src/mesa/drivers/dri/i965/brw_shader.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 DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 extern "C" {
  25 #include "main/macros.h"
  26 #include "brw_context.h"
  27 #include "brw_vs.h"
  28 }
  29 #include "brw_fs.h"
  30 #include "glsl/ir_optimization.h"
  31 #include "glsl/ir_print_visitor.h"
  32
  33 struct gl_shader *
  34 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
  35 {
  36    struct brw_shader *shader;
  37
  38    shader = rzalloc(NULL, struct brw_shader);
  39    if (shader) {
  40       shader->base.Type = type;
  41       shader->base.Name = name;
  42       _mesa_init_shader(ctx, &shader->base);
  43    }
  44
  45    return &shader->base;
  46 }
  47
  48 struct gl_shader_program *
  49 brw_new_shader_program(struct gl_context *ctx, GLuint name)
  50 {
  51    struct brw_shader_program *prog;
  52    prog = rzalloc(NULL, struct brw_shader_program);
  53    if (prog) {
  54       prog->base.Name = name;
  55       _mesa_init_shader_program(ctx, &prog->base);
  56    }
  57    return &prog->base;
  58 }
  59
  60 /**
  61  * Performs a compile of the shader stages even when we don't know
  62  * what non-orthogonal state will be set, in the hope that it reflects
  63  * the eventual NOS used, and thus allows us to produce link failures.
  64  */
  65 bool
  66 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
  67 {
  68    struct brw_context *brw = brw_context(ctx);
  69
  70    if (brw->precompile && !brw_fs_precompile(ctx, prog))
  71       return false;
  72
  73    if (!brw_vs_precompile(ctx, prog))
  74       return false;
  75
  76    return true;
  77 }
  78
  79 GLboolean
  80 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
  81 {
  82    struct brw_context *brw = brw_context(ctx);
  83    struct intel_context *intel = &brw->intel;
  84    unsigned int stage;
  85
  86    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
  87       struct brw_shader *shader =
  88          (struct brw_shader *)shProg->_LinkedShaders[stage];
  89       static const GLenum targets[] = {
  90          GL_VERTEX_PROGRAM_ARB,
  91          GL_FRAGMENT_PROGRAM_ARB,
  92          GL_GEOMETRY_PROGRAM_NV
  93       };
  94
  95       if (!shader)
  96          continue;
  97
  98       struct gl_program *prog =
  99          ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
 100       if (!prog)
 101         return NULL;
 102       prog->Parameters = _mesa_new_parameter_list();
 103
 104       _mesa_generate_parameters_list_for_uniforms(shProg, &shader->base,
 105                                                   prog->Parameters);
 106
 107       if (stage == 0) {
 108          struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 109          vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
 110       }
 111
 112       if (stage == 1) {
 113          class uses_kill_visitor : public ir_hierarchical_visitor {
 114          public:
 115             uses_kill_visitor() : uses_kill(false)
 116             {
 117                /* empty */
 118             }
 119
 120             virtual ir_visitor_status visit_enter(class ir_discard *ir)
 121             {
 122                this->uses_kill = true;
 123                return visit_stop;
 124             }
 125
 126             bool uses_kill;
 127          };
 128
 129          uses_kill_visitor v;
 130
 131          v.run(shader->base.ir);
 132
 133          struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 134          fp->UsesKill = v.uses_kill;
 135       }
 136
 137       void *mem_ctx = ralloc_context(NULL);
 138       bool progress;
 139
 140       if (shader->ir)
 141          ralloc_free(shader->ir);
 142       shader->ir = new(shader) exec_list;
 143       clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
 144
 145       do_mat_op_to_vec(shader->ir);
 146       lower_instructions(shader->ir,
 147                          MOD_TO_FRACT |
 148                          DIV_TO_MUL_RCP |
 149                          SUB_TO_ADD_NEG |
 150                          EXP_TO_EXP2 |
 151                          LOG_TO_LOG2);
 152
 153       /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 154        * if-statements need to be flattened.
 155        */
 156       if (intel->gen < 6)
 157          lower_if_to_cond_assign(shader->ir, 16);
 158
 159       do_lower_texture_projection(shader->ir);
 160       do_vec_index_to_cond_assign(shader->ir);
 161       brw_do_cubemap_normalize(shader->ir);
 162       lower_noise(shader->ir);
 163       lower_quadop_vector(shader->ir, false);
 164
 165       bool input = true;
 166       bool output = stage == MESA_SHADER_FRAGMENT;
 167       bool temp = stage == MESA_SHADER_FRAGMENT;
 168       bool uniform = stage == MESA_SHADER_FRAGMENT;
 169
 170       lower_variable_index_to_cond_assign(shader->ir,
 171                                           input, output, temp, uniform);
 172
 173       do {
 174          progress = false;
 175
 176          if (stage == MESA_SHADER_FRAGMENT) {
 177             brw_do_channel_expressions(shader->ir);
 178             brw_do_vector_splitting(shader->ir);
 179          }
 180
 181          progress = do_lower_jumps(shader->ir, true, true,
 182                                    true, /* main return */
 183                                    false, /* continue */
 184                                    false /* loops */
 185                                    ) || progress;
 186
 187          progress = do_common_optimization(shader->ir, true, true, 32)
 188            || progress;
 189       } while (progress);
 190
 191       /* Make a pass over the IR to add state references for any built-in
 192        * uniforms that are used.  This has to be done now (during linking).
 193        * Code generation doesn't happen until the first time this shader is
 194        * used for rendering.  Waiting until then to generate the parameters is
 195        * too late.  At that point, the values for the built-in informs won't
 196        * get sent to the shader.
 197        */
 198       foreach_list(node, shader->ir) {
 199          ir_variable *var = ((ir_instruction *) node)->as_variable();
 200
 201          if ((var == NULL) || (var->mode != ir_var_uniform)
 202              || (strncmp(var->name, "gl_", 3) != 0))
 203             continue;
 204
 205          const ir_state_slot *const slots = var->state_slots;
 206          assert(var->state_slots != NULL);
 207
 208          for (unsigned int i = 0; i < var->num_state_slots; i++) {
 209             _mesa_add_state_reference(prog->Parameters,
 210                                       (gl_state_index *) slots[i].tokens);
 211          }
 212       }
 213
 214       validate_ir_tree(shader->ir);
 215
 216       reparent_ir(shader->ir, shader->ir);
 217       ralloc_free(mem_ctx);
 218
 219       do_set_program_inouts(shader->ir, prog,
 220                             shader->base.Type == GL_FRAGMENT_SHADER);
 221
 222       prog->SamplersUsed = shader->base.active_samplers;
 223       _mesa_update_shader_textures_used(shProg, prog);
 224
 225       _mesa_reference_program(ctx, &shader->base.Program, prog);
 226
 227       /* This has to be done last.  Any operation that can cause
 228        * prog->ParameterValues to get reallocated (e.g., anything that adds a
 229        * program constant) has to happen before creating this linkage.
 230        */
 231       _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
 232    }
 233
 234    if (!brw_shader_precompile(ctx, shProg))
 235       return false;
 236
 237    return true;
 238 }
 239
 240
 241 int
 242 brw_type_for_base_type(const struct glsl_type *type)
 243 {
 244    switch (type->base_type) {
 245    case GLSL_TYPE_FLOAT:
 246       return BRW_REGISTER_TYPE_F;
 247    case GLSL_TYPE_INT:
 248    case GLSL_TYPE_BOOL:
 249       return BRW_REGISTER_TYPE_D;
 250    case GLSL_TYPE_UINT:
 251       return BRW_REGISTER_TYPE_UD;
 252    case GLSL_TYPE_ARRAY:
 253       return brw_type_for_base_type(type->fields.array);
 254    case GLSL_TYPE_STRUCT:
 255    case GLSL_TYPE_SAMPLER:
 256       /* These should be overridden with the type of the member when
 257        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
 258        * way to trip up if we don't.
 259        */
 260       return BRW_REGISTER_TYPE_UD;
 261    default:
 262       assert(!"not reached");
 263       return BRW_REGISTER_TYPE_F;
 264    }
 265 }
 266
 267 uint32_t
 268 brw_conditional_for_comparison(unsigned int op)
 269 {
 270    switch (op) {
 271    case ir_binop_less:
 272       return BRW_CONDITIONAL_L;
 273    case ir_binop_greater:
 274       return BRW_CONDITIONAL_G;
 275    case ir_binop_lequal:
 276       return BRW_CONDITIONAL_LE;
 277    case ir_binop_gequal:
 278       return BRW_CONDITIONAL_GE;
 279    case ir_binop_equal:
 280    case ir_binop_all_equal: /* same as equal for scalars */
 281       return BRW_CONDITIONAL_Z;
 282    case ir_binop_nequal:
 283    case ir_binop_any_nequal: /* same as nequal for scalars */
 284       return BRW_CONDITIONAL_NZ;
 285    default:
 286       assert(!"not reached: bad operation for comparison");
 287       return BRW_CONDITIONAL_NZ;
 288    }
 289 }
 290
 291 uint32_t
 292 brw_math_function(enum opcode op)
 293 {
 294    switch (op) {
 295    case SHADER_OPCODE_RCP:
 296       return BRW_MATH_FUNCTION_INV;
 297    case SHADER_OPCODE_RSQ:
 298       return BRW_MATH_FUNCTION_RSQ;
 299    case SHADER_OPCODE_SQRT:
 300       return BRW_MATH_FUNCTION_SQRT;
 301    case SHADER_OPCODE_EXP2:
 302       return BRW_MATH_FUNCTION_EXP;
 303    case SHADER_OPCODE_LOG2:
 304       return BRW_MATH_FUNCTION_LOG;
 305    case SHADER_OPCODE_POW:
 306       return BRW_MATH_FUNCTION_POW;
 307    case SHADER_OPCODE_SIN:
 308       return BRW_MATH_FUNCTION_SIN;
 309    case SHADER_OPCODE_COS:
 310       return BRW_MATH_FUNCTION_COS;
 311    case SHADER_OPCODE_INT_QUOTIENT:
 312       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 313    case SHADER_OPCODE_INT_REMAINDER:
 314       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 315    default:
 316       assert(!"not reached: unknown math function");
 317       return 0;
 318    }
 319 }
 320
 321 uint32_t
 322 brw_texture_offset(ir_constant *offset)
 323 {
 324    assert(offset != NULL);
 325
 326    signed char offsets[3];
 327    for (unsigned i = 0; i < offset->type->vector_elements; i++)
 328       offsets[i] = (signed char) offset->value.i[i];
 329
 330    /* Combine all three offsets into a single unsigned dword:
 331     *
 332     *    bits 11:8 - U Offset (X component)
 333     *    bits  7:4 - V Offset (Y component)
 334     *    bits  3:0 - R Offset (Z component)
 335     */
 336    unsigned offset_bits = 0;
 337    for (unsigned i = 0; i < offset->type->vector_elements; i++) {
 338       const unsigned shift = 4 * (2 - i);
 339       offset_bits |= (offsets[i] << shift) & (0xF << shift);
 340    }
 341    return offset_bits;
 342 }