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 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 "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->data.mode == ir_var_shader_in ||
  79           var->data.mode == ir_var_shader_out ||
  80           var->data.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 =
  97          BITFIELD64_BIT(var->data.location + var->data.index + offset + i);
  98       if (var->data.mode == ir_var_shader_in) {
  99          prog->InputsRead |= bitfield;
 100          if (is_fragment_shader) {
 101             gl_fragment_program *fprog = (gl_fragment_program *) prog;
 102             fprog->InterpQualifier[var->data.location +
 103                                    var->data.index + offset + i] =
 104                (glsl_interp_qualifier) var->data.interpolation;
 105             if (var->data.centroid)
 106                fprog->IsCentroid |= bitfield;
 107             if (var->data.sample)
 108                fprog->IsSample |= bitfield;
 109          }
 110       } else if (var->data.mode == ir_var_system_value) {
 111          prog->SystemValuesRead |= bitfield;
 112       } else {
 113          assert(var->data.mode == ir_var_shader_out);
 114          prog->OutputsWritten |= bitfield;
 115       }
 116    }
 117 }
 118
 119 /**
 120  * Mark an entire variable as used.  Caller must ensure that the variable
 121  * represents a shader input or output.
 122  */
 123 void
 124 ir_set_program_inouts_visitor::mark_whole_variable(ir_variable *var)
 125 {
 126    const glsl_type *type = var->type;
 127    if (this->shader_type == GL_GEOMETRY_SHADER &&
 128        var->data.mode == ir_var_shader_in && type->is_array()) {
 129       type = type->fields.array;
 130    }
 131
 132    mark(this->prog, var, 0, type->count_attribute_slots(),
 133         this->shader_type == GL_FRAGMENT_SHADER);
 134 }
 135
 136 /* Default handler: Mark all the locations in the variable as used. */
 137 ir_visitor_status
 138 ir_set_program_inouts_visitor::visit(ir_dereference_variable *ir)
 139 {
 140    if (!is_shader_inout(ir->var))
 141       return visit_continue;
 142
 143    mark_whole_variable(ir->var);
 144
 145    return visit_continue;
 146 }
 147
 148 /**
 149  * Try to mark a portion of the given variable as used.  Caller must ensure
 150  * that the variable represents a shader input or output which can be indexed
 151  * into in array fashion (an array or matrix).  For the purpose of geometry
 152  * shader inputs (which are always arrays*), this means that the array element
 153  * must be something that can be indexed into in array fashion.
 154  *
 155  * *Except gl_PrimitiveIDIn, as noted below.
 156  *
 157  * If the index can't be interpreted as a constant, or some other problem
 158  * occurs, then nothing will be marked and false will be returned.
 159  */
 160 bool
 161 ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable *var,
 162                                                          ir_rvalue *index)
 163 {
 164    const glsl_type *type = var->type;
 165
 166    if (this->shader_type == GL_GEOMETRY_SHADER &&
 167        var->data.mode == ir_var_shader_in) {
 168       /* The only geometry shader input that is not an array is
 169        * gl_PrimitiveIDIn, and in that case, this code will never be reached,
 170        * because gl_PrimitiveIDIn can't be indexed into in array fashion.
 171        */
 172       assert(type->is_array());
 173       type = type->fields.array;
 174    }
 175
 176    /* The code below only handles:
 177     *
 178     * - Indexing into matrices
 179     * - Indexing into arrays of (matrices, vectors, or scalars)
 180     *
 181     * All other possibilities are either prohibited by GLSL (vertex inputs and
 182     * fragment outputs can't be structs) or should have been eliminated by
 183     * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into
 184     * vectors, and lower_packed_varyings() gets rid of structs that occur in
 185     * varyings).
 186     */
 187    if (!(type->is_matrix() ||
 188         (type->is_array() &&
 189          (type->fields.array->is_numeric() ||
 190           type->fields.array->is_boolean())))) {
 191       assert(!"Unexpected indexing in ir_set_program_inouts");
 192
 193       /* For safety in release builds, in case we ever encounter unexpected
 194        * indexing, give up and let the caller mark the whole variable as used.
 195        */
 196       return false;
 197    }
 198
 199    ir_constant *index_as_constant = index->as_constant();
 200    if (!index_as_constant)
 201       return false;
 202
 203    unsigned elem_width;
 204    unsigned num_elems;
 205    if (type->is_array()) {
 206       num_elems = type->length;
 207       if (type->fields.array->is_matrix())
 208          elem_width = type->fields.array->matrix_columns;
 209       else
 210          elem_width = 1;
 211    } else {
 212       num_elems = type->matrix_columns;
 213       elem_width = 1;
 214    }
 215
 216    if (index_as_constant->value.u[0] >= num_elems) {
 217       /* Constant index outside the bounds of the matrix/array.  This could
 218        * arise as a result of constant folding of a legal GLSL program.
 219        *
 220        * Even though the spec says that indexing outside the bounds of a
 221        * matrix/array results in undefined behaviour, we don't want to pass
 222        * out-of-range values to mark() (since this could result in slots that
 223        * don't exist being marked as used), so just let the caller mark the
 224        * whole variable as used.
 225        */
 226       return false;
 227    }
 228
 229    mark(this->prog, var, index_as_constant->value.u[0] * elem_width,
 230         elem_width, this->shader_type == GL_FRAGMENT_SHADER);
 231    return true;
 232 }
 233
 234 ir_visitor_status
 235 ir_set_program_inouts_visitor::visit_enter(ir_dereference_array *ir)
 236 {
 237    /* Note: for geometry shader inputs, lower_named_interface_blocks may
 238     * create 2D arrays, so we need to be able to handle those.  2D arrays
 239     * shouldn't be able to crop up for any other reason.
 240     */
 241    if (ir_dereference_array * const inner_array =
 242        ir->array->as_dereference_array()) {
 243       /*          ir => foo[i][j]
 244        * inner_array => foo[i]
 245        */
 246       if (ir_dereference_variable * const deref_var =
 247           inner_array->array->as_dereference_variable()) {
 248          if (this->shader_type == GL_GEOMETRY_SHADER &&
 249              deref_var->var->data.mode == ir_var_shader_in) {
 250             /* foo is a geometry shader input, so i is the vertex, and j the
 251              * part of the input we're accessing.
 252              */
 253             if (try_mark_partial_variable(deref_var->var, ir->array_index))
 254             {
 255                /* We've now taken care of foo and j, but i might contain a
 256                 * subexpression that accesses shader inputs.  So manually
 257                 * visit i and then continue with the parent.
 258                 */
 259                inner_array->array_index->accept(this);
 260                return visit_continue_with_parent;
 261             }
 262          }
 263       }
 264    } else if (ir_dereference_variable * const deref_var =
 265               ir->array->as_dereference_variable()) {
 266       /* ir => foo[i], where foo is a variable. */
 267       if (this->shader_type == GL_GEOMETRY_SHADER &&
 268           deref_var->var->data.mode == ir_var_shader_in) {
 269          /* foo is a geometry shader input, so i is the vertex, and we're
 270           * accessing the entire input.
 271           */
 272          mark_whole_variable(deref_var->var);
 273          /* We've now taken care of foo, but i might contain a subexpression
 274           * that accesses shader inputs.  So manually visit i and then
 275           * continue with the parent.
 276           */
 277          ir->array_index->accept(this);
 278          return visit_continue_with_parent;
 279       } else if (is_shader_inout(deref_var->var)) {
 280          /* foo is a shader input/output, but not a geometry shader input,
 281           * so i is the part of the input we're accessing.
 282           */
 283          if (try_mark_partial_variable(deref_var->var, ir->array_index))
 284             return visit_continue_with_parent;
 285       }
 286    }
 287
 288    /* The expression is something we don't recognize.  Just visit its
 289     * subexpressions.
 290     */
 291    return visit_continue;
 292 }
 293
 294 ir_visitor_status
 295 ir_set_program_inouts_visitor::visit_enter(ir_function_signature *ir)
 296 {
 297    /* We don't want to descend into the function parameters and
 298     * consider them as shader inputs or outputs.
 299     */
 300    visit_list_elements(this, &ir->body);
 301    return visit_continue_with_parent;
 302 }
 303
 304 ir_visitor_status
 305 ir_set_program_inouts_visitor::visit_enter(ir_expression *ir)
 306 {
 307    if (this->shader_type == GL_FRAGMENT_SHADER &&
 308        ir->operation == ir_unop_dFdy) {
 309       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 310       fprog->UsesDFdy = true;
 311    }
 312    return visit_continue;
 313 }
 314
 315 ir_visitor_status
 316 ir_set_program_inouts_visitor::visit_enter(ir_discard *)
 317 {
 318    /* discards are only allowed in fragment shaders. */
 319    assert(this->shader_type == GL_FRAGMENT_SHADER);
 320
 321    gl_fragment_program *fprog = (gl_fragment_program *) prog;
 322    fprog->UsesKill = true;
 323
 324    return visit_continue;
 325 }
 326
 327 ir_visitor_status
 328 ir_set_program_inouts_visitor::visit_enter(ir_texture *ir)
 329 {
 330    if (ir->op == ir_tg4)
 331       prog->UsesGather = true;
 332    return visit_continue;
 333 }
 334
 335 void
 336 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
 337                       GLenum shader_type)
 338 {
 339    ir_set_program_inouts_visitor v(prog, shader_type);
 340
 341    prog->InputsRead = 0;
 342    prog->OutputsWritten = 0;
 343    prog->SystemValuesRead = 0;
 344    if (shader_type == GL_FRAGMENT_SHADER) {
 345       gl_fragment_program *fprog = (gl_fragment_program *) prog;
 346       memset(fprog->InterpQualifier, 0, sizeof(fprog->InterpQualifier));
 347       fprog->IsCentroid = 0;
 348       fprog->IsSample = 0;
 349       fprog->UsesDFdy = false;
 350       fprog->UsesKill = false;
 351    }
 352    visit_list_elements(&v, instructions);
 353 }