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