src/mesa/drivers/dri/i965/brw_tcs.c

   1 /*
   2  * Copyright © 2014 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 /**
  25  * \file brw_tcs.c
  26  *
  27  * Tessellation control shader state upload code.
  28  */
  29
  30 #include "brw_context.h"
  31 #include "brw_nir.h"
  32 #include "brw_program.h"
  33 #include "brw_shader.h"
  34 #include "brw_state.h"
  35 #include "program/prog_parameter.h"
  36
  37 static void
  38 brw_tcs_debug_recompile(struct brw_context *brw,
  39                        struct gl_shader_program *shader_prog,
  40                        const struct brw_tcs_prog_key *key)
  41 {
  42    struct brw_cache_item *c = NULL;
  43    const struct brw_tcs_prog_key *old_key = NULL;
  44    bool found = false;
  45
  46    perf_debug("Recompiling tessellation control shader for program %d\n",
  47               shader_prog->Name);
  48
  49    for (unsigned int i = 0; i < brw->cache.size; i++) {
  50       for (c = brw->cache.items[i]; c; c = c->next) {
  51          if (c->cache_id == BRW_CACHE_TCS_PROG) {
  52             old_key = c->key;
  53
  54             if (old_key->program_string_id == key->program_string_id)
  55                break;
  56          }
  57       }
  58       if (c)
  59          break;
  60    }
  61
  62    if (!c) {
  63       perf_debug("  Didn't find previous compile in the shader cache for "
  64                  "debug\n");
  65       return;
  66    }
  67
  68    found |= key_debug(brw, "input vertices", old_key->input_vertices,
  69                       key->input_vertices);
  70    found |= key_debug(brw, "outputs written", old_key->outputs_written,
  71                       key->outputs_written);
  72    found |= key_debug(brw, "patch outputs written", old_key->patch_outputs_written,
  73                       key->patch_outputs_written);
  74    found |= key_debug(brw, "TES primitive mode", old_key->tes_primitive_mode,
  75                       key->tes_primitive_mode);
  76    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  77
  78    if (!found) {
  79       perf_debug("  Something else\n");
  80    }
  81 }
  82
  83 static bool
  84 brw_codegen_tcs_prog(struct brw_context *brw,
  85                      struct gl_shader_program *shader_prog,
  86                      struct brw_tess_ctrl_program *tcp,
  87                      struct brw_tcs_prog_key *key)
  88 {
  89    struct gl_context *ctx = &brw->ctx;
  90    const struct brw_compiler *compiler = brw->intelScreen->compiler;
  91    struct brw_stage_state *stage_state = &brw->tcs.base;
  92    nir_shader *nir;
  93    struct brw_tcs_prog_data prog_data;
  94    bool start_busy = false;
  95    double start_time = 0;
  96
  97    if (tcp) {
  98       nir = tcp->program.Base.nir;
  99    } else {
 100       /* Create a dummy nir_shader.  We won't actually use NIR code to
 101        * generate assembly (it's easier to generate assembly directly),
 102        * but the whole compiler assumes one of these exists.
 103        */
 104       const nir_shader_compiler_options *options =
 105          ctx->Const.ShaderCompilerOptions[MESA_SHADER_TESS_CTRL].NirOptions;
 106       nir = nir_shader_create(NULL, MESA_SHADER_TESS_CTRL, options);
 107       nir->num_uniforms = 2; /* both halves of the patch header */
 108       nir->info.outputs_written = key->outputs_written;
 109       nir->info.inputs_read = key->outputs_written;
 110       nir->info.tcs.vertices_out = key->input_vertices;
 111       nir->info.name = ralloc_strdup(nir, "passthrough");
 112    }
 113
 114    memset(&prog_data, 0, sizeof(prog_data));
 115
 116    /* Allocate the references to the uniforms that will end up in the
 117     * prog_data associated with the compiled program, and which will be freed
 118     * by the state cache.
 119     *
 120     * Note: param_count needs to be num_uniform_components * 4, since we add
 121     * padding around uniform values below vec4 size, so the worst case is that
 122     * every uniform is a float which gets padded to the size of a vec4.
 123     */
 124    struct gl_shader *tcs = shader_prog ?
 125       shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL] : NULL;
 126    int param_count = nir->num_uniforms;
 127    if (!compiler->scalar_stage[MESA_SHADER_TESS_CTRL])
 128       param_count *= 4;
 129
 130    prog_data.base.base.param =
 131       rzalloc_array(NULL, const gl_constant_value *, param_count);
 132    prog_data.base.base.pull_param =
 133       rzalloc_array(NULL, const gl_constant_value *, param_count);
 134    prog_data.base.base.nr_params = param_count;
 135
 136    if (tcs) {
 137       prog_data.base.base.image_param =
 138          rzalloc_array(NULL, struct brw_image_param, tcs->NumImages);
 139       prog_data.base.base.nr_image_params = tcs->NumImages;
 140
 141       brw_nir_setup_glsl_uniforms(nir, shader_prog, &tcp->program.Base,
 142                                   &prog_data.base.base, false);
 143    } else {
 144       /* Upload the Patch URB Header as the first two uniforms.
 145        * Do the annoying scrambling so the shader doesn't have to.
 146        */
 147       const float **param = (const float **) prog_data.base.base.param;
 148       static float zero = 0.0f;
 149       for (int i = 0; i < 4; i++) {
 150          param[7 - i] = &ctx->TessCtrlProgram.patch_default_outer_level[i];
 151       }
 152
 153       if (key->tes_primitive_mode == GL_QUADS) {
 154          param[3] = &ctx->TessCtrlProgram.patch_default_inner_level[0];
 155          param[2] = &ctx->TessCtrlProgram.patch_default_inner_level[1];
 156          param[1] = &zero;
 157          param[0] = &zero;
 158       } else if (key->tes_primitive_mode == GL_TRIANGLES) {
 159          param[4] = &ctx->TessCtrlProgram.patch_default_inner_level[0];
 160          for (int i = 0; i < 4; i++)
 161             param[i] = &zero;
 162       }
 163    }
 164
 165    if (unlikely(INTEL_DEBUG & DEBUG_TCS) && tcs)
 166       brw_dump_ir("tessellation control", shader_prog, tcs, NULL);
 167
 168    int st_index = -1;
 169    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 170       st_index = brw_get_shader_time_index(brw, shader_prog, NULL, ST_TCS);
 171
 172    if (unlikely(brw->perf_debug)) {
 173       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 174       start_time = get_time();
 175    }
 176
 177    void *mem_ctx = ralloc_context(NULL);
 178    unsigned program_size;
 179    char *error_str;
 180    const unsigned *program =
 181       brw_compile_tcs(compiler, brw, mem_ctx, key, &prog_data, nir, st_index,
 182                       &program_size, &error_str);
 183    if (program == NULL) {
 184       if (shader_prog) {
 185          shader_prog->LinkStatus = false;
 186          ralloc_strcat(&shader_prog->InfoLog, error_str);
 187       } else {
 188          ralloc_free(nir);
 189       }
 190
 191       _mesa_problem(NULL, "Failed to compile tessellation control shader: "
 192                     "%s\n", error_str);
 193
 194       ralloc_free(mem_ctx);
 195       return false;
 196    }
 197
 198    if (unlikely(brw->perf_debug)) {
 199       struct brw_shader *btcs = (struct brw_shader *) tcs;
 200       if (btcs->compiled_once) {
 201          brw_tcs_debug_recompile(brw, shader_prog, key);
 202       }
 203       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 204          perf_debug("TCS compile took %.03f ms and stalled the GPU\n",
 205                     (get_time() - start_time) * 1000);
 206       }
 207       btcs->compiled_once = true;
 208    }
 209
 210    /* Scratch space is used for register spilling */
 211    if (prog_data.base.base.total_scratch) {
 212       brw_get_scratch_bo(brw, &stage_state->scratch_bo,
 213                          prog_data.base.base.total_scratch *
 214                          brw->max_hs_threads);
 215    }
 216
 217    brw_upload_cache(&brw->cache, BRW_CACHE_TCS_PROG,
 218                     key, sizeof(*key),
 219                     program, program_size,
 220                     &prog_data, sizeof(prog_data),
 221                     &stage_state->prog_offset, &brw->tcs.prog_data);
 222    ralloc_free(mem_ctx);
 223    if (!tcs)
 224       ralloc_free(nir);
 225
 226    return true;
 227 }
 228
 229
 230 void
 231 brw_upload_tcs_prog(struct brw_context *brw,
 232                     uint64_t per_vertex_slots,
 233                     uint32_t per_patch_slots)
 234 {
 235    struct gl_context *ctx = &brw->ctx;
 236    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 237    struct brw_stage_state *stage_state = &brw->tcs.base;
 238    struct brw_tcs_prog_key key;
 239    /* BRW_NEW_TESS_PROGRAMS */
 240    struct brw_tess_ctrl_program *tcp =
 241       (struct brw_tess_ctrl_program *) brw->tess_ctrl_program;
 242    struct brw_tess_eval_program *tep =
 243       (struct brw_tess_eval_program *) brw->tess_eval_program;
 244    assert(tep);
 245
 246    if (!brw_state_dirty(brw,
 247                         _NEW_TEXTURE,
 248                         BRW_NEW_PATCH_PRIMITIVE |
 249                         BRW_NEW_TESS_PROGRAMS))
 250       return;
 251
 252    struct gl_program *prog = &tcp->program.Base;
 253
 254    memset(&key, 0, sizeof(key));
 255
 256    key.input_vertices = ctx->TessCtrlProgram.patch_vertices;
 257    key.outputs_written = per_vertex_slots;
 258    key.patch_outputs_written = per_patch_slots;
 259
 260    /* We need to specialize our code generation for tessellation levels
 261     * based on the domain the DS is expecting to tessellate.
 262     */
 263    key.tes_primitive_mode = tep->program.PrimitiveMode;
 264
 265    if (tcp) {
 266       key.program_string_id = tcp->id;
 267
 268       /* _NEW_TEXTURE */
 269       brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
 270                                          &key.tex);
 271    } else {
 272       key.outputs_written = tep->program.Base.InputsRead;
 273    }
 274
 275
 276    if (!brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG,
 277                          &key, sizeof(key),
 278                          &stage_state->prog_offset, &brw->tcs.prog_data)) {
 279       bool success = brw_codegen_tcs_prog(brw, current[MESA_SHADER_TESS_CTRL],
 280                                           tcp, &key);
 281       assert(success);
 282       (void)success;
 283    }
 284    brw->tcs.base.prog_data = &brw->tcs.prog_data->base.base;
 285 }
 286
 287
 288 bool
 289 brw_tcs_precompile(struct gl_context *ctx,
 290                    struct gl_shader_program *shader_prog,
 291                    struct gl_program *prog)
 292 {
 293    struct brw_context *brw = brw_context(ctx);
 294    struct brw_tcs_prog_key key;
 295    uint32_t old_prog_offset = brw->tcs.base.prog_offset;
 296    struct brw_tcs_prog_data *old_prog_data = brw->tcs.prog_data;
 297    bool success;
 298
 299    struct gl_tess_ctrl_program *tcp = (struct gl_tess_ctrl_program *)prog;
 300    struct brw_tess_ctrl_program *btcp = brw_tess_ctrl_program(tcp);
 301
 302    memset(&key, 0, sizeof(key));
 303
 304    key.program_string_id = btcp->id;
 305    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 306
 307    /* Guess that the input and output patches have the same dimensionality. */
 308    key.input_vertices = shader_prog->TessCtrl.VerticesOut;
 309
 310    key.tes_primitive_mode =
 311       shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL] ?
 312       shader_prog->TessEval.PrimitiveMode : GL_TRIANGLES;
 313
 314    key.outputs_written = prog->OutputsWritten;
 315    key.patch_outputs_written = prog->PatchOutputsWritten;
 316
 317    success = brw_codegen_tcs_prog(brw, shader_prog, btcp, &key);
 318
 319    brw->tcs.base.prog_offset = old_prog_offset;
 320    brw->tcs.prog_data = old_prog_data;
 321
 322    return success;
 323 }