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 brw_nir_lower_inputs(nir_shader *nir, bool is_scalar)
  32 {
  33    switch (nir->stage) {
  34    case MESA_SHADER_GEOMETRY:
  35       foreach_list_typed(nir_variable, var, node, &nir->inputs) {
  36          var->data.driver_location = var->data.location;
  37       }
  38       break;
  39    default:
  40       nir_assign_var_locations(&nir->inputs, &nir->num_inputs,
  41                                is_scalar ? type_size_scalar : type_size_vec4);
  42       break;
  43    }
  44 }
  45
  46 static void
  47 brw_nir_lower_outputs(nir_shader *nir, bool is_scalar)
  48 {
  49    if (is_scalar) {
  50       nir_assign_var_locations(&nir->outputs, &nir->num_outputs, type_size_scalar);
  51    } else {
  52       nir_foreach_variable(var, &nir->outputs)
  53          var->data.driver_location = var->data.location;
  54    }
  55 }
  56
  57 static void
  58 nir_optimize(nir_shader *nir, bool is_scalar)
  59 {
  60    bool progress;
  61    do {
  62       progress = false;
  63       nir_lower_vars_to_ssa(nir);
  64       nir_validate_shader(nir);
  65
  66       if (is_scalar) {
  67          nir_lower_alu_to_scalar(nir);
  68          nir_validate_shader(nir);
  69       }
  70
  71       progress |= nir_copy_prop(nir);
  72       nir_validate_shader(nir);
  73
  74       if (is_scalar) {
  75          nir_lower_phis_to_scalar(nir);
  76          nir_validate_shader(nir);
  77       }
  78
  79       progress |= nir_copy_prop(nir);
  80       nir_validate_shader(nir);
  81       progress |= nir_opt_dce(nir);
  82       nir_validate_shader(nir);
  83       progress |= nir_opt_cse(nir);
  84       nir_validate_shader(nir);
  85       progress |= nir_opt_peephole_select(nir);
  86       nir_validate_shader(nir);
  87       progress |= nir_opt_algebraic(nir);
  88       nir_validate_shader(nir);
  89       progress |= nir_opt_constant_folding(nir);
  90       nir_validate_shader(nir);
  91       progress |= nir_opt_dead_cf(nir);
  92       nir_validate_shader(nir);
  93       progress |= nir_opt_remove_phis(nir);
  94       nir_validate_shader(nir);
  95       progress |= nir_opt_undef(nir);
  96       nir_validate_shader(nir);
  97    } while (progress);
  98 }
  99
 100 nir_shader *
 101 brw_create_nir(struct brw_context *brw,
 102                const struct gl_shader_program *shader_prog,
 103                const struct gl_program *prog,
 104                gl_shader_stage stage,
 105                bool is_scalar)
 106 {
 107    struct gl_context *ctx = &brw->ctx;
 108    const nir_shader_compiler_options *options =
 109       ctx->Const.ShaderCompilerOptions[stage].NirOptions;
 110    nir_shader *nir;
 111
 112    /* First, lower the GLSL IR or Mesa IR to NIR */
 113    if (shader_prog) {
 114       nir = glsl_to_nir(shader_prog, stage, options);
 115    } else {
 116       nir = prog_to_nir(prog, options);
 117       nir_convert_to_ssa(nir); /* turn registers into SSA */
 118    }
 119    nir_validate_shader(nir);
 120
 121    brw_process_nir(nir, brw->intelScreen->devinfo, shader_prog, stage, is_scalar);
 122
 123    static GLuint msg_id = 0;
 124    _mesa_gl_debug(&brw->ctx, &msg_id,
 125                   MESA_DEBUG_SOURCE_SHADER_COMPILER,
 126                   MESA_DEBUG_TYPE_OTHER,
 127                   MESA_DEBUG_SEVERITY_NOTIFICATION,
 128                   "%s NIR shader:\n",
 129                   _mesa_shader_stage_to_abbrev(stage));
 130
 131    return nir;
 132 }
 133
 134 void
 135 brw_process_nir(nir_shader *nir,
 136                 const struct brw_device_info *devinfo,
 137                 const struct gl_shader_program *shader_prog,
 138                 gl_shader_stage stage, bool is_scalar)
 139 {
 140    bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 141    static const nir_lower_tex_options tex_options = {
 142       .lower_txp = ~0,
 143    };
 144
 145    if (stage == MESA_SHADER_GEOMETRY) {
 146       nir_lower_gs_intrinsics(nir);
 147       nir_validate_shader(nir);
 148    }
 149
 150    nir_lower_global_vars_to_local(nir);
 151    nir_validate_shader(nir);
 152
 153    nir_lower_tex(nir, &tex_options);
 154    nir_validate_shader(nir);
 155
 156    nir_normalize_cubemap_coords(nir);
 157    nir_validate_shader(nir);
 158
 159    nir_split_var_copies(nir);
 160    nir_validate_shader(nir);
 161
 162    nir_optimize(nir, is_scalar);
 163
 164    /* Lower a bunch of stuff */
 165    nir_lower_var_copies(nir);
 166    nir_validate_shader(nir);
 167
 168    /* Get rid of split copies */
 169    nir_optimize(nir, is_scalar);
 170
 171    brw_nir_lower_inputs(nir, is_scalar);
 172    brw_nir_lower_outputs(nir, is_scalar);
 173    nir_assign_var_locations(&nir->uniforms,
 174                             &nir->num_uniforms,
 175                             is_scalar ? type_size_scalar : type_size_vec4);
 176    nir_lower_io(nir, -1, is_scalar ? type_size_scalar : type_size_vec4);
 177    nir_validate_shader(nir);
 178
 179    nir_remove_dead_variables(nir);
 180    nir_validate_shader(nir);
 181
 182    if (shader_prog) {
 183       nir_lower_samplers(nir, shader_prog);
 184       nir_validate_shader(nir);
 185    }
 186
 187    nir_lower_system_values(nir);
 188    nir_validate_shader(nir);
 189
 190    nir_lower_atomics(nir);
 191    nir_validate_shader(nir);
 192
 193    nir_optimize(nir, is_scalar);
 194
 195    if (devinfo->gen >= 6) {
 196       /* Try and fuse multiply-adds */
 197       nir_opt_peephole_ffma(nir);
 198       nir_validate_shader(nir);
 199    }
 200
 201    nir_opt_algebraic_late(nir);
 202    nir_validate_shader(nir);
 203
 204    nir_lower_locals_to_regs(nir);
 205    nir_validate_shader(nir);
 206
 207    nir_lower_to_source_mods(nir);
 208    nir_validate_shader(nir);
 209    nir_copy_prop(nir);
 210    nir_validate_shader(nir);
 211    nir_opt_dce(nir);
 212    nir_validate_shader(nir);
 213
 214    if (unlikely(debug_enabled)) {
 215       /* Re-index SSA defs so we print more sensible numbers. */
 216       nir_foreach_overload(nir, overload) {
 217          if (overload->impl)
 218             nir_index_ssa_defs(overload->impl);
 219       }
 220
 221       fprintf(stderr, "NIR (SSA form) for %s shader:\n",
 222               _mesa_shader_stage_to_string(stage));
 223       nir_print_shader(nir, stderr);
 224    }
 225
 226    nir_convert_from_ssa(nir, true);
 227    nir_validate_shader(nir);
 228
 229    if (!is_scalar) {
 230       nir_move_vec_src_uses_to_dest(nir);
 231       nir_validate_shader(nir);
 232
 233       nir_lower_vec_to_movs(nir);
 234       nir_validate_shader(nir);
 235    }
 236
 237    /* This is the last pass we run before we start emitting stuff.  It
 238     * determines when we need to insert boolean resolves on Gen <= 5.  We
 239     * run it last because it stashes data in instr->pass_flags and we don't
 240     * want that to be squashed by other NIR passes.
 241     */
 242    if (devinfo->gen <= 5)
 243       brw_nir_analyze_boolean_resolves(nir);
 244
 245    nir_sweep(nir);
 246
 247    if (unlikely(debug_enabled)) {
 248       fprintf(stderr, "NIR (final form) for %s shader:\n",
 249               _mesa_shader_stage_to_string(stage));
 250       nir_print_shader(nir, stderr);
 251    }
 252 }
 253
 254 enum brw_reg_type
 255 brw_type_for_nir_type(nir_alu_type type)
 256 {
 257    switch (type) {
 258    case nir_type_unsigned:
 259       return BRW_REGISTER_TYPE_UD;
 260    case nir_type_bool:
 261    case nir_type_int:
 262       return BRW_REGISTER_TYPE_D;
 263    case nir_type_float:
 264       return BRW_REGISTER_TYPE_F;
 265    default:
 266       unreachable("unknown type");
 267    }
 268
 269    return BRW_REGISTER_TYPE_F;
 270 }
 271
 272 /* Returns the glsl_base_type corresponding to a nir_alu_type.
 273  * This is used by both brw_vec4_nir and brw_fs_nir.
 274  */
 275 enum glsl_base_type
 276 brw_glsl_base_type_for_nir_type(nir_alu_type type)
 277 {
 278    switch (type) {
 279    case nir_type_float:
 280       return GLSL_TYPE_FLOAT;
 281
 282    case nir_type_int:
 283       return GLSL_TYPE_INT;
 284
 285    case nir_type_unsigned:
 286       return GLSL_TYPE_UINT;
 287
 288    default:
 289       unreachable("bad type");
 290    }
 291 }