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