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