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

   1 /*
   2  * Copyright © 2013 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_vec4_gs.c
  26  *
  27  * State atom for client-programmable geometry shaders, and support code.
  28  */
  29
  30 #include "brw_gs.h"
  31 #include "brw_context.h"
  32 #include "brw_vec4_gs_visitor.h"
  33 #include "brw_state.h"
  34 #include "brw_ff_gs.h"
  35 #include "brw_nir.h"
  36 #include "brw_program.h"
  37 #include "compiler/glsl/ir_uniform.h"
  38
  39 static void
  40 brw_gs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  41                        const struct brw_gs_prog_key *key)
  42 {
  43    struct brw_cache_item *c = NULL;
  44    const struct brw_gs_prog_key *old_key = NULL;
  45    bool found = false;
  46
  47    perf_debug("Recompiling geometry shader for program %d\n", prog->Id);
  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_GS_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 |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  69
  70    if (!found) {
  71       perf_debug("  Something else\n");
  72    }
  73 }
  74
  75 static void
  76 assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
  77                                 const struct gl_program *prog,
  78                                 struct brw_gs_prog_data *prog_data)
  79 {
  80    /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
  81     * feedback surfaces.
  82     */
  83    uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
  84
  85    brw_assign_common_binding_table_offsets(devinfo, prog,
  86                                            &prog_data->base.base, reserved);
  87 }
  88
  89 static bool
  90 brw_codegen_gs_prog(struct brw_context *brw,
  91                     struct brw_program *gp,
  92                     struct brw_gs_prog_key *key)
  93 {
  94    struct brw_compiler *compiler = brw->screen->compiler;
  95    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  96    struct brw_stage_state *stage_state = &brw->gs.base;
  97    struct brw_gs_prog_data prog_data;
  98    bool start_busy = false;
  99    double start_time = 0;
 100
 101    memset(&prog_data, 0, sizeof(prog_data));
 102
 103    assign_gs_binding_table_offsets(devinfo, &gp->program, &prog_data);
 104
 105    /* Allocate the references to the uniforms that will end up in the
 106     * prog_data associated with the compiled program, and which will be freed
 107     * by the state cache.
 108     *
 109     * Note: param_count needs to be num_uniform_components * 4, since we add
 110     * padding around uniform values below vec4 size, so the worst case is that
 111     * every uniform is a float which gets padded to the size of a vec4.
 112     */
 113    int param_count = gp->program.nir->num_uniforms / 4;
 114
 115    prog_data.base.base.param =
 116       rzalloc_array(NULL, const gl_constant_value *, param_count);
 117    prog_data.base.base.pull_param =
 118       rzalloc_array(NULL, const gl_constant_value *, param_count);
 119    prog_data.base.base.image_param =
 120       rzalloc_array(NULL, struct brw_image_param,
 121                     gp->program.info.num_images);
 122    prog_data.base.base.nr_params = param_count;
 123    prog_data.base.base.nr_image_params = gp->program.info.num_images;
 124
 125    brw_nir_setup_glsl_uniforms(gp->program.nir, &gp->program,
 126                                &prog_data.base.base,
 127                                compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
 128
 129    uint64_t outputs_written = gp->program.info.outputs_written;
 130
 131    brw_compute_vue_map(devinfo,
 132                        &prog_data.base.vue_map, outputs_written,
 133                        gp->program.info.separate_shader);
 134
 135    int st_index = -1;
 136    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 137       st_index = brw_get_shader_time_index(brw, &gp->program, ST_GS, true);
 138
 139    if (unlikely(brw->perf_debug)) {
 140       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 141       start_time = get_time();
 142    }
 143
 144    void *mem_ctx = ralloc_context(NULL);
 145    unsigned program_size;
 146    char *error_str;
 147    const unsigned *program =
 148       brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
 149                      &prog_data, gp->program.nir, &gp->program,
 150                      st_index, &program_size, &error_str);
 151    if (program == NULL) {
 152       ralloc_strcat(&gp->program.sh.data->InfoLog, error_str);
 153       _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
 154
 155       ralloc_free(mem_ctx);
 156       return false;
 157    }
 158
 159    if (unlikely(brw->perf_debug)) {
 160       if (gp->compiled_once) {
 161          brw_gs_debug_recompile(brw, &gp->program, key);
 162       }
 163       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 164          perf_debug("GS compile took %.03f ms and stalled the GPU\n",
 165                     (get_time() - start_time) * 1000);
 166       }
 167       gp->compiled_once = true;
 168    }
 169
 170    /* Scratch space is used for register spilling */
 171    brw_alloc_stage_scratch(brw, stage_state,
 172                            prog_data.base.base.total_scratch,
 173                            devinfo->max_gs_threads);
 174
 175    brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
 176                     key, sizeof(*key),
 177                     program, program_size,
 178                     &prog_data, sizeof(prog_data),
 179                     &stage_state->prog_offset, &brw->gs.base.prog_data);
 180    ralloc_free(mem_ctx);
 181
 182    return true;
 183 }
 184
 185 static bool
 186 brw_gs_state_dirty(const struct brw_context *brw)
 187 {
 188    return brw_state_dirty(brw,
 189                           _NEW_TEXTURE,
 190                           BRW_NEW_GEOMETRY_PROGRAM |
 191                           BRW_NEW_TRANSFORM_FEEDBACK);
 192 }
 193
 194 void
 195 brw_gs_populate_key(struct brw_context *brw,
 196                     struct brw_gs_prog_key *key)
 197 {
 198    struct gl_context *ctx = &brw->ctx;
 199    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
 200
 201    memset(key, 0, sizeof(*key));
 202
 203    key->program_string_id = gp->id;
 204
 205    /* _NEW_TEXTURE */
 206    brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
 207 }
 208
 209 void
 210 brw_upload_gs_prog(struct brw_context *brw)
 211 {
 212    struct brw_stage_state *stage_state = &brw->gs.base;
 213    struct brw_gs_prog_key key;
 214    /* BRW_NEW_GEOMETRY_PROGRAM */
 215    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
 216
 217    if (!brw_gs_state_dirty(brw))
 218       return;
 219
 220    if (gp == NULL) {
 221       /* No geometry shader.  Vertex data just passes straight through. */
 222       if (brw->gen == 6 &&
 223           (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
 224          gen6_brw_upload_ff_gs_prog(brw);
 225          return;
 226       }
 227
 228       /* Other state atoms had better not try to access prog_data, since
 229        * there's no GS program.
 230        */
 231       brw->gs.base.prog_data = NULL;
 232
 233       return;
 234    }
 235
 236    brw_gs_populate_key(brw, &key);
 237
 238    if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
 239                          &key, sizeof(key),
 240                          &stage_state->prog_offset,
 241                          &brw->gs.base.prog_data)) {
 242       bool success = brw_codegen_gs_prog(brw, gp, &key);
 243       assert(success);
 244       (void)success;
 245    }
 246 }
 247
 248 bool
 249 brw_gs_precompile(struct gl_context *ctx, struct gl_program *prog)
 250 {
 251    struct brw_context *brw = brw_context(ctx);
 252    struct brw_gs_prog_key key;
 253    uint32_t old_prog_offset = brw->gs.base.prog_offset;
 254    struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
 255    bool success;
 256
 257    struct brw_program *bgp = brw_program(prog);
 258
 259    memset(&key, 0, sizeof(key));
 260
 261    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 262    key.program_string_id = bgp->id;
 263
 264    success = brw_codegen_gs_prog(brw, bgp, &key);
 265
 266    brw->gs.base.prog_offset = old_prog_offset;
 267    brw->gs.base.prog_data = old_prog_data;
 268
 269    return success;
 270 }