir_type_unsigned:
 241       return GLSL_TYPE_UINT;
 242
 243    default:
 244       unreachable("bad type");
 245    }
 246 }