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