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 "glsl/glsl_parser_extras.h"
  26 #include "glsl/nir/glsl_to_nir.h"
  27 #include "program/prog_to_nir.h"
  28
  29 static void
  30 nir_optimize(nir_shader *nir, bool is_scalar)
  31 {
  32    bool progress;
  33    do {
  34       progress = false;
  35       nir_lower_vars_to_ssa(nir);
  36       nir_validate_shader(nir);
  37
  38       if (is_scalar) {
  39          nir_lower_alu_to_scalar(nir);
  40          nir_validate_shader(nir);
  41       }
  42
  43       progress |= nir_copy_prop(nir);
  44       nir_validate_shader(nir);
  45
  46       if (is_scalar) {
  47          nir_lower_phis_to_scalar(nir);
  48          nir_validate_shader(nir);
  49       }
  50
  51       progress |= nir_copy_prop(nir);
  52       nir_validate_shader(nir);
  53       progress |= nir_opt_dce(nir);
  54       nir_validate_shader(nir);
  55       progress |= nir_opt_cse(nir);
  56       nir_validate_shader(nir);
  57       progress |= nir_opt_peephole_select(nir);
  58       nir_validate_shader(nir);
  59       progress |= nir_opt_algebraic(nir);
  60       nir_validate_shader(nir);
  61       progress |= nir_opt_constant_folding(nir);
  62       nir_validate_shader(nir);
  63       progress |= nir_opt_remove_phis(nir);
  64       nir_validate_shader(nir);
  65    } while (progress);
  66 }
  67
  68 nir_shader *
  69 brw_create_nir(struct brw_context *brw,
  70                const struct gl_shader_program *shader_prog,
  71                const struct gl_program *prog,
  72                gl_shader_stage stage,
  73                bool is_scalar)
  74 {
  75    struct gl_context *ctx = &brw->ctx;
  76    const nir_shader_compiler_options *options =
  77       ctx->Const.ShaderCompilerOptions[stage].NirOptions;
  78    struct gl_shader *shader = shader_prog ? shader_prog->_LinkedShaders[stage] : NULL;
  79    nir_shader *nir;
  80
  81    /* First, lower the GLSL IR or Mesa IR to NIR */
  82    if (shader_prog) {
  83       nir = glsl_to_nir(shader, options);
  84    } else {
  85       nir = prog_to_nir(prog, options);
  86       nir_convert_to_ssa(nir); /* turn registers into SSA */
  87    }
  88    nir_validate_shader(nir);
  89
  90    brw_process_nir(nir, brw->intelScreen->devinfo, shader_prog, stage, is_scalar);
  91
  92    static GLuint msg_id = 0;
  93    _mesa_gl_debug(&brw->ctx, &msg_id,
  94                   MESA_DEBUG_SOURCE_SHADER_COMPILER,
  95                   MESA_DEBUG_TYPE_OTHER,
  96                   MESA_DEBUG_SEVERITY_NOTIFICATION,
  97                   "%s NIR shader:\n",
  98                   _mesa_shader_stage_to_abbrev(stage));
  99
 100    return nir;
 101 }
 102
 103 void
 104 brw_process_nir(nir_shader *nir,
 105                 const struct brw_device_info *devinfo,
 106                 const struct gl_shader_program *shader_prog,
 107                 gl_shader_stage stage, bool is_scalar)
 108 {
 109    bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 110
 111    nir_lower_global_vars_to_local(nir);
 112    nir_validate_shader(nir);
 113
 114    nir_lower_tex_projector(nir);
 115    nir_validate_shader(nir);
 116
 117    nir_normalize_cubemap_coords(nir);
 118    nir_validate_shader(nir);
 119
 120    nir_split_var_copies(nir);
 121    nir_validate_shader(nir);
 122
 123    nir_optimize(nir, is_scalar);
 124
 125    /* Lower a bunch of stuff */
 126    nir_lower_var_copies(nir);
 127    nir_validate_shader(nir);
 128
 129    /* Get rid of split copies */
 130    nir_optimize(nir, is_scalar);
 131
 132    if (is_scalar) {
 133       nir_assign_var_locations_direct_first(nir, &nir->uniforms,
 134                                             &nir->num_direct_uniforms,
 135                                             &nir->num_uniforms,
 136                                             is_scalar);
 137       nir_assign_var_locations(&nir->outputs, &nir->num_outputs, is_scalar);
 138    } else {
 139       nir_assign_var_locations(&nir->uniforms,
 140                                &nir->num_uniforms,
 141                                is_scalar);
 142
 143       foreach_list_typed(nir_variable, var, node, &nir->outputs)
 144          var->data.driver_location = var->data.location;
 145    }
 146    nir_assign_var_locations(&nir->inputs, &nir->num_inputs, is_scalar);
 147
 148    nir_lower_io(nir, is_scalar);
 149
 150    nir_validate_shader(nir);
 151
 152    nir_remove_dead_variables(nir);
 153    nir_validate_shader(nir);
 154
 155    if (shader_prog) {
 156       nir_lower_samplers(nir, shader_prog, stage);
 157    } else {
 158       nir_lower_samplers_for_vk(nir);
 159    }
 160    nir_validate_shader(nir);
 161
 162    nir_lower_system_values(nir);
 163    nir_validate_shader(nir);
 164
 165    nir_lower_atomics(nir);
 166    nir_validate_shader(nir);
 167
 168    nir_optimize(nir, is_scalar);
 169
 170    if (devinfo->gen >= 6) {
 171       /* Try and fuse multiply-adds */
 172       nir_opt_peephole_ffma(nir);
 173       nir_validate_shader(nir);
 174    }
 175
 176    nir_opt_algebraic_late(nir);
 177    nir_validate_shader(nir);
 178
 179    nir_lower_locals_to_regs(nir);
 180    nir_validate_shader(nir);
 181
 182    nir_lower_to_source_mods(nir);
 183    nir_validate_shader(nir);
 184    nir_copy_prop(nir);
 185    nir_validate_shader(nir);
 186    nir_opt_dce(nir);
 187    nir_validate_shader(nir);
 188
 189    if (unlikely(debug_enabled)) {
 190       /* Re-index SSA defs so we print more sensible numbers. */
 191       nir_foreach_overload(nir, overload) {
 192          if (overload->impl)
 193             nir_index_ssa_defs(overload->impl);
 194       }
 195
 196       fprintf(stderr, "NIR (SSA form) for %s shader:\n",
 197               _mesa_shader_stage_to_string(stage));
 198       nir_print_shader(nir, stderr);
 199    }
 200
 201    nir_convert_from_ssa(nir, is_scalar);
 202    nir_validate_shader(nir);
 203
 204    if (!is_scalar) {
 205       nir_lower_vec_to_movs(nir);
 206       nir_validate_shader(nir);
 207    }
 208
 209    /* This is the last pass we run before we start emitting stuff.  It
 210     * determines when we need to insert boolean resolves on Gen <= 5.  We
 211     * run it last because it stashes data in instr->pass_flags and we don't
 212     * want that to be squashed by other NIR passes.
 213     */
 214    if (devinfo->gen <= 5)
 215       brw_nir_analyze_boolean_resolves(nir);
 216
 217    nir_sweep(nir);
 218
 219    if (unlikely(debug_enabled)) {
 220       fprintf(stderr, "NIR (final form) for %s shader:\n",
 221               _mesa_shader_stage_to_string(stage));
 222       nir_print_shader(nir, stderr);
 223    }
 224 }
 225
 226 enum brw_reg_type
 227 brw_type_for_nir_type(nir_alu_type type)
 228 {
 229    switch (type) {
 230    case nir_type_unsigned:
 231       return BRW_REGISTER_TYPE_UD;
 232    case nir_type_bool:
 233    case nir_type_int:
 234       return BRW_REGISTER_TYPE_D;
 235    case nir_type_float:
 236       return BRW_REGISTER_TYPE_F;
 237    default:
 238       unreachable("unknown type");
 239    }
 240
 241    return BRW_REGISTER_TYPE_F;
 242 }
 243
 244 /* Returns the glsl_base_type corresponding to a nir_alu_type.
 245  * This is used by both brw_vec4_nir and brw_fs_nir.
 246  */
 247 enum glsl_base_type
 248 brw_glsl_base_type_for_nir_type(nir_alu_type type)
 249 {
 250    switch (type) {
 251    case nir_type_float:
 252       return GLSL_TYPE_FLOAT;
 253
 254    case nir_type_int:
 255       return GLSL_TYPE_INT;
 256
 257    case nir_type_unsigned:
 258       return GLSL_TYPE_UINT;
 259
 260    default:
 261       unreachable("bad type");
 262    }
 263 }