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 "compiler/brw_nir.h"
  26 #include "brw_program.h"
  27 #include "compiler/glsl/ir.h"
  28 #include "compiler/glsl/ir_optimization.h"
  29 #include "compiler/glsl/program.h"
  30 #include "program/program.h"
  31 #include "main/mtypes.h"
  32 #include "main/shaderapi.h"
  33 #include "main/shaderobj.h"
  34 #include "main/uniforms.h"
  35
  36 /**
  37  * Performs a compile of the shader stages even when we don't know
  38  * what non-orthogonal state will be set, in the hope that it reflects
  39  * the eventual NOS used, and thus allows us to produce link failures.
  40  */
  41 static bool
  42 brw_shader_precompile(struct gl_context *ctx,
  43                       struct gl_shader_program *sh_prog)
  44 {
  45    struct gl_linked_shader *vs = sh_prog->_LinkedShaders[MESA_SHADER_VERTEX];
  46    struct gl_linked_shader *tcs = sh_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
  47    struct gl_linked_shader *tes = sh_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
  48    struct gl_linked_shader *gs = sh_prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
  49    struct gl_linked_shader *fs = sh_prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  50    struct gl_linked_shader *cs = sh_prog->_LinkedShaders[MESA_SHADER_COMPUTE];
  51
  52    if (fs && !brw_fs_precompile(ctx, fs->Program))
  53       return false;
  54
  55    if (gs && !brw_gs_precompile(ctx, gs->Program))
  56       return false;
  57
  58    if (tes && !brw_tes_precompile(ctx, sh_prog, tes->Program))
  59       return false;
  60
  61    if (tcs && !brw_tcs_precompile(ctx, sh_prog, tcs->Program))
  62       return false;
  63
  64    if (vs && !brw_vs_precompile(ctx, vs->Program))
  65       return false;
  66
  67    if (cs && !brw_cs_precompile(ctx, cs->Program))
  68       return false;
  69
  70    return true;
  71 }
  72
  73 static void
  74 brw_lower_packing_builtins(struct brw_context *brw,
  75                            exec_list *ir)
  76 {
  77    /* Gens < 7 don't have instructions to convert to or from half-precision,
  78     * and Gens < 6 don't expose that functionality.
  79     */
  80    if (brw->gen != 6)
  81       return;
  82
  83    lower_packing_builtins(ir, LOWER_PACK_HALF_2x16 | LOWER_UNPACK_HALF_2x16);
  84 }
  85
  86 static void
  87 process_glsl_ir(struct brw_context *brw,
  88                 struct gl_shader_program *shader_prog,
  89                 struct gl_linked_shader *shader)
  90 {
  91    struct gl_context *ctx = &brw->ctx;
  92
  93    /* Temporary memory context for any new IR. */
  94    void *mem_ctx = ralloc_context(NULL);
  95
  96    ralloc_adopt(mem_ctx, shader->ir);
  97
  98    lower_blend_equation_advanced(shader);
  99
 100    /* lower_packing_builtins() inserts arithmetic instructions, so it
 101     * must precede lower_instructions().
 102     */
 103    brw_lower_packing_builtins(brw, shader->ir);
 104    do_mat_op_to_vec(shader->ir);
 105
 106    unsigned instructions_to_lower = (DIV_TO_MUL_RCP |
 107                                      SUB_TO_ADD_NEG |
 108                                      EXP_TO_EXP2 |
 109                                      LOG_TO_LOG2 |
 110                                      DFREXP_DLDEXP_TO_ARITH);
 111    if (brw->gen < 7) {
 112       instructions_to_lower |= BIT_COUNT_TO_MATH |
 113                                EXTRACT_TO_SHIFTS |
 114                                INSERT_TO_SHIFTS |
 115                                REVERSE_TO_SHIFTS;
 116    }
 117
 118    lower_instructions(shader->ir, instructions_to_lower);
 119
 120    /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 121     * if-statements need to be flattened.
 122     */
 123    if (brw->gen < 6)
 124       lower_if_to_cond_assign(shader->Stage, shader->ir, 16);
 125
 126    do_lower_texture_projection(shader->ir);
 127    do_vec_index_to_cond_assign(shader->ir);
 128    lower_vector_insert(shader->ir, true);
 129    lower_offset_arrays(shader->ir);
 130    lower_noise(shader->ir);
 131    lower_quadop_vector(shader->ir, false);
 132
 133    validate_ir_tree(shader->ir);
 134
 135    /* Now that we've finished altering the linked IR, reparent any live IR back
 136     * to the permanent memory context, and free the temporary one (discarding any
 137     * junk we optimized away).
 138     */
 139    reparent_ir(shader->ir, shader->ir);
 140    ralloc_free(mem_ctx);
 141
 142    if (ctx->_Shader->Flags & GLSL_DUMP) {
 143       fprintf(stderr, "\n");
 144       if (shader->ir) {
 145          fprintf(stderr, "GLSL IR for linked %s program %d:\n",
 146                  _mesa_shader_stage_to_string(shader->Stage),
 147                  shader_prog->Name);
 148          _mesa_print_ir(stderr, shader->ir, NULL);
 149       } else {
 150          fprintf(stderr, "No GLSL IR for linked %s program %d (shader may be "
 151                  "from cache)\n", _mesa_shader_stage_to_string(shader->Stage),
 152                  shader_prog->Name);
 153       }
 154       fprintf(stderr, "\n");
 155    }
 156 }
 157
 158 static void
 159 unify_interfaces(struct shader_info **infos)
 160 {
 161    struct shader_info *prev_info = NULL;
 162
 163    for (unsigned i = MESA_SHADER_VERTEX; i < MESA_SHADER_FRAGMENT; i++) {
 164       if (!infos[i])
 165          continue;
 166
 167       if (prev_info) {
 168          prev_info->outputs_written |= infos[i]->inputs_read &
 169             ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 170          infos[i]->inputs_read |= prev_info->outputs_written &
 171             ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 172
 173          prev_info->patch_outputs_written |= infos[i]->patch_inputs_read;
 174          infos[i]->patch_inputs_read |= prev_info->patch_outputs_written;
 175       }
 176       prev_info = infos[i];
 177    }
 178 }
 179
 180 static void
 181 update_xfb_info(struct gl_transform_feedback_info *xfb_info)
 182 {
 183    if (!xfb_info)
 184       return;
 185
 186    for (unsigned i = 0; i < xfb_info->NumOutputs; i++) {
 187       struct gl_transform_feedback_output *output = &xfb_info->Outputs[i];
 188
 189       /* The VUE header contains three scalar fields packed together:
 190        * - gl_PointSize is stored in VARYING_SLOT_PSIZ.w
 191        * - gl_Layer is stored in VARYING_SLOT_PSIZ.y
 192        * - gl_ViewportIndex is stored in VARYING_SLOT_PSIZ.z
 193        */
 194       switch (output->OutputRegister) {
 195       case VARYING_SLOT_LAYER:
 196          assert(output->NumComponents == 1);
 197          output->OutputRegister = VARYING_SLOT_PSIZ;
 198          output->ComponentOffset = 1;
 199          break;
 200       case VARYING_SLOT_VIEWPORT:
 201          assert(output->NumComponents == 1);
 202          output->OutputRegister = VARYING_SLOT_PSIZ;
 203          output->ComponentOffset = 2;
 204          break;
 205       case VARYING_SLOT_PSIZ:
 206          assert(output->NumComponents == 1);
 207          output->ComponentOffset = 3;
 208          break;
 209       }
 210    }
 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    struct shader_info *infos[MESA_SHADER_STAGES] = { 0, };
 220
 221    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 222       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 223       if (!shader)
 224          continue;
 225
 226       struct gl_program *prog = shader->Program;
 227       prog->Parameters = _mesa_new_parameter_list();
 228
 229       process_glsl_ir(brw, shProg, shader);
 230
 231       _mesa_copy_linked_program_data(shProg, shader);
 232
 233       prog->ShadowSamplers = shader->shadow_samplers;
 234       _mesa_update_shader_textures_used(shProg, prog);
 235
 236       update_xfb_info(prog->sh.LinkedTransformFeedback);
 237
 238       bool debug_enabled =
 239          (INTEL_DEBUG & intel_debug_flag_for_shader_stage(shader->Stage));
 240
 241       if (debug_enabled && shader->ir) {
 242          fprintf(stderr, "GLSL IR for native %s shader %d:\n",
 243                  _mesa_shader_stage_to_string(shader->Stage), shProg->Name);
 244          _mesa_print_ir(stderr, shader->ir, NULL);
 245          fprintf(stderr, "\n\n");
 246       }
 247
 248       prog->nir = brw_create_nir(brw, shProg, prog, (gl_shader_stage) stage,
 249                                  compiler->scalar_stage[stage]);
 250       infos[stage] = &prog->nir->info;
 251
 252       /* Make a pass over the IR to add state references for any built-in
 253        * uniforms that are used.  This has to be done now (during linking).
 254        * Code generation doesn't happen until the first time this shader is
 255        * used for rendering.  Waiting until then to generate the parameters is
 256        * too late.  At that point, the values for the built-in uniforms won't
 257        * get sent to the shader.
 258        */
 259       nir_foreach_variable(var, &prog->nir->uniforms) {
 260          if (strncmp(var->name, "gl_", 3) == 0) {
 261             const nir_state_slot *const slots = var->state_slots;
 262             assert(var->state_slots != NULL);
 263
 264             for (unsigned int i = 0; i < var->num_state_slots; i++) {
 265                _mesa_add_state_reference(prog->Parameters,
 266                                          (gl_state_index *)slots[i].tokens);
 267             }
 268          }
 269       }
 270    }
 271
 272    /* The linker tries to dead code eliminate unused varying components,
 273     * and make sure interfaces match.  But it isn't able to do so in all
 274     * cases.  So, explicitly make the interfaces match by OR'ing together
 275     * the inputs_read/outputs_written bitfields of adjacent stages.
 276     */
 277    if (!shProg->SeparateShader)
 278       unify_interfaces(infos);
 279
 280    if ((ctx->_Shader->Flags & GLSL_DUMP) && shProg->Name != 0) {
 281       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 282          const struct gl_shader *sh = shProg->Shaders[i];
 283          if (!sh)
 284             continue;
 285
 286          fprintf(stderr, "GLSL %s shader %d source for linked program %d:\n",
 287                  _mesa_shader_stage_to_string(sh->Stage),
 288                  i, shProg->Name);
 289          fprintf(stderr, "%s", sh->Source);
 290          fprintf(stderr, "\n");
 291       }
 292    }
 293
 294    if (brw->precompile && !brw_shader_precompile(ctx, shProg))
 295       return false;
 296
 297    build_program_resource_list(ctx, shProg);
 298
 299    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 300       struct gl_linked_shader *shader = shProg->_LinkedShaders[stage];
 301       if (!shader)
 302          continue;
 303
 304       /* The GLSL IR won't be needed anymore. */
 305       ralloc_free(shader->ir);
 306       shader->ir = NULL;
 307    }
 308
 309    return true;
 310 }