src/glsl/lower_ubo_reference.cpp

   1 /*
   2  * Copyright © 2012 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 lower_ubo_reference.cpp
  26  *
  27  * IR lower pass to replace dereferences of variables in a uniform
  28  * buffer object with usage of ir_binop_ubo_load expressions, each of
  29  * which can read data up to the size of a vec4.
  30  *
  31  * This relieves drivers of the responsibility to deal with tricky UBO
  32  * layout issues like std140 structures and row_major matrices on
  33  * their own.
  34  */
  35
  36 #include "ir.h"
  37 #include "ir_builder.h"
  38 #include "ir_rvalue_visitor.h"
  39 #include "main/macros.h"
  40
  41 using namespace ir_builder;
  42
  43 namespace {
  44 class lower_ubo_reference_visitor : public ir_rvalue_enter_visitor {
  45 public:
  46    lower_ubo_reference_visitor(struct gl_shader *shader)
  47    : shader(shader)
  48    {
  49    }
  50
  51    void handle_rvalue(ir_rvalue **rvalue);
  52    void emit_ubo_loads(ir_dereference *deref, ir_variable *base_offset,
  53                        unsigned int deref_offset);
  54    ir_expression *ubo_load(const struct glsl_type *type,
  55                            ir_rvalue *offset);
  56
  57    void *mem_ctx;
  58    struct gl_shader *shader;
  59    struct gl_uniform_buffer_variable *ubo_var;
  60    ir_rvalue *uniform_block;
  61    bool progress;
  62 };
  63
  64 /**
  65  * Determine the name of the interface block field
  66  *
  67  * This is the name of the specific member as it would appear in the
  68  * \c gl_uniform_buffer_variable::Name field in the shader's
  69  * \c UniformBlocks array.
  70  */
  71 static const char *
  72 interface_field_name(void *mem_ctx, char *base_name, ir_dereference *d,
  73                      ir_rvalue **nonconst_block_index)
  74 {
  75    ir_rvalue *previous_index = NULL;
  76    *nonconst_block_index = NULL;
  77
  78    while (d != NULL) {
  79       switch (d->ir_type) {
  80       case ir_type_dereference_variable: {
  81          ir_dereference_variable *v = (ir_dereference_variable *) d;
  82          if (previous_index
  83              && v->var->is_interface_instance()
  84              && v->var->type->is_array()) {
  85
  86             ir_constant *const_index = previous_index->as_constant();
  87             if (!const_index) {
  88                *nonconst_block_index = previous_index;
  89                return ralloc_asprintf(mem_ctx, "%s[0]", base_name);
  90             } else {
  91                return ralloc_asprintf(mem_ctx,
  92                                       "%s[%d]",
  93                                       base_name,
  94                                       const_index->get_uint_component(0));
  95             }
  96          } else {
  97             return base_name;
  98          }
  99
 100          break;
 101       }
 102
 103       case ir_type_dereference_record: {
 104          ir_dereference_record *r = (ir_dereference_record *) d;
 105
 106          d = r->record->as_dereference();
 107          break;
 108       }
 109
 110       case ir_type_dereference_array: {
 111          ir_dereference_array *a = (ir_dereference_array *) d;
 112
 113          d = a->array->as_dereference();
 114          previous_index = a->array_index;
 115
 116          break;
 117       }
 118
 119       default:
 120          assert(!"Should not get here.");
 121          break;
 122       }
 123    }
 124
 125    assert(!"Should not get here.");
 126    return NULL;
 127 }
 128
 129 void
 130 lower_ubo_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
 131 {
 132    if (!*rvalue)
 133       return;
 134
 135    ir_dereference *deref = (*rvalue)->as_dereference();
 136    if (!deref)
 137       return;
 138
 139    ir_variable *var = deref->variable_referenced();
 140    if (!var || !var->is_in_uniform_block())
 141       return;
 142
 143    mem_ctx = ralloc_parent(*rvalue);
 144
 145    ir_rvalue *nonconst_block_index;
 146    const char *const field_name =
 147       interface_field_name(mem_ctx, (char *) var->get_interface_type()->name,
 148                            deref, &nonconst_block_index);
 149
 150    this->uniform_block = NULL;
 151    for (unsigned i = 0; i < shader->NumUniformBlocks; i++) {
 152       if (strcmp(field_name, shader->UniformBlocks[i].Name) == 0) {
 153
 154          ir_constant *index = new(mem_ctx) ir_constant(i);
 155
 156          if (nonconst_block_index) {
 157             if (nonconst_block_index->type != glsl_type::uint_type)
 158                nonconst_block_index = i2u(nonconst_block_index);
 159             this->uniform_block = add(nonconst_block_index, index);
 160          } else {
 161             this->uniform_block = index;
 162          }
 163
 164          struct gl_uniform_block *block = &shader->UniformBlocks[i];
 165
 166          this->ubo_var = var->is_interface_instance()
 167             ? &block->Uniforms[0] : &block->Uniforms[var->data.location];
 168
 169          break;
 170       }
 171    }
 172
 173    assert(this->uniform_block);
 174
 175    ir_rvalue *offset = new(mem_ctx) ir_constant(0u);
 176    unsigned const_offset = 0;
 177    bool row_major = ubo_var->RowMajor;
 178
 179    /* Calculate the offset to the start of the region of the UBO
 180     * dereferenced by *rvalue.  This may be a variable offset if an
 181     * array dereference has a variable index.
 182     */
 183    while (deref) {
 184       switch (deref->ir_type) {
 185       case ir_type_dereference_variable: {
 186          const_offset += ubo_var->Offset;
 187          deref = NULL;
 188          break;
 189       }
 190
 191       case ir_type_dereference_array: {
 192          ir_dereference_array *deref_array = (ir_dereference_array *)deref;
 193          unsigned array_stride;
 194          if (deref_array->array->type->is_matrix() && row_major) {
 195             /* When loading a vector out of a row major matrix, the
 196              * step between the columns (vectors) is the size of a
 197              * float, while the step between the rows (elements of a
 198              * vector) is handled below in emit_ubo_loads.
 199              */
 200             array_stride = 4;
 201          } else if (deref_array->type->is_interface()) {
 202             /* We're processing an array dereference of an interface instance
 203              * array.  The thing being dereferenced *must* be a variable
 204              * dereference because intefaces cannot be embedded an other
 205              * types.  In terms of calculating the offsets for the lowering
 206              * pass, we don't care about the array index.  All elements of an
 207              * interface instance array will have the same offsets relative to
 208              * the base of the block that backs them.
 209              */
 210             assert(deref_array->array->as_dereference_variable());
 211             deref = deref_array->array->as_dereference();
 212             break;
 213          } else {
 214             array_stride = deref_array->type->std140_size(row_major);
 215             array_stride = glsl_align(array_stride, 16);
 216          }
 217
 218          ir_rvalue *array_index = deref_array->array_index;
 219          if (array_index->type->base_type == GLSL_TYPE_INT)
 220             array_index = i2u(array_index);
 221
 222          ir_constant *const_index =
 223             array_index->constant_expression_value(NULL);
 224          if (const_index) {
 225             const_offset += array_stride * const_index->value.u[0];
 226          } else {
 227             offset = add(offset,
 228                          mul(array_index,
 229                              new(mem_ctx) ir_constant(array_stride)));
 230          }
 231          deref = deref_array->array->as_dereference();
 232          break;
 233       }
 234
 235       case ir_type_dereference_record: {
 236          ir_dereference_record *deref_record = (ir_dereference_record *)deref;
 237          const glsl_type *struct_type = deref_record->record->type;
 238          unsigned intra_struct_offset = 0;
 239
 240          unsigned max_field_align = 16;
 241          for (unsigned int i = 0; i < struct_type->length; i++) {
 242             const glsl_type *type = struct_type->fields.structure[i].type;
 243             unsigned field_align = type->std140_base_alignment(row_major);
 244             max_field_align = MAX2(field_align, max_field_align);
 245             intra_struct_offset = glsl_align(intra_struct_offset, field_align);
 246
 247             if (strcmp(struct_type->fields.structure[i].name,
 248                        deref_record->field) == 0)
 249                break;
 250             intra_struct_offset += type->std140_size(row_major);
 251          }
 252
 253          const_offset = glsl_align(const_offset, max_field_align);
 254          const_offset += intra_struct_offset;
 255
 256          deref = deref_record->record->as_dereference();
 257          break;
 258       }
 259       default:
 260          assert(!"not reached");
 261          deref = NULL;
 262          break;
 263       }
 264    }
 265
 266    /* Now that we've calculated the offset to the start of the
 267     * dereference, walk over the type and emit loads into a temporary.
 268     */
 269    const glsl_type *type = (*rvalue)->type;
 270    ir_variable *load_var = new(mem_ctx) ir_variable(type,
 271                                                     "ubo_load_temp",
 272                                                     ir_var_temporary);
 273    base_ir->insert_before(load_var);
 274
 275    ir_variable *load_offset = new(mem_ctx) ir_variable(glsl_type::uint_type,
 276                                                        "ubo_load_temp_offset",
 277                                                        ir_var_temporary);
 278    base_ir->insert_before(load_offset);
 279    base_ir->insert_before(assign(load_offset, offset));
 280
 281    deref = new(mem_ctx) ir_dereference_variable(load_var);
 282    emit_ubo_loads(deref, load_offset, const_offset);
 283    *rvalue = deref;
 284
 285    progress = true;
 286 }
 287
 288 ir_expression *
 289 lower_ubo_reference_visitor::ubo_load(const glsl_type *type,
 290                                       ir_rvalue *offset)
 291 {
 292    ir_rvalue *block_ref = this->uniform_block->clone(mem_ctx, NULL);
 293    return new(mem_ctx)
 294       ir_expression(ir_binop_ubo_load,
 295                     type,
 296                     block_ref,
 297                     offset);
 298
 299 }
 300
 301 /**
 302  * Takes LHS and emits a series of assignments into its components
 303  * from the UBO variable at variable_offset + deref_offset.
 304  *
 305  * Recursively calls itself to break the deref down to the point that
 306  * the ir_binop_ubo_load expressions generated are contiguous scalars
 307  * or vectors.
 308  */
 309 void
 310 lower_ubo_reference_visitor::emit_ubo_loads(ir_dereference *deref,
 311                                             ir_variable *base_offset,
 312                                             unsigned int deref_offset)
 313 {
 314    if (deref->type->is_record()) {
 315       unsigned int field_offset = 0;
 316
 317       for (unsigned i = 0; i < deref->type->length; i++) {
 318          const struct glsl_struct_field *field =
 319             &deref->type->fields.structure[i];
 320          ir_dereference *field_deref =
 321             new(mem_ctx) ir_dereference_record(deref->clone(mem_ctx, NULL),
 322                                                field->name);
 323
 324          field_offset =
 325             glsl_align(field_offset,
 326                        field->type->std140_base_alignment(ubo_var->RowMajor));
 327
 328          emit_ubo_loads(field_deref, base_offset, deref_offset + field_offset);
 329
 330          field_offset += field->type->std140_size(ubo_var->RowMajor);
 331       }
 332       return;
 333    }
 334
 335    if (deref->type->is_array()) {
 336       unsigned array_stride =
 337          glsl_align(deref->type->fields.array->std140_size(ubo_var->RowMajor),
 338                     16);
 339
 340       for (unsigned i = 0; i < deref->type->length; i++) {
 341          ir_constant *element = new(mem_ctx) ir_constant(i);
 342          ir_dereference *element_deref =
 343             new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
 344                                               element);
 345          emit_ubo_loads(element_deref, base_offset,
 346                         deref_offset + i * array_stride);
 347       }
 348       return;
 349    }
 350
 351    if (deref->type->is_matrix()) {
 352       for (unsigned i = 0; i < deref->type->matrix_columns; i++) {
 353          ir_constant *col = new(mem_ctx) ir_constant(i);
 354          ir_dereference *col_deref =
 355             new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
 356                                               col);
 357
 358          if (ubo_var->RowMajor) {
 359             /* For a row-major matrix, the next column starts at the next
 360              * element.
 361              */
 362             emit_ubo_loads(col_deref, base_offset, deref_offset + i * 4);
 363          } else {
 364             /* std140 always rounds the stride of arrays (and matrices) to a
 365              * vec4, so matrices are always 16 between columns/rows.
 366              */
 367             emit_ubo_loads(col_deref, base_offset, deref_offset + i * 16);
 368          }
 369       }
 370       return;
 371    }
 372
 373    assert(deref->type->is_scalar() ||
 374           deref->type->is_vector());
 375
 376    if (!ubo_var->RowMajor) {
 377       ir_rvalue *offset = add(base_offset,
 378                               new(mem_ctx) ir_constant(deref_offset));
 379       base_ir->insert_before(assign(deref->clone(mem_ctx, NULL),
 380                                     ubo_load(deref->type, offset)));
 381    } else {
 382       /* We're dereffing a column out of a row-major matrix, so we
 383        * gather the vector from each stored row.
 384       */
 385       assert(deref->type->base_type == GLSL_TYPE_FLOAT);
 386       /* Matrices, row_major or not, are stored as if they were
 387        * arrays of vectors of the appropriate size in std140.
 388        * Arrays have their strides rounded up to a vec4, so the
 389        * matrix stride is always 16.
 390        */
 391       unsigned matrix_stride = 16;
 392
 393       for (unsigned i = 0; i < deref->type->vector_elements; i++) {
 394          ir_rvalue *chan_offset =
 395             add(base_offset,
 396                 new(mem_ctx) ir_constant(deref_offset + i * matrix_stride));
 397
 398          base_ir->insert_before(assign(deref->clone(mem_ctx, NULL),
 399                                        ubo_load(glsl_type::float_type,
 400                                                 chan_offset),
 401                                        (1U << i)));
 402       }
 403    }
 404 }
 405
 406 } /* unnamed namespace */
 407
 408 void
 409 lower_ubo_reference(struct gl_shader *shader, exec_list *instructions)
 410 {
 411    lower_ubo_reference_visitor v(shader);
 412
 413    /* Loop over the instructions lowering references, because we take
 414     * a deref of a UBO array using a UBO dereference as the index will
 415     * produce a collection of instructions all of which have cloned
 416     * UBO dereferences for that array index.
 417     */
 418    do {
 419       v.progress = false;
 420       visit_list_elements(&v, instructions);
 421    } while (v.progress);
 422 }