src/mesa/drivers/dri/i965/brw_link.cpp

   1 /*
   2  * Copyright © 2015 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 #include "brw_context.h"
  25 #include "brw_shader.h"
  26 #include "brw_fs.h"
  27 #include "brw_nir.h"
  28 #include "brw_program.h"
  29 #include "compiler/glsl/ir.h"
  30 #include "compiler/glsl/ir_optimization.h"
  31 #include "compiler/glsl/program.h"
  32 #include "program/program.h"
  33 #include "main/shaderapi.h"
  34 #include "main/shaderobj.h"
  35 #include "main/uniforms.h"
  36
  37 /**
  38  * Performs a compile of the shader stages even when we don't know
  39  * what non-orthogonal state will be set, in the hope that it reflects
  40  * the eventual NOS used, and thus allows us to produce link failures.
  41  */
  42 static bool
  43 brw_shader_precompile(struct gl_context *ctx,
  44                       struct gl_shader_program *sh_prog)
  45 {
  46    struct gl_linked_shader *vs = sh_prog->_LinkedShaders[MESA_SHADER_VERTEX];
  47    struct gl_linked_shader *tcs = sh_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
  48    struct gl_linked_shader *tes = sh_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
  49    struct gl_linked_shader *gs = sh_prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
  50    struct gl_linked_shader *fs = sh_prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  51    struct gl_linked_shader *cs = sh_prog->_LinkedShaders[MESA_SHADER_COMPUTE];
  52
  53    if (fs && !brw_fs_precompile(ctx, sh_prog, fs->Program))
  54       return false;
  55
  56    if (gs && !brw_gs_precompile(ctx, sh_prog, gs->Program))
  57       return false;
  58
  59    if (tes && !brw_tes_precompile(ctx, sh_prog, tes->Program))
  60       return false;
  61
  62    if (tcs && !brw_tcs_precompile(ctx, sh_prog, tcs->Program))
  63       return false;
  64
  65    if (vs && !brw_vs_precompile(ctx, sh_prog, vs->Program))
  66       return false;
  67
  68    if (cs && !brw_cs_precompile(ctx, sh_prog, cs->Program))
  69       return false;
  70
  71    return true;
  72 }
  73
  74 static void
  75 brw_lower_packing_builtins(struct brw_context *brw,
  76                            gl_shader_stage shader_type,
  77                            exec_list *ir)
  78 {
  79    /* Gens < 7 don't have instructions to convert to or from half-precision,
  80     * and Gens < 6 don't expose that functionality.
  81     */
  82    if (brw->gen != 6)
  83       return;
  84
  85    lower_packing_builtins(ir, LOWER_PACK_HALF_2x16 | LOWER_UNPACK_HALF_2x16);
  86 }
  87
  88 static void
  89 process_glsl_ir(struct brw_context *brw,
  90                 struct gl_shader_program *shader_prog,
  91                 struct gl_linked_shader *shader)
  92 {
  93    struct gl_context *ctx = &brw->ctx;
  94    const struct brw_compiler *compiler = brw->screen->compiler;
  95    const struct gl_shader_compiler_options *options =
  96       &ctx->Const.ShaderCompilerOptions[shader->Stage];
  97
  98    /* Temporary memory context for any new IR. */
  99    void *mem_ctx = ralloc_context(NULL);
 100
 101    ralloc_adopt(mem_ctx, shader->ir);
 102
 103    lower_blend_equation_advanced(shader);
 104
 105    /* lower_packing_builtins() inserts arithmetic instructions, so it
 106     * must precede lower_instructions().
 107     */
 108    brw_lower_packing_builtins(brw, shader->Stage, shader->ir);
 109    do_mat_op_to_vec(shader->ir);
 110
 111    unsigned instructions_to_lower = (DIV_TO_MUL_RCP |
 112                                      SUB_TO_ADD_NEG |
 113                                      EXP_TO_EXP2 |
 114                                      LOG_TO_LOG2 |
 115                                      DFREXP_DLDEXP_TO_ARITH);
 116    if (brw->gen < 7) {
 117       instructions_to_lower |= BIT_COUNT_TO_MATH |
 118                                EXTRACT_TO_SHIFTS |
 119                                INSERT_TO_SHIFTS |
 120                                REVERSE_TO_SHIFTS;
 121    }
 122
 123    lower_instructions(shader->ir, instructions_to_lower);
 124
 125    /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 126     * if-statements need to be flattened.
 127     */
 128    if (brw->gen < 6)
 129       lower_if_to_cond_assign(shader->ir, 16);
 130
 131    do_lower_texture_projection(shader->ir);
 132    brw_lower_texture_gradients(brw, shader->ir);
 133    do_vec_index_to_cond_assign(shader->ir);
 134    lower_vector_insert(shader->ir, true);
 135    lower_offset_arrays(shader->ir);
 136    lower_noise(shader->ir);
 137    lower_quadop_vector(shader->ir, false);
 138
 139    do_copy_propagation(shader->ir);
 140
 141    bool lowered_variable_indexing =
 142       lower_variable_index_to_cond_assign(shader->Stage, shader->ir,
 143                                           options->EmitNoIndirectInput,
 144                                           options->EmitNoIndirectOutput,
 145                                           options->EmitNoIndirectTemp,
 146                                           options->EmitNoIndirectUniform);
 147
 148    if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
 149       perf_debug("Unsupported form of variable indexing in %s; falling "
 150                  "back to very inefficient code generation\n",
 151                  _mesa_shader_stage_to_abbrev(shader->Stage));
 152    }
 153
 154    bool progress;
 155    do {
 156       progress = false;
 157
 158       if (compiler->scalar_stage[shader->Stage]) {
 159          if (shader->Stage == MESA_SHADER_VERTEX ||
 160              shader->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,
 172                                         options, ctx->Const.NativeIntegers) || progress;
 173    } while (progress);
 174
 175    validate_ir_tree(shader->ir);
 176
 177    /* Now that we've finished altering the linked IR, reparent any live IR back
 178     * to the permanent memory context, and free the temporary one (discarding any
 179     * junk we optimized away).
 180     */
 181    reparent_ir(shader->ir, shader->ir);
 182    ralloc_free(mem_ctx);
 183
 184    if (ctx->_Shader->Flags & GLSL_DUMP) {
 185       fprintf(stderr, "\n");
 186       if (shader->ir) {
 187          fprintf(stderr, "GLSL IR for linked %s program %d:\n",
 188                  _mesa_shader_stage_to_string(shader->Stage),
 189                  shader_prog->Name);
 190          _mesa_print_ir(stderr, shader->ir, NULL);
 191       } else {
 192          fprintf(stderr, "No GLSL IR for linked %s program %d (shader may be "
 193                  "from cache)\n", _mesa_shader_stage_to_string(shader->Stage),
 194                  shader_prog->Name);
 195       }
 196       fprintf(stderr, "\n");
 197    }
 198 }
 199
 200 extern "C" struct gl_linked_shader *
 201 brw_new_shader(gl_shader_stage stage)
 202 {
 203    struct brw_shader *shader;
 204
 205    shader = rzalloc(NULL, struct brw_shader);
 206    if (shader) {
 207       shader->base.Stage = stage;
 208    }
 209
 210    return &shader->base;
 211 }
 212
 213 extern "C" GLboolean
 214 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 215 {
 216    struct brw_context *brw = brw_context(ctx);
 217    const struct brw_compiler *compiler = brw->screen->compiler;
 218    unsigned int stage;
 219
 220    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 221       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 222       if (!shader)
 223          continue;
 224
 225       struct gl_program *prog =
 226          ctx->Driver.NewProgram(ctx, _mesa_shader_stage_to_program(stage),
 227                                 0);
 228       if (!prog)
 229         return false;
 230       prog->Parameters = _mesa_new_parameter_list();
 231
 232       _mesa_copy_linked_program_data((gl_shader_stage) stage, shProg, prog);
 233
 234       process_glsl_ir(brw, shProg, shader);
 235
 236       /* Make a pass over the IR to add state references for any built-in
 237        * uniforms that are used.  This has to be done now (during linking).
 238        * Code generation doesn't happen until the first time this shader is
 239        * used for rendering.  Waiting until then to generate the parameters is
 240        * too late.  At that point, the values for the built-in uniforms won't
 241        * get sent to the shader.
 242        */
 243       foreach_in_list(ir_instruction, node, shader->ir) {
 244          ir_variable *var = node->as_variable();
 245
 246          if ((var == NULL) || (var->data.mode != ir_var_uniform)
 247              || (strncmp(var->name, "gl_", 3) != 0))
 248             continue;
 249
 250          const ir_state_slot *const slots = var->get_state_slots();
 251          assert(slots != NULL);
 252
 253          for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
 254             _mesa_add_state_reference(prog->Parameters,
 255                                       (gl_state_index *) slots[i].tokens);
 256          }
 257       }
 258
 259       do_set_program_inouts(shader->ir, prog, shader->Stage);
 260
 261       prog->SamplersUsed = shader->active_samplers;
 262       prog->ShadowSamplers = shader->shadow_samplers;
 263       _mesa_update_shader_textures_used(shProg, prog);
 264
 265       _mesa_reference_program(ctx, &shader->Program, prog);
 266
 267       brw_add_texrect_params(prog);
 268
 269       prog->nir = brw_create_nir(brw, shProg, prog, (gl_shader_stage) stage,
 270                                  compiler->scalar_stage[stage]);
 271
 272       _mesa_reference_program(ctx, &prog, NULL);
 273    }
 274
 275    if ((ctx->_Shader->Flags & GLSL_DUMP) && shProg->Name != 0) {
 276       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 277          const struct gl_shader *sh = shProg->Shaders[i];
 278          if (!sh)
 279             continue;
 280
 281          fprintf(stderr, "GLSL %s shader %d source for linked program %d:\n",
 282                  _mesa_shader_stage_to_string(sh->Stage),
 283                  i, shProg->Name);
 284          fprintf(stderr, "%s", sh->Source);
 285          fprintf(stderr, "\n");
 286       }
 287    }
 288
 289    if (brw->precompile && !brw_shader_precompile(ctx, shProg))
 290       return false;
 291
 292    build_program_resource_list(ctx, shProg);
 293
 294    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 295       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 296       if (!shader)
 297          continue;
 298
 299       /* The GLSL IR won't be needed anymore. */
 300       ralloc_free(shader->ir);
 301       shader->ir = NULL;
 302    }
 303
 304    return true;
 305 }