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/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_shader *vs = sh_prog->_LinkedShaders[MESA_SHADER_VERTEX];
  46    struct gl_shader *tcs = sh_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
  47    struct gl_shader *tes = sh_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
  48    struct gl_shader *gs = sh_prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
  49    struct gl_shader *fs = sh_prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  50    struct gl_shader *cs = sh_prog->_LinkedShaders[MESA_SHADER_COMPUTE];
  51
  52    if (fs && !brw_fs_precompile(ctx, sh_prog, fs->Program))
  53       return false;
  54
  55    if (gs && !brw_gs_precompile(ctx, sh_prog, 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, sh_prog, vs->Program))
  65       return false;
  66
  67    if (cs && !brw_cs_precompile(ctx, sh_prog, cs->Program))
  68       return false;
  69
  70    return true;
  71 }
  72
  73 static void
  74 brw_lower_packing_builtins(struct brw_context *brw,
  75                            gl_shader_stage shader_type,
  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(gl_shader_stage stage,
  89                 struct brw_context *brw,
  90                 struct gl_shader_program *shader_prog,
  91                 struct gl_shader *shader)
  92 {
  93    struct gl_context *ctx = &brw->ctx;
  94    const struct brw_compiler *compiler = brw->intelScreen->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_packing_builtins() inserts arithmetic instructions, so it
 104     * must precede lower_instructions().
 105     */
 106    brw_lower_packing_builtins(brw, shader->Stage, shader->ir);
 107    do_mat_op_to_vec(shader->ir);
 108    lower_instructions(shader->ir,
 109                       DIV_TO_MUL_RCP |
 110                       SUB_TO_ADD_NEG |
 111                       EXP_TO_EXP2 |
 112                       LOG_TO_LOG2 |
 113                       DFREXP_DLDEXP_TO_ARITH |
 114                       CARRY_TO_ARITH |
 115                       BORROW_TO_ARITH);
 116
 117    /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 118     * if-statements need to be flattened.
 119     */
 120    if (brw->gen < 6)
 121       lower_if_to_cond_assign(shader->ir, 16);
 122
 123    do_lower_texture_projection(shader->ir);
 124    brw_lower_texture_gradients(brw, shader->ir);
 125    do_vec_index_to_cond_assign(shader->ir);
 126    lower_vector_insert(shader->ir, true);
 127    lower_offset_arrays(shader->ir);
 128    brw_do_lower_unnormalized_offset(shader->ir);
 129    lower_noise(shader->ir);
 130    lower_quadop_vector(shader->ir, false);
 131
 132    bool lowered_variable_indexing =
 133       lower_variable_index_to_cond_assign((gl_shader_stage)stage,
 134                                           shader->ir,
 135                                           options->EmitNoIndirectInput,
 136                                           options->EmitNoIndirectOutput,
 137                                           options->EmitNoIndirectTemp,
 138                                           options->EmitNoIndirectUniform);
 139
 140    if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
 141       perf_debug("Unsupported form of variable indexing in %s; falling "
 142                  "back to very inefficient code generation\n",
 143                  _mesa_shader_stage_to_abbrev(shader->Stage));
 144    }
 145
 146    bool progress;
 147    do {
 148       progress = false;
 149
 150       if (compiler->scalar_stage[shader->Stage]) {
 151          if (shader->Stage == MESA_SHADER_VERTEX ||
 152              shader->Stage == MESA_SHADER_FRAGMENT)
 153             brw_do_channel_expressions(shader->ir);
 154          brw_do_vector_splitting(shader->ir);
 155       }
 156
 157       progress = do_lower_jumps(shader->ir, true, true,
 158                                 true, /* main return */
 159                                 false, /* continue */
 160                                 false /* loops */
 161                                 ) || progress;
 162
 163       progress = do_common_optimization(shader->ir, true, true,
 164                                         options, ctx->Const.NativeIntegers) || progress;
 165    } while (progress);
 166
 167    validate_ir_tree(shader->ir);
 168
 169    /* Now that we've finished altering the linked IR, reparent any live IR back
 170     * to the permanent memory context, and free the temporary one (discarding any
 171     * junk we optimized away).
 172     */
 173    reparent_ir(shader->ir, shader->ir);
 174    ralloc_free(mem_ctx);
 175
 176    if (ctx->_Shader->Flags & GLSL_DUMP) {
 177       fprintf(stderr, "\n");
 178       fprintf(stderr, "GLSL IR for linked %s program %d:\n",
 179               _mesa_shader_stage_to_string(shader->Stage),
 180               shader_prog->Name);
 181       _mesa_print_ir(stderr, shader->ir, NULL);
 182       fprintf(stderr, "\n");
 183    }
 184 }
 185
 186 extern "C" GLboolean
 187 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 188 {
 189    struct brw_context *brw = brw_context(ctx);
 190    const struct brw_compiler *compiler = brw->intelScreen->compiler;
 191    unsigned int stage;
 192
 193    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 194       struct gl_shader *shader = shProg->_LinkedShaders[stage];
 195       if (!shader)
 196          continue;
 197
 198       struct gl_program *prog =
 199          ctx->Driver.NewProgram(ctx, _mesa_shader_stage_to_program(stage),
 200                                 shader->Name);
 201       if (!prog)
 202         return false;
 203       prog->Parameters = _mesa_new_parameter_list();
 204
 205       _mesa_copy_linked_program_data((gl_shader_stage) stage, shProg, prog);
 206
 207       process_glsl_ir((gl_shader_stage) stage, brw, shProg, shader);
 208
 209       /* Make a pass over the IR to add state references for any built-in
 210        * uniforms that are used.  This has to be done now (during linking).
 211        * Code generation doesn't happen until the first time this shader is
 212        * used for rendering.  Waiting until then to generate the parameters is
 213        * too late.  At that point, the values for the built-in uniforms won't
 214        * get sent to the shader.
 215        */
 216       foreach_in_list(ir_instruction, node, shader->ir) {
 217          ir_variable *var = node->as_variable();
 218
 219          if ((var == NULL) || (var->data.mode != ir_var_uniform)
 220              || (strncmp(var->name, "gl_", 3) != 0))
 221             continue;
 222
 223          const ir_state_slot *const slots = var->get_state_slots();
 224          assert(slots != NULL);
 225
 226          for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
 227             _mesa_add_state_reference(prog->Parameters,
 228                                       (gl_state_index *) slots[i].tokens);
 229          }
 230       }
 231
 232       do_set_program_inouts(shader->ir, prog, shader->Stage);
 233
 234       prog->SamplersUsed = shader->active_samplers;
 235       prog->ShadowSamplers = shader->shadow_samplers;
 236       _mesa_update_shader_textures_used(shProg, prog);
 237
 238       _mesa_reference_program(ctx, &shader->Program, prog);
 239
 240       brw_add_texrect_params(prog);
 241
 242       prog->nir = brw_create_nir(brw, shProg, prog, (gl_shader_stage) stage,
 243                                  compiler->scalar_stage[stage]);
 244
 245       _mesa_reference_program(ctx, &prog, NULL);
 246    }
 247
 248    if ((ctx->_Shader->Flags & GLSL_DUMP) && shProg->Name != 0) {
 249       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 250          const struct gl_shader *sh = shProg->Shaders[i];
 251          if (!sh)
 252             continue;
 253
 254          fprintf(stderr, "GLSL %s shader %d source for linked program %d:\n",
 255                  _mesa_shader_stage_to_string(sh->Stage),
 256                  i, shProg->Name);
 257          fprintf(stderr, "%s", sh->Source);
 258          fprintf(stderr, "\n");
 259       }
 260    }
 261
 262    if (brw->precompile && !brw_shader_precompile(ctx, shProg))
 263       return false;
 264
 265    build_program_resource_list(ctx, shProg);
 266    return true;
 267 }