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                            exec_list *ir)
  77 {
  78    /* Gens < 7 don't have instructions to convert to or from half-precision,
  79     * and Gens < 6 don't expose that functionality.
  80     */
  81    if (brw->gen != 6)
  82       return;
  83
  84    lower_packing_builtins(ir, LOWER_PACK_HALF_2x16 | LOWER_UNPACK_HALF_2x16);
  85 }
  86
  87 static void
  88 process_glsl_ir(struct brw_context *brw,
  89                 struct gl_shader_program *shader_prog,
  90                 struct gl_linked_shader *shader)
  91 {
  92    struct gl_context *ctx = &brw->ctx;
  93    const struct brw_compiler *compiler = brw->screen->compiler;
  94    const struct gl_shader_compiler_options *options =
  95       &ctx->Const.ShaderCompilerOptions[shader->Stage];
  96
  97    /* Temporary memory context for any new IR. */
  98    void *mem_ctx = ralloc_context(NULL);
  99
 100    ralloc_adopt(mem_ctx, shader->ir);
 101
 102    lower_blend_equation_advanced(shader);
 103
 104    /* lower_packing_builtins() inserts arithmetic instructions, so it
 105     * must precede lower_instructions().
 106     */
 107    brw_lower_packing_builtins(brw, shader->ir);
 108    do_mat_op_to_vec(shader->ir);
 109
 110    unsigned instructions_to_lower = (DIV_TO_MUL_RCP |
 111                                      SUB_TO_ADD_NEG |
 112                                      EXP_TO_EXP2 |
 113                                      LOG_TO_LOG2 |
 114                                      DFREXP_DLDEXP_TO_ARITH);
 115    if (brw->gen < 7) {
 116       instructions_to_lower |= BIT_COUNT_TO_MATH |
 117                                EXTRACT_TO_SHIFTS |
 118                                INSERT_TO_SHIFTS |
 119                                REVERSE_TO_SHIFTS;
 120    }
 121
 122    lower_instructions(shader->ir, instructions_to_lower);
 123
 124    /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 125     * if-statements need to be flattened.
 126     */
 127    if (brw->gen < 6)
 128       lower_if_to_cond_assign(shader->Stage, shader->ir, 16);
 129
 130    do_lower_texture_projection(shader->ir);
 131    brw_lower_texture_gradients(brw, shader->ir);
 132    do_vec_index_to_cond_assign(shader->ir);
 133    lower_vector_insert(shader->ir, true);
 134    lower_offset_arrays(shader->ir);
 135    lower_noise(shader->ir);
 136    lower_quadop_vector(shader->ir, false);
 137
 138    do_copy_propagation(shader->ir);
 139
 140    bool lowered_variable_indexing =
 141       lower_variable_index_to_cond_assign(shader->Stage, shader->ir,
 142                                           options->EmitNoIndirectInput,
 143                                           options->EmitNoIndirectOutput,
 144                                           options->EmitNoIndirectTemp,
 145                                           options->EmitNoIndirectUniform);
 146
 147    if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
 148       perf_debug("Unsupported form of variable indexing in %s; falling "
 149                  "back to very inefficient code generation\n",
 150                  _mesa_shader_stage_to_abbrev(shader->Stage));
 151    }
 152
 153    bool progress;
 154    do {
 155       progress = false;
 156
 157       if (compiler->scalar_stage[shader->Stage]) {
 158          if (shader->Stage == MESA_SHADER_VERTEX ||
 159              shader->Stage == MESA_SHADER_FRAGMENT)
 160             brw_do_channel_expressions(shader->ir);
 161          brw_do_vector_splitting(shader->ir);
 162       }
 163
 164       progress = do_lower_jumps(shader->ir, true, true,
 165                                 true, /* main return */
 166                                 false, /* continue */
 167                                 false /* loops */
 168                                 ) || progress;
 169
 170       progress = do_common_optimization(shader->ir, true, true,
 171                                         options, ctx->Const.NativeIntegers) || progress;
 172    } while (progress);
 173
 174    validate_ir_tree(shader->ir);
 175
 176    /* Now that we've finished altering the linked IR, reparent any live IR back
 177     * to the permanent memory context, and free the temporary one (discarding any
 178     * junk we optimized away).
 179     */
 180    reparent_ir(shader->ir, shader->ir);
 181    ralloc_free(mem_ctx);
 182
 183    if (ctx->_Shader->Flags & GLSL_DUMP) {
 184       fprintf(stderr, "\n");
 185       if (shader->ir) {
 186          fprintf(stderr, "GLSL IR for linked %s program %d:\n",
 187                  _mesa_shader_stage_to_string(shader->Stage),
 188                  shader_prog->Name);
 189          _mesa_print_ir(stderr, shader->ir, NULL);
 190       } else {
 191          fprintf(stderr, "No GLSL IR for linked %s program %d (shader may be "
 192                  "from cache)\n", _mesa_shader_stage_to_string(shader->Stage),
 193                  shader_prog->Name);
 194       }
 195       fprintf(stderr, "\n");
 196    }
 197 }
 198
 199 extern "C" struct gl_linked_shader *
 200 brw_new_shader(gl_shader_stage stage)
 201 {
 202    struct brw_shader *shader;
 203
 204    shader = rzalloc(NULL, struct brw_shader);
 205    if (shader) {
 206       shader->base.Stage = stage;
 207    }
 208
 209    return &shader->base;
 210 }
 211
 212 extern "C" GLboolean
 213 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 214 {
 215    struct brw_context *brw = brw_context(ctx);
 216    const struct brw_compiler *compiler = brw->screen->compiler;
 217    unsigned int stage;
 218
 219    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 220       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 221       if (!shader)
 222          continue;
 223
 224       struct gl_program *prog = shader->Program;
 225       prog->Parameters = _mesa_new_parameter_list();
 226
 227       process_glsl_ir(brw, shProg, shader);
 228
 229       _mesa_copy_linked_program_data(shProg, shader);
 230
 231       /* Make a pass over the IR to add state references for any built-in
 232        * uniforms that are used.  This has to be done now (during linking).
 233        * Code generation doesn't happen until the first time this shader is
 234        * used for rendering.  Waiting until then to generate the parameters is
 235        * too late.  At that point, the values for the built-in uniforms won't
 236        * get sent to the shader.
 237        */
 238       foreach_in_list(ir_instruction, node, shader->ir) {
 239          ir_variable *var = node->as_variable();
 240
 241          if ((var == NULL) || (var->data.mode != ir_var_uniform)
 242              || (strncmp(var->name, "gl_", 3) != 0))
 243             continue;
 244
 245          const ir_state_slot *const slots = var->get_state_slots();
 246          assert(slots != NULL);
 247
 248          for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
 249             _mesa_add_state_reference(prog->Parameters,
 250                                       (gl_state_index *) slots[i].tokens);
 251          }
 252       }
 253
 254       prog->SamplersUsed = shader->active_samplers;
 255       prog->ShadowSamplers = shader->shadow_samplers;
 256       _mesa_update_shader_textures_used(shProg, prog);
 257
 258       brw_add_texrect_params(prog);
 259
 260       bool debug_enabled =
 261          (INTEL_DEBUG & intel_debug_flag_for_shader_stage(shader->Stage));
 262
 263       if (debug_enabled && shader->ir) {
 264          fprintf(stderr, "GLSL IR for native %s shader %d:\n",
 265                  _mesa_shader_stage_to_string(shader->Stage), shProg->Name);
 266          _mesa_print_ir(stderr, shader->ir, NULL);
 267          fprintf(stderr, "\n\n");
 268       }
 269
 270       prog->nir = brw_create_nir(brw, shProg, prog, (gl_shader_stage) stage,
 271                                  compiler->scalar_stage[stage]);
 272    }
 273
 274    if ((ctx->_Shader->Flags & GLSL_DUMP) && shProg->Name != 0) {
 275       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 276          const struct gl_shader *sh = shProg->Shaders[i];
 277          if (!sh)
 278             continue;
 279
 280          fprintf(stderr, "GLSL %s shader %d source for linked program %d:\n",
 281                  _mesa_shader_stage_to_string(sh->Stage),
 282                  i, shProg->Name);
 283          fprintf(stderr, "%s", sh->Source);
 284          fprintf(stderr, "\n");
 285       }
 286    }
 287
 288    if (brw->precompile && !brw_shader_precompile(ctx, shProg))
 289       return false;
 290
 291    build_program_resource_list(ctx, shProg);
 292
 293    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 294       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 295       if (!shader)
 296          continue;
 297
 298       /* The GLSL IR won't be needed anymore. */
 299       ralloc_free(shader->ir);
 300       shader->ir = NULL;
 301    }
 302
 303    return true;
 304 }