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_state.h"
  33 #include "brw_ff_gs.h"
  34 #include "compiler/brw_nir.h"
  35 #include "brw_program.h"
  36 #include "compiler/glsl/ir_uniform.h"
  37
  38 static void
  39 brw_gs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  40                        const struct brw_gs_prog_key *key)
  41 {
  42    perf_debug("Recompiling geometry shader for program %d\n", prog->Id);
  43
  44    bool found = false;
  45    const struct brw_gs_prog_key *old_key =
  46       brw_find_previous_compile(&brw->cache, BRW_CACHE_GS_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
  57    if (!found) {
  58       perf_debug("  Something else\n");
  59    }
  60 }
  61
  62 static void
  63 assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
  64                                 const struct gl_program *prog,
  65                                 struct brw_gs_prog_data *prog_data)
  66 {
  67    /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
  68     * feedback surfaces.
  69     */
  70    uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
  71
  72    brw_assign_common_binding_table_offsets(devinfo, prog,
  73                                            &prog_data->base.base, reserved);
  74 }
  75
  76 static bool
  77 brw_codegen_gs_prog(struct brw_context *brw,
  78                     struct brw_program *gp,
  79                     struct brw_gs_prog_key *key)
  80 {
  81    struct brw_compiler *compiler = brw->screen->compiler;
  82    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  83    struct brw_stage_state *stage_state = &brw->gs.base;
  84    struct brw_gs_prog_data prog_data;
  85    bool start_busy = false;
  86    double start_time = 0;
  87
  88    memset(&prog_data, 0, sizeof(prog_data));
  89
  90    assign_gs_binding_table_offsets(devinfo, &gp->program, &prog_data);
  91
  92    /* Allocate the references to the uniforms that will end up in the
  93     * prog_data associated with the compiled program, and which will be freed
  94     * by the state cache.
  95     *
  96     * Note: param_count needs to be num_uniform_components * 4, since we add
  97     * padding around uniform values below vec4 size, so the worst case is that
  98     * every uniform is a float which gets padded to the size of a vec4.
  99     */
 100    int param_count = gp->program.nir->num_uniforms / 4;
 101
 102    prog_data.base.base.param =
 103       rzalloc_array(NULL, const gl_constant_value *, param_count);
 104    prog_data.base.base.pull_param =
 105       rzalloc_array(NULL, const gl_constant_value *, param_count);
 106    prog_data.base.base.image_param =
 107       rzalloc_array(NULL, struct brw_image_param,
 108                     gp->program.info.num_images);
 109    prog_data.base.base.nr_params = param_count;
 110    prog_data.base.base.nr_image_params = gp->program.info.num_images;
 111
 112    brw_nir_setup_glsl_uniforms(gp->program.nir, &gp->program,
 113                                &prog_data.base.base,
 114                                compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
 115    brw_nir_analyze_ubo_ranges(compiler, gp->program.nir,
 116                               prog_data.base.base.ubo_ranges);
 117
 118    uint64_t outputs_written = gp->program.nir->info.outputs_written;
 119
 120    brw_compute_vue_map(devinfo,
 121                        &prog_data.base.vue_map, outputs_written,
 122                        gp->program.info.separate_shader);
 123
 124    int st_index = -1;
 125    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 126       st_index = brw_get_shader_time_index(brw, &gp->program, ST_GS, true);
 127
 128    if (unlikely(brw->perf_debug)) {
 129       start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo);
 130       start_time = get_time();
 131    }
 132
 133    void *mem_ctx = ralloc_context(NULL);
 134    unsigned program_size;
 135    char *error_str;
 136    const unsigned *program =
 137       brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
 138                      &prog_data, gp->program.nir, &gp->program,
 139                      st_index, &program_size, &error_str);
 140    if (program == NULL) {
 141       ralloc_strcat(&gp->program.sh.data->InfoLog, error_str);
 142       _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
 143
 144       ralloc_free(mem_ctx);
 145       return false;
 146    }
 147
 148    if (unlikely(brw->perf_debug)) {
 149       if (gp->compiled_once) {
 150          brw_gs_debug_recompile(brw, &gp->program, key);
 151       }
 152       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 153          perf_debug("GS compile took %.03f ms and stalled the GPU\n",
 154                     (get_time() - start_time) * 1000);
 155       }
 156       gp->compiled_once = true;
 157    }
 158
 159    /* Scratch space is used for register spilling */
 160    brw_alloc_stage_scratch(brw, stage_state,
 161                            prog_data.base.base.total_scratch,
 162                            devinfo->max_gs_threads);
 163
 164    brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
 165                     key, sizeof(*key),
 166                     program, program_size,
 167                     &prog_data, sizeof(prog_data),
 168                     &stage_state->prog_offset, &brw->gs.base.prog_data);
 169    ralloc_free(mem_ctx);
 170
 171    return true;
 172 }
 173
 174 static bool
 175 brw_gs_state_dirty(const struct brw_context *brw)
 176 {
 177    return brw_state_dirty(brw,
 178                           _NEW_TEXTURE,
 179                           BRW_NEW_GEOMETRY_PROGRAM |
 180                           BRW_NEW_TRANSFORM_FEEDBACK);
 181 }
 182
 183 void
 184 brw_gs_populate_key(struct brw_context *brw,
 185                     struct brw_gs_prog_key *key)
 186 {
 187    struct gl_context *ctx = &brw->ctx;
 188    struct brw_program *gp =
 189       (struct brw_program *) brw->programs[MESA_SHADER_GEOMETRY];
 190
 191    memset(key, 0, sizeof(*key));
 192
 193    key->program_string_id = gp->id;
 194
 195    /* _NEW_TEXTURE */
 196    brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
 197 }
 198
 199 void
 200 brw_upload_gs_prog(struct brw_context *brw)
 201 {
 202    struct brw_stage_state *stage_state = &brw->gs.base;
 203    struct brw_gs_prog_key key;
 204    /* BRW_NEW_GEOMETRY_PROGRAM */
 205    struct brw_program *gp =
 206       (struct brw_program *) brw->programs[MESA_SHADER_GEOMETRY];
 207
 208    if (!brw_gs_state_dirty(brw))
 209       return;
 210
 211    brw_gs_populate_key(brw, &key);
 212
 213    if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
 214                          &key, sizeof(key),
 215                          &stage_state->prog_offset,
 216                          &brw->gs.base.prog_data)) {
 217       bool success = brw_codegen_gs_prog(brw, gp, &key);
 218       assert(success);
 219       (void)success;
 220    }
 221 }
 222
 223 bool
 224 brw_gs_precompile(struct gl_context *ctx, struct gl_program *prog)
 225 {
 226    struct brw_context *brw = brw_context(ctx);
 227    struct brw_gs_prog_key key;
 228    uint32_t old_prog_offset = brw->gs.base.prog_offset;
 229    struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
 230    bool success;
 231
 232    struct brw_program *bgp = brw_program(prog);
 233
 234    memset(&key, 0, sizeof(key));
 235
 236    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 237    key.program_string_id = bgp->id;
 238
 239    success = brw_codegen_gs_prog(brw, bgp, &key);
 240
 241    brw->gs.base.prog_offset = old_prog_offset;
 242    brw->gs.base.prog_data = old_prog_data;
 243
 244    return success;
 245 }