src/compiler/glsl/ir_set_program_inouts.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 ir_set_program_inouts.cpp
  26  *
  27  * Sets the InputsRead and OutputsWritten of Mesa programs.
  28  *
  29  * Additionally, for fragment shaders, sets the InterpQualifier array, the
  30  * IsCentroid and IsSample bitfields, and the UsesDFdy flag.
  31  *
  32  * Mesa programs (gl_program, not gl_shader_program) have a set of
  33  * flags indicating which varyings are read and written.  Computing
  34  * which are actually read from some sort of backend code can be
  35  * tricky when variable array indexing involved.  So this pass
  36  * provides support for setting InputsRead and OutputsWritten right
  37  * from the GLSL IR.
  38  */
  39
  40 #include "main/core.h" /* for struct gl_program */
  41 #include "ir.h"
  42 #include "ir_visitor.h"
  43 #include "compiler/glsl_types.h"
  44
  45 namespace {
  46
  47 class ir_set_program_inouts_visitor : public ir_hierarchical_visitor {
  48 public:
  49    ir_set_program_inouts_visitor(struct gl_program *prog,
  50                                  gl_shader_stage shader_stage)
  51    {
  52       this->prog = prog;
  53       this->shader_stage = shader_stage;
  54    }
  55    ~ir_set_program_inouts_visitor()
  56    {
  57    }
  58
  59    virtual ir_visitor_status visit_enter(ir_dereference_array *);
  60    virtual ir_visitor_status visit_enter(ir_function_signature *);
  61    virtual ir_visitor_status visit_enter(ir_expression *);
  62    virtual ir_visitor_status visit_enter(ir_discard *);
  63    virtual ir_visitor_status visit_enter(ir_texture *);
  64    virtual ir_visitor_status visit(ir_dereference_variable *);
  65
  66 private:
  67    void mark_whole_variable(ir_variable *var);
  68    bool try_mark_partial_variable(ir_variable *var, ir_rvalue *index);
  69
  70    struct gl_program *prog;
  71    gl_shader_stage shader_stage;
  72 };
  73
  74 } /* anonymous namespace */
  75
  76 static inline bool
  77 is_shader_inout(ir_variable *var)
  78 {
  79    return var->data.mode == ir_var_shader_in ||
  80           var->data.mode == ir_var_shader_out ||
  81           var->data.mode == ir_var_system_value;
  82 }
  83
  84 static void
  85 mark(struct gl_program *prog, ir_variable *var, int offset, int len,
  86      gl_shader_stage stage)
  87 {
  88    /* As of GLSL 1.20, varyings can only be floats, floating-point
  89     * vectors or matrices, or arrays of them.  For Mesa programs using
  90     * InputsRead/OutputsWritten, everything but matrices uses one
  91     * slot, while matrices use a slot per column.  Presumably
  92     * something doing a more clever packing would use something other
  93     * than InputsRead/OutputsWritten.
  94     */
  95
  96    for (int i = 0; i < len; i++) {
  97       assert(var->data.location != -1);
  98
  99       int idx = var->data.location + offset + i;
 100       bool is_patch_generic = var->data.patch &&
 101                               idx != VARYING_SLOT_TESS_LEVEL_INNER &&
 102                               idx != VARYING_SLOT_TESS_LEVEL_OUTER &&
 103                               idx != VARYING_SLOT_BOUNDING_BOX0 &&
 104                               idx != VARYING_SLOT_BOUNDING_BOX1;
 105       GLbitfield64 bitfield;
 106
 107       if (is_patch_generic) {
 108          assert(idx >= VARYING_SLOT_PATCH0 && idx < VARYING_SLOT_TESS_MAX);
 109          bitfield = BITFIELD64_BIT(idx - VARYING_SLOT_PATCH0);
 110       }
 111       else {
 112          assert(idx < VARYING_SLOT_MAX);
 113          bitfield = BITFIELD64_BIT(idx);
 114       }
 115
 116       if (var->data.mode == ir_var_shader_in) {
 117          if (is_patch_generic)
 118             prog->PatchInputsRead |= bitfield;
 119          else
 120             prog->InputsRead |= bitfield;
 121
 122          /* double inputs read is only for vertex inputs */
 123          if (stage == MESA_SHADER_VERTEX &&
 124              var->type->without_array()->is_dual_slot())
 125             prog->DoubleInputsRead |= bitfield;
 126
 127          if (stage == MESA_SHADER_FRAGMENT) {
 128             gl_fragment_program *fprog = (gl_fragment_program *) prog;
 129             fprog->InterpQualifier[idx] =
 130                (glsl_interp_mode) var->data.interpolation;
 131             if (var->data.centroid)
 132                fprog->IsCentroid |= bitfield;
 133             if (var->data.sample)
 134                fprog->IsSample |= bitfield;
 135          }
 136       } else if (var->data.mode == ir_var_system_value) {
 137          prog->SystemValuesRead |= bitfield;
 138       } else {
 139          assert(var->data.mode == ir_var_shader_out);
 140          if (is_patch_generic) {
 141             prog->PatchOutputsWritten |= bitfield;
 142          } else if (!var->data.read_only) {
 143             prog->OutputsWritten |= bitfield;
 144             if (var->data.index > 0)
 145                prog->SecondaryOutputsWritten |= bitfield;
 146          }
 147
 148          if (var->data.fb_fetch_output)
 149             prog->OutputsRead |= bitfield;
 150       }
 151    }
 152 }
 153
 154 /**
 155  * Mark an entire variable as used.  Caller must ensure that the variable
 156  * represents a shader input or output.
 157  */
 158 void
 159 ir_set_program_inouts_visitor::mark_whole_variable(ir_variable *var)
 160 {
 161    const glsl_type *type = var->type;
 162    bool is_vertex_input = false;
 163    if (this->shader_stage == MESA_SHADER_GEOMETRY &&
 164        var->data.mode == ir_var_shader_in && type->is_array()) {
 165       type = type->fields.array;
 166    }
 167
 168    if (this->shader_stage == MESA_SHADER_TESS_CTRL &&
 169        var->data.mode == ir_var_shader_in) {
 170       assert(type->is_array());
 171       type = type->fields.array;
 172    }
 173
 174    if (this->shader_stage == MESA_SHADER_TESS_CTRL &&
 175        var->data.mode == ir_var_shader_out && !var->data.patch) {
 176       assert(type->is_array());
 177       type = type->fields.array;
 178    }
 179
 180    if (this->shader_stage == MESA_SHADER_TESS_EVAL &&
 181        var->data.mode == ir_var_shader_in && !var->data.patch) {
 182       assert(type->is_array());
 183       type = type->fields.array;
 184    }
 185
 186    if (this->shader_stage == MESA_SHADER_VERTEX &&
 187        var->data.mode == ir_var_shader_in)
 188       is_vertex_input = true;
 189
 190    mark(this->prog, var, 0, type->count_attribute_slots(is_vertex_input),
 191         this->shader_stage);
 192 }
 193
 194 /* Default handler: Mark all the locations in the variable as used. */
 195 ir_visitor_status
 196 ir_set_program_inouts_visitor::visit(ir_dereference_variable *ir)
 197 {
 198    if (!is_shader_inout(ir->var))
 199       return visit_continue;
 200
 201    mark_whole_variable(ir->var);
 202
 203    return visit_continue;
 204 }
 205
 206 /**
 207  * Try to mark a portion of the given variable as used.  Caller must ensure
 208  * that the variable represents a shader input or output which can be indexed
 209  * into in array fashion (an array or matrix).  For the purpose of geometry
 210  * shader inputs (which are always arrays*), this means that the array element
 211  * must be something that can be indexed into in array fashion.
 212  *
 213  * *Except gl_PrimitiveIDIn, as noted below.
 214  *
 215  * For tessellation control shaders all inputs and non-patch outputs are
 216  * arrays. For tessellation evaluation shaders non-patch inputs are arrays.
 217  *
 218  * If the index can't be interpreted as a constant, or some other problem
 219  * occurs, then nothing will be marked and false will be returned.
 220  */
 221 bool
 222 ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable *var,
 223                                                          ir_rvalue *index)
 224 {
 225    const glsl_type *type = var->type;
 226
 227    if (this->shader_stage == MESA_SHADER_GEOMETRY &&
 228        var->data.mode == ir_var_shader_in) {
 229       /* The only geometry shader input that is not an array is
 230        * gl_PrimitiveIDIn, and in that case, this code will never be reached,
 231        * because gl_PrimitiveIDIn can't be indexed into in array fashion.
 232        */
 233       assert(type->is_array());
 234       type = type->fields.array;
 235    }
 236
 237    if (this->shader_stage == MESA_SHADER_TESS_CTRL &&
 238        var->data.mode == ir_var_shader_in) {
 239       assert(type->is_array());
 240       type = type->fields.array;
 241    }
 242
 243    if (this->shader_stage == MESA_SHADER_TESS_CTRL &&
 244        var->data.mode == ir_var_shader_out && !var->data.patch) {
 245       assert(type->is_array());
 246       type = type->fields.array;
 247    }
 248
 249    if (this->shader_stage == MESA_SHADER_TESS_EVAL &&
 250        var->data.mode == ir_var_shader_in && !var->data.patch) {
 251       assert(type->is_array());
 252       type = type->fields.array;
 253    }
 254
 255    /* TODO: implement proper arrays of arrays support
 256     * for now let the caller mark whole variable as used.
 257     */
 258    if (type->is_array() && type->fields.array->is_array())
 259       return false;
 260
 261    /* The code below only handles:
 262     *
 263     * - Indexing into matrices
 264     * - Indexing into arrays of (matrices, vectors, or scalars)
 265     *
 266     * All other possibilities are either prohibited by GLSL (vertex inputs and
 267     * fragment outputs can't be structs) or should have been eliminated by
 268     * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into
 269     * vectors, and lower_packed_varyings() gets rid of structs that occur in
 270     * varyings).
 271     *
 272     * However, we don't use varying packing in all cases - tessellation
 273     * shaders bypass it.  This means we'll see varying structs and arrays
 274     * of structs here.  For now, we just give up so the caller marks the
 275     * entire variable as used.
 276     */
 277    if (!(type->is_matrix() ||
 278         (type->is_array() &&
 279          (type->fields.array->is_numeric() ||
 280           type->fields.array->is_boolean())))) {
 281
 282       /* If we don't know how to handle this case, give up and let the
 283        * caller mark the whole variable as used.
 284        */
 285       return false;
 286    }
 287
 288    ir_constant *index_as_constant = index->as_constant();
 289    if (!index_as_constant)
 290       return false;
 291
 292    unsigned elem_width;
 293    unsigned num_elems;
 294    if (type->is_array()) {
 295       num_elems = type->length;
 296       if (type->fields.array->is_matrix())
 297          elem_width = type->fields.array->matrix_columns;
 298       else
 299          elem_width = 1;
 300    } else {
 301       num_elems = type->matrix_columns;
 302       elem_width = 1;
 303    }
 304
 305    if (index_as_constant->value.u[0] >= num_elems) {
 306       /* Constant index outside the bounds of the matrix/array.  This could
 307        * arise as a result of constant folding of a legal GLSL program.
 308        *
 309        * Even though the spec says that indexing outside the bounds of a
 310        * matrix/array results in undefined behaviour, we don't want to pass
 311        * out-of-range values to mark() (since this could result in slots that
 312        * don't exist being marked as used), so just let the caller mark the
 313        * whole variable as used.
 314        */
 315       return false;
 316    }
 317
 318    /* double element width for double types that takes two slots */
 319    if (this->shader_stage != MESA_SHADER_VERTEX ||
 320        var->data.mode != ir_var_shader_in) {
 321       if (type->without_array()->is_dual_slot())
 322          elem_width *= 2;
 323    }
 324
 325    mark(this->prog, var, index_as_constant->value.u[0] * elem_width,
 326         elem_width, this->shader_stage);
 327    return true;
 328 }
 329
 330 static bool
 331 is_multiple_vertices(gl_shader_stage stage, ir_variable *var)
 332 {
 333    if (var->data.patch)
 334       return false;
 335
 336    if (var->data.mode == ir_var_shader_in)
 337       return stage == MESA_SHADER_GEOMETRY ||
 338              stage == MESA_SHADER_TESS_CTRL ||
 339              stage == MESA_SHADER_TESS_EVAL;
 340    if (var->data.mode == ir_var_shader_out)
 341       return stage == MESA_SHADER_TESS_CTRL;
 342
 343    return false;
 344 }
 345
 346 ir_visitor_status
 347 ir_set_program_inouts_visitor::visit_enter(ir_dereference_array *ir)
 348 {
 349    /* Note: for geometry shader inputs, lower_named_interface_blocks may
 350     * create 2D arrays, so we need to be able to handle those.  2D arrays
 351     * shouldn't be able to crop up for any other reason.
 352     */
 353    if (ir_dereference_array * const inner_array =
 354        ir->array->as_dereference_array()) {
 355       /*          ir => foo[i][j]
 356        * inner_array => foo[i]
 357        */
 358       if (ir_dereference_variable * const deref_var =
 359           inner_array->array->as_dereference_variable()) {
 360          if (is_multiple_vertices(this->shader_stage, deref_var->var)) {
 361             /* foo is a geometry or tessellation shader input, so i is
 362              * the vertex, and j the part of the input we're accessing.
 363              */
 364             if (try_mark_partial_variable(deref_var->var, ir->array_index))
 365             {
 366                /* We've now taken care of foo and j, but i might contain a
 367                 * subexpression that accesses shader inputs.  So manually
 368                 * visit i and then continue with the parent.
 369                 */
 370                inner_array->array_index->accept(this);
 371                return visit_continue_with_parent;
 372             }
 373          }
 374       }
 375    } else if (ir_dereference_variable * const deref_var =
 376               ir->array->as_dereference_variable()) {
 377       /* ir => foo[i], where foo is a variable. */
 378       if (is_multiple_vertices(this->shader_stage, deref_var->var)) {
 379          /* foo is a geometry or tessellation shader input, so i is
 380           * the vertex, and we're accessing the entire input.
 381           */
 382          mark_whole_variable(deref_var->var);
 383          /* We've now taken care of foo, but i might contain a subexpression
 384           * that accesses shader inputs.  So manually visit i and then
 385           * continue with the parent.
 386           */
 387          ir->array_index->accept(this);
 388          return visit_continue_with_parent;
 389       } else if (is_shader_inout(deref_var->var)) {
 390          /* foo is a shader input/output, but not a geometry shader input,
 391           * so i is the part of the input we're accessing.
 392           */
 393          if (try_mark_partial_variable(deref_var->var, ir->array_index))
 394             return visit_continue_with_parent;
 395       }
 396    }
 397
 398    /* The expression is something we don't recognize.  Just visit its
 399     * subexpressions.
 400     */
 401    return visit_continue;
 402 }
 403
 404 ir_visitor_status
 405 ir_set_program_inouts_visitor::visit_enter(ir_function_signature *ir)
 406 {
 407    /* We don't want to descend into the function parameters and
 408     * consider them as shader inputs or outputs.
 409     */
 410    visit_list_elements(this, &ir->body);
 411    return visit_continue_with_parent;
 412 }
 413
 414 ir_visitor_status
 415 ir_set_program_inouts_visitor::visit_enter(ir_expression *ir)
 416 {
 417    if (this->shader_stage == MESA_SHADER_FRAGMENT &&
 418        (ir->operation == ir_unop_dFdy ||
 419         ir->operation == ir_unop_dFdy_coarse ||
 420         ir->operation == ir_unop_dFdy_fine)) {
 421       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 422       fprog->UsesDFdy = true;
 423    }
 424    return visit_continue;
 425 }
 426
 427 ir_visitor_status
 428 ir_set_program_inouts_visitor::visit_enter(ir_discard *)
 429 {
 430    /* discards are only allowed in fragment shaders. */
 431    assert(this->shader_stage == MESA_SHADER_FRAGMENT);
 432
 433    gl_fragment_program *fprog = (gl_fragment_program *) prog;
 434    fprog->UsesKill = true;
 435
 436    return visit_continue;
 437 }
 438
 439 ir_visitor_status
 440 ir_set_program_inouts_visitor::visit_enter(ir_texture *ir)
 441 {
 442    if (ir->op == ir_tg4)
 443       prog->UsesGather = true;
 444    return visit_continue;
 445 }
 446
 447 void
 448 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
 449                       gl_shader_stage shader_stage)
 450 {
 451    ir_set_program_inouts_visitor v(prog, shader_stage);
 452
 453    prog->InputsRead = 0;
 454    prog->OutputsWritten = 0;
 455    prog->SecondaryOutputsWritten = 0;
 456    prog->OutputsRead = 0;
 457    prog->PatchInputsRead = 0;
 458    prog->PatchOutputsWritten = 0;
 459    prog->SystemValuesRead = 0;
 460    if (shader_stage == MESA_SHADER_FRAGMENT) {
 461       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 462       memset(fprog->InterpQualifier, 0, sizeof(fprog->InterpQualifier));
 463       fprog->IsCentroid = 0;
 464       fprog->IsSample = 0;
 465       fprog->UsesDFdy = false;
 466       fprog->UsesKill = false;
 467    }
 468    visit_list_elements(&v, instructions);
 469 }