src/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 bitfield, 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 "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, GLenum shader_type)
  50    {
  51       this->prog = prog;
  52       this->shader_type = shader_type;
  53    }
  54    ~ir_set_program_inouts_visitor()
  55    {
  56    }
  57
  58    virtual ir_visitor_status visit_enter(ir_dereference_array *);
  59    virtual ir_visitor_status visit_enter(ir_function_signature *);
  60    virtual ir_visitor_status visit_enter(ir_expression *);
  61    virtual ir_visitor_status visit_enter(ir_discard *);
  62    virtual ir_visitor_status visit_enter(ir_texture *);
  63    virtual ir_visitor_status visit(ir_dereference_variable *);
  64
  65 private:
  66    void mark_whole_variable(ir_variable *var);
  67    bool try_mark_partial_variable(ir_variable *var, ir_rvalue *index);
  68
  69    struct gl_program *prog;
  70    GLenum shader_type;
  71 };
  72
  73 } /* anonymous namespace */
  74
  75 static inline bool
  76 is_shader_inout(ir_variable *var)
  77 {
  78    return var->mode == ir_var_shader_in ||
  79           var->mode == ir_var_shader_out ||
  80           var->mode == ir_var_system_value;
  81 }
  82
  83 static void
  84 mark(struct gl_program *prog, ir_variable *var, int offset, int len,
  85      bool is_fragment_shader)
  86 {
  87    /* As of GLSL 1.20, varyings can only be floats, floating-point
  88     * vectors or matrices, or arrays of them.  For Mesa programs using
  89     * InputsRead/OutputsWritten, everything but matrices uses one
  90     * slot, while matrices use a slot per column.  Presumably
  91     * something doing a more clever packing would use something other
  92     * than InputsRead/OutputsWritten.
  93     */
  94
  95    for (int i = 0; i < len; i++) {
  96       GLbitfield64 bitfield = BITFIELD64_BIT(var->location + var->index + offset + i);
  97       if (var->mode == ir_var_shader_in) {
  98          prog->InputsRead |= bitfield;
  99          if (is_fragment_shader) {
 100             gl_fragment_program *fprog = (gl_fragment_program *) prog;
 101             fprog->InterpQualifier[var->location + var->index + offset + i] =
 102                (glsl_interp_qualifier) var->interpolation;
 103             if (var->centroid)
 104                fprog->IsCentroid |= bitfield;
 105          }
 106       } else if (var->mode == ir_var_system_value) {
 107          prog->SystemValuesRead |= bitfield;
 108       } else {
 109          assert(var->mode == ir_var_shader_out);
 110          prog->OutputsWritten |= bitfield;
 111       }
 112    }
 113 }
 114
 115 /**
 116  * Mark an entire variable as used.  Caller must ensure that the variable
 117  * represents a shader input or output.
 118  */
 119 void
 120 ir_set_program_inouts_visitor::mark_whole_variable(ir_variable *var)
 121 {
 122    const glsl_type *type = var->type;
 123    if (this->shader_type == GL_GEOMETRY_SHADER &&
 124        var->mode == ir_var_shader_in && type->is_array()) {
 125       type = type->fields.array;
 126    }
 127
 128    mark(this->prog, var, 0, type->count_attribute_slots(),
 129         this->shader_type == GL_FRAGMENT_SHADER);
 130 }
 131
 132 /* Default handler: Mark all the locations in the variable as used. */
 133 ir_visitor_status
 134 ir_set_program_inouts_visitor::visit(ir_dereference_variable *ir)
 135 {
 136    if (!is_shader_inout(ir->var))
 137       return visit_continue;
 138
 139    mark_whole_variable(ir->var);
 140
 141    return visit_continue;
 142 }
 143
 144 /**
 145  * Try to mark a portion of the given variable as used.  Caller must ensure
 146  * that the variable represents a shader input or output which can be indexed
 147  * into in array fashion (an array or matrix).  For the purpose of geometry
 148  * shader inputs (which are always arrays*), this means that the array element
 149  * must be something that can be indexed into in array fashion.
 150  *
 151  * *Except gl_PrimitiveIDIn, as noted below.
 152  *
 153  * If the index can't be interpreted as a constant, or some other problem
 154  * occurs, then nothing will be marked and false will be returned.
 155  */
 156 bool
 157 ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable *var,
 158                                                          ir_rvalue *index)
 159 {
 160    const glsl_type *type = var->type;
 161
 162    if (this->shader_type == GL_GEOMETRY_SHADER &&
 163        var->mode == ir_var_shader_in) {
 164       /* The only geometry shader input that is not an array is
 165        * gl_PrimitiveIDIn, and in that case, this code will never be reached,
 166        * because gl_PrimitiveIDIn can't be indexed into in array fashion.
 167        */
 168       assert(type->is_array());
 169       type = type->fields.array;
 170    }
 171
 172    /* The code below only handles:
 173     *
 174     * - Indexing into matrices
 175     * - Indexing into arrays of (matrices, vectors, or scalars)
 176     *
 177     * All other possibilities are either prohibited by GLSL (vertex inputs and
 178     * fragment outputs can't be structs) or should have been eliminated by
 179     * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into
 180     * vectors, and lower_packed_varyings() gets rid of structs that occur in
 181     * varyings).
 182     */
 183    if (!(type->is_matrix() ||
 184         (type->is_array() &&
 185          (type->fields.array->is_numeric() ||
 186           type->fields.array->is_boolean())))) {
 187       assert(!"Unexpected indexing in ir_set_program_inouts");
 188
 189       /* For safety in release builds, in case we ever encounter unexpected
 190        * indexing, give up and let the caller mark the whole variable as used.
 191        */
 192       return false;
 193    }
 194
 195    ir_constant *index_as_constant = index->as_constant();
 196    if (!index_as_constant)
 197       return false;
 198
 199    unsigned elem_width;
 200    unsigned num_elems;
 201    if (type->is_array()) {
 202       num_elems = type->length;
 203       if (type->fields.array->is_matrix())
 204          elem_width = type->fields.array->matrix_columns;
 205       else
 206          elem_width = 1;
 207    } else {
 208       num_elems = type->matrix_columns;
 209       elem_width = 1;
 210    }
 211
 212    if (index_as_constant->value.u[0] >= num_elems) {
 213       /* Constant index outside the bounds of the matrix/array.  This could
 214        * arise as a result of constant folding of a legal GLSL program.
 215        *
 216        * Even though the spec says that indexing outside the bounds of a
 217        * matrix/array results in undefined behaviour, we don't want to pass
 218        * out-of-range values to mark() (since this could result in slots that
 219        * don't exist being marked as used), so just let the caller mark the
 220        * whole variable as used.
 221        */
 222       return false;
 223    }
 224
 225    mark(this->prog, var, index_as_constant->value.u[0] * elem_width,
 226         elem_width, this->shader_type == GL_FRAGMENT_SHADER);
 227    return true;
 228 }
 229
 230 ir_visitor_status
 231 ir_set_program_inouts_visitor::visit_enter(ir_dereference_array *ir)
 232 {
 233    /* Note: for geometry shader inputs, lower_named_interface_blocks may
 234     * create 2D arrays, so we need to be able to handle those.  2D arrays
 235     * shouldn't be able to crop up for any other reason.
 236     */
 237    if (ir_dereference_array * const inner_array =
 238        ir->array->as_dereference_array()) {
 239       /*          ir => foo[i][j]
 240        * inner_array => foo[i]
 241        */
 242       if (ir_dereference_variable * const deref_var =
 243           inner_array->array->as_dereference_variable()) {
 244          if (this->shader_type == GL_GEOMETRY_SHADER &&
 245              deref_var->var->mode == ir_var_shader_in) {
 246             /* foo is a geometry shader input, so i is the vertex, and j the
 247              * part of the input we're accessing.
 248              */
 249             if (try_mark_partial_variable(deref_var->var, ir->array_index))
 250             {
 251                /* We've now taken care of foo and j, but i might contain a
 252                 * subexpression that accesses shader inputs.  So manually
 253                 * visit i and then continue with the parent.
 254                 */
 255                inner_array->array_index->accept(this);
 256                return visit_continue_with_parent;
 257             }
 258          }
 259       }
 260    } else if (ir_dereference_variable * const deref_var =
 261               ir->array->as_dereference_variable()) {
 262       /* ir => foo[i], where foo is a variable. */
 263       if (this->shader_type == GL_GEOMETRY_SHADER &&
 264           deref_var->var->mode == ir_var_shader_in) {
 265          /* foo is a geometry shader input, so i is the vertex, and we're
 266           * accessing the entire input.
 267           */
 268          mark_whole_variable(deref_var->var);
 269          /* We've now taken care of foo, but i might contain a subexpression
 270           * that accesses shader inputs.  So manually visit i and then
 271           * continue with the parent.
 272           */
 273          ir->array_index->accept(this);
 274          return visit_continue_with_parent;
 275       } else if (is_shader_inout(deref_var->var)) {
 276          /* foo is a shader input/output, but not a geometry shader input,
 277           * so i is the part of the input we're accessing.
 278           */
 279          if (try_mark_partial_variable(deref_var->var, ir->array_index))
 280             return visit_continue_with_parent;
 281       }
 282    }
 283
 284    /* The expression is something we don't recognize.  Just visit its
 285     * subexpressions.
 286     */
 287    return visit_continue;
 288 }
 289
 290 ir_visitor_status
 291 ir_set_program_inouts_visitor::visit_enter(ir_function_signature *ir)
 292 {
 293    /* We don't want to descend into the function parameters and
 294     * consider them as shader inputs or outputs.
 295     */
 296    visit_list_elements(this, &ir->body);
 297    return visit_continue_with_parent;
 298 }
 299
 300 ir_visitor_status
 301 ir_set_program_inouts_visitor::visit_enter(ir_expression *ir)
 302 {
 303    if (this->shader_type == GL_FRAGMENT_SHADER &&
 304        ir->operation == ir_unop_dFdy) {
 305       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 306       fprog->UsesDFdy = true;
 307    }
 308    return visit_continue;
 309 }
 310
 311 ir_visitor_status
 312 ir_set_program_inouts_visitor::visit_enter(ir_discard *)
 313 {
 314    /* discards are only allowed in fragment shaders. */
 315    assert(this->shader_type == GL_FRAGMENT_SHADER);
 316
 317    gl_fragment_program *fprog = (gl_fragment_program *) prog;
 318    fprog->UsesKill = true;
 319
 320    return visit_continue;
 321 }
 322
 323 ir_visitor_status
 324 ir_set_program_inouts_visitor::visit_enter(ir_texture *ir)
 325 {
 326    if (ir->op == ir_tg4)
 327       prog->UsesGather = true;
 328    return visit_continue;
 329 }
 330
 331 void
 332 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
 333                       GLenum shader_type)
 334 {
 335    ir_set_program_inouts_visitor v(prog, shader_type);
 336
 337    prog->InputsRead = 0;
 338    prog->OutputsWritten = 0;
 339    prog->SystemValuesRead = 0;
 340    if (shader_type == GL_FRAGMENT_SHADER) {
 341       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 342       memset(fprog->InterpQualifier, 0, sizeof(fprog->InterpQualifier));
 343       fprog->IsCentroid = 0;
 344       fprog->UsesDFdy = false;
 345       fprog->UsesKill = false;
 346    }
 347    visit_list_elements(&v, instructions);
 348 }