src/compiler/nir/nir_lower_io_to_scalar.c

   1 /*
   2  * Copyright © 2016 Broadcom
   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 "nir.h"
  25 #include "nir_builder.h"
  26 #include "nir_deref.h"
  27
  28 /** @file nir_lower_io_to_scalar.c
  29  *
  30  * Replaces nir_load_input/nir_store_output operations with num_components !=
  31  * 1 with individual per-channel operations.
  32  */
  33
  34 static void
  35 lower_load_input_to_scalar(nir_builder *b, nir_intrinsic_instr *intr)
  36 {
  37    b->cursor = nir_before_instr(&intr->instr);
  38
  39    assert(intr->dest.is_ssa);
  40
  41    nir_ssa_def *loads[NIR_MAX_VEC_COMPONENTS];
  42
  43    for (unsigned i = 0; i < intr->num_components; i++) {
  44       nir_intrinsic_instr *chan_intr =
  45          nir_intrinsic_instr_create(b->shader, intr->intrinsic);
  46       nir_ssa_dest_init(&chan_intr->instr, &chan_intr->dest,
  47                         1, intr->dest.ssa.bit_size, NULL);
  48       chan_intr->num_components = 1;
  49
  50       nir_intrinsic_set_base(chan_intr, nir_intrinsic_base(intr));
  51       nir_intrinsic_set_component(chan_intr, nir_intrinsic_component(intr) + i);
  52       nir_intrinsic_set_type(chan_intr, nir_intrinsic_type(intr));
  53       /* offset */
  54       nir_src_copy(&chan_intr->src[0], &intr->src[0], chan_intr);
  55
  56       nir_builder_instr_insert(b, &chan_intr->instr);
  57
  58       loads[i] = &chan_intr->dest.ssa;
  59    }
  60
  61    nir_ssa_def_rewrite_uses(&intr->dest.ssa,
  62                             nir_src_for_ssa(nir_vec(b, loads,
  63                                                     intr->num_components)));
  64    nir_instr_remove(&intr->instr);
  65 }
  66
  67 static void
  68 lower_store_output_to_scalar(nir_builder *b, nir_intrinsic_instr *intr)
  69 {
  70    b->cursor = nir_before_instr(&intr->instr);
  71
  72    nir_ssa_def *value = nir_ssa_for_src(b, intr->src[0], intr->num_components);
  73
  74    for (unsigned i = 0; i < intr->num_components; i++) {
  75       if (!(nir_intrinsic_write_mask(intr) & (1 << i)))
  76          continue;
  77
  78       nir_intrinsic_instr *chan_intr =
  79          nir_intrinsic_instr_create(b->shader, intr->intrinsic);
  80       chan_intr->num_components = 1;
  81
  82       nir_intrinsic_set_base(chan_intr, nir_intrinsic_base(intr));
  83       nir_intrinsic_set_write_mask(chan_intr, 0x1);
  84       nir_intrinsic_set_component(chan_intr, nir_intrinsic_component(intr) + i);
  85       nir_intrinsic_set_type(chan_intr, nir_intrinsic_type(intr));
  86
  87       /* value */
  88       chan_intr->src[0] = nir_src_for_ssa(nir_channel(b, value, i));
  89       /* offset */
  90       nir_src_copy(&chan_intr->src[1], &intr->src[1], chan_intr);
  91
  92       nir_builder_instr_insert(b, &chan_intr->instr);
  93    }
  94
  95    nir_instr_remove(&intr->instr);
  96 }
  97
  98 void
  99 nir_lower_io_to_scalar(nir_shader *shader, nir_variable_mode mask)
 100 {
 101    nir_foreach_function(function, shader) {
 102       if (function->impl) {
 103          nir_builder b;
 104          nir_builder_init(&b, function->impl);
 105
 106          nir_foreach_block(block, function->impl) {
 107             nir_foreach_instr_safe(instr, block) {
 108                if (instr->type != nir_instr_type_intrinsic)
 109                   continue;
 110
 111                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 112
 113                if (intr->num_components == 1)
 114                   continue;
 115
 116                switch (intr->intrinsic) {
 117                case nir_intrinsic_load_input:
 118                   if (mask & nir_var_shader_in)
 119                      lower_load_input_to_scalar(&b, intr);
 120                   break;
 121                case nir_intrinsic_store_output:
 122                   if (mask & nir_var_shader_out)
 123                      lower_store_output_to_scalar(&b, intr);
 124                   break;
 125                default:
 126                   break;
 127                }
 128             }
 129          }
 130       }
 131    }
 132 }
 133
 134 static nir_variable **
 135 get_channel_variables(struct hash_table *ht, nir_variable *var)
 136 {
 137    nir_variable **chan_vars;
 138    struct hash_entry *entry = _mesa_hash_table_search(ht, var);
 139    if (!entry) {
 140       chan_vars = (nir_variable **) calloc(4, sizeof(nir_variable *));
 141       _mesa_hash_table_insert(ht, var, chan_vars);
 142    } else {
 143       chan_vars = (nir_variable **) entry->data;
 144    }
 145
 146    return chan_vars;
 147 }
 148
 149 /*
 150  * Note that the src deref that we are cloning is the head of the
 151  * chain of deref instructions from the original intrinsic, but
 152  * the dst we are cloning to is the tail (because chains of deref
 153  * instructions are created back to front)
 154  */
 155
 156 static nir_deref_instr *
 157 clone_deref_array(nir_builder *b, nir_deref_instr *dst_tail,
 158                   const nir_deref_instr *src_head)
 159 {
 160    const nir_deref_instr *parent = nir_deref_instr_parent(src_head);
 161
 162    if (!parent)
 163       return dst_tail;
 164
 165    assert(src_head->deref_type == nir_deref_type_array);
 166
 167    dst_tail = clone_deref_array(b, dst_tail, parent);
 168
 169    return nir_build_deref_array(b, dst_tail,
 170                                 nir_ssa_for_src(b, src_head->arr.index, 1));
 171 }
 172
 173 static void
 174 lower_load_to_scalar_early(nir_builder *b, nir_intrinsic_instr *intr,
 175                            nir_variable *var, struct hash_table *split_inputs,
 176                            struct hash_table *split_outputs)
 177 {
 178    b->cursor = nir_before_instr(&intr->instr);
 179
 180    assert(intr->dest.is_ssa);
 181
 182    nir_ssa_def *loads[NIR_MAX_VEC_COMPONENTS];
 183
 184    nir_variable **chan_vars;
 185    if (var->data.mode == nir_var_shader_in) {
 186       chan_vars = get_channel_variables(split_inputs, var);
 187    } else {
 188       chan_vars = get_channel_variables(split_outputs, var);
 189    }
 190
 191    for (unsigned i = 0; i < intr->num_components; i++) {
 192       nir_variable *chan_var = chan_vars[var->data.location_frac + i];
 193       if (!chan_vars[var->data.location_frac + i]) {
 194          chan_var = nir_variable_clone(var, b->shader);
 195          chan_var->data.location_frac =  var->data.location_frac + i;
 196          chan_var->type = glsl_channel_type(chan_var->type);
 197          if (var->data.explicit_offset) {
 198             unsigned comp_size = glsl_get_bit_size(chan_var->type) / 8;
 199             chan_var->data.offset = var->data.offset + i * comp_size;
 200          }
 201
 202          chan_vars[var->data.location_frac + i] = chan_var;
 203
 204          nir_shader_add_variable(b->shader, chan_var);
 205       }
 206
 207       nir_intrinsic_instr *chan_intr =
 208          nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 209       nir_ssa_dest_init(&chan_intr->instr, &chan_intr->dest,
 210                         1, intr->dest.ssa.bit_size, NULL);
 211       chan_intr->num_components = 1;
 212
 213       nir_deref_instr *deref = nir_build_deref_var(b, chan_var);
 214
 215       deref = clone_deref_array(b, deref, nir_src_as_deref(intr->src[0]));
 216
 217       chan_intr->src[0] = nir_src_for_ssa(&deref->dest.ssa);
 218
 219       if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
 220           intr->intrinsic == nir_intrinsic_interp_deref_at_sample ||
 221           intr->intrinsic == nir_intrinsic_interp_deref_at_vertex)
 222          nir_src_copy(&chan_intr->src[1], &intr->src[1], &chan_intr->instr);
 223
 224       nir_builder_instr_insert(b, &chan_intr->instr);
 225
 226       loads[i] = &chan_intr->dest.ssa;
 227    }
 228
 229    nir_ssa_def_rewrite_uses(&intr->dest.ssa,
 230                             nir_src_for_ssa(nir_vec(b, loads,
 231                                                     intr->num_components)));
 232
 233    /* Remove the old load intrinsic */
 234    nir_instr_remove(&intr->instr);
 235 }
 236
 237 static void
 238 lower_store_output_to_scalar_early(nir_builder *b, nir_intrinsic_instr *intr,
 239                                    nir_variable *var,
 240                                    struct hash_table *split_outputs)
 241 {
 242    b->cursor = nir_before_instr(&intr->instr);
 243
 244    nir_ssa_def *value = nir_ssa_for_src(b, intr->src[1], intr->num_components);
 245
 246    nir_variable **chan_vars = get_channel_variables(split_outputs, var);
 247    for (unsigned i = 0; i < intr->num_components; i++) {
 248       if (!(nir_intrinsic_write_mask(intr) & (1 << i)))
 249          continue;
 250
 251       nir_variable *chan_var = chan_vars[var->data.location_frac + i];
 252       if (!chan_vars[var->data.location_frac + i]) {
 253          chan_var = nir_variable_clone(var, b->shader);
 254          chan_var->data.location_frac =  var->data.location_frac + i;
 255          chan_var->type = glsl_channel_type(chan_var->type);
 256          if (var->data.explicit_offset) {
 257             unsigned comp_size = glsl_get_bit_size(chan_var->type) / 8;
 258             chan_var->data.offset = var->data.offset + i * comp_size;
 259          }
 260
 261          chan_vars[var->data.location_frac + i] = chan_var;
 262
 263          nir_shader_add_variable(b->shader, chan_var);
 264       }
 265
 266       nir_intrinsic_instr *chan_intr =
 267          nir_intrinsic_instr_create(b->shader, intr->intrinsic);
 268       chan_intr->num_components = 1;
 269
 270       nir_intrinsic_set_write_mask(chan_intr, 0x1);
 271
 272       nir_deref_instr *deref = nir_build_deref_var(b, chan_var);
 273
 274       deref = clone_deref_array(b, deref, nir_src_as_deref(intr->src[0]));
 275
 276       chan_intr->src[0] = nir_src_for_ssa(&deref->dest.ssa);
 277       chan_intr->src[1] = nir_src_for_ssa(nir_channel(b, value, i));
 278
 279       nir_builder_instr_insert(b, &chan_intr->instr);
 280    }
 281
 282    /* Remove the old store intrinsic */
 283    nir_instr_remove(&intr->instr);
 284 }
 285
 286 /*
 287  * This function is intended to be called earlier than nir_lower_io_to_scalar()
 288  * i.e. before nir_lower_io() is called.
 289  */
 290 void
 291 nir_lower_io_to_scalar_early(nir_shader *shader, nir_variable_mode mask)
 292 {
 293    struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
 294    struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
 295
 296    nir_foreach_function(function, shader) {
 297       if (function->impl) {
 298          nir_builder b;
 299          nir_builder_init(&b, function->impl);
 300
 301          nir_foreach_block(block, function->impl) {
 302             nir_foreach_instr_safe(instr, block) {
 303                if (instr->type != nir_instr_type_intrinsic)
 304                   continue;
 305
 306                nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 307
 308                if (intr->num_components == 1)
 309                   continue;
 310
 311                if (intr->intrinsic != nir_intrinsic_load_deref &&
 312                    intr->intrinsic != nir_intrinsic_store_deref &&
 313                    intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 314                    intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 315                    intr->intrinsic != nir_intrinsic_interp_deref_at_offset &&
 316                    intr->intrinsic != nir_intrinsic_interp_deref_at_vertex)
 317                   continue;
 318
 319                nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
 320                nir_variable_mode mode = deref->mode;
 321                if (!(mode & mask))
 322                   continue;
 323
 324                nir_variable *var = nir_deref_instr_get_variable(deref);
 325
 326                /* TODO: add patch support */
 327                if (var->data.patch)
 328                   continue;
 329
 330                /* TODO: add doubles support */
 331                if (glsl_type_is_64bit(glsl_without_array(var->type)))
 332                   continue;
 333
 334                if (!(shader->info.stage == MESA_SHADER_VERTEX &&
 335                      mode == nir_var_shader_in) &&
 336                    var->data.location < VARYING_SLOT_VAR0 &&
 337                    var->data.location >= 0)
 338                   continue;
 339
 340                /* Don't bother splitting if we can't opt away any unused
 341                 * components.
 342                 */
 343                if (var->data.always_active_io)
 344                   continue;
 345
 346               /* Skip types we cannot split */
 347               if (glsl_type_is_matrix(glsl_without_array(var->type)) ||
 348                   glsl_type_is_struct_or_ifc(glsl_without_array(var->type)))
 349                  continue;
 350
 351                switch (intr->intrinsic) {
 352                case nir_intrinsic_interp_deref_at_centroid:
 353                case nir_intrinsic_interp_deref_at_sample:
 354                case nir_intrinsic_interp_deref_at_offset:
 355                case nir_intrinsic_interp_deref_at_vertex:
 356                case nir_intrinsic_load_deref:
 357                   if ((mask & nir_var_shader_in && mode == nir_var_shader_in) ||
 358                       (mask & nir_var_shader_out && mode == nir_var_shader_out))
 359                      lower_load_to_scalar_early(&b, intr, var, split_inputs,
 360                                                 split_outputs);
 361                   break;
 362                case nir_intrinsic_store_deref:
 363                   if (mask & nir_var_shader_out &&
 364                       mode == nir_var_shader_out)
 365                      lower_store_output_to_scalar_early(&b, intr, var,
 366                                                         split_outputs);
 367                   break;
 368                default:
 369                   break;
 370                }
 371             }
 372          }
 373       }
 374    }
 375
 376    /* Remove old input from the shaders inputs list */
 377    hash_table_foreach(split_inputs, entry) {
 378       nir_variable *var = (nir_variable *) entry->key;
 379       exec_node_remove(&var->node);
 380
 381       free(entry->data);
 382    }
 383
 384    /* Remove old output from the shaders outputs list */
 385    hash_table_foreach(split_outputs, entry) {
 386       nir_variable *var = (nir_variable *) entry->key;
 387       exec_node_remove(&var->node);
 388
 389       free(entry->data);
 390    }
 391
 392    _mesa_hash_table_destroy(split_inputs, NULL);
 393    _mesa_hash_table_destroy(split_outputs, NULL);
 394
 395    nir_remove_dead_derefs(shader);
 396 }