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