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
 116    uint64_t outputs_written = gp->program.info.outputs_written;
 117
 118    brw_compute_vue_map(devinfo,
 119                        &prog_data.base.vue_map, outputs_written,
 120                        gp->program.info.separate_shader);
 121
 122    int st_index = -1;
 123    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 124       st_index = brw_get_shader_time_index(brw, &gp->program, ST_GS, true);
 125
 126    if (unlikely(brw->perf_debug)) {
 127       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 128       start_time = get_time();
 129    }
 130
 131    void *mem_ctx = ralloc_context(NULL);
 132    unsigned program_size;
 133    char *error_str;
 134    const unsigned *program =
 135       brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
 136                      &prog_data, gp->program.nir, &gp->program,
 137                      st_index, &program_size, &error_str);
 138    if (program == NULL) {
 139       ralloc_strcat(&gp->program.sh.data->InfoLog, error_str);
 140       _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
 141
 142       ralloc_free(mem_ctx);
 143       return false;
 144    }
 145
 146    if (unlikely(brw->perf_debug)) {
 147       if (gp->compiled_once) {
 148          brw_gs_debug_recompile(brw, &gp->program, key);
 149       }
 150       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 151          perf_debug("GS compile took %.03f ms and stalled the GPU\n",
 152                     (get_time() - start_time) * 1000);
 153       }
 154       gp->compiled_once = true;
 155    }
 156
 157    /* Scratch space is used for register spilling */
 158    brw_alloc_stage_scratch(brw, stage_state,
 159                            prog_data.base.base.total_scratch,
 160                            devinfo->max_gs_threads);
 161
 162    brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
 163                     key, sizeof(*key),
 164                     program, program_size,
 165                     &prog_data, sizeof(prog_data),
 166                     &stage_state->prog_offset, &brw->gs.base.prog_data);
 167    ralloc_free(mem_ctx);
 168
 169    return true;
 170 }
 171
 172 static bool
 173 brw_gs_state_dirty(const struct brw_context *brw)
 174 {
 175    return brw_state_dirty(brw,
 176                           _NEW_TEXTURE,
 177                           BRW_NEW_GEOMETRY_PROGRAM |
 178                           BRW_NEW_TRANSFORM_FEEDBACK);
 179 }
 180
 181 void
 182 brw_gs_populate_key(struct brw_context *brw,
 183                     struct brw_gs_prog_key *key)
 184 {
 185    struct gl_context *ctx = &brw->ctx;
 186    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
 187
 188    memset(key, 0, sizeof(*key));
 189
 190    key->program_string_id = gp->id;
 191
 192    /* _NEW_TEXTURE */
 193    brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
 194 }
 195
 196 void
 197 brw_upload_gs_prog(struct brw_context *brw)
 198 {
 199    struct brw_stage_state *stage_state = &brw->gs.base;
 200    struct brw_gs_prog_key key;
 201    /* BRW_NEW_GEOMETRY_PROGRAM */
 202    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
 203
 204    if (!brw_gs_state_dirty(brw))
 205       return;
 206
 207    if (gp == NULL) {
 208       /* No geometry shader.  Vertex data just passes straight through. */
 209       if (brw->gen == 6 &&
 210           (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
 211          gen6_brw_upload_ff_gs_prog(brw);
 212          return;
 213       }
 214
 215       /* Other state atoms had better not try to access prog_data, since
 216        * there's no GS program.
 217        */
 218       brw->gs.base.prog_data = NULL;
 219
 220       return;
 221    }
 222
 223    brw_gs_populate_key(brw, &key);
 224
 225    if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
 226                          &key, sizeof(key),
 227                          &stage_state->prog_offset,
 228                          &brw->gs.base.prog_data)) {
 229       bool success = brw_codegen_gs_prog(brw, gp, &key);
 230       assert(success);
 231       (void)success;
 232    }
 233 }
 234
 235 bool
 236 brw_gs_precompile(struct gl_context *ctx, struct gl_program *prog)
 237 {
 238    struct brw_context *brw = brw_context(ctx);
 239    struct brw_gs_prog_key key;
 240    uint32_t old_prog_offset = brw->gs.base.prog_offset;
 241    struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
 242    bool success;
 243
 244    struct brw_program *bgp = brw_program(prog);
 245
 246    memset(&key, 0, sizeof(key));
 247
 248    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 249    key.program_string_id = bgp->id;
 250
 251    success = brw_codegen_gs_prog(brw, bgp, &key);
 252
 253    brw->gs.base.prog_offset = old_prog_offset;
 254    brw->gs.base.prog_data = old_prog_data;
 255
 256    return success;
 257 }