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

   1 /*
   2  * Copyright (c) 2014 - 2015 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 #include "util/ralloc.h"
  25 #include "brw_context.h"
  26 #include "brw_cs.h"
  27 #include "brw_eu.h"
  28 #include "brw_wm.h"
  29 #include "brw_shader.h"
  30 #include "intel_mipmap_tree.h"
  31 #include "brw_state.h"
  32 #include "intel_batchbuffer.h"
  33
  34 bool
  35 brw_cs_prog_data_compare(const void *in_a, const void *in_b)
  36 {
  37    const struct brw_cs_prog_data *a =
  38       (const struct brw_cs_prog_data *)in_a;
  39    const struct brw_cs_prog_data *b =
  40       (const struct brw_cs_prog_data *)in_b;
  41
  42    /* Compare the base structure. */
  43    if (!brw_stage_prog_data_compare(&a->base, &b->base))
  44       return false;
  45
  46    /* Compare the rest of the structure. */
  47    const unsigned offset = sizeof(struct brw_stage_prog_data);
  48    if (memcmp(((char *) a) + offset, ((char *) b) + offset,
  49               sizeof(struct brw_cs_prog_data) - offset))
  50       return false;
  51
  52    return true;
  53 }
  54
  55 static bool
  56 brw_codegen_cs_prog(struct brw_context *brw,
  57                     struct gl_shader_program *prog,
  58                     struct brw_compute_program *cp,
  59                     struct brw_cs_prog_key *key)
  60 {
  61    struct gl_context *ctx = &brw->ctx;
  62    const GLuint *program;
  63    void *mem_ctx = ralloc_context(NULL);
  64    GLuint program_size;
  65    struct brw_cs_prog_data prog_data;
  66    bool start_busy = false;
  67    double start_time = 0;
  68
  69    struct brw_shader *cs =
  70       (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
  71    assert (cs);
  72
  73    memset(&prog_data, 0, sizeof(prog_data));
  74
  75    /* Allocate the references to the uniforms that will end up in the
  76     * prog_data associated with the compiled program, and which will be freed
  77     * by the state cache.
  78     */
  79    int param_count = cs->base.num_uniform_components +
  80                      cs->base.NumImages * BRW_IMAGE_PARAM_SIZE;
  81
  82    /* The backend also sometimes adds params for texture size. */
  83    param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
  84    prog_data.base.param =
  85       rzalloc_array(NULL, const gl_constant_value *, param_count);
  86    prog_data.base.pull_param =
  87       rzalloc_array(NULL, const gl_constant_value *, param_count);
  88    prog_data.base.image_param =
  89       rzalloc_array(NULL, struct brw_image_param, cs->base.NumImages);
  90    prog_data.base.nr_params = param_count;
  91    prog_data.base.nr_image_params = cs->base.NumImages;
  92
  93    if (unlikely(brw->perf_debug)) {
  94       start_busy = (brw->batch.last_bo &&
  95                     drm_intel_bo_busy(brw->batch.last_bo));
  96       start_time = get_time();
  97    }
  98
  99    program = brw_cs_emit(brw, mem_ctx, key, &prog_data,
 100                          &cp->program, prog, &program_size);
 101    if (program == NULL) {
 102       ralloc_free(mem_ctx);
 103       return false;
 104    }
 105
 106    if (unlikely(brw->perf_debug) && cs) {
 107       if (cs->compiled_once) {
 108          _mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
 109       }
 110       cs->compiled_once = true;
 111
 112       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 113          perf_debug("CS compile took %.03f ms and stalled the GPU\n",
 114                     (get_time() - start_time) * 1000);
 115       }
 116    }
 117
 118    if (prog_data.base.total_scratch) {
 119       brw_get_scratch_bo(brw, &brw->cs.base.scratch_bo,
 120                          prog_data.base.total_scratch * brw->max_cs_threads);
 121    }
 122
 123    if (unlikely(INTEL_DEBUG & DEBUG_CS))
 124       fprintf(stderr, "\n");
 125
 126    brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
 127                     key, sizeof(*key),
 128                     program, program_size,
 129                     &prog_data, sizeof(prog_data),
 130                     &brw->cs.base.prog_offset, &brw->cs.prog_data);
 131    ralloc_free(mem_ctx);
 132
 133    return true;
 134 }
 135
 136
 137 static void
 138 brw_cs_populate_key(struct brw_context *brw, struct brw_cs_prog_key *key)
 139 {
 140    struct gl_context *ctx = &brw->ctx;
 141    /* BRW_NEW_COMPUTE_PROGRAM */
 142    const struct brw_compute_program *cp =
 143       (struct brw_compute_program *) brw->compute_program;
 144    const struct gl_program *prog = (struct gl_program *) cp;
 145
 146    memset(key, 0, sizeof(*key));
 147
 148    /* _NEW_TEXTURE */
 149    brw_populate_sampler_prog_key_data(ctx, prog, brw->cs.base.sampler_count,
 150                                       &key->tex);
 151
 152    /* The unique compute program ID */
 153    key->program_string_id = cp->id;
 154 }
 155
 156
 157 void
 158 brw_upload_cs_prog(struct brw_context *brw)
 159 {
 160    struct gl_context *ctx = &brw->ctx;
 161    struct brw_cs_prog_key key;
 162    struct brw_compute_program *cp = (struct brw_compute_program *)
 163       brw->compute_program;
 164
 165    if (!cp)
 166       return;
 167
 168    if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
 169       return;
 170
 171    brw->cs.base.sampler_count =
 172       _mesa_fls(ctx->ComputeProgram._Current->Base.SamplersUsed);
 173
 174    brw_cs_populate_key(brw, &key);
 175
 176    if (!brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG,
 177                          &key, sizeof(key),
 178                          &brw->cs.base.prog_offset, &brw->cs.prog_data)) {
 179       bool success =
 180          brw_codegen_cs_prog(brw,
 181                              ctx->Shader.CurrentProgram[MESA_SHADER_COMPUTE],
 182                              cp, &key);
 183       (void) success;
 184       assert(success);
 185    }
 186    brw->cs.base.prog_data = &brw->cs.prog_data->base;
 187 }
 188
 189
 190 bool
 191 brw_cs_precompile(struct gl_context *ctx,
 192                   struct gl_shader_program *shader_prog,
 193                   struct gl_program *prog)
 194 {
 195    struct brw_context *brw = brw_context(ctx);
 196    struct brw_cs_prog_key key;
 197
 198    struct gl_compute_program *cp = (struct gl_compute_program *) prog;
 199    struct brw_compute_program *bcp = brw_compute_program(cp);
 200
 201    memset(&key, 0, sizeof(key));
 202    key.program_string_id = bcp->id;
 203
 204    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 205
 206    uint32_t old_prog_offset = brw->cs.base.prog_offset;
 207    struct brw_cs_prog_data *old_prog_data = brw->cs.prog_data;
 208
 209    bool success = brw_codegen_cs_prog(brw, shader_prog, bcp, &key);
 210
 211    brw->cs.base.prog_offset = old_prog_offset;
 212    brw->cs.prog_data = old_prog_data;
 213
 214    return success;
 215 }