src/mesa/drivers/dri/i965/brw_tes.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_tes.c
  26  *
  27  * Tessellation evaluation 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_tes_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  39                        const struct brw_tes_prog_key *key)
  40 {
  41    perf_debug("Recompiling tessellation evaluation shader for program %d\n",
  42               prog->Id);
  43
  44    bool found = false;
  45    const struct brw_tes_prog_key *old_key =
  46       brw_find_previous_compile(&brw->cache, BRW_CACHE_TES_PROG,
  47                                 key->program_string_id);
  48
  49    if (!old_key) {
  50       perf_debug("  Didn't find previous compile in the shader cache for "
  51                  "debug\n");
  52       return;
  53    }
  54
  55    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  56    found |= key_debug(brw, "inputs read", old_key->inputs_read,
  57                       key->inputs_read);
  58    found |= key_debug(brw, "patch inputs read", old_key->patch_inputs_read,
  59                       key->patch_inputs_read);
  60
  61    if (!found) {
  62       perf_debug("  Something else\n");
  63    }
  64 }
  65
  66 static bool
  67 brw_codegen_tes_prog(struct brw_context *brw,
  68                      struct brw_program *tep,
  69                      struct brw_tes_prog_key *key)
  70 {
  71    const struct brw_compiler *compiler = brw->screen->compiler;
  72    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  73    struct brw_stage_state *stage_state = &brw->tes.base;
  74    nir_shader *nir = tep->program.nir;
  75    struct brw_tes_prog_data prog_data;
  76    bool start_busy = false;
  77    double start_time = 0;
  78
  79    memset(&prog_data, 0, sizeof(prog_data));
  80
  81    brw_assign_common_binding_table_offsets(devinfo, &tep->program,
  82                                            &prog_data.base.base, 0);
  83
  84    /* Allocate the references to the uniforms that will end up in the
  85     * prog_data associated with the compiled program, and which will be freed
  86     * by the state cache.
  87     *
  88     * Note: param_count needs to be num_uniform_components * 4, since we add
  89     * padding around uniform values below vec4 size, so the worst case is that
  90     * every uniform is a float which gets padded to the size of a vec4.
  91     */
  92    int param_count = nir->num_uniforms / 4;
  93
  94    prog_data.base.base.param =
  95       rzalloc_array(NULL, const gl_constant_value *, param_count);
  96    prog_data.base.base.pull_param =
  97       rzalloc_array(NULL, const gl_constant_value *, param_count);
  98    prog_data.base.base.image_param =
  99       rzalloc_array(NULL, struct brw_image_param,
 100                     tep->program.info.num_images);
 101    prog_data.base.base.nr_params = param_count;
 102    prog_data.base.base.nr_image_params = tep->program.info.num_images;
 103
 104    brw_nir_setup_glsl_uniforms(nir, &tep->program, &prog_data.base.base,
 105                                compiler->scalar_stage[MESA_SHADER_TESS_EVAL]);
 106
 107    int st_index = -1;
 108    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 109       st_index = brw_get_shader_time_index(brw, &tep->program, ST_TES, true);
 110
 111    if (unlikely(brw->perf_debug)) {
 112       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 113       start_time = get_time();
 114    }
 115
 116    struct brw_vue_map input_vue_map;
 117    brw_compute_tess_vue_map(&input_vue_map, key->inputs_read,
 118                             key->patch_inputs_read);
 119
 120    void *mem_ctx = ralloc_context(NULL);
 121    unsigned program_size;
 122    char *error_str;
 123    const unsigned *program =
 124       brw_compile_tes(compiler, brw, mem_ctx, key, &input_vue_map, &prog_data,
 125                       nir, &tep->program, st_index, &program_size, &error_str);
 126    if (program == NULL) {
 127       tep->program.sh.data->LinkStatus = false;
 128       ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
 129
 130       _mesa_problem(NULL, "Failed to compile tessellation evaluation shader: "
 131                     "%s\n", error_str);
 132
 133       ralloc_free(mem_ctx);
 134       return false;
 135    }
 136
 137    if (unlikely(brw->perf_debug)) {
 138       if (tep->compiled_once) {
 139          brw_tes_debug_recompile(brw, &tep->program, key);
 140       }
 141       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 142          perf_debug("TES compile took %.03f ms and stalled the GPU\n",
 143                     (get_time() - start_time) * 1000);
 144       }
 145       tep->compiled_once = true;
 146    }
 147
 148    /* Scratch space is used for register spilling */
 149    brw_alloc_stage_scratch(brw, stage_state,
 150                            prog_data.base.base.total_scratch,
 151                            devinfo->max_tes_threads);
 152
 153    brw_upload_cache(&brw->cache, BRW_CACHE_TES_PROG,
 154                     key, sizeof(*key),
 155                     program, program_size,
 156                     &prog_data, sizeof(prog_data),
 157                     &stage_state->prog_offset, &brw->tes.base.prog_data);
 158    ralloc_free(mem_ctx);
 159
 160    return true;
 161 }
 162
 163 void
 164 brw_tes_populate_key(struct brw_context *brw,
 165                      struct brw_tes_prog_key *key)
 166 {
 167    struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program;
 168    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 169    struct gl_program *prog = &tep->program;
 170
 171    uint64_t per_vertex_slots = prog->info.inputs_read;
 172    uint32_t per_patch_slots = prog->info.patch_inputs_read;
 173
 174    memset(key, 0, sizeof(*key));
 175
 176    key->program_string_id = tep->id;
 177
 178    /* The TCS may have additional outputs which aren't read by the
 179     * TES (possibly for cross-thread communication).  These need to
 180     * be stored in the Patch URB Entry as well.
 181     */
 182    if (tcp) {
 183       struct gl_program *tcp_prog = &tcp->program;
 184       per_vertex_slots |= tcp_prog->info.outputs_written &
 185          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 186       per_patch_slots |= tcp_prog->info.patch_outputs_written;
 187    }
 188
 189    key->inputs_read = per_vertex_slots;
 190    key->patch_inputs_read = per_patch_slots;
 191
 192    /* _NEW_TEXTURE */
 193    brw_populate_sampler_prog_key_data(&brw->ctx, prog, &key->tex);
 194 }
 195
 196 void
 197 brw_upload_tes_prog(struct brw_context *brw)
 198 {
 199    struct brw_stage_state *stage_state = &brw->tes.base;
 200    struct brw_tes_prog_key key;
 201    /* BRW_NEW_TESS_PROGRAMS */
 202    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 203
 204    if (!brw_state_dirty(brw,
 205                         _NEW_TEXTURE,
 206                         BRW_NEW_TESS_PROGRAMS))
 207       return;
 208
 209    brw_tes_populate_key(brw, &key);
 210
 211    if (!brw_search_cache(&brw->cache, BRW_CACHE_TES_PROG,
 212                          &key, sizeof(key),
 213                          &stage_state->prog_offset,
 214                          &brw->tes.base.prog_data)) {
 215       bool success = brw_codegen_tes_prog(brw, tep, &key);
 216       assert(success);
 217       (void)success;
 218    }
 219 }
 220
 221
 222 bool
 223 brw_tes_precompile(struct gl_context *ctx,
 224                    struct gl_shader_program *shader_prog,
 225                    struct gl_program *prog)
 226 {
 227    struct brw_context *brw = brw_context(ctx);
 228    struct brw_tes_prog_key key;
 229    uint32_t old_prog_offset = brw->tes.base.prog_offset;
 230    struct brw_stage_prog_data *old_prog_data = brw->tes.base.prog_data;
 231    bool success;
 232
 233    struct brw_program *btep = brw_program(prog);
 234
 235    memset(&key, 0, sizeof(key));
 236
 237    key.program_string_id = btep->id;
 238    key.inputs_read = prog->nir->info->inputs_read;
 239    key.patch_inputs_read = prog->nir->info->patch_inputs_read;
 240
 241    if (shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]) {
 242       struct gl_program *tcp =
 243          shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]->Program;
 244       key.inputs_read |= tcp->nir->info->outputs_written &
 245          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 246       key.patch_inputs_read |= tcp->nir->info->patch_outputs_written;
 247    }
 248
 249    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 250
 251    success = brw_codegen_tes_prog(brw, btep, &key);
 252
 253    brw->tes.base.prog_offset = old_prog_offset;
 254    brw->tes.base.prog_data = old_prog_data;
 255
 256    return success;
 257 }