src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp

   1 /*
   2  * Copyright © 2010 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 /**
  25  * \file brw_wm_vector_splitting.cpp
  26  *
  27  * If a vector is only ever referenced by its components, then
  28  * split those components out to individual variables so they can be
  29  * handled normally by other optimization passes.
  30  *
  31  * This skips vectors in uniforms and varyings, which need to be
  32  * accessible as vectors for their access by the GL.  Also, vector
  33  * results of non-variable-derefs in assignments aren't handled
  34  * because to do so we would have to store the vector result to a
  35  * temporary in order to unload each channel, and to do so would just
  36  * loop us back to where we started.  For the 965, this is exactly the
  37  * behavior we want for the results of texture lookups, but probably not for
  38  */
  39
  40 extern "C" {
  41 #include "main/core.h"
  42 #include "intel_context.h"
  43 }
  44 #include "../glsl/ir.h"
  45 #include "../glsl/ir_visitor.h"
  46 #include "../glsl/ir_print_visitor.h"
  47 #include "../glsl/ir_rvalue_visitor.h"
  48 #include "../glsl/glsl_types.h"
  49
  50 static bool debug = false;
  51
  52 class variable_entry : public exec_node
  53 {
  54 public:
  55    variable_entry(ir_variable *var)
  56    {
  57       this->var = var;
  58       this->whole_vector_access = 0;
  59       this->declaration = false;
  60       this->mem_ctx = NULL;
  61    }
  62
  63    ir_variable *var; /* The key: the variable's pointer. */
  64
  65    /** Number of times the variable is referenced, including assignments. */
  66    unsigned whole_vector_access;
  67
  68    bool declaration; /* If the variable had a decl in the instruction stream */
  69
  70    ir_variable *components[4];
  71
  72    /** talloc_parent(this->var) -- the shader's talloc context. */
  73    void *mem_ctx;
  74 };
  75
  76 class ir_vector_reference_visitor : public ir_hierarchical_visitor {
  77 public:
  78    ir_vector_reference_visitor(void)
  79    {
  80       this->mem_ctx = talloc_new(NULL);
  81       this->variable_list.make_empty();
  82    }
  83
  84    ~ir_vector_reference_visitor(void)
  85    {
  86       talloc_free(mem_ctx);
  87    }
  88
  89    virtual ir_visitor_status visit(ir_variable *);
  90    virtual ir_visitor_status visit(ir_dereference_variable *);
  91    virtual ir_visitor_status visit_enter(ir_swizzle *);
  92    virtual ir_visitor_status visit_enter(ir_assignment *);
  93    virtual ir_visitor_status visit_enter(ir_function_signature *);
  94
  95    variable_entry *get_variable_entry(ir_variable *var);
  96
  97    /* List of variable_entry */
  98    exec_list variable_list;
  99
 100    void *mem_ctx;
 101 };
 102
 103 variable_entry *
 104 ir_vector_reference_visitor::get_variable_entry(ir_variable *var)
 105 {
 106    assert(var);
 107
 108    if (!var->type->is_vector())
 109       return NULL;
 110
 111    switch (var->mode) {
 112    case ir_var_uniform:
 113    case ir_var_in:
 114    case ir_var_out:
 115    case ir_var_inout:
 116       /* Can't split varyings or uniforms.  Function in/outs won't get split
 117        * either, so don't care about the ambiguity.
 118        */
 119       return NULL;
 120    case ir_var_auto:
 121    case ir_var_temporary:
 122       break;
 123    }
 124
 125    foreach_iter(exec_list_iterator, iter, this->variable_list) {
 126       variable_entry *entry = (variable_entry *)iter.get();
 127       if (entry->var == var)
 128          return entry;
 129    }
 130
 131    variable_entry *entry = new(mem_ctx) variable_entry(var);
 132    this->variable_list.push_tail(entry);
 133    return entry;
 134 }
 135
 136
 137 ir_visitor_status
 138 ir_vector_reference_visitor::visit(ir_variable *ir)
 139 {
 140    variable_entry *entry = this->get_variable_entry(ir);
 141
 142    if (entry)
 143       entry->declaration = true;
 144
 145    return visit_continue;
 146 }
 147
 148 ir_visitor_status
 149 ir_vector_reference_visitor::visit(ir_dereference_variable *ir)
 150 {
 151    ir_variable *const var = ir->var;
 152    variable_entry *entry = this->get_variable_entry(var);
 153
 154    if (entry)
 155       entry->whole_vector_access++;
 156
 157    return visit_continue;
 158 }
 159
 160 ir_visitor_status
 161 ir_vector_reference_visitor::visit_enter(ir_swizzle *ir)
 162 {
 163    /* Don't descend into a vector ir_dereference_variable below. */
 164    if (ir->val->as_dereference_variable() && ir->type->is_scalar())
 165       return visit_continue_with_parent;
 166
 167    return visit_continue;
 168 }
 169
 170 ir_visitor_status
 171 ir_vector_reference_visitor::visit_enter(ir_assignment *ir)
 172 {
 173    if (ir->lhs->as_dereference_variable() &&
 174        ir->rhs->as_dereference_variable() &&
 175        !ir->condition) {
 176       /* We'll split copies of a vector to copies of channels, so don't
 177        * descend to the ir_dereference_variables.
 178        */
 179       return visit_continue_with_parent;
 180    }
 181    if (ir->lhs->as_dereference_variable() &&
 182        is_power_of_two(ir->write_mask) &&
 183        !ir->condition) {
 184       /* If we're writing just a channel, then channel-splitting the LHS is OK.
 185        */
 186       ir->rhs->accept(this);
 187       return visit_continue_with_parent;
 188    }
 189    return visit_continue;
 190 }
 191
 192 ir_visitor_status
 193 ir_vector_reference_visitor::visit_enter(ir_function_signature *ir)
 194 {
 195    /* We don't want to descend into the function parameters and
 196     * split them, so just accept the body here.
 197     */
 198    visit_list_elements(this, &ir->body);
 199    return visit_continue_with_parent;
 200 }
 201
 202 class ir_vector_splitting_visitor : public ir_rvalue_visitor {
 203 public:
 204    ir_vector_splitting_visitor(exec_list *vars)
 205    {
 206       this->variable_list = vars;
 207    }
 208
 209    virtual ir_visitor_status visit_leave(ir_assignment *);
 210
 211    void handle_rvalue(ir_rvalue **rvalue);
 212    struct variable_entry *get_splitting_entry(ir_variable *var);
 213
 214    exec_list *variable_list;
 215    void *mem_ctx;
 216 };
 217
 218 struct variable_entry *
 219 ir_vector_splitting_visitor::get_splitting_entry(ir_variable *var)
 220 {
 221    assert(var);
 222
 223    if (!var->type->is_vector())
 224       return NULL;
 225
 226    foreach_iter(exec_list_iterator, iter, *this->variable_list) {
 227       variable_entry *entry = (variable_entry *)iter.get();
 228       if (entry->var == var) {
 229          return entry;
 230       }
 231    }
 232
 233    return NULL;
 234 }
 235
 236 void
 237 ir_vector_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 238 {
 239    if (!*rvalue)
 240       return;
 241
 242    ir_swizzle *swiz = (*rvalue)->as_swizzle();
 243    if (!swiz || !swiz->type->is_scalar())
 244       return;
 245
 246    ir_dereference_variable *deref_var = swiz->val->as_dereference_variable();
 247    if (!deref_var)
 248       return;
 249
 250    variable_entry *entry = get_splitting_entry(deref_var->var);
 251    if (!entry)
 252       return;
 253
 254    ir_variable *var = entry->components[swiz->mask.x];
 255    *rvalue = new(entry->mem_ctx) ir_dereference_variable(var);
 256 }
 257
 258 ir_visitor_status
 259 ir_vector_splitting_visitor::visit_leave(ir_assignment *ir)
 260 {
 261    ir_dereference_variable *lhs_deref = ir->lhs->as_dereference_variable();
 262    ir_dereference_variable *rhs_deref = ir->rhs->as_dereference_variable();
 263    variable_entry *lhs = lhs_deref ? get_splitting_entry(lhs_deref->var) : NULL;
 264    variable_entry *rhs = rhs_deref ? get_splitting_entry(rhs_deref->var) : NULL;
 265
 266    if (lhs_deref && rhs_deref && (lhs || rhs) && !ir->condition) {
 267       /* Straight assignment of vector variables. */
 268       for (unsigned int i = 0; i < ir->rhs->type->vector_elements; i++) {
 269          ir_dereference *new_lhs;
 270          ir_rvalue *new_rhs;
 271          void *mem_ctx = lhs ? lhs->mem_ctx : rhs->mem_ctx;
 272          unsigned int writemask;
 273
 274          if (lhs) {
 275             new_lhs = new(mem_ctx) ir_dereference_variable(lhs->components[i]);
 276             writemask = (ir->write_mask >> i) & 1;
 277          } else {
 278             new_lhs = ir->lhs->clone(mem_ctx, NULL);
 279             writemask = ir->write_mask & (1 << i);
 280          }
 281
 282          if (rhs) {
 283             new_rhs = new(mem_ctx) ir_dereference_variable(rhs->components[i]);
 284             /* If we're writing into a writemask, smear it out to that channel. */
 285             if (!lhs)
 286                new_rhs = new(mem_ctx) ir_swizzle(new_rhs, i, i, i, i, i + 1);
 287          } else {
 288             new_rhs = new(mem_ctx) ir_swizzle(ir->rhs->clone(mem_ctx, NULL),
 289                                               i, i, i, i, 1);
 290          }
 291
 292          ir->insert_before(new(mem_ctx) ir_assignment(new_lhs,
 293                                                       new_rhs,
 294                                                       NULL, writemask));
 295       }
 296       ir->remove();
 297    } else if (lhs) {
 298       int elem = -1;
 299
 300       switch (ir->write_mask) {
 301       case (1 << 0):
 302          elem = 0;
 303          break;
 304       case (1 << 1):
 305          elem = 1;
 306          break;
 307       case (1 << 2):
 308          elem = 2;
 309          break;
 310       case (1 << 3):
 311          elem = 3;
 312          break;
 313       default:
 314          ir->print();
 315          assert(!"not reached: non-channelwise dereference of LHS.");
 316       }
 317
 318       ir->lhs = new(mem_ctx) ir_dereference_variable(lhs->components[elem]);
 319       ir->write_mask = (1 << 0);
 320
 321       handle_rvalue(&ir->rhs);
 322       ir->rhs = new(mem_ctx) ir_swizzle(ir->rhs,
 323                                         elem, elem, elem, elem, 1);
 324    } else {
 325       handle_rvalue(&ir->rhs);
 326    }
 327
 328    handle_rvalue(&ir->condition);
 329
 330    return visit_continue;
 331 }
 332
 333 extern "C" {
 334 bool
 335 brw_do_vector_splitting(exec_list *instructions)
 336 {
 337    ir_vector_reference_visitor refs;
 338
 339    visit_list_elements(&refs, instructions);
 340
 341    /* Trim out variables we can't split. */
 342    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 343       variable_entry *entry = (variable_entry *)iter.get();
 344
 345       if (debug) {
 346          printf("vector %s@%p: decl %d, whole_access %d\n",
 347                 entry->var->name, (void *) entry->var, entry->declaration,
 348                 entry->whole_vector_access);
 349       }
 350
 351       if (!entry->declaration || entry->whole_vector_access) {
 352          entry->remove();
 353       }
 354    }
 355
 356    if (refs.variable_list.is_empty())
 357       return false;
 358
 359    void *mem_ctx = talloc_new(NULL);
 360
 361    /* Replace the decls of the vectors to be split with their split
 362     * components.
 363     */
 364    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 365       variable_entry *entry = (variable_entry *)iter.get();
 366       const struct glsl_type *type;
 367       type = glsl_type::get_instance(entry->var->type->base_type, 1, 1);
 368
 369       entry->mem_ctx = talloc_parent(entry->var);
 370
 371       for (unsigned int i = 0; i < entry->var->type->vector_elements; i++) {
 372          const char *name = talloc_asprintf(mem_ctx, "%s_%c",
 373                                             entry->var->name,
 374                                             "xyzw"[i]);
 375
 376          entry->components[i] = new(entry->mem_ctx) ir_variable(type, name,
 377                                                                 ir_var_temporary);
 378          entry->var->insert_before(entry->components[i]);
 379       }
 380
 381       entry->var->remove();
 382    }
 383
 384    ir_vector_splitting_visitor split(&refs.variable_list);
 385    visit_list_elements(&split, instructions);
 386
 387    talloc_free(mem_ctx);
 388
 389    return true;
 390 }
 391 }